]>
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 | ||
f89e09cf AJ |
133 | /* |
134 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
135 | * bytes including terminating null byte. | |
136 | */ | |
137 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
138 | { | |
139 | int i; | |
140 | int ret; | |
141 | char *bp = buf; | |
142 | u64 flags = bg_flags; | |
143 | u32 size_bp = size_buf; | |
144 | ||
145 | if (!flags) { | |
146 | strcpy(bp, "NONE"); | |
147 | return; | |
148 | } | |
149 | ||
150 | #define DESCRIBE_FLAG(flag, desc) \ | |
151 | do { \ | |
152 | if (flags & (flag)) { \ | |
153 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
154 | if (ret < 0 || ret >= size_bp) \ | |
155 | goto out_overflow; \ | |
156 | size_bp -= ret; \ | |
157 | bp += ret; \ | |
158 | flags &= ~(flag); \ | |
159 | } \ | |
160 | } while (0) | |
161 | ||
162 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
163 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
164 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
165 | ||
166 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
167 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
168 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
169 | btrfs_raid_array[i].raid_name); | |
170 | #undef DESCRIBE_FLAG | |
171 | ||
172 | if (flags) { | |
173 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
174 | size_bp -= ret; | |
175 | } | |
176 | ||
177 | if (size_bp < size_buf) | |
178 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
179 | ||
180 | /* | |
181 | * The text is trimmed, it's up to the caller to provide sufficiently | |
182 | * large buffer | |
183 | */ | |
184 | out_overflow:; | |
185 | } | |
186 | ||
2b82032c | 187 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 188 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 189 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 190 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 191 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 192 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
193 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
194 | enum btrfs_map_op op, | |
195 | u64 logical, u64 *length, | |
196 | struct btrfs_bio **bbio_ret, | |
197 | int mirror_num, int need_raid_map); | |
2b82032c | 198 | |
9c6b1c4d DS |
199 | /* |
200 | * Device locking | |
201 | * ============== | |
202 | * | |
203 | * There are several mutexes that protect manipulation of devices and low-level | |
204 | * structures like chunks but not block groups, extents or files | |
205 | * | |
206 | * uuid_mutex (global lock) | |
207 | * ------------------------ | |
208 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
209 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
210 | * device) or requested by the device= mount option | |
211 | * | |
212 | * the mutex can be very coarse and can cover long-running operations | |
213 | * | |
214 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
215 | * seeding, structure cloning, openning/closing devices at mount/umount time | |
216 | * | |
217 | * global::fs_devs - add, remove, updates to the global list | |
218 | * | |
219 | * does not protect: manipulation of the fs_devices::devices list! | |
220 | * | |
221 | * btrfs_device::name - renames (write side), read is RCU | |
222 | * | |
223 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
224 | * ------------------------------------------------ | |
225 | * protects updates to fs_devices::devices, ie. adding and deleting | |
226 | * | |
227 | * simple list traversal with read-only actions can be done with RCU protection | |
228 | * | |
229 | * may be used to exclude some operations from running concurrently without any | |
230 | * modifications to the list (see write_all_supers) | |
231 | * | |
9c6b1c4d DS |
232 | * balance_mutex |
233 | * ------------- | |
234 | * protects balance structures (status, state) and context accessed from | |
235 | * several places (internally, ioctl) | |
236 | * | |
237 | * chunk_mutex | |
238 | * ----------- | |
239 | * protects chunks, adding or removing during allocation, trim or when a new | |
240 | * device is added/removed | |
241 | * | |
242 | * cleaner_mutex | |
243 | * ------------- | |
244 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
245 | * cleaning together with relocation or delayed iputs | |
246 | * | |
247 | * | |
248 | * Lock nesting | |
249 | * ============ | |
250 | * | |
251 | * uuid_mutex | |
252 | * volume_mutex | |
253 | * device_list_mutex | |
254 | * chunk_mutex | |
255 | * balance_mutex | |
89595e80 AJ |
256 | * |
257 | * | |
258 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
259 | * ====================================== | |
260 | * | |
261 | * Maintains the exclusivity of the following operations that apply to the | |
262 | * whole filesystem and cannot run in parallel. | |
263 | * | |
264 | * - Balance (*) | |
265 | * - Device add | |
266 | * - Device remove | |
267 | * - Device replace (*) | |
268 | * - Resize | |
269 | * | |
270 | * The device operations (as above) can be in one of the following states: | |
271 | * | |
272 | * - Running state | |
273 | * - Paused state | |
274 | * - Completed state | |
275 | * | |
276 | * Only device operations marked with (*) can go into the Paused state for the | |
277 | * following reasons: | |
278 | * | |
279 | * - ioctl (only Balance can be Paused through ioctl) | |
280 | * - filesystem remounted as read-only | |
281 | * - filesystem unmounted and mounted as read-only | |
282 | * - system power-cycle and filesystem mounted as read-only | |
283 | * - filesystem or device errors leading to forced read-only | |
284 | * | |
285 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
286 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
287 | * A device operation in Paused or Running state can be canceled or resumed | |
288 | * either by ioctl (Balance only) or when remounted as read-write. | |
289 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
290 | * completed. | |
9c6b1c4d DS |
291 | */ |
292 | ||
67a2c45e | 293 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 294 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
295 | struct list_head *btrfs_get_fs_uuids(void) |
296 | { | |
297 | return &fs_uuids; | |
298 | } | |
8a4b83cc | 299 | |
2dfeca9b DS |
300 | /* |
301 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
302 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
303 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
304 | * |
305 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
306 | * The returned struct is not linked onto any lists and can be destroyed with | |
307 | * kfree() right away. | |
308 | */ | |
7239ff4b NB |
309 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
310 | const u8 *metadata_fsid) | |
2208a378 ID |
311 | { |
312 | struct btrfs_fs_devices *fs_devs; | |
313 | ||
78f2c9e6 | 314 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
315 | if (!fs_devs) |
316 | return ERR_PTR(-ENOMEM); | |
317 | ||
318 | mutex_init(&fs_devs->device_list_mutex); | |
319 | ||
320 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 321 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 | 322 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
c4babc5e | 323 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
324 | if (fsid) |
325 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 326 | |
7239ff4b NB |
327 | if (metadata_fsid) |
328 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
329 | else if (fsid) | |
330 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
331 | ||
2208a378 ID |
332 | return fs_devs; |
333 | } | |
334 | ||
a425f9d4 | 335 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf DS |
336 | { |
337 | rcu_string_free(device->name); | |
338 | bio_put(device->flush_bio); | |
339 | kfree(device); | |
340 | } | |
341 | ||
e4404d6e YZ |
342 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
343 | { | |
344 | struct btrfs_device *device; | |
345 | WARN_ON(fs_devices->opened); | |
346 | while (!list_empty(&fs_devices->devices)) { | |
347 | device = list_entry(fs_devices->devices.next, | |
348 | struct btrfs_device, dev_list); | |
349 | list_del(&device->dev_list); | |
a425f9d4 | 350 | btrfs_free_device(device); |
e4404d6e YZ |
351 | } |
352 | kfree(fs_devices); | |
353 | } | |
354 | ||
b8b8ff59 LC |
355 | static void btrfs_kobject_uevent(struct block_device *bdev, |
356 | enum kobject_action action) | |
357 | { | |
358 | int ret; | |
359 | ||
360 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
361 | if (ret) | |
efe120a0 | 362 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
363 | action, |
364 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
365 | &disk_to_dev(bdev->bd_disk)->kobj); | |
366 | } | |
367 | ||
ffc5a379 | 368 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
369 | { |
370 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 371 | |
2b82032c YZ |
372 | while (!list_empty(&fs_uuids)) { |
373 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
374 | struct btrfs_fs_devices, fs_list); |
375 | list_del(&fs_devices->fs_list); | |
e4404d6e | 376 | free_fs_devices(fs_devices); |
8a4b83cc | 377 | } |
8a4b83cc CM |
378 | } |
379 | ||
48dae9cf DS |
380 | /* |
381 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
382 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 383 | * btrfs_free_device. |
48dae9cf | 384 | */ |
12bd2fc0 ID |
385 | static struct btrfs_device *__alloc_device(void) |
386 | { | |
387 | struct btrfs_device *dev; | |
388 | ||
78f2c9e6 | 389 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
390 | if (!dev) |
391 | return ERR_PTR(-ENOMEM); | |
392 | ||
e0ae9994 DS |
393 | /* |
394 | * Preallocate a bio that's always going to be used for flushing device | |
395 | * barriers and matches the device lifespan | |
396 | */ | |
397 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
398 | if (!dev->flush_bio) { | |
399 | kfree(dev); | |
400 | return ERR_PTR(-ENOMEM); | |
401 | } | |
e0ae9994 | 402 | |
12bd2fc0 ID |
403 | INIT_LIST_HEAD(&dev->dev_list); |
404 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 405 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
406 | |
407 | spin_lock_init(&dev->io_lock); | |
408 | ||
12bd2fc0 | 409 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 410 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 411 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 412 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 413 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
414 | |
415 | return dev; | |
416 | } | |
417 | ||
35c70103 DS |
418 | /* |
419 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
420 | * return NULL. | |
421 | * | |
422 | * If devid and uuid are both specified, the match must be exact, otherwise | |
423 | * only devid is used. | |
424 | */ | |
425 | static struct btrfs_device *find_device(struct btrfs_fs_devices *fs_devices, | |
426 | u64 devid, const u8 *uuid) | |
8a4b83cc CM |
427 | { |
428 | struct btrfs_device *dev; | |
8a4b83cc | 429 | |
636d2c9d | 430 | list_for_each_entry(dev, &fs_devices->devices, dev_list) { |
a443755f | 431 | if (dev->devid == devid && |
8f18cf13 | 432 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 433 | return dev; |
a443755f | 434 | } |
8a4b83cc CM |
435 | } |
436 | return NULL; | |
437 | } | |
438 | ||
7239ff4b NB |
439 | static noinline struct btrfs_fs_devices *find_fsid( |
440 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 441 | { |
8a4b83cc CM |
442 | struct btrfs_fs_devices *fs_devices; |
443 | ||
7239ff4b NB |
444 | ASSERT(fsid); |
445 | ||
7a62d0f0 NB |
446 | if (metadata_fsid) { |
447 | /* | |
448 | * Handle scanned device having completed its fsid change but | |
449 | * belonging to a fs_devices that was created by first scanning | |
450 | * a device which didn't have its fsid/metadata_uuid changed | |
451 | * at all and the CHANGING_FSID_V2 flag set. | |
452 | */ | |
453 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
454 | if (fs_devices->fsid_change && | |
455 | memcmp(metadata_fsid, fs_devices->fsid, | |
456 | BTRFS_FSID_SIZE) == 0 && | |
457 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
458 | BTRFS_FSID_SIZE) == 0) { | |
459 | return fs_devices; | |
460 | } | |
461 | } | |
cc5de4e7 NB |
462 | /* |
463 | * Handle scanned device having completed its fsid change but | |
464 | * belonging to a fs_devices that was created by a device that | |
465 | * has an outdated pair of fsid/metadata_uuid and | |
466 | * CHANGING_FSID_V2 flag set. | |
467 | */ | |
468 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
469 | if (fs_devices->fsid_change && | |
470 | memcmp(fs_devices->metadata_uuid, | |
471 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
472 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
473 | BTRFS_FSID_SIZE) == 0) { | |
474 | return fs_devices; | |
475 | } | |
476 | } | |
7a62d0f0 NB |
477 | } |
478 | ||
479 | /* Handle non-split brain cases */ | |
c4babc5e | 480 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
481 | if (metadata_fsid) { |
482 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
483 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
484 | BTRFS_FSID_SIZE) == 0) | |
485 | return fs_devices; | |
486 | } else { | |
487 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
488 | return fs_devices; | |
489 | } | |
8a4b83cc CM |
490 | } |
491 | return NULL; | |
492 | } | |
493 | ||
beaf8ab3 SB |
494 | static int |
495 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
496 | int flush, struct block_device **bdev, | |
497 | struct buffer_head **bh) | |
498 | { | |
499 | int ret; | |
500 | ||
501 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
502 | ||
503 | if (IS_ERR(*bdev)) { | |
504 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
505 | goto error; |
506 | } | |
507 | ||
508 | if (flush) | |
509 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 510 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
511 | if (ret) { |
512 | blkdev_put(*bdev, flags); | |
513 | goto error; | |
514 | } | |
515 | invalidate_bdev(*bdev); | |
516 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
517 | if (IS_ERR(*bh)) { |
518 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
519 | blkdev_put(*bdev, flags); |
520 | goto error; | |
521 | } | |
522 | ||
523 | return 0; | |
524 | ||
525 | error: | |
526 | *bdev = NULL; | |
527 | *bh = NULL; | |
528 | return ret; | |
529 | } | |
530 | ||
ffbd517d CM |
531 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
532 | struct bio *head, struct bio *tail) | |
533 | { | |
534 | ||
535 | struct bio *old_head; | |
536 | ||
537 | old_head = pending_bios->head; | |
538 | pending_bios->head = head; | |
539 | if (pending_bios->tail) | |
540 | tail->bi_next = old_head; | |
541 | else | |
542 | pending_bios->tail = tail; | |
543 | } | |
544 | ||
8b712842 CM |
545 | /* |
546 | * we try to collect pending bios for a device so we don't get a large | |
547 | * number of procs sending bios down to the same device. This greatly | |
548 | * improves the schedulers ability to collect and merge the bios. | |
549 | * | |
550 | * But, it also turns into a long list of bios to process and that is sure | |
551 | * to eventually make the worker thread block. The solution here is to | |
552 | * make some progress and then put this work struct back at the end of | |
553 | * the list if the block device is congested. This way, multiple devices | |
554 | * can make progress from a single worker thread. | |
555 | */ | |
143bede5 | 556 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 557 | { |
0b246afa | 558 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
559 | struct bio *pending; |
560 | struct backing_dev_info *bdi; | |
ffbd517d | 561 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
562 | struct bio *tail; |
563 | struct bio *cur; | |
564 | int again = 0; | |
ffbd517d | 565 | unsigned long num_run; |
d644d8a1 | 566 | unsigned long batch_run = 0; |
b765ead5 | 567 | unsigned long last_waited = 0; |
d84275c9 | 568 | int force_reg = 0; |
0e588859 | 569 | int sync_pending = 0; |
211588ad CM |
570 | struct blk_plug plug; |
571 | ||
572 | /* | |
573 | * this function runs all the bios we've collected for | |
574 | * a particular device. We don't want to wander off to | |
575 | * another device without first sending all of these down. | |
576 | * So, setup a plug here and finish it off before we return | |
577 | */ | |
578 | blk_start_plug(&plug); | |
8b712842 | 579 | |
efa7c9f9 | 580 | bdi = device->bdev->bd_bdi; |
b64a2851 | 581 | |
8b712842 CM |
582 | loop: |
583 | spin_lock(&device->io_lock); | |
584 | ||
a6837051 | 585 | loop_lock: |
d84275c9 | 586 | num_run = 0; |
ffbd517d | 587 | |
8b712842 CM |
588 | /* take all the bios off the list at once and process them |
589 | * later on (without the lock held). But, remember the | |
590 | * tail and other pointers so the bios can be properly reinserted | |
591 | * into the list if we hit congestion | |
592 | */ | |
d84275c9 | 593 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 594 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
595 | force_reg = 1; |
596 | } else { | |
ffbd517d | 597 | pending_bios = &device->pending_bios; |
d84275c9 CM |
598 | force_reg = 0; |
599 | } | |
ffbd517d CM |
600 | |
601 | pending = pending_bios->head; | |
602 | tail = pending_bios->tail; | |
8b712842 | 603 | WARN_ON(pending && !tail); |
8b712842 CM |
604 | |
605 | /* | |
606 | * if pending was null this time around, no bios need processing | |
607 | * at all and we can stop. Otherwise it'll loop back up again | |
608 | * and do an additional check so no bios are missed. | |
609 | * | |
610 | * device->running_pending is used to synchronize with the | |
611 | * schedule_bio code. | |
612 | */ | |
ffbd517d CM |
613 | if (device->pending_sync_bios.head == NULL && |
614 | device->pending_bios.head == NULL) { | |
8b712842 CM |
615 | again = 0; |
616 | device->running_pending = 0; | |
ffbd517d CM |
617 | } else { |
618 | again = 1; | |
619 | device->running_pending = 1; | |
8b712842 | 620 | } |
ffbd517d CM |
621 | |
622 | pending_bios->head = NULL; | |
623 | pending_bios->tail = NULL; | |
624 | ||
8b712842 CM |
625 | spin_unlock(&device->io_lock); |
626 | ||
d397712b | 627 | while (pending) { |
ffbd517d CM |
628 | |
629 | rmb(); | |
d84275c9 CM |
630 | /* we want to work on both lists, but do more bios on the |
631 | * sync list than the regular list | |
632 | */ | |
633 | if ((num_run > 32 && | |
634 | pending_bios != &device->pending_sync_bios && | |
635 | device->pending_sync_bios.head) || | |
636 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
637 | device->pending_bios.head)) { | |
ffbd517d CM |
638 | spin_lock(&device->io_lock); |
639 | requeue_list(pending_bios, pending, tail); | |
640 | goto loop_lock; | |
641 | } | |
642 | ||
8b712842 CM |
643 | cur = pending; |
644 | pending = pending->bi_next; | |
645 | cur->bi_next = NULL; | |
b64a2851 | 646 | |
dac56212 | 647 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 648 | |
2ab1ba68 CM |
649 | /* |
650 | * if we're doing the sync list, record that our | |
651 | * plug has some sync requests on it | |
652 | * | |
653 | * If we're doing the regular list and there are | |
654 | * sync requests sitting around, unplug before | |
655 | * we add more | |
656 | */ | |
657 | if (pending_bios == &device->pending_sync_bios) { | |
658 | sync_pending = 1; | |
659 | } else if (sync_pending) { | |
660 | blk_finish_plug(&plug); | |
661 | blk_start_plug(&plug); | |
662 | sync_pending = 0; | |
663 | } | |
664 | ||
4e49ea4a | 665 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
666 | num_run++; |
667 | batch_run++; | |
853d8ec4 DS |
668 | |
669 | cond_resched(); | |
8b712842 CM |
670 | |
671 | /* | |
672 | * we made progress, there is more work to do and the bdi | |
673 | * is now congested. Back off and let other work structs | |
674 | * run instead | |
675 | */ | |
57fd5a5f | 676 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 677 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 678 | struct io_context *ioc; |
8b712842 | 679 | |
b765ead5 CM |
680 | ioc = current->io_context; |
681 | ||
682 | /* | |
683 | * the main goal here is that we don't want to | |
684 | * block if we're going to be able to submit | |
685 | * more requests without blocking. | |
686 | * | |
687 | * This code does two great things, it pokes into | |
688 | * the elevator code from a filesystem _and_ | |
689 | * it makes assumptions about how batching works. | |
690 | */ | |
691 | if (ioc && ioc->nr_batch_requests > 0 && | |
692 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
693 | (last_waited == 0 || | |
694 | ioc->last_waited == last_waited)) { | |
695 | /* | |
696 | * we want to go through our batch of | |
697 | * requests and stop. So, we copy out | |
698 | * the ioc->last_waited time and test | |
699 | * against it before looping | |
700 | */ | |
701 | last_waited = ioc->last_waited; | |
853d8ec4 | 702 | cond_resched(); |
b765ead5 CM |
703 | continue; |
704 | } | |
8b712842 | 705 | spin_lock(&device->io_lock); |
ffbd517d | 706 | requeue_list(pending_bios, pending, tail); |
a6837051 | 707 | device->running_pending = 1; |
8b712842 CM |
708 | |
709 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
710 | btrfs_queue_work(fs_info->submit_workers, |
711 | &device->work); | |
8b712842 CM |
712 | goto done; |
713 | } | |
714 | } | |
ffbd517d | 715 | |
51684082 CM |
716 | cond_resched(); |
717 | if (again) | |
718 | goto loop; | |
719 | ||
720 | spin_lock(&device->io_lock); | |
721 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
722 | goto loop_lock; | |
723 | spin_unlock(&device->io_lock); | |
724 | ||
8b712842 | 725 | done: |
211588ad | 726 | blk_finish_plug(&plug); |
8b712842 CM |
727 | } |
728 | ||
b2950863 | 729 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
730 | { |
731 | struct btrfs_device *device; | |
732 | ||
733 | device = container_of(work, struct btrfs_device, work); | |
734 | run_scheduled_bios(device); | |
735 | } | |
736 | ||
d8367db3 AJ |
737 | /* |
738 | * Search and remove all stale (devices which are not mounted) devices. | |
739 | * When both inputs are NULL, it will search and release all stale devices. | |
740 | * path: Optional. When provided will it release all unmounted devices | |
741 | * matching this path only. | |
742 | * skip_dev: Optional. Will skip this device when searching for the stale | |
743 | * devices. | |
744 | */ | |
745 | static void btrfs_free_stale_devices(const char *path, | |
fa6d2ae5 | 746 | struct btrfs_device *skip_device) |
4fde46f0 | 747 | { |
fa6d2ae5 AJ |
748 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
749 | struct btrfs_device *device, *tmp_device; | |
4fde46f0 | 750 | |
fa6d2ae5 | 751 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
7bcb8164 AJ |
752 | mutex_lock(&fs_devices->device_list_mutex); |
753 | if (fs_devices->opened) { | |
754 | mutex_unlock(&fs_devices->device_list_mutex); | |
4fde46f0 | 755 | continue; |
7bcb8164 | 756 | } |
4fde46f0 | 757 | |
fa6d2ae5 AJ |
758 | list_for_each_entry_safe(device, tmp_device, |
759 | &fs_devices->devices, dev_list) { | |
522f1b45 | 760 | int not_found = 0; |
4fde46f0 | 761 | |
fa6d2ae5 | 762 | if (skip_device && skip_device == device) |
d8367db3 | 763 | continue; |
fa6d2ae5 | 764 | if (path && !device->name) |
4fde46f0 AJ |
765 | continue; |
766 | ||
4fde46f0 | 767 | rcu_read_lock(); |
d8367db3 | 768 | if (path) |
fa6d2ae5 | 769 | not_found = strcmp(rcu_str_deref(device->name), |
d8367db3 | 770 | path); |
4fde46f0 | 771 | rcu_read_unlock(); |
38cf665d AJ |
772 | if (not_found) |
773 | continue; | |
4fde46f0 | 774 | |
4fde46f0 | 775 | /* delete the stale device */ |
7bcb8164 AJ |
776 | fs_devices->num_devices--; |
777 | list_del(&device->dev_list); | |
778 | btrfs_free_device(device); | |
779 | ||
780 | if (fs_devices->num_devices == 0) | |
fd649f10 | 781 | break; |
7bcb8164 AJ |
782 | } |
783 | mutex_unlock(&fs_devices->device_list_mutex); | |
784 | if (fs_devices->num_devices == 0) { | |
785 | btrfs_sysfs_remove_fsid(fs_devices); | |
786 | list_del(&fs_devices->fs_list); | |
787 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
788 | } |
789 | } | |
790 | } | |
791 | ||
0fb08bcc AJ |
792 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
793 | struct btrfs_device *device, fmode_t flags, | |
794 | void *holder) | |
795 | { | |
796 | struct request_queue *q; | |
797 | struct block_device *bdev; | |
798 | struct buffer_head *bh; | |
799 | struct btrfs_super_block *disk_super; | |
800 | u64 devid; | |
801 | int ret; | |
802 | ||
803 | if (device->bdev) | |
804 | return -EINVAL; | |
805 | if (!device->name) | |
806 | return -EINVAL; | |
807 | ||
808 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
809 | &bdev, &bh); | |
810 | if (ret) | |
811 | return ret; | |
812 | ||
813 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
814 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
815 | if (devid != device->devid) | |
816 | goto error_brelse; | |
817 | ||
818 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
819 | goto error_brelse; | |
820 | ||
821 | device->generation = btrfs_super_generation(disk_super); | |
822 | ||
823 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
824 | if (btrfs_super_incompat_flags(disk_super) & |
825 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
826 | pr_err( | |
827 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
828 | goto error_brelse; | |
829 | } | |
830 | ||
ebbede42 | 831 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
832 | fs_devices->seeding = 1; |
833 | } else { | |
ebbede42 AJ |
834 | if (bdev_read_only(bdev)) |
835 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
836 | else | |
837 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
838 | } |
839 | ||
840 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
841 | if (!blk_queue_nonrot(q)) |
842 | fs_devices->rotating = 1; | |
843 | ||
844 | device->bdev = bdev; | |
e12c9621 | 845 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
846 | device->mode = flags; |
847 | ||
848 | fs_devices->open_devices++; | |
ebbede42 AJ |
849 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
850 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 851 | fs_devices->rw_devices++; |
b1b8e386 | 852 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
853 | } |
854 | brelse(bh); | |
855 | ||
856 | return 0; | |
857 | ||
858 | error_brelse: | |
859 | brelse(bh); | |
860 | blkdev_put(bdev, flags); | |
861 | ||
862 | return -EINVAL; | |
863 | } | |
864 | ||
7a62d0f0 NB |
865 | /* |
866 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
867 | * being created with a disk that has already completed its fsid change. | |
868 | */ | |
869 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
870 | struct btrfs_super_block *disk_super) | |
871 | { | |
872 | struct btrfs_fs_devices *fs_devices; | |
873 | ||
874 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
875 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
876 | BTRFS_FSID_SIZE) != 0 && | |
877 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
878 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
879 | return fs_devices; | |
880 | } | |
881 | } | |
882 | ||
883 | return NULL; | |
884 | } | |
885 | ||
cc5de4e7 NB |
886 | |
887 | static struct btrfs_fs_devices *find_fsid_changed( | |
888 | struct btrfs_super_block *disk_super) | |
889 | { | |
890 | struct btrfs_fs_devices *fs_devices; | |
891 | ||
892 | /* | |
893 | * Handles the case where scanned device is part of an fs that had | |
894 | * multiple successful changes of FSID but curently device didn't | |
895 | * observe it. Meaning our fsid will be different than theirs. | |
896 | */ | |
897 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
898 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
899 | BTRFS_FSID_SIZE) != 0 && | |
900 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
901 | BTRFS_FSID_SIZE) == 0 && | |
902 | memcmp(fs_devices->fsid, disk_super->fsid, | |
903 | BTRFS_FSID_SIZE) != 0) { | |
904 | return fs_devices; | |
905 | } | |
906 | } | |
907 | ||
908 | return NULL; | |
909 | } | |
60999ca4 DS |
910 | /* |
911 | * Add new device to list of registered devices | |
912 | * | |
913 | * Returns: | |
e124ece5 AJ |
914 | * device pointer which was just added or updated when successful |
915 | * error pointer when failed | |
60999ca4 | 916 | */ |
e124ece5 | 917 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
918 | struct btrfs_super_block *disk_super, |
919 | bool *new_device_added) | |
8a4b83cc CM |
920 | { |
921 | struct btrfs_device *device; | |
7a62d0f0 | 922 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 923 | struct rcu_string *name; |
8a4b83cc | 924 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 925 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
926 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
927 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
928 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
929 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 930 | |
cc5de4e7 NB |
931 | if (fsid_change_in_progress) { |
932 | if (!has_metadata_uuid) { | |
933 | /* | |
934 | * When we have an image which has CHANGING_FSID_V2 set | |
935 | * it might belong to either a filesystem which has | |
936 | * disks with completed fsid change or it might belong | |
937 | * to fs with no UUID changes in effect, handle both. | |
938 | */ | |
939 | fs_devices = find_fsid_inprogress(disk_super); | |
940 | if (!fs_devices) | |
941 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
942 | } else { | |
943 | fs_devices = find_fsid_changed(disk_super); | |
944 | } | |
7a62d0f0 NB |
945 | } else if (has_metadata_uuid) { |
946 | fs_devices = find_fsid(disk_super->fsid, | |
947 | disk_super->metadata_uuid); | |
948 | } else { | |
7239ff4b | 949 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
950 | } |
951 | ||
8a4b83cc | 952 | |
8a4b83cc | 953 | if (!fs_devices) { |
7239ff4b NB |
954 | if (has_metadata_uuid) |
955 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
956 | disk_super->metadata_uuid); | |
957 | else | |
958 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
959 | ||
d1a63002 NB |
960 | fs_devices->fsid_change = fsid_change_in_progress; |
961 | ||
2208a378 | 962 | if (IS_ERR(fs_devices)) |
e124ece5 | 963 | return ERR_CAST(fs_devices); |
2208a378 | 964 | |
9c6d173e | 965 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 966 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 967 | |
8a4b83cc CM |
968 | device = NULL; |
969 | } else { | |
9c6d173e | 970 | mutex_lock(&fs_devices->device_list_mutex); |
35c70103 DS |
971 | device = find_device(fs_devices, devid, |
972 | disk_super->dev_item.uuid); | |
7a62d0f0 NB |
973 | |
974 | /* | |
975 | * If this disk has been pulled into an fs devices created by | |
976 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
977 | * metadata_uuid/fsid values of the fs_devices. | |
978 | */ | |
979 | if (has_metadata_uuid && fs_devices->fsid_change && | |
980 | found_transid > fs_devices->latest_generation) { | |
981 | memcpy(fs_devices->fsid, disk_super->fsid, | |
982 | BTRFS_FSID_SIZE); | |
983 | memcpy(fs_devices->metadata_uuid, | |
984 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
985 | ||
986 | fs_devices->fsid_change = false; | |
987 | } | |
8a4b83cc | 988 | } |
443f24fe | 989 | |
8a4b83cc | 990 | if (!device) { |
9c6d173e AJ |
991 | if (fs_devices->opened) { |
992 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 993 | return ERR_PTR(-EBUSY); |
9c6d173e | 994 | } |
2b82032c | 995 | |
12bd2fc0 ID |
996 | device = btrfs_alloc_device(NULL, &devid, |
997 | disk_super->dev_item.uuid); | |
998 | if (IS_ERR(device)) { | |
9c6d173e | 999 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 1000 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 1001 | return device; |
8a4b83cc | 1002 | } |
606686ee JB |
1003 | |
1004 | name = rcu_string_strdup(path, GFP_NOFS); | |
1005 | if (!name) { | |
a425f9d4 | 1006 | btrfs_free_device(device); |
9c6d173e | 1007 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1008 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 1009 | } |
606686ee | 1010 | rcu_assign_pointer(device->name, name); |
90519d66 | 1011 | |
1f78160c | 1012 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 1013 | fs_devices->num_devices++; |
e5e9a520 | 1014 | |
2b82032c | 1015 | device->fs_devices = fs_devices; |
4306a974 | 1016 | *new_device_added = true; |
327f18cc AJ |
1017 | |
1018 | if (disk_super->label[0]) | |
1019 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
1020 | disk_super->label, devid, found_transid, path); | |
1021 | else | |
1022 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
1023 | disk_super->fsid, devid, found_transid, path); | |
1024 | ||
606686ee | 1025 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
1026 | /* |
1027 | * When FS is already mounted. | |
1028 | * 1. If you are here and if the device->name is NULL that | |
1029 | * means this device was missing at time of FS mount. | |
1030 | * 2. If you are here and if the device->name is different | |
1031 | * from 'path' that means either | |
1032 | * a. The same device disappeared and reappeared with | |
1033 | * different name. or | |
1034 | * b. The missing-disk-which-was-replaced, has | |
1035 | * reappeared now. | |
1036 | * | |
1037 | * We must allow 1 and 2a above. But 2b would be a spurious | |
1038 | * and unintentional. | |
1039 | * | |
1040 | * Further in case of 1 and 2a above, the disk at 'path' | |
1041 | * would have missed some transaction when it was away and | |
1042 | * in case of 2a the stale bdev has to be updated as well. | |
1043 | * 2b must not be allowed at all time. | |
1044 | */ | |
1045 | ||
1046 | /* | |
0f23ae74 CM |
1047 | * For now, we do allow update to btrfs_fs_device through the |
1048 | * btrfs dev scan cli after FS has been mounted. We're still | |
1049 | * tracking a problem where systems fail mount by subvolume id | |
1050 | * when we reject replacement on a mounted FS. | |
b96de000 | 1051 | */ |
0f23ae74 | 1052 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
1053 | /* |
1054 | * That is if the FS is _not_ mounted and if you | |
1055 | * are here, that means there is more than one | |
1056 | * disk with same uuid and devid.We keep the one | |
1057 | * with larger generation number or the last-in if | |
1058 | * generation are equal. | |
1059 | */ | |
9c6d173e | 1060 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1061 | return ERR_PTR(-EEXIST); |
77bdae4d | 1062 | } |
b96de000 | 1063 | |
a9261d41 AJ |
1064 | /* |
1065 | * We are going to replace the device path for a given devid, | |
1066 | * make sure it's the same device if the device is mounted | |
1067 | */ | |
1068 | if (device->bdev) { | |
1069 | struct block_device *path_bdev; | |
1070 | ||
1071 | path_bdev = lookup_bdev(path); | |
1072 | if (IS_ERR(path_bdev)) { | |
1073 | mutex_unlock(&fs_devices->device_list_mutex); | |
1074 | return ERR_CAST(path_bdev); | |
1075 | } | |
1076 | ||
1077 | if (device->bdev != path_bdev) { | |
1078 | bdput(path_bdev); | |
1079 | mutex_unlock(&fs_devices->device_list_mutex); | |
1080 | btrfs_warn_in_rcu(device->fs_info, | |
1081 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1082 | disk_super->fsid, devid, | |
1083 | rcu_str_deref(device->name), path); | |
1084 | return ERR_PTR(-EEXIST); | |
1085 | } | |
1086 | bdput(path_bdev); | |
1087 | btrfs_info_in_rcu(device->fs_info, | |
1088 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1089 | disk_super->fsid, devid, | |
1090 | rcu_str_deref(device->name), path); | |
1091 | } | |
1092 | ||
606686ee | 1093 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1094 | if (!name) { |
1095 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1096 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1097 | } |
606686ee JB |
1098 | rcu_string_free(device->name); |
1099 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1100 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1101 | fs_devices->missing_devices--; |
e6e674bd | 1102 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1103 | } |
8a4b83cc CM |
1104 | } |
1105 | ||
77bdae4d AJ |
1106 | /* |
1107 | * Unmount does not free the btrfs_device struct but would zero | |
1108 | * generation along with most of the other members. So just update | |
1109 | * it back. We need it to pick the disk with largest generation | |
1110 | * (as above). | |
1111 | */ | |
d1a63002 | 1112 | if (!fs_devices->opened) { |
77bdae4d | 1113 | device->generation = found_transid; |
d1a63002 NB |
1114 | fs_devices->latest_generation = max_t(u64, found_transid, |
1115 | fs_devices->latest_generation); | |
1116 | } | |
77bdae4d | 1117 | |
f2788d2f AJ |
1118 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1119 | ||
9c6d173e | 1120 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1121 | return device; |
8a4b83cc CM |
1122 | } |
1123 | ||
e4404d6e YZ |
1124 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1125 | { | |
1126 | struct btrfs_fs_devices *fs_devices; | |
1127 | struct btrfs_device *device; | |
1128 | struct btrfs_device *orig_dev; | |
1129 | ||
7239ff4b | 1130 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1131 | if (IS_ERR(fs_devices)) |
1132 | return fs_devices; | |
e4404d6e | 1133 | |
adbbb863 | 1134 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1135 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 1136 | |
46224705 | 1137 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 1138 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
1139 | struct rcu_string *name; |
1140 | ||
12bd2fc0 ID |
1141 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1142 | orig_dev->uuid); | |
1143 | if (IS_ERR(device)) | |
e4404d6e YZ |
1144 | goto error; |
1145 | ||
606686ee JB |
1146 | /* |
1147 | * This is ok to do without rcu read locked because we hold the | |
1148 | * uuid mutex so nothing we touch in here is going to disappear. | |
1149 | */ | |
e755f780 | 1150 | if (orig_dev->name) { |
78f2c9e6 DS |
1151 | name = rcu_string_strdup(orig_dev->name->str, |
1152 | GFP_KERNEL); | |
e755f780 | 1153 | if (!name) { |
a425f9d4 | 1154 | btrfs_free_device(device); |
e755f780 AJ |
1155 | goto error; |
1156 | } | |
1157 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1158 | } |
e4404d6e | 1159 | |
e4404d6e YZ |
1160 | list_add(&device->dev_list, &fs_devices->devices); |
1161 | device->fs_devices = fs_devices; | |
1162 | fs_devices->num_devices++; | |
1163 | } | |
adbbb863 | 1164 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1165 | return fs_devices; |
1166 | error: | |
adbbb863 | 1167 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1168 | free_fs_devices(fs_devices); |
1169 | return ERR_PTR(-ENOMEM); | |
1170 | } | |
1171 | ||
9b99b115 AJ |
1172 | /* |
1173 | * After we have read the system tree and know devids belonging to | |
1174 | * this filesystem, remove the device which does not belong there. | |
1175 | */ | |
1176 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1177 | { |
c6e30871 | 1178 | struct btrfs_device *device, *next; |
443f24fe | 1179 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1180 | |
dfe25020 CM |
1181 | mutex_lock(&uuid_mutex); |
1182 | again: | |
46224705 | 1183 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1184 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1185 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1186 | &device->dev_state)) { | |
401e29c1 AJ |
1187 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1188 | &device->dev_state) && | |
1189 | (!latest_dev || | |
1190 | device->generation > latest_dev->generation)) { | |
443f24fe | 1191 | latest_dev = device; |
a6b0d5c8 | 1192 | } |
2b82032c | 1193 | continue; |
a6b0d5c8 | 1194 | } |
2b82032c | 1195 | |
8dabb742 SB |
1196 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1197 | /* | |
1198 | * In the first step, keep the device which has | |
1199 | * the correct fsid and the devid that is used | |
1200 | * for the dev_replace procedure. | |
1201 | * In the second step, the dev_replace state is | |
1202 | * read from the device tree and it is known | |
1203 | * whether the procedure is really active or | |
1204 | * not, which means whether this device is | |
1205 | * used or whether it should be removed. | |
1206 | */ | |
401e29c1 AJ |
1207 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1208 | &device->dev_state)) { | |
8dabb742 SB |
1209 | continue; |
1210 | } | |
1211 | } | |
2b82032c | 1212 | if (device->bdev) { |
d4d77629 | 1213 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1214 | device->bdev = NULL; |
1215 | fs_devices->open_devices--; | |
1216 | } | |
ebbede42 | 1217 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1218 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1219 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1220 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1221 | &device->dev_state)) | |
8dabb742 | 1222 | fs_devices->rw_devices--; |
2b82032c | 1223 | } |
e4404d6e YZ |
1224 | list_del_init(&device->dev_list); |
1225 | fs_devices->num_devices--; | |
a425f9d4 | 1226 | btrfs_free_device(device); |
dfe25020 | 1227 | } |
2b82032c YZ |
1228 | |
1229 | if (fs_devices->seed) { | |
1230 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1231 | goto again; |
1232 | } | |
1233 | ||
443f24fe | 1234 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1235 | |
dfe25020 | 1236 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1237 | } |
a0af469b | 1238 | |
f06c5965 | 1239 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
1240 | { |
1241 | struct btrfs_device *device; | |
1242 | ||
9f5316c1 | 1243 | device = container_of(head, struct btrfs_device, rcu); |
a425f9d4 | 1244 | btrfs_free_device(device); |
1f78160c XG |
1245 | } |
1246 | ||
14238819 AJ |
1247 | static void btrfs_close_bdev(struct btrfs_device *device) |
1248 | { | |
08ffcae8 DS |
1249 | if (!device->bdev) |
1250 | return; | |
1251 | ||
ebbede42 | 1252 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1253 | sync_blockdev(device->bdev); |
1254 | invalidate_bdev(device->bdev); | |
1255 | } | |
1256 | ||
08ffcae8 | 1257 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1258 | } |
1259 | ||
959b1c04 | 1260 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1261 | { |
1262 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1263 | struct btrfs_device *new_device; | |
1264 | struct rcu_string *name; | |
1265 | ||
1266 | if (device->bdev) | |
1267 | fs_devices->open_devices--; | |
1268 | ||
ebbede42 | 1269 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1270 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1271 | list_del_init(&device->dev_alloc_list); | |
1272 | fs_devices->rw_devices--; | |
1273 | } | |
1274 | ||
e6e674bd | 1275 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1276 | fs_devices->missing_devices--; |
1277 | ||
959b1c04 NB |
1278 | btrfs_close_bdev(device); |
1279 | ||
f448341a AJ |
1280 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1281 | device->uuid); | |
1282 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1283 | ||
1284 | /* Safe because we are under uuid_mutex */ | |
1285 | if (device->name) { | |
1286 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1287 | BUG_ON(!name); /* -ENOMEM */ | |
1288 | rcu_assign_pointer(new_device->name, name); | |
1289 | } | |
1290 | ||
1291 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1292 | new_device->fs_devices = device->fs_devices; | |
959b1c04 NB |
1293 | |
1294 | call_rcu(&device->rcu, free_device_rcu); | |
f448341a AJ |
1295 | } |
1296 | ||
0226e0eb | 1297 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1298 | { |
2037a093 | 1299 | struct btrfs_device *device, *tmp; |
e4404d6e | 1300 | |
2b82032c YZ |
1301 | if (--fs_devices->opened > 0) |
1302 | return 0; | |
8a4b83cc | 1303 | |
c9513edb | 1304 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1305 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1306 | btrfs_close_one_device(device); |
8a4b83cc | 1307 | } |
c9513edb XG |
1308 | mutex_unlock(&fs_devices->device_list_mutex); |
1309 | ||
e4404d6e YZ |
1310 | WARN_ON(fs_devices->open_devices); |
1311 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1312 | fs_devices->opened = 0; |
1313 | fs_devices->seeding = 0; | |
2b82032c | 1314 | |
8a4b83cc CM |
1315 | return 0; |
1316 | } | |
1317 | ||
2b82032c YZ |
1318 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1319 | { | |
e4404d6e | 1320 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1321 | int ret; |
1322 | ||
1323 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1324 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1325 | if (!fs_devices->opened) { |
1326 | seed_devices = fs_devices->seed; | |
1327 | fs_devices->seed = NULL; | |
1328 | } | |
2b82032c | 1329 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1330 | |
1331 | while (seed_devices) { | |
1332 | fs_devices = seed_devices; | |
1333 | seed_devices = fs_devices->seed; | |
0226e0eb | 1334 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1335 | free_fs_devices(fs_devices); |
1336 | } | |
2b82032c YZ |
1337 | return ret; |
1338 | } | |
1339 | ||
897fb573 | 1340 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1341 | fmode_t flags, void *holder) |
8a4b83cc | 1342 | { |
8a4b83cc | 1343 | struct btrfs_device *device; |
443f24fe | 1344 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1345 | int ret = 0; |
8a4b83cc | 1346 | |
d4d77629 TH |
1347 | flags |= FMODE_EXCL; |
1348 | ||
f117e290 | 1349 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1350 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1351 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1352 | continue; |
a0af469b | 1353 | |
9f050db4 AJ |
1354 | if (!latest_dev || |
1355 | device->generation > latest_dev->generation) | |
1356 | latest_dev = device; | |
8a4b83cc | 1357 | } |
a0af469b | 1358 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1359 | ret = -EINVAL; |
a0af469b CM |
1360 | goto out; |
1361 | } | |
2b82032c | 1362 | fs_devices->opened = 1; |
443f24fe | 1363 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1364 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1365 | out: |
2b82032c YZ |
1366 | return ret; |
1367 | } | |
1368 | ||
f8e10cd3 AJ |
1369 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1370 | { | |
1371 | struct btrfs_device *dev1, *dev2; | |
1372 | ||
1373 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1374 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1375 | ||
1376 | if (dev1->devid < dev2->devid) | |
1377 | return -1; | |
1378 | else if (dev1->devid > dev2->devid) | |
1379 | return 1; | |
1380 | return 0; | |
1381 | } | |
1382 | ||
2b82032c | 1383 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1384 | fmode_t flags, void *holder) |
2b82032c YZ |
1385 | { |
1386 | int ret; | |
1387 | ||
f5194e34 DS |
1388 | lockdep_assert_held(&uuid_mutex); |
1389 | ||
542c5908 | 1390 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1391 | if (fs_devices->opened) { |
e4404d6e YZ |
1392 | fs_devices->opened++; |
1393 | ret = 0; | |
2b82032c | 1394 | } else { |
f8e10cd3 | 1395 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1396 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1397 | } |
542c5908 AJ |
1398 | mutex_unlock(&fs_devices->device_list_mutex); |
1399 | ||
8a4b83cc CM |
1400 | return ret; |
1401 | } | |
1402 | ||
c9162bdf | 1403 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1404 | { |
1405 | kunmap(page); | |
1406 | put_page(page); | |
1407 | } | |
1408 | ||
c9162bdf OS |
1409 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1410 | struct page **page, | |
1411 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1412 | { |
1413 | void *p; | |
1414 | pgoff_t index; | |
1415 | ||
1416 | /* make sure our super fits in the device */ | |
1417 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1418 | return 1; | |
1419 | ||
1420 | /* make sure our super fits in the page */ | |
1421 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1422 | return 1; | |
1423 | ||
1424 | /* make sure our super doesn't straddle pages on disk */ | |
1425 | index = bytenr >> PAGE_SHIFT; | |
1426 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1427 | return 1; | |
1428 | ||
1429 | /* pull in the page with our super */ | |
1430 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1431 | index, GFP_KERNEL); | |
1432 | ||
1433 | if (IS_ERR_OR_NULL(*page)) | |
1434 | return 1; | |
1435 | ||
1436 | p = kmap(*page); | |
1437 | ||
1438 | /* align our pointer to the offset of the super block */ | |
1439 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1440 | ||
1441 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1442 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1443 | btrfs_release_disk_super(*page); | |
1444 | return 1; | |
1445 | } | |
1446 | ||
1447 | if ((*disk_super)->label[0] && | |
1448 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1449 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1450 | ||
1451 | return 0; | |
1452 | } | |
1453 | ||
6f60cbd3 DS |
1454 | /* |
1455 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1456 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1457 | * is read via pagecache | |
1458 | */ | |
36350e95 GJ |
1459 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1460 | void *holder) | |
8a4b83cc CM |
1461 | { |
1462 | struct btrfs_super_block *disk_super; | |
4306a974 | 1463 | bool new_device_added = false; |
36350e95 | 1464 | struct btrfs_device *device = NULL; |
8a4b83cc | 1465 | struct block_device *bdev; |
6f60cbd3 | 1466 | struct page *page; |
6f60cbd3 | 1467 | u64 bytenr; |
8a4b83cc | 1468 | |
899f9307 DS |
1469 | lockdep_assert_held(&uuid_mutex); |
1470 | ||
6f60cbd3 DS |
1471 | /* |
1472 | * we would like to check all the supers, but that would make | |
1473 | * a btrfs mount succeed after a mkfs from a different FS. | |
1474 | * So, we need to add a special mount option to scan for | |
1475 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1476 | */ | |
1477 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1478 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1479 | |
1480 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1481 | if (IS_ERR(bdev)) |
36350e95 | 1482 | return ERR_CAST(bdev); |
6f60cbd3 | 1483 | |
05a5c55d | 1484 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1485 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1486 | goto error_bdev_put; |
05a5c55d | 1487 | } |
6f60cbd3 | 1488 | |
4306a974 | 1489 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1490 | if (!IS_ERR(device)) { |
4306a974 AJ |
1491 | if (new_device_added) |
1492 | btrfs_free_stale_devices(path, device); | |
1493 | } | |
6f60cbd3 | 1494 | |
6cf86a00 | 1495 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1496 | |
1497 | error_bdev_put: | |
d4d77629 | 1498 | blkdev_put(bdev, flags); |
b6ed73bc | 1499 | |
36350e95 | 1500 | return device; |
8a4b83cc | 1501 | } |
0b86a832 | 1502 | |
499f377f | 1503 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1504 | struct btrfs_device *device, |
1505 | u64 *start, u64 len) | |
1506 | { | |
fb456252 | 1507 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1508 | struct extent_map *em; |
499f377f | 1509 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1510 | int ret = 0; |
1b984508 | 1511 | u64 physical_start = *start; |
6df9a95e | 1512 | |
499f377f JM |
1513 | if (transaction) |
1514 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1515 | again: |
1516 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1517 | struct map_lookup *map; |
1518 | int i; | |
1519 | ||
95617d69 | 1520 | map = em->map_lookup; |
6df9a95e | 1521 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1522 | u64 end; |
1523 | ||
6df9a95e JB |
1524 | if (map->stripes[i].dev != device) |
1525 | continue; | |
1b984508 | 1526 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1527 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1528 | physical_start) |
6df9a95e | 1529 | continue; |
c152b63e FM |
1530 | /* |
1531 | * Make sure that while processing the pinned list we do | |
1532 | * not override our *start with a lower value, because | |
1533 | * we can have pinned chunks that fall within this | |
1534 | * device hole and that have lower physical addresses | |
1535 | * than the pending chunks we processed before. If we | |
1536 | * do not take this special care we can end up getting | |
1537 | * 2 pending chunks that start at the same physical | |
1538 | * device offsets because the end offset of a pinned | |
1539 | * chunk can be equal to the start offset of some | |
1540 | * pending chunk. | |
1541 | */ | |
1542 | end = map->stripes[i].physical + em->orig_block_len; | |
1543 | if (end > *start) { | |
1544 | *start = end; | |
1545 | ret = 1; | |
1546 | } | |
6df9a95e JB |
1547 | } |
1548 | } | |
499f377f JM |
1549 | if (search_list != &fs_info->pinned_chunks) { |
1550 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1551 | goto again; |
1552 | } | |
6df9a95e JB |
1553 | |
1554 | return ret; | |
1555 | } | |
1556 | ||
1557 | ||
0b86a832 | 1558 | /* |
499f377f JM |
1559 | * find_free_dev_extent_start - find free space in the specified device |
1560 | * @device: the device which we search the free space in | |
1561 | * @num_bytes: the size of the free space that we need | |
1562 | * @search_start: the position from which to begin the search | |
1563 | * @start: store the start of the free space. | |
1564 | * @len: the size of the free space. that we find, or the size | |
1565 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1566 | * |
0b86a832 CM |
1567 | * this uses a pretty simple search, the expectation is that it is |
1568 | * called very infrequently and that a given device has a small number | |
1569 | * of extents | |
7bfc837d MX |
1570 | * |
1571 | * @start is used to store the start of the free space if we find. But if we | |
1572 | * don't find suitable free space, it will be used to store the start position | |
1573 | * of the max free space. | |
1574 | * | |
1575 | * @len is used to store the size of the free space that we find. | |
1576 | * But if we don't find suitable free space, it is used to store the size of | |
1577 | * the max free space. | |
0b86a832 | 1578 | */ |
499f377f JM |
1579 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1580 | struct btrfs_device *device, u64 num_bytes, | |
1581 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1582 | { |
0b246afa JM |
1583 | struct btrfs_fs_info *fs_info = device->fs_info; |
1584 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1585 | struct btrfs_key key; |
7bfc837d | 1586 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1587 | struct btrfs_path *path; |
7bfc837d MX |
1588 | u64 hole_size; |
1589 | u64 max_hole_start; | |
1590 | u64 max_hole_size; | |
1591 | u64 extent_end; | |
0b86a832 CM |
1592 | u64 search_end = device->total_bytes; |
1593 | int ret; | |
7bfc837d | 1594 | int slot; |
0b86a832 | 1595 | struct extent_buffer *l; |
8cdc7c5b FM |
1596 | |
1597 | /* | |
1598 | * We don't want to overwrite the superblock on the drive nor any area | |
1599 | * used by the boot loader (grub for example), so we make sure to start | |
1600 | * at an offset of at least 1MB. | |
1601 | */ | |
0d0c71b3 | 1602 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1603 | |
6df9a95e JB |
1604 | path = btrfs_alloc_path(); |
1605 | if (!path) | |
1606 | return -ENOMEM; | |
f2ab7618 | 1607 | |
7bfc837d MX |
1608 | max_hole_start = search_start; |
1609 | max_hole_size = 0; | |
1610 | ||
f2ab7618 | 1611 | again: |
401e29c1 AJ |
1612 | if (search_start >= search_end || |
1613 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1614 | ret = -ENOSPC; |
6df9a95e | 1615 | goto out; |
7bfc837d MX |
1616 | } |
1617 | ||
e4058b54 | 1618 | path->reada = READA_FORWARD; |
6df9a95e JB |
1619 | path->search_commit_root = 1; |
1620 | path->skip_locking = 1; | |
7bfc837d | 1621 | |
0b86a832 CM |
1622 | key.objectid = device->devid; |
1623 | key.offset = search_start; | |
1624 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1625 | |
125ccb0a | 1626 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1627 | if (ret < 0) |
7bfc837d | 1628 | goto out; |
1fcbac58 YZ |
1629 | if (ret > 0) { |
1630 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1631 | if (ret < 0) | |
7bfc837d | 1632 | goto out; |
1fcbac58 | 1633 | } |
7bfc837d | 1634 | |
0b86a832 CM |
1635 | while (1) { |
1636 | l = path->nodes[0]; | |
1637 | slot = path->slots[0]; | |
1638 | if (slot >= btrfs_header_nritems(l)) { | |
1639 | ret = btrfs_next_leaf(root, path); | |
1640 | if (ret == 0) | |
1641 | continue; | |
1642 | if (ret < 0) | |
7bfc837d MX |
1643 | goto out; |
1644 | ||
1645 | break; | |
0b86a832 CM |
1646 | } |
1647 | btrfs_item_key_to_cpu(l, &key, slot); | |
1648 | ||
1649 | if (key.objectid < device->devid) | |
1650 | goto next; | |
1651 | ||
1652 | if (key.objectid > device->devid) | |
7bfc837d | 1653 | break; |
0b86a832 | 1654 | |
962a298f | 1655 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1656 | goto next; |
9779b72f | 1657 | |
7bfc837d MX |
1658 | if (key.offset > search_start) { |
1659 | hole_size = key.offset - search_start; | |
9779b72f | 1660 | |
6df9a95e JB |
1661 | /* |
1662 | * Have to check before we set max_hole_start, otherwise | |
1663 | * we could end up sending back this offset anyway. | |
1664 | */ | |
499f377f | 1665 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1666 | &search_start, |
1b984508 FL |
1667 | hole_size)) { |
1668 | if (key.offset >= search_start) { | |
1669 | hole_size = key.offset - search_start; | |
1670 | } else { | |
1671 | WARN_ON_ONCE(1); | |
1672 | hole_size = 0; | |
1673 | } | |
1674 | } | |
6df9a95e | 1675 | |
7bfc837d MX |
1676 | if (hole_size > max_hole_size) { |
1677 | max_hole_start = search_start; | |
1678 | max_hole_size = hole_size; | |
1679 | } | |
9779b72f | 1680 | |
7bfc837d MX |
1681 | /* |
1682 | * If this free space is greater than which we need, | |
1683 | * it must be the max free space that we have found | |
1684 | * until now, so max_hole_start must point to the start | |
1685 | * of this free space and the length of this free space | |
1686 | * is stored in max_hole_size. Thus, we return | |
1687 | * max_hole_start and max_hole_size and go back to the | |
1688 | * caller. | |
1689 | */ | |
1690 | if (hole_size >= num_bytes) { | |
1691 | ret = 0; | |
1692 | goto out; | |
0b86a832 CM |
1693 | } |
1694 | } | |
0b86a832 | 1695 | |
0b86a832 | 1696 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1697 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1698 | dev_extent); | |
1699 | if (extent_end > search_start) | |
1700 | search_start = extent_end; | |
0b86a832 CM |
1701 | next: |
1702 | path->slots[0]++; | |
1703 | cond_resched(); | |
1704 | } | |
0b86a832 | 1705 | |
38c01b96 | 1706 | /* |
1707 | * At this point, search_start should be the end of | |
1708 | * allocated dev extents, and when shrinking the device, | |
1709 | * search_end may be smaller than search_start. | |
1710 | */ | |
f2ab7618 | 1711 | if (search_end > search_start) { |
38c01b96 | 1712 | hole_size = search_end - search_start; |
1713 | ||
499f377f | 1714 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1715 | hole_size)) { |
1716 | btrfs_release_path(path); | |
1717 | goto again; | |
1718 | } | |
0b86a832 | 1719 | |
f2ab7618 ZL |
1720 | if (hole_size > max_hole_size) { |
1721 | max_hole_start = search_start; | |
1722 | max_hole_size = hole_size; | |
1723 | } | |
6df9a95e JB |
1724 | } |
1725 | ||
7bfc837d | 1726 | /* See above. */ |
f2ab7618 | 1727 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1728 | ret = -ENOSPC; |
1729 | else | |
1730 | ret = 0; | |
1731 | ||
1732 | out: | |
2b82032c | 1733 | btrfs_free_path(path); |
7bfc837d | 1734 | *start = max_hole_start; |
b2117a39 | 1735 | if (len) |
7bfc837d | 1736 | *len = max_hole_size; |
0b86a832 CM |
1737 | return ret; |
1738 | } | |
1739 | ||
499f377f JM |
1740 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1741 | struct btrfs_device *device, u64 num_bytes, | |
1742 | u64 *start, u64 *len) | |
1743 | { | |
499f377f | 1744 | /* FIXME use last free of some kind */ |
499f377f | 1745 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1746 | num_bytes, 0, start, len); |
499f377f JM |
1747 | } |
1748 | ||
b2950863 | 1749 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1750 | struct btrfs_device *device, |
2196d6e8 | 1751 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1752 | { |
0b246afa JM |
1753 | struct btrfs_fs_info *fs_info = device->fs_info; |
1754 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1755 | int ret; |
1756 | struct btrfs_path *path; | |
8f18cf13 | 1757 | struct btrfs_key key; |
a061fc8d CM |
1758 | struct btrfs_key found_key; |
1759 | struct extent_buffer *leaf = NULL; | |
1760 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1761 | |
1762 | path = btrfs_alloc_path(); | |
1763 | if (!path) | |
1764 | return -ENOMEM; | |
1765 | ||
1766 | key.objectid = device->devid; | |
1767 | key.offset = start; | |
1768 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1769 | again: |
8f18cf13 | 1770 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1771 | if (ret > 0) { |
1772 | ret = btrfs_previous_item(root, path, key.objectid, | |
1773 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1774 | if (ret) |
1775 | goto out; | |
a061fc8d CM |
1776 | leaf = path->nodes[0]; |
1777 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1778 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1779 | struct btrfs_dev_extent); | |
1780 | BUG_ON(found_key.offset > start || found_key.offset + | |
1781 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1782 | key = found_key; |
1783 | btrfs_release_path(path); | |
1784 | goto again; | |
a061fc8d CM |
1785 | } else if (ret == 0) { |
1786 | leaf = path->nodes[0]; | |
1787 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1788 | struct btrfs_dev_extent); | |
79787eaa | 1789 | } else { |
0b246afa | 1790 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1791 | goto out; |
a061fc8d | 1792 | } |
8f18cf13 | 1793 | |
2196d6e8 MX |
1794 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1795 | ||
8f18cf13 | 1796 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1797 | if (ret) { |
0b246afa JM |
1798 | btrfs_handle_fs_error(fs_info, ret, |
1799 | "Failed to remove dev extent item"); | |
13212b54 | 1800 | } else { |
3204d33c | 1801 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1802 | } |
b0b802d7 | 1803 | out: |
8f18cf13 CM |
1804 | btrfs_free_path(path); |
1805 | return ret; | |
1806 | } | |
1807 | ||
48a3b636 ES |
1808 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1809 | struct btrfs_device *device, | |
48a3b636 | 1810 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1811 | { |
1812 | int ret; | |
1813 | struct btrfs_path *path; | |
0b246afa JM |
1814 | struct btrfs_fs_info *fs_info = device->fs_info; |
1815 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1816 | struct btrfs_dev_extent *extent; |
1817 | struct extent_buffer *leaf; | |
1818 | struct btrfs_key key; | |
1819 | ||
e12c9621 | 1820 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1821 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1822 | path = btrfs_alloc_path(); |
1823 | if (!path) | |
1824 | return -ENOMEM; | |
1825 | ||
0b86a832 | 1826 | key.objectid = device->devid; |
2b82032c | 1827 | key.offset = start; |
0b86a832 CM |
1828 | key.type = BTRFS_DEV_EXTENT_KEY; |
1829 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1830 | sizeof(*extent)); | |
2cdcecbc MF |
1831 | if (ret) |
1832 | goto out; | |
0b86a832 CM |
1833 | |
1834 | leaf = path->nodes[0]; | |
1835 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1836 | struct btrfs_dev_extent); | |
b5d9071c NB |
1837 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1838 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1839 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1840 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1841 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1842 | ||
0b86a832 CM |
1843 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1844 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1845 | out: |
0b86a832 CM |
1846 | btrfs_free_path(path); |
1847 | return ret; | |
1848 | } | |
1849 | ||
6df9a95e | 1850 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1851 | { |
6df9a95e JB |
1852 | struct extent_map_tree *em_tree; |
1853 | struct extent_map *em; | |
1854 | struct rb_node *n; | |
1855 | u64 ret = 0; | |
0b86a832 | 1856 | |
6df9a95e JB |
1857 | em_tree = &fs_info->mapping_tree.map_tree; |
1858 | read_lock(&em_tree->lock); | |
07e1ce09 | 1859 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1860 | if (n) { |
1861 | em = rb_entry(n, struct extent_map, rb_node); | |
1862 | ret = em->start + em->len; | |
0b86a832 | 1863 | } |
6df9a95e JB |
1864 | read_unlock(&em_tree->lock); |
1865 | ||
0b86a832 CM |
1866 | return ret; |
1867 | } | |
1868 | ||
53f10659 ID |
1869 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1870 | u64 *devid_ret) | |
0b86a832 CM |
1871 | { |
1872 | int ret; | |
1873 | struct btrfs_key key; | |
1874 | struct btrfs_key found_key; | |
2b82032c YZ |
1875 | struct btrfs_path *path; |
1876 | ||
2b82032c YZ |
1877 | path = btrfs_alloc_path(); |
1878 | if (!path) | |
1879 | return -ENOMEM; | |
0b86a832 CM |
1880 | |
1881 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1882 | key.type = BTRFS_DEV_ITEM_KEY; | |
1883 | key.offset = (u64)-1; | |
1884 | ||
53f10659 | 1885 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1886 | if (ret < 0) |
1887 | goto error; | |
1888 | ||
79787eaa | 1889 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1890 | |
53f10659 ID |
1891 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1892 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1893 | BTRFS_DEV_ITEM_KEY); |
1894 | if (ret) { | |
53f10659 | 1895 | *devid_ret = 1; |
0b86a832 CM |
1896 | } else { |
1897 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1898 | path->slots[0]); | |
53f10659 | 1899 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1900 | } |
1901 | ret = 0; | |
1902 | error: | |
2b82032c | 1903 | btrfs_free_path(path); |
0b86a832 CM |
1904 | return ret; |
1905 | } | |
1906 | ||
1907 | /* | |
1908 | * the device information is stored in the chunk root | |
1909 | * the btrfs_device struct should be fully filled in | |
1910 | */ | |
c74a0b02 | 1911 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1912 | struct btrfs_device *device) |
0b86a832 CM |
1913 | { |
1914 | int ret; | |
1915 | struct btrfs_path *path; | |
1916 | struct btrfs_dev_item *dev_item; | |
1917 | struct extent_buffer *leaf; | |
1918 | struct btrfs_key key; | |
1919 | unsigned long ptr; | |
0b86a832 | 1920 | |
0b86a832 CM |
1921 | path = btrfs_alloc_path(); |
1922 | if (!path) | |
1923 | return -ENOMEM; | |
1924 | ||
0b86a832 CM |
1925 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1926 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1927 | key.offset = device->devid; |
0b86a832 | 1928 | |
8e87e856 NB |
1929 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1930 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1931 | if (ret) |
1932 | goto out; | |
1933 | ||
1934 | leaf = path->nodes[0]; | |
1935 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1936 | ||
1937 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1938 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1939 | btrfs_set_device_type(leaf, dev_item, device->type); |
1940 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1941 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1942 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1943 | btrfs_set_device_total_bytes(leaf, dev_item, |
1944 | btrfs_device_get_disk_total_bytes(device)); | |
1945 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1946 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1947 | btrfs_set_device_group(leaf, dev_item, 0); |
1948 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1949 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1950 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1951 | |
410ba3a2 | 1952 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1953 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1954 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1955 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1956 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1957 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1958 | |
2b82032c | 1959 | ret = 0; |
0b86a832 CM |
1960 | out: |
1961 | btrfs_free_path(path); | |
1962 | return ret; | |
1963 | } | |
8f18cf13 | 1964 | |
5a1972bd QW |
1965 | /* |
1966 | * Function to update ctime/mtime for a given device path. | |
1967 | * Mainly used for ctime/mtime based probe like libblkid. | |
1968 | */ | |
da353f6b | 1969 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1970 | { |
1971 | struct file *filp; | |
1972 | ||
1973 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1974 | if (IS_ERR(filp)) |
5a1972bd QW |
1975 | return; |
1976 | file_update_time(filp); | |
1977 | filp_close(filp, NULL); | |
5a1972bd QW |
1978 | } |
1979 | ||
5b4aacef | 1980 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1981 | struct btrfs_device *device) |
1982 | { | |
5b4aacef | 1983 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1984 | int ret; |
1985 | struct btrfs_path *path; | |
a061fc8d | 1986 | struct btrfs_key key; |
a061fc8d CM |
1987 | struct btrfs_trans_handle *trans; |
1988 | ||
a061fc8d CM |
1989 | path = btrfs_alloc_path(); |
1990 | if (!path) | |
1991 | return -ENOMEM; | |
1992 | ||
a22285a6 | 1993 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1994 | if (IS_ERR(trans)) { |
1995 | btrfs_free_path(path); | |
1996 | return PTR_ERR(trans); | |
1997 | } | |
a061fc8d CM |
1998 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1999 | key.type = BTRFS_DEV_ITEM_KEY; | |
2000 | key.offset = device->devid; | |
2001 | ||
2002 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
2003 | if (ret) { |
2004 | if (ret > 0) | |
2005 | ret = -ENOENT; | |
2006 | btrfs_abort_transaction(trans, ret); | |
2007 | btrfs_end_transaction(trans); | |
a061fc8d CM |
2008 | goto out; |
2009 | } | |
2010 | ||
2011 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
2012 | if (ret) { |
2013 | btrfs_abort_transaction(trans, ret); | |
2014 | btrfs_end_transaction(trans); | |
2015 | } | |
2016 | ||
a061fc8d CM |
2017 | out: |
2018 | btrfs_free_path(path); | |
5e9f2ad5 NB |
2019 | if (!ret) |
2020 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
2021 | return ret; |
2022 | } | |
2023 | ||
3cc31a0d DS |
2024 | /* |
2025 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
2026 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
2027 | * replace. | |
2028 | */ | |
2029 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
2030 | u64 num_devices) | |
a061fc8d | 2031 | { |
a061fc8d | 2032 | u64 all_avail; |
de98ced9 | 2033 | unsigned seq; |
418775a2 | 2034 | int i; |
a061fc8d | 2035 | |
de98ced9 | 2036 | do { |
bd45ffbc | 2037 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2038 | |
bd45ffbc AJ |
2039 | all_avail = fs_info->avail_data_alloc_bits | |
2040 | fs_info->avail_system_alloc_bits | | |
2041 | fs_info->avail_metadata_alloc_bits; | |
2042 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2043 | |
418775a2 | 2044 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2045 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2046 | continue; |
a061fc8d | 2047 | |
418775a2 | 2048 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2049 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2050 | |
418775a2 DS |
2051 | if (ret) |
2052 | return ret; | |
2053 | } | |
53b381b3 DW |
2054 | } |
2055 | ||
bd45ffbc | 2056 | return 0; |
f1fa7f26 AJ |
2057 | } |
2058 | ||
c9162bdf OS |
2059 | static struct btrfs_device * btrfs_find_next_active_device( |
2060 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2061 | { |
2b82032c | 2062 | struct btrfs_device *next_device; |
88acff64 AJ |
2063 | |
2064 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2065 | if (next_device != device && | |
e6e674bd AJ |
2066 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2067 | && next_device->bdev) | |
88acff64 AJ |
2068 | return next_device; |
2069 | } | |
2070 | ||
2071 | return NULL; | |
2072 | } | |
2073 | ||
2074 | /* | |
2075 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2076 | * and replace it with the provided or the next active device, in the context | |
2077 | * where this function called, there should be always be another device (or | |
2078 | * this_dev) which is active. | |
2079 | */ | |
d6507cf1 NB |
2080 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2081 | struct btrfs_device *this_dev) | |
88acff64 | 2082 | { |
d6507cf1 | 2083 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2084 | struct btrfs_device *next_device; |
2085 | ||
2086 | if (this_dev) | |
2087 | next_device = this_dev; | |
2088 | else | |
2089 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2090 | device); | |
2091 | ASSERT(next_device); | |
2092 | ||
2093 | if (fs_info->sb->s_bdev && | |
2094 | (fs_info->sb->s_bdev == device->bdev)) | |
2095 | fs_info->sb->s_bdev = next_device->bdev; | |
2096 | ||
2097 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2098 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2099 | } | |
2100 | ||
1da73967 AJ |
2101 | /* |
2102 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2103 | * currently replaced. | |
2104 | */ | |
2105 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2106 | { | |
2107 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2108 | ||
cb5583dd | 2109 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2110 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2111 | ASSERT(num_devices > 1); | |
2112 | num_devices--; | |
2113 | } | |
cb5583dd | 2114 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2115 | |
2116 | return num_devices; | |
2117 | } | |
2118 | ||
da353f6b DS |
2119 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2120 | u64 devid) | |
f1fa7f26 AJ |
2121 | { |
2122 | struct btrfs_device *device; | |
1f78160c | 2123 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2124 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2125 | u64 num_devices; |
a061fc8d CM |
2126 | int ret = 0; |
2127 | ||
a061fc8d CM |
2128 | mutex_lock(&uuid_mutex); |
2129 | ||
1da73967 | 2130 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2131 | |
0b246afa | 2132 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2133 | if (ret) |
a061fc8d | 2134 | goto out; |
a061fc8d | 2135 | |
a27a94c2 NB |
2136 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2137 | ||
2138 | if (IS_ERR(device)) { | |
2139 | if (PTR_ERR(device) == -ENOENT && | |
2140 | strcmp(device_path, "missing") == 0) | |
2141 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2142 | else | |
2143 | ret = PTR_ERR(device); | |
53b381b3 | 2144 | goto out; |
a27a94c2 | 2145 | } |
dfe25020 | 2146 | |
eede2bf3 OS |
2147 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2148 | btrfs_warn_in_rcu(fs_info, | |
2149 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2150 | rcu_str_deref(device->name), device->devid); | |
2151 | ret = -ETXTBSY; | |
2152 | goto out; | |
2153 | } | |
2154 | ||
401e29c1 | 2155 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2156 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2157 | goto out; |
63a212ab SB |
2158 | } |
2159 | ||
ebbede42 AJ |
2160 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2161 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2162 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2163 | goto out; |
2b82032c YZ |
2164 | } |
2165 | ||
ebbede42 | 2166 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2167 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2168 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2169 | device->fs_devices->rw_devices--; |
34441361 | 2170 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2171 | } |
a061fc8d | 2172 | |
d7901554 | 2173 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2174 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2175 | mutex_lock(&uuid_mutex); |
a061fc8d | 2176 | if (ret) |
9b3517e9 | 2177 | goto error_undo; |
a061fc8d | 2178 | |
63a212ab SB |
2179 | /* |
2180 | * TODO: the superblock still includes this device in its num_devices | |
2181 | * counter although write_all_supers() is not locked out. This | |
2182 | * could give a filesystem state which requires a degraded mount. | |
2183 | */ | |
0b246afa | 2184 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 2185 | if (ret) |
9b3517e9 | 2186 | goto error_undo; |
a061fc8d | 2187 | |
e12c9621 | 2188 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 2189 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
2190 | |
2191 | /* | |
2192 | * the device list mutex makes sure that we don't change | |
2193 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2194 | * the device supers. Whoever is writing all supers, should |
2195 | * lock the device list mutex before getting the number of | |
2196 | * devices in the super block (super_copy). Conversely, | |
2197 | * whoever updates the number of devices in the super block | |
2198 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2199 | */ |
1f78160c | 2200 | |
41a52a0f AJ |
2201 | /* |
2202 | * In normal cases the cur_devices == fs_devices. But in case | |
2203 | * of deleting a seed device, the cur_devices should point to | |
2204 | * its own fs_devices listed under the fs_devices->seed. | |
2205 | */ | |
1f78160c | 2206 | cur_devices = device->fs_devices; |
b5185197 | 2207 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2208 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2209 | |
41a52a0f AJ |
2210 | cur_devices->num_devices--; |
2211 | cur_devices->total_devices--; | |
b4993e64 AJ |
2212 | /* Update total_devices of the parent fs_devices if it's seed */ |
2213 | if (cur_devices != fs_devices) | |
2214 | fs_devices->total_devices--; | |
2b82032c | 2215 | |
e6e674bd | 2216 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2217 | cur_devices->missing_devices--; |
cd02dca5 | 2218 | |
d6507cf1 | 2219 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2220 | |
0bfaa9c5 | 2221 | if (device->bdev) { |
41a52a0f | 2222 | cur_devices->open_devices--; |
0bfaa9c5 | 2223 | /* remove sysfs entry */ |
b5185197 | 2224 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2225 | } |
99994cde | 2226 | |
0b246afa JM |
2227 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2228 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2229 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2230 | |
cea67ab9 JM |
2231 | /* |
2232 | * at this point, the device is zero sized and detached from | |
2233 | * the devices list. All that's left is to zero out the old | |
2234 | * supers and free the device. | |
2235 | */ | |
ebbede42 | 2236 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2237 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2238 | ||
2239 | btrfs_close_bdev(device); | |
f06c5965 | 2240 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2241 | |
1f78160c | 2242 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2243 | while (fs_devices) { |
8321cf25 RS |
2244 | if (fs_devices->seed == cur_devices) { |
2245 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2246 | break; |
8321cf25 | 2247 | } |
e4404d6e | 2248 | fs_devices = fs_devices->seed; |
2b82032c | 2249 | } |
1f78160c | 2250 | cur_devices->seed = NULL; |
0226e0eb | 2251 | close_fs_devices(cur_devices); |
1f78160c | 2252 | free_fs_devices(cur_devices); |
2b82032c YZ |
2253 | } |
2254 | ||
a061fc8d CM |
2255 | out: |
2256 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2257 | return ret; |
24fc572f | 2258 | |
9b3517e9 | 2259 | error_undo: |
ebbede42 | 2260 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2261 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2262 | list_add(&device->dev_alloc_list, |
b5185197 | 2263 | &fs_devices->alloc_list); |
c3929c36 | 2264 | device->fs_devices->rw_devices++; |
34441361 | 2265 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2266 | } |
24fc572f | 2267 | goto out; |
a061fc8d CM |
2268 | } |
2269 | ||
68a9db5f | 2270 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2271 | { |
d51908ce AJ |
2272 | struct btrfs_fs_devices *fs_devices; |
2273 | ||
68a9db5f | 2274 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2275 | |
25e8e911 AJ |
2276 | /* |
2277 | * in case of fs with no seed, srcdev->fs_devices will point | |
2278 | * to fs_devices of fs_info. However when the dev being replaced is | |
2279 | * a seed dev it will point to the seed's local fs_devices. In short | |
2280 | * srcdev will have its correct fs_devices in both the cases. | |
2281 | */ | |
2282 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2283 | |
e93c89c1 | 2284 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2285 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2286 | fs_devices->num_devices--; |
e6e674bd | 2287 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2288 | fs_devices->missing_devices--; |
e93c89c1 | 2289 | |
ebbede42 | 2290 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2291 | fs_devices->rw_devices--; |
1357272f | 2292 | |
82372bc8 | 2293 | if (srcdev->bdev) |
d51908ce | 2294 | fs_devices->open_devices--; |
084b6e7c QW |
2295 | } |
2296 | ||
2297 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2298 | struct btrfs_device *srcdev) | |
2299 | { | |
2300 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2301 | |
ebbede42 | 2302 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2303 | /* zero out the old super if it is writable */ |
2304 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2305 | } | |
14238819 AJ |
2306 | |
2307 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2308 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2309 | |
94d5f0c2 AJ |
2310 | /* if this is no devs we rather delete the fs_devices */ |
2311 | if (!fs_devices->num_devices) { | |
2312 | struct btrfs_fs_devices *tmp_fs_devices; | |
2313 | ||
6dd38f81 AJ |
2314 | /* |
2315 | * On a mounted FS, num_devices can't be zero unless it's a | |
2316 | * seed. In case of a seed device being replaced, the replace | |
2317 | * target added to the sprout FS, so there will be no more | |
2318 | * device left under the seed FS. | |
2319 | */ | |
2320 | ASSERT(fs_devices->seeding); | |
2321 | ||
94d5f0c2 AJ |
2322 | tmp_fs_devices = fs_info->fs_devices; |
2323 | while (tmp_fs_devices) { | |
2324 | if (tmp_fs_devices->seed == fs_devices) { | |
2325 | tmp_fs_devices->seed = fs_devices->seed; | |
2326 | break; | |
2327 | } | |
2328 | tmp_fs_devices = tmp_fs_devices->seed; | |
2329 | } | |
2330 | fs_devices->seed = NULL; | |
0226e0eb | 2331 | close_fs_devices(fs_devices); |
8bef8401 | 2332 | free_fs_devices(fs_devices); |
94d5f0c2 | 2333 | } |
e93c89c1 SB |
2334 | } |
2335 | ||
4f5ad7bd | 2336 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2337 | { |
4f5ad7bd | 2338 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2339 | |
e93c89c1 | 2340 | WARN_ON(!tgtdev); |
d9a071f0 | 2341 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2342 | |
d9a071f0 | 2343 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2344 | |
779bf3fe | 2345 | if (tgtdev->bdev) |
d9a071f0 | 2346 | fs_devices->open_devices--; |
779bf3fe | 2347 | |
d9a071f0 | 2348 | fs_devices->num_devices--; |
e93c89c1 | 2349 | |
d6507cf1 | 2350 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2351 | |
e93c89c1 | 2352 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2353 | |
d9a071f0 | 2354 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2355 | |
2356 | /* | |
2357 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2358 | * may lead to a call to btrfs_show_devname() which will try | |
2359 | * to hold device_list_mutex. And here this device | |
2360 | * is already out of device list, so we don't have to hold | |
2361 | * the device_list_mutex lock. | |
2362 | */ | |
2363 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2364 | |
2365 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2366 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2367 | } |
2368 | ||
b444ad46 NB |
2369 | static struct btrfs_device *btrfs_find_device_by_path( |
2370 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2371 | { |
2372 | int ret = 0; | |
2373 | struct btrfs_super_block *disk_super; | |
2374 | u64 devid; | |
2375 | u8 *dev_uuid; | |
2376 | struct block_device *bdev; | |
2377 | struct buffer_head *bh; | |
b444ad46 | 2378 | struct btrfs_device *device; |
7ba15b7d | 2379 | |
7ba15b7d | 2380 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2381 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2382 | if (ret) |
b444ad46 | 2383 | return ERR_PTR(ret); |
7ba15b7d SB |
2384 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2385 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2386 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b NB |
2387 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
2388 | device = btrfs_find_device(fs_info, devid, dev_uuid, | |
2389 | disk_super->metadata_uuid); | |
2390 | else | |
2391 | device = btrfs_find_device(fs_info, devid, | |
2392 | dev_uuid, disk_super->fsid); | |
2393 | ||
7ba15b7d | 2394 | brelse(bh); |
b444ad46 NB |
2395 | if (!device) |
2396 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2397 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2398 | return device; |
7ba15b7d SB |
2399 | } |
2400 | ||
6c050407 NB |
2401 | static struct btrfs_device *btrfs_find_device_missing_or_by_path( |
2402 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d | 2403 | { |
6c050407 | 2404 | struct btrfs_device *device = NULL; |
7ba15b7d SB |
2405 | if (strcmp(device_path, "missing") == 0) { |
2406 | struct list_head *devices; | |
2407 | struct btrfs_device *tmp; | |
2408 | ||
0b246afa | 2409 | devices = &fs_info->fs_devices->devices; |
7ba15b7d | 2410 | list_for_each_entry(tmp, devices, dev_list) { |
e12c9621 AJ |
2411 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
2412 | &tmp->dev_state) && !tmp->bdev) { | |
6c050407 | 2413 | device = tmp; |
7ba15b7d SB |
2414 | break; |
2415 | } | |
2416 | } | |
2417 | ||
6c050407 NB |
2418 | if (!device) |
2419 | return ERR_PTR(-ENOENT); | |
7ba15b7d | 2420 | } else { |
6c050407 | 2421 | device = btrfs_find_device_by_path(fs_info, device_path); |
7ba15b7d | 2422 | } |
b444ad46 | 2423 | |
6c050407 | 2424 | return device; |
7ba15b7d SB |
2425 | } |
2426 | ||
5c5c0df0 DS |
2427 | /* |
2428 | * Lookup a device given by device id, or the path if the id is 0. | |
2429 | */ | |
a27a94c2 NB |
2430 | struct btrfs_device *btrfs_find_device_by_devspec( |
2431 | struct btrfs_fs_info *fs_info, u64 devid, const char *devpath) | |
24e0474b | 2432 | { |
a27a94c2 | 2433 | struct btrfs_device *device; |
24e0474b | 2434 | |
5c5c0df0 | 2435 | if (devid) { |
a27a94c2 NB |
2436 | device = btrfs_find_device(fs_info, devid, NULL, NULL); |
2437 | if (!device) | |
2438 | return ERR_PTR(-ENOENT); | |
24e0474b | 2439 | } else { |
5c5c0df0 | 2440 | if (!devpath || !devpath[0]) |
a27a94c2 NB |
2441 | return ERR_PTR(-EINVAL); |
2442 | device = btrfs_find_device_missing_or_by_path(fs_info, devpath); | |
24e0474b | 2443 | } |
a27a94c2 | 2444 | return device; |
24e0474b AJ |
2445 | } |
2446 | ||
2b82032c YZ |
2447 | /* |
2448 | * does all the dirty work required for changing file system's UUID. | |
2449 | */ | |
2ff7e61e | 2450 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2451 | { |
0b246afa | 2452 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2453 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2454 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2455 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2456 | struct btrfs_device *device; |
2457 | u64 super_flags; | |
2458 | ||
a32bf9a3 | 2459 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2460 | if (!fs_devices->seeding) |
2b82032c YZ |
2461 | return -EINVAL; |
2462 | ||
7239ff4b | 2463 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2464 | if (IS_ERR(seed_devices)) |
2465 | return PTR_ERR(seed_devices); | |
2b82032c | 2466 | |
e4404d6e YZ |
2467 | old_devices = clone_fs_devices(fs_devices); |
2468 | if (IS_ERR(old_devices)) { | |
2469 | kfree(seed_devices); | |
2470 | return PTR_ERR(old_devices); | |
2b82032c | 2471 | } |
e4404d6e | 2472 | |
c4babc5e | 2473 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2474 | |
e4404d6e YZ |
2475 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2476 | seed_devices->opened = 1; | |
2477 | INIT_LIST_HEAD(&seed_devices->devices); | |
2478 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2479 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2480 | |
321a4bf7 | 2481 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2482 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2483 | synchronize_rcu); | |
2196d6e8 MX |
2484 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2485 | device->fs_devices = seed_devices; | |
c9513edb | 2486 | |
34441361 | 2487 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2488 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2489 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2490 | |
2b82032c YZ |
2491 | fs_devices->seeding = 0; |
2492 | fs_devices->num_devices = 0; | |
2493 | fs_devices->open_devices = 0; | |
69611ac8 | 2494 | fs_devices->missing_devices = 0; |
69611ac8 | 2495 | fs_devices->rotating = 0; |
e4404d6e | 2496 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2497 | |
2498 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2499 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2500 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2501 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2502 | |
2b82032c YZ |
2503 | super_flags = btrfs_super_flags(disk_super) & |
2504 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2505 | btrfs_set_super_flags(disk_super, super_flags); | |
2506 | ||
2507 | return 0; | |
2508 | } | |
2509 | ||
2510 | /* | |
01327610 | 2511 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2512 | */ |
2513 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2514 | struct btrfs_fs_info *fs_info) |
2b82032c | 2515 | { |
5b4aacef | 2516 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2517 | struct btrfs_path *path; |
2518 | struct extent_buffer *leaf; | |
2519 | struct btrfs_dev_item *dev_item; | |
2520 | struct btrfs_device *device; | |
2521 | struct btrfs_key key; | |
44880fdc | 2522 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2523 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2524 | u64 devid; | |
2525 | int ret; | |
2526 | ||
2527 | path = btrfs_alloc_path(); | |
2528 | if (!path) | |
2529 | return -ENOMEM; | |
2530 | ||
2b82032c YZ |
2531 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2532 | key.offset = 0; | |
2533 | key.type = BTRFS_DEV_ITEM_KEY; | |
2534 | ||
2535 | while (1) { | |
2536 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2537 | if (ret < 0) | |
2538 | goto error; | |
2539 | ||
2540 | leaf = path->nodes[0]; | |
2541 | next_slot: | |
2542 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2543 | ret = btrfs_next_leaf(root, path); | |
2544 | if (ret > 0) | |
2545 | break; | |
2546 | if (ret < 0) | |
2547 | goto error; | |
2548 | leaf = path->nodes[0]; | |
2549 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2550 | btrfs_release_path(path); |
2b82032c YZ |
2551 | continue; |
2552 | } | |
2553 | ||
2554 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2555 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2556 | key.type != BTRFS_DEV_ITEM_KEY) | |
2557 | break; | |
2558 | ||
2559 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2560 | struct btrfs_dev_item); | |
2561 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2562 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2563 | BTRFS_UUID_SIZE); |
1473b24e | 2564 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2565 | BTRFS_FSID_SIZE); |
0b246afa | 2566 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2567 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2568 | |
2569 | if (device->fs_devices->seeding) { | |
2570 | btrfs_set_device_generation(leaf, dev_item, | |
2571 | device->generation); | |
2572 | btrfs_mark_buffer_dirty(leaf); | |
2573 | } | |
2574 | ||
2575 | path->slots[0]++; | |
2576 | goto next_slot; | |
2577 | } | |
2578 | ret = 0; | |
2579 | error: | |
2580 | btrfs_free_path(path); | |
2581 | return ret; | |
2582 | } | |
2583 | ||
da353f6b | 2584 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2585 | { |
5112febb | 2586 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2587 | struct request_queue *q; |
788f20eb CM |
2588 | struct btrfs_trans_handle *trans; |
2589 | struct btrfs_device *device; | |
2590 | struct block_device *bdev; | |
0b246afa | 2591 | struct super_block *sb = fs_info->sb; |
606686ee | 2592 | struct rcu_string *name; |
5da54bc1 | 2593 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2594 | u64 orig_super_total_bytes; |
2595 | u64 orig_super_num_devices; | |
2b82032c | 2596 | int seeding_dev = 0; |
788f20eb | 2597 | int ret = 0; |
7132a262 | 2598 | bool unlocked = false; |
788f20eb | 2599 | |
5da54bc1 | 2600 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2601 | return -EROFS; |
788f20eb | 2602 | |
a5d16333 | 2603 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2604 | fs_info->bdev_holder); |
7f59203a JB |
2605 | if (IS_ERR(bdev)) |
2606 | return PTR_ERR(bdev); | |
a2135011 | 2607 | |
5da54bc1 | 2608 | if (fs_devices->seeding) { |
2b82032c YZ |
2609 | seeding_dev = 1; |
2610 | down_write(&sb->s_umount); | |
2611 | mutex_lock(&uuid_mutex); | |
2612 | } | |
2613 | ||
8c8bee1d | 2614 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2615 | |
5da54bc1 | 2616 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2617 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2618 | if (device->bdev == bdev) { |
2619 | ret = -EEXIST; | |
d25628bd | 2620 | mutex_unlock( |
5da54bc1 | 2621 | &fs_devices->device_list_mutex); |
2b82032c | 2622 | goto error; |
788f20eb CM |
2623 | } |
2624 | } | |
5da54bc1 | 2625 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2626 | |
0b246afa | 2627 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2628 | if (IS_ERR(device)) { |
788f20eb | 2629 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2630 | ret = PTR_ERR(device); |
2b82032c | 2631 | goto error; |
788f20eb CM |
2632 | } |
2633 | ||
78f2c9e6 | 2634 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2635 | if (!name) { |
2b82032c | 2636 | ret = -ENOMEM; |
5c4cf6c9 | 2637 | goto error_free_device; |
788f20eb | 2638 | } |
606686ee | 2639 | rcu_assign_pointer(device->name, name); |
2b82032c | 2640 | |
a22285a6 | 2641 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2642 | if (IS_ERR(trans)) { |
98d5dc13 | 2643 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2644 | goto error_free_device; |
98d5dc13 TI |
2645 | } |
2646 | ||
d5e2003c | 2647 | q = bdev_get_queue(bdev); |
ebbede42 | 2648 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2649 | device->generation = trans->transid; |
0b246afa JM |
2650 | device->io_width = fs_info->sectorsize; |
2651 | device->io_align = fs_info->sectorsize; | |
2652 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2653 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2654 | fs_info->sectorsize); | |
2cc3c559 | 2655 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2656 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2657 | device->fs_info = fs_info; |
788f20eb | 2658 | device->bdev = bdev; |
e12c9621 | 2659 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2660 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2661 | device->mode = FMODE_EXCL; |
27087f37 | 2662 | device->dev_stats_valid = 1; |
9f6d2510 | 2663 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2664 | |
2b82032c | 2665 | if (seeding_dev) { |
1751e8a6 | 2666 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2667 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2668 | if (ret) { |
2669 | btrfs_abort_transaction(trans, ret); | |
2670 | goto error_trans; | |
2671 | } | |
2b82032c | 2672 | } |
788f20eb | 2673 | |
5da54bc1 | 2674 | device->fs_devices = fs_devices; |
e5e9a520 | 2675 | |
5da54bc1 | 2676 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2677 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2678 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2679 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2680 | fs_devices->num_devices++; | |
2681 | fs_devices->open_devices++; | |
2682 | fs_devices->rw_devices++; | |
2683 | fs_devices->total_devices++; | |
2684 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2685 | |
a5ed45f8 | 2686 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2687 | |
e884f4f0 | 2688 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2689 | fs_devices->rotating = 1; |
c289811c | 2690 | |
39379faa | 2691 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2692 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2693 | round_down(orig_super_total_bytes + device->total_bytes, |
2694 | fs_info->sectorsize)); | |
788f20eb | 2695 | |
39379faa NA |
2696 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2697 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2698 | orig_super_num_devices + 1); | |
0d39376a AJ |
2699 | |
2700 | /* add sysfs device entry */ | |
5da54bc1 | 2701 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2702 | |
2196d6e8 MX |
2703 | /* |
2704 | * we've got more storage, clear any full flags on the space | |
2705 | * infos | |
2706 | */ | |
0b246afa | 2707 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2708 | |
34441361 | 2709 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2710 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2711 | |
2b82032c | 2712 | if (seeding_dev) { |
34441361 | 2713 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2714 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2715 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2716 | if (ret) { |
66642832 | 2717 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2718 | goto error_sysfs; |
005d6427 | 2719 | } |
2196d6e8 MX |
2720 | } |
2721 | ||
8e87e856 | 2722 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2723 | if (ret) { |
66642832 | 2724 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2725 | goto error_sysfs; |
2196d6e8 MX |
2726 | } |
2727 | ||
2728 | if (seeding_dev) { | |
2729 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2730 | ||
0b246afa | 2731 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2732 | if (ret) { |
66642832 | 2733 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2734 | goto error_sysfs; |
005d6427 | 2735 | } |
b2373f25 AJ |
2736 | |
2737 | /* Sprouting would change fsid of the mounted root, | |
2738 | * so rename the fsid on the sysfs | |
2739 | */ | |
2740 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
de37aa51 | 2741 | fs_info->fs_devices->fsid); |
5da54bc1 | 2742 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2743 | btrfs_warn(fs_info, |
2744 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2745 | } |
2746 | ||
3a45bb20 | 2747 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2748 | |
2b82032c YZ |
2749 | if (seeding_dev) { |
2750 | mutex_unlock(&uuid_mutex); | |
2751 | up_write(&sb->s_umount); | |
7132a262 | 2752 | unlocked = true; |
788f20eb | 2753 | |
79787eaa JM |
2754 | if (ret) /* transaction commit */ |
2755 | return ret; | |
2756 | ||
2ff7e61e | 2757 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2758 | if (ret < 0) |
0b246afa | 2759 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2760 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2761 | trans = btrfs_attach_transaction(root); |
2762 | if (IS_ERR(trans)) { | |
2763 | if (PTR_ERR(trans) == -ENOENT) | |
2764 | return 0; | |
7132a262 AJ |
2765 | ret = PTR_ERR(trans); |
2766 | trans = NULL; | |
2767 | goto error_sysfs; | |
671415b7 | 2768 | } |
3a45bb20 | 2769 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2770 | } |
c9e9f97b | 2771 | |
5a1972bd QW |
2772 | /* Update ctime/mtime for libblkid */ |
2773 | update_dev_time(device_path); | |
2b82032c | 2774 | return ret; |
79787eaa | 2775 | |
d31c32f6 | 2776 | error_sysfs: |
5da54bc1 | 2777 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2778 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2779 | mutex_lock(&fs_info->chunk_mutex); | |
2780 | list_del_rcu(&device->dev_list); | |
2781 | list_del(&device->dev_alloc_list); | |
2782 | fs_info->fs_devices->num_devices--; | |
2783 | fs_info->fs_devices->open_devices--; | |
2784 | fs_info->fs_devices->rw_devices--; | |
2785 | fs_info->fs_devices->total_devices--; | |
2786 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2787 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2788 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2789 | orig_super_total_bytes); | |
2790 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2791 | orig_super_num_devices); | |
2792 | mutex_unlock(&fs_info->chunk_mutex); | |
2793 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2794 | error_trans: |
0af2c4bf | 2795 | if (seeding_dev) |
1751e8a6 | 2796 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2797 | if (trans) |
2798 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2799 | error_free_device: |
a425f9d4 | 2800 | btrfs_free_device(device); |
2b82032c | 2801 | error: |
e525fd89 | 2802 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2803 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2804 | mutex_unlock(&uuid_mutex); |
2805 | up_write(&sb->s_umount); | |
2806 | } | |
c9e9f97b | 2807 | return ret; |
788f20eb CM |
2808 | } |
2809 | ||
d397712b CM |
2810 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2811 | struct btrfs_device *device) | |
0b86a832 CM |
2812 | { |
2813 | int ret; | |
2814 | struct btrfs_path *path; | |
0b246afa | 2815 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2816 | struct btrfs_dev_item *dev_item; |
2817 | struct extent_buffer *leaf; | |
2818 | struct btrfs_key key; | |
2819 | ||
0b86a832 CM |
2820 | path = btrfs_alloc_path(); |
2821 | if (!path) | |
2822 | return -ENOMEM; | |
2823 | ||
2824 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2825 | key.type = BTRFS_DEV_ITEM_KEY; | |
2826 | key.offset = device->devid; | |
2827 | ||
2828 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2829 | if (ret < 0) | |
2830 | goto out; | |
2831 | ||
2832 | if (ret > 0) { | |
2833 | ret = -ENOENT; | |
2834 | goto out; | |
2835 | } | |
2836 | ||
2837 | leaf = path->nodes[0]; | |
2838 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2839 | ||
2840 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2841 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2842 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2843 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2844 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2845 | btrfs_set_device_total_bytes(leaf, dev_item, |
2846 | btrfs_device_get_disk_total_bytes(device)); | |
2847 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2848 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2849 | btrfs_mark_buffer_dirty(leaf); |
2850 | ||
2851 | out: | |
2852 | btrfs_free_path(path); | |
2853 | return ret; | |
2854 | } | |
2855 | ||
2196d6e8 | 2856 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2857 | struct btrfs_device *device, u64 new_size) |
2858 | { | |
0b246afa JM |
2859 | struct btrfs_fs_info *fs_info = device->fs_info; |
2860 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2861 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2862 | u64 old_total; |
2863 | u64 diff; | |
8f18cf13 | 2864 | |
ebbede42 | 2865 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2866 | return -EACCES; |
2196d6e8 | 2867 | |
7dfb8be1 NB |
2868 | new_size = round_down(new_size, fs_info->sectorsize); |
2869 | ||
34441361 | 2870 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2871 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2872 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2873 | |
63a212ab | 2874 | if (new_size <= device->total_bytes || |
401e29c1 | 2875 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2876 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2877 | return -EINVAL; |
2196d6e8 | 2878 | } |
2b82032c | 2879 | |
0b246afa | 2880 | fs_devices = fs_info->fs_devices; |
2b82032c | 2881 | |
7dfb8be1 NB |
2882 | btrfs_set_super_total_bytes(super_copy, |
2883 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2884 | device->fs_devices->total_rw_bytes += diff; |
2885 | ||
7cc8e58d MX |
2886 | btrfs_device_set_total_bytes(device, new_size); |
2887 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2888 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2889 | if (list_empty(&device->resized_list)) |
2890 | list_add_tail(&device->resized_list, | |
2891 | &fs_devices->resized_devices); | |
34441361 | 2892 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2893 | |
8f18cf13 CM |
2894 | return btrfs_update_device(trans, device); |
2895 | } | |
2896 | ||
f4208794 | 2897 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2898 | { |
f4208794 | 2899 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2900 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2901 | int ret; |
2902 | struct btrfs_path *path; | |
2903 | struct btrfs_key key; | |
2904 | ||
8f18cf13 CM |
2905 | path = btrfs_alloc_path(); |
2906 | if (!path) | |
2907 | return -ENOMEM; | |
2908 | ||
408fbf19 | 2909 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2910 | key.offset = chunk_offset; |
2911 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2912 | ||
2913 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2914 | if (ret < 0) |
2915 | goto out; | |
2916 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2917 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2918 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2919 | ret = -ENOENT; |
2920 | goto out; | |
2921 | } | |
8f18cf13 CM |
2922 | |
2923 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2924 | if (ret < 0) |
0b246afa JM |
2925 | btrfs_handle_fs_error(fs_info, ret, |
2926 | "Failed to delete chunk item."); | |
79787eaa | 2927 | out: |
8f18cf13 | 2928 | btrfs_free_path(path); |
65a246c5 | 2929 | return ret; |
8f18cf13 CM |
2930 | } |
2931 | ||
408fbf19 | 2932 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2933 | { |
0b246afa | 2934 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2935 | struct btrfs_disk_key *disk_key; |
2936 | struct btrfs_chunk *chunk; | |
2937 | u8 *ptr; | |
2938 | int ret = 0; | |
2939 | u32 num_stripes; | |
2940 | u32 array_size; | |
2941 | u32 len = 0; | |
2942 | u32 cur; | |
2943 | struct btrfs_key key; | |
2944 | ||
34441361 | 2945 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2946 | array_size = btrfs_super_sys_array_size(super_copy); |
2947 | ||
2948 | ptr = super_copy->sys_chunk_array; | |
2949 | cur = 0; | |
2950 | ||
2951 | while (cur < array_size) { | |
2952 | disk_key = (struct btrfs_disk_key *)ptr; | |
2953 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2954 | ||
2955 | len = sizeof(*disk_key); | |
2956 | ||
2957 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2958 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2959 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2960 | len += btrfs_chunk_item_size(num_stripes); | |
2961 | } else { | |
2962 | ret = -EIO; | |
2963 | break; | |
2964 | } | |
408fbf19 | 2965 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2966 | key.offset == chunk_offset) { |
2967 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2968 | array_size -= len; | |
2969 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2970 | } else { | |
2971 | ptr += len; | |
2972 | cur += len; | |
2973 | } | |
2974 | } | |
34441361 | 2975 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2976 | return ret; |
2977 | } | |
2978 | ||
60ca842e OS |
2979 | /* |
2980 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2981 | * @logical: Logical block offset in bytes. | |
2982 | * @length: Length of extent in bytes. | |
2983 | * | |
2984 | * Return: Chunk mapping or ERR_PTR. | |
2985 | */ | |
2986 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2987 | u64 logical, u64 length) | |
592d92ee LB |
2988 | { |
2989 | struct extent_map_tree *em_tree; | |
2990 | struct extent_map *em; | |
2991 | ||
2992 | em_tree = &fs_info->mapping_tree.map_tree; | |
2993 | read_lock(&em_tree->lock); | |
2994 | em = lookup_extent_mapping(em_tree, logical, length); | |
2995 | read_unlock(&em_tree->lock); | |
2996 | ||
2997 | if (!em) { | |
2998 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2999 | logical, length); | |
3000 | return ERR_PTR(-EINVAL); | |
3001 | } | |
3002 | ||
3003 | if (em->start > logical || em->start + em->len < logical) { | |
3004 | btrfs_crit(fs_info, | |
3005 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
3006 | logical, length, em->start, em->start + em->len); | |
3007 | free_extent_map(em); | |
3008 | return ERR_PTR(-EINVAL); | |
3009 | } | |
3010 | ||
3011 | /* callers are responsible for dropping em's ref. */ | |
3012 | return em; | |
3013 | } | |
3014 | ||
97aff912 | 3015 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3016 | { |
97aff912 | 3017 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3018 | struct extent_map *em; |
3019 | struct map_lookup *map; | |
2196d6e8 | 3020 | u64 dev_extent_len = 0; |
47ab2a6c | 3021 | int i, ret = 0; |
0b246afa | 3022 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3023 | |
60ca842e | 3024 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3025 | if (IS_ERR(em)) { |
47ab2a6c JB |
3026 | /* |
3027 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3028 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3029 | * do anything we still error out. |
3030 | */ | |
3031 | ASSERT(0); | |
592d92ee | 3032 | return PTR_ERR(em); |
47ab2a6c | 3033 | } |
95617d69 | 3034 | map = em->map_lookup; |
34441361 | 3035 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 3036 | check_system_chunk(trans, map->type); |
34441361 | 3037 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 3038 | |
57ba4cb8 FM |
3039 | /* |
3040 | * Take the device list mutex to prevent races with the final phase of | |
3041 | * a device replace operation that replaces the device object associated | |
3042 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
3043 | */ | |
3044 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3045 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3046 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3047 | ret = btrfs_free_dev_extent(trans, device, |
3048 | map->stripes[i].physical, | |
3049 | &dev_extent_len); | |
47ab2a6c | 3050 | if (ret) { |
57ba4cb8 | 3051 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3052 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3053 | goto out; |
3054 | } | |
a061fc8d | 3055 | |
2196d6e8 | 3056 | if (device->bytes_used > 0) { |
34441361 | 3057 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3058 | btrfs_device_set_bytes_used(device, |
3059 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3060 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3061 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3062 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3063 | } |
a061fc8d | 3064 | |
64bc6c2a NB |
3065 | ret = btrfs_update_device(trans, device); |
3066 | if (ret) { | |
3067 | mutex_unlock(&fs_devices->device_list_mutex); | |
3068 | btrfs_abort_transaction(trans, ret); | |
3069 | goto out; | |
dfe25020 | 3070 | } |
8f18cf13 | 3071 | } |
57ba4cb8 FM |
3072 | mutex_unlock(&fs_devices->device_list_mutex); |
3073 | ||
f4208794 | 3074 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3075 | if (ret) { |
66642832 | 3076 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3077 | goto out; |
3078 | } | |
8f18cf13 | 3079 | |
6bccf3ab | 3080 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3081 | |
8f18cf13 | 3082 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3083 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3084 | if (ret) { |
66642832 | 3085 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3086 | goto out; |
3087 | } | |
8f18cf13 CM |
3088 | } |
3089 | ||
5a98ec01 | 3090 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3091 | if (ret) { |
66642832 | 3092 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3093 | goto out; |
3094 | } | |
2b82032c | 3095 | |
47ab2a6c | 3096 | out: |
2b82032c YZ |
3097 | /* once for us */ |
3098 | free_extent_map(em); | |
47ab2a6c JB |
3099 | return ret; |
3100 | } | |
2b82032c | 3101 | |
5b4aacef | 3102 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3103 | { |
5b4aacef | 3104 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3105 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3106 | int ret; |
2b82032c | 3107 | |
67c5e7d4 FM |
3108 | /* |
3109 | * Prevent races with automatic removal of unused block groups. | |
3110 | * After we relocate and before we remove the chunk with offset | |
3111 | * chunk_offset, automatic removal of the block group can kick in, | |
3112 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3113 | * | |
3114 | * Make sure to acquire this mutex before doing a tree search (dev | |
3115 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3116 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3117 | * we release the path used to search the chunk/dev tree and before | |
3118 | * the current task acquires this mutex and calls us. | |
3119 | */ | |
a32bf9a3 | 3120 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3121 | |
0b246afa | 3122 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
3123 | if (ret) |
3124 | return -ENOSPC; | |
3125 | ||
3126 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3127 | btrfs_scrub_pause(fs_info); |
0b246afa | 3128 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3129 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3130 | if (ret) |
3131 | return ret; | |
3132 | ||
75cb379d JM |
3133 | /* |
3134 | * We add the kobjects here (and after forcing data chunk creation) | |
3135 | * since relocation is the only place we'll create chunks of a new | |
3136 | * type at runtime. The only place where we'll remove the last | |
3137 | * chunk of a type is the call immediately below this one. Even | |
3138 | * so, we're protected against races with the cleaner thread since | |
3139 | * we're covered by the delete_unused_bgs_mutex. | |
3140 | */ | |
3141 | btrfs_add_raid_kobjects(fs_info); | |
3142 | ||
19c4d2f9 CM |
3143 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3144 | chunk_offset); | |
3145 | if (IS_ERR(trans)) { | |
3146 | ret = PTR_ERR(trans); | |
3147 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3148 | return ret; | |
3149 | } | |
3150 | ||
47ab2a6c | 3151 | /* |
19c4d2f9 CM |
3152 | * step two, delete the device extents and the |
3153 | * chunk tree entries | |
47ab2a6c | 3154 | */ |
97aff912 | 3155 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3156 | btrfs_end_transaction(trans); |
19c4d2f9 | 3157 | return ret; |
2b82032c YZ |
3158 | } |
3159 | ||
2ff7e61e | 3160 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3161 | { |
0b246afa | 3162 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3163 | struct btrfs_path *path; |
3164 | struct extent_buffer *leaf; | |
3165 | struct btrfs_chunk *chunk; | |
3166 | struct btrfs_key key; | |
3167 | struct btrfs_key found_key; | |
2b82032c | 3168 | u64 chunk_type; |
ba1bf481 JB |
3169 | bool retried = false; |
3170 | int failed = 0; | |
2b82032c YZ |
3171 | int ret; |
3172 | ||
3173 | path = btrfs_alloc_path(); | |
3174 | if (!path) | |
3175 | return -ENOMEM; | |
3176 | ||
ba1bf481 | 3177 | again: |
2b82032c YZ |
3178 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3179 | key.offset = (u64)-1; | |
3180 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3181 | ||
3182 | while (1) { | |
0b246afa | 3183 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3184 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3185 | if (ret < 0) { |
0b246afa | 3186 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3187 | goto error; |
67c5e7d4 | 3188 | } |
79787eaa | 3189 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3190 | |
3191 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3192 | key.type); | |
67c5e7d4 | 3193 | if (ret) |
0b246afa | 3194 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3195 | if (ret < 0) |
3196 | goto error; | |
3197 | if (ret > 0) | |
3198 | break; | |
1a40e23b | 3199 | |
2b82032c YZ |
3200 | leaf = path->nodes[0]; |
3201 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3202 | |
2b82032c YZ |
3203 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3204 | struct btrfs_chunk); | |
3205 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3206 | btrfs_release_path(path); |
8f18cf13 | 3207 | |
2b82032c | 3208 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3209 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3210 | if (ret == -ENOSPC) |
3211 | failed++; | |
14586651 HS |
3212 | else |
3213 | BUG_ON(ret); | |
2b82032c | 3214 | } |
0b246afa | 3215 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3216 | |
2b82032c YZ |
3217 | if (found_key.offset == 0) |
3218 | break; | |
3219 | key.offset = found_key.offset - 1; | |
3220 | } | |
3221 | ret = 0; | |
ba1bf481 JB |
3222 | if (failed && !retried) { |
3223 | failed = 0; | |
3224 | retried = true; | |
3225 | goto again; | |
fae7f21c | 3226 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3227 | ret = -ENOSPC; |
3228 | } | |
2b82032c YZ |
3229 | error: |
3230 | btrfs_free_path(path); | |
3231 | return ret; | |
8f18cf13 CM |
3232 | } |
3233 | ||
a6f93c71 LB |
3234 | /* |
3235 | * return 1 : allocate a data chunk successfully, | |
3236 | * return <0: errors during allocating a data chunk, | |
3237 | * return 0 : no need to allocate a data chunk. | |
3238 | */ | |
3239 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3240 | u64 chunk_offset) | |
3241 | { | |
3242 | struct btrfs_block_group_cache *cache; | |
3243 | u64 bytes_used; | |
3244 | u64 chunk_type; | |
3245 | ||
3246 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3247 | ASSERT(cache); | |
3248 | chunk_type = cache->flags; | |
3249 | btrfs_put_block_group(cache); | |
3250 | ||
3251 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3252 | spin_lock(&fs_info->data_sinfo->lock); | |
3253 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3254 | spin_unlock(&fs_info->data_sinfo->lock); | |
3255 | ||
3256 | if (!bytes_used) { | |
3257 | struct btrfs_trans_handle *trans; | |
3258 | int ret; | |
3259 | ||
3260 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3261 | if (IS_ERR(trans)) | |
3262 | return PTR_ERR(trans); | |
3263 | ||
43a7e99d | 3264 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3265 | BTRFS_BLOCK_GROUP_DATA); |
3266 | btrfs_end_transaction(trans); | |
3267 | if (ret < 0) | |
3268 | return ret; | |
3269 | ||
75cb379d JM |
3270 | btrfs_add_raid_kobjects(fs_info); |
3271 | ||
a6f93c71 LB |
3272 | return 1; |
3273 | } | |
3274 | } | |
3275 | return 0; | |
3276 | } | |
3277 | ||
6bccf3ab | 3278 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3279 | struct btrfs_balance_control *bctl) |
3280 | { | |
6bccf3ab | 3281 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3282 | struct btrfs_trans_handle *trans; |
3283 | struct btrfs_balance_item *item; | |
3284 | struct btrfs_disk_balance_args disk_bargs; | |
3285 | struct btrfs_path *path; | |
3286 | struct extent_buffer *leaf; | |
3287 | struct btrfs_key key; | |
3288 | int ret, err; | |
3289 | ||
3290 | path = btrfs_alloc_path(); | |
3291 | if (!path) | |
3292 | return -ENOMEM; | |
3293 | ||
3294 | trans = btrfs_start_transaction(root, 0); | |
3295 | if (IS_ERR(trans)) { | |
3296 | btrfs_free_path(path); | |
3297 | return PTR_ERR(trans); | |
3298 | } | |
3299 | ||
3300 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3301 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3302 | key.offset = 0; |
3303 | ||
3304 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3305 | sizeof(*item)); | |
3306 | if (ret) | |
3307 | goto out; | |
3308 | ||
3309 | leaf = path->nodes[0]; | |
3310 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3311 | ||
b159fa28 | 3312 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3313 | |
3314 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3315 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3316 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3317 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3318 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3319 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3320 | ||
3321 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3322 | ||
3323 | btrfs_mark_buffer_dirty(leaf); | |
3324 | out: | |
3325 | btrfs_free_path(path); | |
3a45bb20 | 3326 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3327 | if (err && !ret) |
3328 | ret = err; | |
3329 | return ret; | |
3330 | } | |
3331 | ||
6bccf3ab | 3332 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3333 | { |
6bccf3ab | 3334 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3335 | struct btrfs_trans_handle *trans; |
3336 | struct btrfs_path *path; | |
3337 | struct btrfs_key key; | |
3338 | int ret, err; | |
3339 | ||
3340 | path = btrfs_alloc_path(); | |
3341 | if (!path) | |
3342 | return -ENOMEM; | |
3343 | ||
3344 | trans = btrfs_start_transaction(root, 0); | |
3345 | if (IS_ERR(trans)) { | |
3346 | btrfs_free_path(path); | |
3347 | return PTR_ERR(trans); | |
3348 | } | |
3349 | ||
3350 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3351 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3352 | key.offset = 0; |
3353 | ||
3354 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3355 | if (ret < 0) | |
3356 | goto out; | |
3357 | if (ret > 0) { | |
3358 | ret = -ENOENT; | |
3359 | goto out; | |
3360 | } | |
3361 | ||
3362 | ret = btrfs_del_item(trans, root, path); | |
3363 | out: | |
3364 | btrfs_free_path(path); | |
3a45bb20 | 3365 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3366 | if (err && !ret) |
3367 | ret = err; | |
3368 | return ret; | |
3369 | } | |
3370 | ||
59641015 ID |
3371 | /* |
3372 | * This is a heuristic used to reduce the number of chunks balanced on | |
3373 | * resume after balance was interrupted. | |
3374 | */ | |
3375 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3376 | { | |
3377 | /* | |
3378 | * Turn on soft mode for chunk types that were being converted. | |
3379 | */ | |
3380 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3381 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3382 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3383 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3384 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3385 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3386 | ||
3387 | /* | |
3388 | * Turn on usage filter if is not already used. The idea is | |
3389 | * that chunks that we have already balanced should be | |
3390 | * reasonably full. Don't do it for chunks that are being | |
3391 | * converted - that will keep us from relocating unconverted | |
3392 | * (albeit full) chunks. | |
3393 | */ | |
3394 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3395 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3396 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3397 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3398 | bctl->data.usage = 90; | |
3399 | } | |
3400 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3401 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3402 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3403 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3404 | bctl->sys.usage = 90; | |
3405 | } | |
3406 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3407 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3408 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3409 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3410 | bctl->meta.usage = 90; | |
3411 | } | |
3412 | } | |
3413 | ||
149196a2 DS |
3414 | /* |
3415 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3416 | */ | |
3417 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3418 | { |
3419 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3420 | int ret; |
c9e9f97b ID |
3421 | |
3422 | BUG_ON(!fs_info->balance_ctl); | |
3423 | ||
3424 | spin_lock(&fs_info->balance_lock); | |
3425 | fs_info->balance_ctl = NULL; | |
3426 | spin_unlock(&fs_info->balance_lock); | |
3427 | ||
3428 | kfree(bctl); | |
149196a2 DS |
3429 | ret = del_balance_item(fs_info); |
3430 | if (ret) | |
3431 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3432 | } |
3433 | ||
ed25e9b2 ID |
3434 | /* |
3435 | * Balance filters. Return 1 if chunk should be filtered out | |
3436 | * (should not be balanced). | |
3437 | */ | |
899c81ea | 3438 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3439 | struct btrfs_balance_args *bargs) |
3440 | { | |
899c81ea ID |
3441 | chunk_type = chunk_to_extended(chunk_type) & |
3442 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3443 | |
899c81ea | 3444 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3445 | return 0; |
3446 | ||
3447 | return 1; | |
3448 | } | |
3449 | ||
dba72cb3 | 3450 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3451 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3452 | { |
3453 | struct btrfs_block_group_cache *cache; | |
3454 | u64 chunk_used; | |
3455 | u64 user_thresh_min; | |
3456 | u64 user_thresh_max; | |
3457 | int ret = 1; | |
3458 | ||
3459 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3460 | chunk_used = btrfs_block_group_used(&cache->item); | |
3461 | ||
3462 | if (bargs->usage_min == 0) | |
3463 | user_thresh_min = 0; | |
3464 | else | |
3465 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3466 | bargs->usage_min); | |
3467 | ||
3468 | if (bargs->usage_max == 0) | |
3469 | user_thresh_max = 1; | |
3470 | else if (bargs->usage_max > 100) | |
3471 | user_thresh_max = cache->key.offset; | |
3472 | else | |
3473 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3474 | bargs->usage_max); | |
3475 | ||
3476 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3477 | ret = 0; | |
3478 | ||
3479 | btrfs_put_block_group(cache); | |
3480 | return ret; | |
3481 | } | |
3482 | ||
dba72cb3 | 3483 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3484 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3485 | { |
3486 | struct btrfs_block_group_cache *cache; | |
3487 | u64 chunk_used, user_thresh; | |
3488 | int ret = 1; | |
3489 | ||
3490 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3491 | chunk_used = btrfs_block_group_used(&cache->item); | |
3492 | ||
bc309467 | 3493 | if (bargs->usage_min == 0) |
3e39cea6 | 3494 | user_thresh = 1; |
a105bb88 ID |
3495 | else if (bargs->usage > 100) |
3496 | user_thresh = cache->key.offset; | |
3497 | else | |
3498 | user_thresh = div_factor_fine(cache->key.offset, | |
3499 | bargs->usage); | |
3500 | ||
5ce5b3c0 ID |
3501 | if (chunk_used < user_thresh) |
3502 | ret = 0; | |
3503 | ||
3504 | btrfs_put_block_group(cache); | |
3505 | return ret; | |
3506 | } | |
3507 | ||
409d404b ID |
3508 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3509 | struct btrfs_chunk *chunk, | |
3510 | struct btrfs_balance_args *bargs) | |
3511 | { | |
3512 | struct btrfs_stripe *stripe; | |
3513 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3514 | int i; | |
3515 | ||
3516 | for (i = 0; i < num_stripes; i++) { | |
3517 | stripe = btrfs_stripe_nr(chunk, i); | |
3518 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3519 | return 0; | |
3520 | } | |
3521 | ||
3522 | return 1; | |
3523 | } | |
3524 | ||
94e60d5a ID |
3525 | /* [pstart, pend) */ |
3526 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3527 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3528 | struct btrfs_balance_args *bargs) |
3529 | { | |
3530 | struct btrfs_stripe *stripe; | |
3531 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3532 | u64 stripe_offset; | |
3533 | u64 stripe_length; | |
3534 | int factor; | |
3535 | int i; | |
3536 | ||
3537 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3538 | return 0; | |
3539 | ||
3540 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3541 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3542 | factor = num_stripes / 2; | |
3543 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3544 | factor = num_stripes - 1; | |
3545 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3546 | factor = num_stripes - 2; | |
3547 | } else { | |
3548 | factor = num_stripes; | |
3549 | } | |
94e60d5a ID |
3550 | |
3551 | for (i = 0; i < num_stripes; i++) { | |
3552 | stripe = btrfs_stripe_nr(chunk, i); | |
3553 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3554 | continue; | |
3555 | ||
3556 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3557 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3558 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3559 | |
3560 | if (stripe_offset < bargs->pend && | |
3561 | stripe_offset + stripe_length > bargs->pstart) | |
3562 | return 0; | |
3563 | } | |
3564 | ||
3565 | return 1; | |
3566 | } | |
3567 | ||
ea67176a ID |
3568 | /* [vstart, vend) */ |
3569 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3570 | struct btrfs_chunk *chunk, | |
3571 | u64 chunk_offset, | |
3572 | struct btrfs_balance_args *bargs) | |
3573 | { | |
3574 | if (chunk_offset < bargs->vend && | |
3575 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3576 | /* at least part of the chunk is inside this vrange */ | |
3577 | return 0; | |
3578 | ||
3579 | return 1; | |
3580 | } | |
3581 | ||
dee32d0a GAP |
3582 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3583 | struct btrfs_chunk *chunk, | |
3584 | struct btrfs_balance_args *bargs) | |
3585 | { | |
3586 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3587 | ||
3588 | if (bargs->stripes_min <= num_stripes | |
3589 | && num_stripes <= bargs->stripes_max) | |
3590 | return 0; | |
3591 | ||
3592 | return 1; | |
3593 | } | |
3594 | ||
899c81ea | 3595 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3596 | struct btrfs_balance_args *bargs) |
3597 | { | |
3598 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3599 | return 0; | |
3600 | ||
899c81ea ID |
3601 | chunk_type = chunk_to_extended(chunk_type) & |
3602 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3603 | |
899c81ea | 3604 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3605 | return 1; |
3606 | ||
3607 | return 0; | |
3608 | } | |
3609 | ||
2ff7e61e | 3610 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3611 | struct extent_buffer *leaf, |
3612 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3613 | { | |
0b246afa | 3614 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3615 | struct btrfs_balance_args *bargs = NULL; |
3616 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3617 | ||
3618 | /* type filter */ | |
3619 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3620 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3621 | return 0; | |
3622 | } | |
3623 | ||
3624 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3625 | bargs = &bctl->data; | |
3626 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3627 | bargs = &bctl->sys; | |
3628 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3629 | bargs = &bctl->meta; | |
3630 | ||
ed25e9b2 ID |
3631 | /* profiles filter */ |
3632 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3633 | chunk_profiles_filter(chunk_type, bargs)) { | |
3634 | return 0; | |
5ce5b3c0 ID |
3635 | } |
3636 | ||
3637 | /* usage filter */ | |
3638 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3639 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3640 | return 0; |
bc309467 | 3641 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3642 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3643 | return 0; |
409d404b ID |
3644 | } |
3645 | ||
3646 | /* devid filter */ | |
3647 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3648 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3649 | return 0; | |
94e60d5a ID |
3650 | } |
3651 | ||
3652 | /* drange filter, makes sense only with devid filter */ | |
3653 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3654 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3655 | return 0; |
ea67176a ID |
3656 | } |
3657 | ||
3658 | /* vrange filter */ | |
3659 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3660 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3661 | return 0; | |
ed25e9b2 ID |
3662 | } |
3663 | ||
dee32d0a GAP |
3664 | /* stripes filter */ |
3665 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3666 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3667 | return 0; | |
3668 | } | |
3669 | ||
cfa4c961 ID |
3670 | /* soft profile changing mode */ |
3671 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3672 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3673 | return 0; | |
3674 | } | |
3675 | ||
7d824b6f DS |
3676 | /* |
3677 | * limited by count, must be the last filter | |
3678 | */ | |
3679 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3680 | if (bargs->limit == 0) | |
3681 | return 0; | |
3682 | else | |
3683 | bargs->limit--; | |
12907fc7 DS |
3684 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3685 | /* | |
3686 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3687 | * determined here because we do not have the global information |
12907fc7 DS |
3688 | * about the count of all chunks that satisfy the filters. |
3689 | */ | |
3690 | if (bargs->limit_max == 0) | |
3691 | return 0; | |
3692 | else | |
3693 | bargs->limit_max--; | |
7d824b6f DS |
3694 | } |
3695 | ||
f43ffb60 ID |
3696 | return 1; |
3697 | } | |
3698 | ||
c9e9f97b | 3699 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3700 | { |
19a39dce | 3701 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3702 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3703 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3704 | struct list_head *devices; | |
ec44a35c CM |
3705 | struct btrfs_device *device; |
3706 | u64 old_size; | |
3707 | u64 size_to_free; | |
12907fc7 | 3708 | u64 chunk_type; |
f43ffb60 | 3709 | struct btrfs_chunk *chunk; |
5a488b9d | 3710 | struct btrfs_path *path = NULL; |
ec44a35c | 3711 | struct btrfs_key key; |
ec44a35c | 3712 | struct btrfs_key found_key; |
c9e9f97b | 3713 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3714 | struct extent_buffer *leaf; |
3715 | int slot; | |
c9e9f97b ID |
3716 | int ret; |
3717 | int enospc_errors = 0; | |
19a39dce | 3718 | bool counting = true; |
12907fc7 | 3719 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3720 | u64 limit_data = bctl->data.limit; |
3721 | u64 limit_meta = bctl->meta.limit; | |
3722 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3723 | u32 count_data = 0; |
3724 | u32 count_meta = 0; | |
3725 | u32 count_sys = 0; | |
2c9fe835 | 3726 | int chunk_reserved = 0; |
ec44a35c | 3727 | |
ec44a35c | 3728 | /* step one make some room on all the devices */ |
c9e9f97b | 3729 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3730 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3731 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3732 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3733 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
ebbede42 | 3734 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) || |
7cc8e58d MX |
3735 | btrfs_device_get_total_bytes(device) - |
3736 | btrfs_device_get_bytes_used(device) > size_to_free || | |
401e29c1 | 3737 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
ec44a35c CM |
3738 | continue; |
3739 | ||
3740 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3741 | if (ret == -ENOSPC) |
3742 | break; | |
5a488b9d LB |
3743 | if (ret) { |
3744 | /* btrfs_shrink_device never returns ret > 0 */ | |
3745 | WARN_ON(ret > 0); | |
3746 | goto error; | |
3747 | } | |
ec44a35c | 3748 | |
a22285a6 | 3749 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3750 | if (IS_ERR(trans)) { |
3751 | ret = PTR_ERR(trans); | |
3752 | btrfs_info_in_rcu(fs_info, | |
3753 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3754 | rcu_str_deref(device->name), ret, | |
3755 | old_size, old_size - size_to_free); | |
3756 | goto error; | |
3757 | } | |
ec44a35c CM |
3758 | |
3759 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d | 3760 | if (ret) { |
3a45bb20 | 3761 | btrfs_end_transaction(trans); |
5a488b9d LB |
3762 | /* btrfs_grow_device never returns ret > 0 */ |
3763 | WARN_ON(ret > 0); | |
3764 | btrfs_info_in_rcu(fs_info, | |
3765 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3766 | rcu_str_deref(device->name), ret, | |
3767 | old_size, old_size - size_to_free); | |
3768 | goto error; | |
3769 | } | |
ec44a35c | 3770 | |
3a45bb20 | 3771 | btrfs_end_transaction(trans); |
ec44a35c CM |
3772 | } |
3773 | ||
3774 | /* step two, relocate all the chunks */ | |
3775 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3776 | if (!path) { |
3777 | ret = -ENOMEM; | |
3778 | goto error; | |
3779 | } | |
19a39dce ID |
3780 | |
3781 | /* zero out stat counters */ | |
3782 | spin_lock(&fs_info->balance_lock); | |
3783 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3784 | spin_unlock(&fs_info->balance_lock); | |
3785 | again: | |
7d824b6f | 3786 | if (!counting) { |
12907fc7 DS |
3787 | /* |
3788 | * The single value limit and min/max limits use the same bytes | |
3789 | * in the | |
3790 | */ | |
7d824b6f DS |
3791 | bctl->data.limit = limit_data; |
3792 | bctl->meta.limit = limit_meta; | |
3793 | bctl->sys.limit = limit_sys; | |
3794 | } | |
ec44a35c CM |
3795 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3796 | key.offset = (u64)-1; | |
3797 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3798 | ||
d397712b | 3799 | while (1) { |
19a39dce | 3800 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3801 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3802 | ret = -ECANCELED; |
3803 | goto error; | |
3804 | } | |
3805 | ||
67c5e7d4 | 3806 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3807 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3808 | if (ret < 0) { |
3809 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3810 | goto error; |
67c5e7d4 | 3811 | } |
ec44a35c CM |
3812 | |
3813 | /* | |
3814 | * this shouldn't happen, it means the last relocate | |
3815 | * failed | |
3816 | */ | |
3817 | if (ret == 0) | |
c9e9f97b | 3818 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3819 | |
3820 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3821 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3822 | if (ret) { |
67c5e7d4 | 3823 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3824 | ret = 0; |
ec44a35c | 3825 | break; |
c9e9f97b | 3826 | } |
7d9eb12c | 3827 | |
f43ffb60 ID |
3828 | leaf = path->nodes[0]; |
3829 | slot = path->slots[0]; | |
3830 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3831 | |
67c5e7d4 FM |
3832 | if (found_key.objectid != key.objectid) { |
3833 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3834 | break; |
67c5e7d4 | 3835 | } |
7d9eb12c | 3836 | |
f43ffb60 | 3837 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3838 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3839 | |
19a39dce ID |
3840 | if (!counting) { |
3841 | spin_lock(&fs_info->balance_lock); | |
3842 | bctl->stat.considered++; | |
3843 | spin_unlock(&fs_info->balance_lock); | |
3844 | } | |
3845 | ||
2ff7e61e | 3846 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3847 | found_key.offset); |
2c9fe835 | 3848 | |
b3b4aa74 | 3849 | btrfs_release_path(path); |
67c5e7d4 FM |
3850 | if (!ret) { |
3851 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3852 | goto loop; |
67c5e7d4 | 3853 | } |
f43ffb60 | 3854 | |
19a39dce | 3855 | if (counting) { |
67c5e7d4 | 3856 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3857 | spin_lock(&fs_info->balance_lock); |
3858 | bctl->stat.expected++; | |
3859 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3860 | |
3861 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3862 | count_data++; | |
3863 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3864 | count_sys++; | |
3865 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3866 | count_meta++; | |
3867 | ||
3868 | goto loop; | |
3869 | } | |
3870 | ||
3871 | /* | |
3872 | * Apply limit_min filter, no need to check if the LIMITS | |
3873 | * filter is used, limit_min is 0 by default | |
3874 | */ | |
3875 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3876 | count_data < bctl->data.limit_min) | |
3877 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3878 | count_meta < bctl->meta.limit_min) | |
3879 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3880 | count_sys < bctl->sys.limit_min)) { | |
3881 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3882 | goto loop; |
3883 | } | |
3884 | ||
a6f93c71 LB |
3885 | if (!chunk_reserved) { |
3886 | /* | |
3887 | * We may be relocating the only data chunk we have, | |
3888 | * which could potentially end up with losing data's | |
3889 | * raid profile, so lets allocate an empty one in | |
3890 | * advance. | |
3891 | */ | |
3892 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3893 | found_key.offset); | |
2c9fe835 ZL |
3894 | if (ret < 0) { |
3895 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3896 | goto error; | |
a6f93c71 LB |
3897 | } else if (ret == 1) { |
3898 | chunk_reserved = 1; | |
2c9fe835 | 3899 | } |
2c9fe835 ZL |
3900 | } |
3901 | ||
5b4aacef | 3902 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3903 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3904 | if (ret == -ENOSPC) { |
c9e9f97b | 3905 | enospc_errors++; |
eede2bf3 OS |
3906 | } else if (ret == -ETXTBSY) { |
3907 | btrfs_info(fs_info, | |
3908 | "skipping relocation of block group %llu due to active swapfile", | |
3909 | found_key.offset); | |
3910 | ret = 0; | |
3911 | } else if (ret) { | |
3912 | goto error; | |
19a39dce ID |
3913 | } else { |
3914 | spin_lock(&fs_info->balance_lock); | |
3915 | bctl->stat.completed++; | |
3916 | spin_unlock(&fs_info->balance_lock); | |
3917 | } | |
f43ffb60 | 3918 | loop: |
795a3321 ID |
3919 | if (found_key.offset == 0) |
3920 | break; | |
ba1bf481 | 3921 | key.offset = found_key.offset - 1; |
ec44a35c | 3922 | } |
c9e9f97b | 3923 | |
19a39dce ID |
3924 | if (counting) { |
3925 | btrfs_release_path(path); | |
3926 | counting = false; | |
3927 | goto again; | |
3928 | } | |
ec44a35c CM |
3929 | error: |
3930 | btrfs_free_path(path); | |
c9e9f97b | 3931 | if (enospc_errors) { |
efe120a0 | 3932 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3933 | enospc_errors); |
c9e9f97b ID |
3934 | if (!ret) |
3935 | ret = -ENOSPC; | |
3936 | } | |
3937 | ||
ec44a35c CM |
3938 | return ret; |
3939 | } | |
3940 | ||
0c460c0d ID |
3941 | /** |
3942 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3943 | * @flags: profile to validate | |
3944 | * @extended: if true @flags is treated as an extended profile | |
3945 | */ | |
3946 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3947 | { | |
3948 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3949 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3950 | ||
3951 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3952 | ||
3953 | /* 1) check that all other bits are zeroed */ | |
3954 | if (flags & ~mask) | |
3955 | return 0; | |
3956 | ||
3957 | /* 2) see if profile is reduced */ | |
3958 | if (flags == 0) | |
3959 | return !extended; /* "0" is valid for usual profiles */ | |
3960 | ||
3961 | /* true if exactly one bit set */ | |
818255fe | 3962 | return is_power_of_2(flags); |
0c460c0d ID |
3963 | } |
3964 | ||
837d5b6e ID |
3965 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3966 | { | |
a7e99c69 ID |
3967 | /* cancel requested || normal exit path */ |
3968 | return atomic_read(&fs_info->balance_cancel_req) || | |
3969 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3970 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3971 | } |
3972 | ||
bdcd3c97 AM |
3973 | /* Non-zero return value signifies invalidity */ |
3974 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3975 | u64 allowed) | |
3976 | { | |
3977 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3978 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3979 | (bctl_arg->target & ~allowed))); | |
3980 | } | |
3981 | ||
56fc37d9 AJ |
3982 | /* |
3983 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3984 | * @size_buf including the terminating null. The output may be trimmed if it | |
3985 | * does not fit into the provided buffer. | |
3986 | */ | |
3987 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3988 | u32 size_buf) | |
3989 | { | |
3990 | int ret; | |
3991 | u32 size_bp = size_buf; | |
3992 | char *bp = buf; | |
3993 | u64 flags = bargs->flags; | |
3994 | char tmp_buf[128] = {'\0'}; | |
3995 | ||
3996 | if (!flags) | |
3997 | return; | |
3998 | ||
3999 | #define CHECK_APPEND_NOARG(a) \ | |
4000 | do { \ | |
4001 | ret = snprintf(bp, size_bp, (a)); \ | |
4002 | if (ret < 0 || ret >= size_bp) \ | |
4003 | goto out_overflow; \ | |
4004 | size_bp -= ret; \ | |
4005 | bp += ret; \ | |
4006 | } while (0) | |
4007 | ||
4008 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4009 | do { \ | |
4010 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4011 | if (ret < 0 || ret >= size_bp) \ | |
4012 | goto out_overflow; \ | |
4013 | size_bp -= ret; \ | |
4014 | bp += ret; \ | |
4015 | } while (0) | |
4016 | ||
4017 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
4018 | do { \ | |
4019 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
4020 | if (ret < 0 || ret >= size_bp) \ | |
4021 | goto out_overflow; \ | |
4022 | size_bp -= ret; \ | |
4023 | bp += ret; \ | |
4024 | } while (0) | |
4025 | ||
4026 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4027 | int index = btrfs_bg_flags_to_raid_index(bargs->target); | |
4028 | ||
4029 | CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index)); | |
4030 | } | |
4031 | ||
4032 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
4033 | CHECK_APPEND_NOARG("soft,"); | |
4034 | ||
4035 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
4036 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
4037 | sizeof(tmp_buf)); | |
4038 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
4039 | } | |
4040 | ||
4041 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
4042 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
4043 | ||
4044 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
4045 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
4046 | bargs->usage_min, bargs->usage_max); | |
4047 | ||
4048 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
4049 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
4050 | ||
4051 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4052 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4053 | bargs->pstart, bargs->pend); | |
4054 | ||
4055 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4056 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4057 | bargs->vstart, bargs->vend); | |
4058 | ||
4059 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4060 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4061 | ||
4062 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4063 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4064 | bargs->limit_min, bargs->limit_max); | |
4065 | ||
4066 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4067 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4068 | bargs->stripes_min, bargs->stripes_max); | |
4069 | ||
4070 | #undef CHECK_APPEND_2ARG | |
4071 | #undef CHECK_APPEND_1ARG | |
4072 | #undef CHECK_APPEND_NOARG | |
4073 | ||
4074 | out_overflow: | |
4075 | ||
4076 | if (size_bp < size_buf) | |
4077 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4078 | else | |
4079 | buf[0] = '\0'; | |
4080 | } | |
4081 | ||
4082 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4083 | { | |
4084 | u32 size_buf = 1024; | |
4085 | char tmp_buf[192] = {'\0'}; | |
4086 | char *buf; | |
4087 | char *bp; | |
4088 | u32 size_bp = size_buf; | |
4089 | int ret; | |
4090 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4091 | ||
4092 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4093 | if (!buf) | |
4094 | return; | |
4095 | ||
4096 | bp = buf; | |
4097 | ||
4098 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4099 | do { \ | |
4100 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4101 | if (ret < 0 || ret >= size_bp) \ | |
4102 | goto out_overflow; \ | |
4103 | size_bp -= ret; \ | |
4104 | bp += ret; \ | |
4105 | } while (0) | |
4106 | ||
4107 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4108 | CHECK_APPEND_1ARG("%s", "-f "); | |
4109 | ||
4110 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4111 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4112 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4113 | } | |
4114 | ||
4115 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4116 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4117 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4118 | } | |
4119 | ||
4120 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4121 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4122 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4123 | } | |
4124 | ||
4125 | #undef CHECK_APPEND_1ARG | |
4126 | ||
4127 | out_overflow: | |
4128 | ||
4129 | if (size_bp < size_buf) | |
4130 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4131 | btrfs_info(fs_info, "balance: %s %s", | |
4132 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4133 | "resume" : "start", buf); | |
4134 | ||
4135 | kfree(buf); | |
4136 | } | |
4137 | ||
c9e9f97b | 4138 | /* |
dccdb07b | 4139 | * Should be called with balance mutexe held |
c9e9f97b | 4140 | */ |
6fcf6e2b DS |
4141 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4142 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4143 | struct btrfs_ioctl_balance_args *bargs) |
4144 | { | |
14506127 | 4145 | u64 meta_target, data_target; |
f43ffb60 | 4146 | u64 allowed; |
e4837f8f | 4147 | int mixed = 0; |
c9e9f97b | 4148 | int ret; |
8dabb742 | 4149 | u64 num_devices; |
de98ced9 | 4150 | unsigned seq; |
5a8067c0 | 4151 | bool reducing_integrity; |
c9e9f97b | 4152 | |
837d5b6e | 4153 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4154 | atomic_read(&fs_info->balance_pause_req) || |
4155 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4156 | ret = -EINVAL; |
4157 | goto out; | |
4158 | } | |
4159 | ||
e4837f8f ID |
4160 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4161 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4162 | mixed = 1; | |
4163 | ||
f43ffb60 ID |
4164 | /* |
4165 | * In case of mixed groups both data and meta should be picked, | |
4166 | * and identical options should be given for both of them. | |
4167 | */ | |
e4837f8f ID |
4168 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4169 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4170 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4171 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4172 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4173 | btrfs_err(fs_info, |
6dac13f8 | 4174 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4175 | ret = -EINVAL; |
4176 | goto out; | |
4177 | } | |
4178 | } | |
4179 | ||
1da73967 AJ |
4180 | num_devices = btrfs_num_devices(fs_info); |
4181 | ||
88be159c AH |
4182 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
4183 | if (num_devices > 1) | |
e4d8ec0f | 4184 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
4185 | if (num_devices > 2) |
4186 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
4187 | if (num_devices > 3) | |
4188 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
4189 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 4190 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
4191 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
4192 | ||
5d163e0e | 4193 | btrfs_err(fs_info, |
6dac13f8 AJ |
4194 | "balance: invalid convert data profile %s", |
4195 | get_raid_name(index)); | |
e4d8ec0f ID |
4196 | ret = -EINVAL; |
4197 | goto out; | |
4198 | } | |
bdcd3c97 | 4199 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
4200 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
4201 | ||
efe120a0 | 4202 | btrfs_err(fs_info, |
6dac13f8 AJ |
4203 | "balance: invalid convert metadata profile %s", |
4204 | get_raid_name(index)); | |
e4d8ec0f ID |
4205 | ret = -EINVAL; |
4206 | goto out; | |
4207 | } | |
bdcd3c97 | 4208 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
4209 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
4210 | ||
efe120a0 | 4211 | btrfs_err(fs_info, |
6dac13f8 AJ |
4212 | "balance: invalid convert system profile %s", |
4213 | get_raid_name(index)); | |
e4d8ec0f ID |
4214 | ret = -EINVAL; |
4215 | goto out; | |
4216 | } | |
4217 | ||
e4d8ec0f ID |
4218 | /* allow to reduce meta or sys integrity only if force set */ |
4219 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
4220 | BTRFS_BLOCK_GROUP_RAID10 | |
4221 | BTRFS_BLOCK_GROUP_RAID5 | | |
4222 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
4223 | do { |
4224 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4225 | ||
4226 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4227 | (fs_info->avail_system_alloc_bits & allowed) && | |
4228 | !(bctl->sys.target & allowed)) || | |
4229 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4230 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4231 | !(bctl->meta.target & allowed))) |
4232 | reducing_integrity = true; | |
4233 | else | |
4234 | reducing_integrity = false; | |
4235 | ||
4236 | /* if we're not converting, the target field is uninitialized */ | |
4237 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4238 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4239 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4240 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4241 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4242 | |
5a8067c0 FM |
4243 | if (reducing_integrity) { |
4244 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4245 | btrfs_info(fs_info, | |
4246 | "balance: force reducing metadata integrity"); | |
4247 | } else { | |
4248 | btrfs_err(fs_info, | |
4249 | "balance: reduces metadata integrity, use --force if you want this"); | |
4250 | ret = -EINVAL; | |
4251 | goto out; | |
4252 | } | |
4253 | } | |
4254 | ||
14506127 AB |
4255 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4256 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
4257 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
4258 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
4259 | ||
ee592d07 | 4260 | btrfs_warn(fs_info, |
6dac13f8 AJ |
4261 | "balance: metadata profile %s has lower redundancy than data profile %s", |
4262 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
4263 | } |
4264 | ||
6bccf3ab | 4265 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4266 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4267 | goto out; |
4268 | ||
59641015 ID |
4269 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4270 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4271 | BUG_ON(fs_info->balance_ctl); |
4272 | spin_lock(&fs_info->balance_lock); | |
4273 | fs_info->balance_ctl = bctl; | |
4274 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4275 | } else { |
4276 | BUG_ON(ret != -EEXIST); | |
4277 | spin_lock(&fs_info->balance_lock); | |
4278 | update_balance_args(bctl); | |
4279 | spin_unlock(&fs_info->balance_lock); | |
4280 | } | |
c9e9f97b | 4281 | |
3009a62f DS |
4282 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4283 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4284 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4285 | mutex_unlock(&fs_info->balance_mutex); |
4286 | ||
4287 | ret = __btrfs_balance(fs_info); | |
4288 | ||
4289 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4290 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4291 | btrfs_info(fs_info, "balance: paused"); | |
4292 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4293 | btrfs_info(fs_info, "balance: canceled"); | |
4294 | else | |
4295 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4296 | ||
3009a62f | 4297 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4298 | |
4299 | if (bargs) { | |
4300 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4301 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4302 | } |
4303 | ||
3a01aa7a ID |
4304 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4305 | balance_need_close(fs_info)) { | |
149196a2 | 4306 | reset_balance_state(fs_info); |
a17c95df | 4307 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4308 | } |
4309 | ||
837d5b6e | 4310 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4311 | |
4312 | return ret; | |
4313 | out: | |
59641015 | 4314 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4315 | reset_balance_state(fs_info); |
a17c95df | 4316 | else |
59641015 | 4317 | kfree(bctl); |
a17c95df DS |
4318 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4319 | ||
59641015 ID |
4320 | return ret; |
4321 | } | |
4322 | ||
4323 | static int balance_kthread(void *data) | |
4324 | { | |
2b6ba629 | 4325 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4326 | int ret = 0; |
59641015 | 4327 | |
59641015 | 4328 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4329 | if (fs_info->balance_ctl) |
6fcf6e2b | 4330 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4331 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4332 | |
59641015 ID |
4333 | return ret; |
4334 | } | |
4335 | ||
2b6ba629 ID |
4336 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4337 | { | |
4338 | struct task_struct *tsk; | |
4339 | ||
1354e1a1 | 4340 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4341 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4342 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4343 | return 0; |
4344 | } | |
1354e1a1 | 4345 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4346 | |
3cdde224 | 4347 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4348 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4349 | return 0; |
4350 | } | |
4351 | ||
02ee654d AJ |
4352 | /* |
4353 | * A ro->rw remount sequence should continue with the paused balance | |
4354 | * regardless of who pauses it, system or the user as of now, so set | |
4355 | * the resume flag. | |
4356 | */ | |
4357 | spin_lock(&fs_info->balance_lock); | |
4358 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4359 | spin_unlock(&fs_info->balance_lock); | |
4360 | ||
2b6ba629 | 4361 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4362 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4363 | } |
4364 | ||
68310a5e | 4365 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4366 | { |
59641015 ID |
4367 | struct btrfs_balance_control *bctl; |
4368 | struct btrfs_balance_item *item; | |
4369 | struct btrfs_disk_balance_args disk_bargs; | |
4370 | struct btrfs_path *path; | |
4371 | struct extent_buffer *leaf; | |
4372 | struct btrfs_key key; | |
4373 | int ret; | |
4374 | ||
4375 | path = btrfs_alloc_path(); | |
4376 | if (!path) | |
4377 | return -ENOMEM; | |
4378 | ||
59641015 | 4379 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4380 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4381 | key.offset = 0; |
4382 | ||
68310a5e | 4383 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4384 | if (ret < 0) |
68310a5e | 4385 | goto out; |
59641015 ID |
4386 | if (ret > 0) { /* ret = -ENOENT; */ |
4387 | ret = 0; | |
68310a5e ID |
4388 | goto out; |
4389 | } | |
4390 | ||
4391 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4392 | if (!bctl) { | |
4393 | ret = -ENOMEM; | |
4394 | goto out; | |
59641015 ID |
4395 | } |
4396 | ||
4397 | leaf = path->nodes[0]; | |
4398 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4399 | ||
68310a5e ID |
4400 | bctl->flags = btrfs_balance_flags(leaf, item); |
4401 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4402 | |
4403 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4404 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4405 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4406 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4407 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4408 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4409 | ||
eee95e3f DS |
4410 | /* |
4411 | * This should never happen, as the paused balance state is recovered | |
4412 | * during mount without any chance of other exclusive ops to collide. | |
4413 | * | |
4414 | * This gives the exclusive op status to balance and keeps in paused | |
4415 | * state until user intervention (cancel or umount). If the ownership | |
4416 | * cannot be assigned, show a message but do not fail. The balance | |
4417 | * is in a paused state and must have fs_info::balance_ctl properly | |
4418 | * set up. | |
4419 | */ | |
4420 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4421 | btrfs_warn(fs_info, | |
6dac13f8 | 4422 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4423 | |
68310a5e | 4424 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4425 | BUG_ON(fs_info->balance_ctl); |
4426 | spin_lock(&fs_info->balance_lock); | |
4427 | fs_info->balance_ctl = bctl; | |
4428 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4429 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4430 | out: |
4431 | btrfs_free_path(path); | |
ec44a35c CM |
4432 | return ret; |
4433 | } | |
4434 | ||
837d5b6e ID |
4435 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4436 | { | |
4437 | int ret = 0; | |
4438 | ||
4439 | mutex_lock(&fs_info->balance_mutex); | |
4440 | if (!fs_info->balance_ctl) { | |
4441 | mutex_unlock(&fs_info->balance_mutex); | |
4442 | return -ENOTCONN; | |
4443 | } | |
4444 | ||
3009a62f | 4445 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4446 | atomic_inc(&fs_info->balance_pause_req); |
4447 | mutex_unlock(&fs_info->balance_mutex); | |
4448 | ||
4449 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4450 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4451 | |
4452 | mutex_lock(&fs_info->balance_mutex); | |
4453 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4454 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4455 | atomic_dec(&fs_info->balance_pause_req); |
4456 | } else { | |
4457 | ret = -ENOTCONN; | |
4458 | } | |
4459 | ||
4460 | mutex_unlock(&fs_info->balance_mutex); | |
4461 | return ret; | |
4462 | } | |
4463 | ||
a7e99c69 ID |
4464 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4465 | { | |
4466 | mutex_lock(&fs_info->balance_mutex); | |
4467 | if (!fs_info->balance_ctl) { | |
4468 | mutex_unlock(&fs_info->balance_mutex); | |
4469 | return -ENOTCONN; | |
4470 | } | |
4471 | ||
cf7d20f4 DS |
4472 | /* |
4473 | * A paused balance with the item stored on disk can be resumed at | |
4474 | * mount time if the mount is read-write. Otherwise it's still paused | |
4475 | * and we must not allow cancelling as it deletes the item. | |
4476 | */ | |
4477 | if (sb_rdonly(fs_info->sb)) { | |
4478 | mutex_unlock(&fs_info->balance_mutex); | |
4479 | return -EROFS; | |
4480 | } | |
4481 | ||
a7e99c69 ID |
4482 | atomic_inc(&fs_info->balance_cancel_req); |
4483 | /* | |
4484 | * if we are running just wait and return, balance item is | |
4485 | * deleted in btrfs_balance in this case | |
4486 | */ | |
3009a62f | 4487 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4488 | mutex_unlock(&fs_info->balance_mutex); |
4489 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4490 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4491 | mutex_lock(&fs_info->balance_mutex); |
4492 | } else { | |
a7e99c69 | 4493 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4494 | /* |
4495 | * Lock released to allow other waiters to continue, we'll | |
4496 | * reexamine the status again. | |
4497 | */ | |
a7e99c69 ID |
4498 | mutex_lock(&fs_info->balance_mutex); |
4499 | ||
a17c95df | 4500 | if (fs_info->balance_ctl) { |
149196a2 | 4501 | reset_balance_state(fs_info); |
a17c95df | 4502 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4503 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4504 | } |
a7e99c69 ID |
4505 | } |
4506 | ||
3009a62f DS |
4507 | BUG_ON(fs_info->balance_ctl || |
4508 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4509 | atomic_dec(&fs_info->balance_cancel_req); |
4510 | mutex_unlock(&fs_info->balance_mutex); | |
4511 | return 0; | |
4512 | } | |
4513 | ||
803b2f54 SB |
4514 | static int btrfs_uuid_scan_kthread(void *data) |
4515 | { | |
4516 | struct btrfs_fs_info *fs_info = data; | |
4517 | struct btrfs_root *root = fs_info->tree_root; | |
4518 | struct btrfs_key key; | |
803b2f54 SB |
4519 | struct btrfs_path *path = NULL; |
4520 | int ret = 0; | |
4521 | struct extent_buffer *eb; | |
4522 | int slot; | |
4523 | struct btrfs_root_item root_item; | |
4524 | u32 item_size; | |
f45388f3 | 4525 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4526 | |
4527 | path = btrfs_alloc_path(); | |
4528 | if (!path) { | |
4529 | ret = -ENOMEM; | |
4530 | goto out; | |
4531 | } | |
4532 | ||
4533 | key.objectid = 0; | |
4534 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4535 | key.offset = 0; | |
4536 | ||
803b2f54 | 4537 | while (1) { |
7c829b72 AJ |
4538 | ret = btrfs_search_forward(root, &key, path, |
4539 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4540 | if (ret) { |
4541 | if (ret > 0) | |
4542 | ret = 0; | |
4543 | break; | |
4544 | } | |
4545 | ||
4546 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4547 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4548 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4549 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4550 | goto skip; | |
4551 | ||
4552 | eb = path->nodes[0]; | |
4553 | slot = path->slots[0]; | |
4554 | item_size = btrfs_item_size_nr(eb, slot); | |
4555 | if (item_size < sizeof(root_item)) | |
4556 | goto skip; | |
4557 | ||
803b2f54 SB |
4558 | read_extent_buffer(eb, &root_item, |
4559 | btrfs_item_ptr_offset(eb, slot), | |
4560 | (int)sizeof(root_item)); | |
4561 | if (btrfs_root_refs(&root_item) == 0) | |
4562 | goto skip; | |
f45388f3 FDBM |
4563 | |
4564 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4565 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4566 | if (trans) | |
4567 | goto update_tree; | |
4568 | ||
4569 | btrfs_release_path(path); | |
803b2f54 SB |
4570 | /* |
4571 | * 1 - subvol uuid item | |
4572 | * 1 - received_subvol uuid item | |
4573 | */ | |
4574 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4575 | if (IS_ERR(trans)) { | |
4576 | ret = PTR_ERR(trans); | |
4577 | break; | |
4578 | } | |
f45388f3 FDBM |
4579 | continue; |
4580 | } else { | |
4581 | goto skip; | |
4582 | } | |
4583 | update_tree: | |
4584 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4585 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4586 | BTRFS_UUID_KEY_SUBVOL, |
4587 | key.objectid); | |
4588 | if (ret < 0) { | |
efe120a0 | 4589 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4590 | ret); |
803b2f54 SB |
4591 | break; |
4592 | } | |
4593 | } | |
4594 | ||
4595 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4596 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4597 | root_item.received_uuid, |
4598 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4599 | key.objectid); | |
4600 | if (ret < 0) { | |
efe120a0 | 4601 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4602 | ret); |
803b2f54 SB |
4603 | break; |
4604 | } | |
4605 | } | |
4606 | ||
f45388f3 | 4607 | skip: |
803b2f54 | 4608 | if (trans) { |
3a45bb20 | 4609 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4610 | trans = NULL; |
803b2f54 SB |
4611 | if (ret) |
4612 | break; | |
4613 | } | |
4614 | ||
803b2f54 SB |
4615 | btrfs_release_path(path); |
4616 | if (key.offset < (u64)-1) { | |
4617 | key.offset++; | |
4618 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4619 | key.offset = 0; | |
4620 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4621 | } else if (key.objectid < (u64)-1) { | |
4622 | key.offset = 0; | |
4623 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4624 | key.objectid++; | |
4625 | } else { | |
4626 | break; | |
4627 | } | |
4628 | cond_resched(); | |
4629 | } | |
4630 | ||
4631 | out: | |
4632 | btrfs_free_path(path); | |
f45388f3 | 4633 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4634 | btrfs_end_transaction(trans); |
803b2f54 | 4635 | if (ret) |
efe120a0 | 4636 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4637 | else |
afcdd129 | 4638 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4639 | up(&fs_info->uuid_tree_rescan_sem); |
4640 | return 0; | |
4641 | } | |
4642 | ||
70f80175 SB |
4643 | /* |
4644 | * Callback for btrfs_uuid_tree_iterate(). | |
4645 | * returns: | |
4646 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4647 | * < 0 if an error occurred. |
70f80175 SB |
4648 | * > 0 if the check failed, which means the caller shall remove the entry. |
4649 | */ | |
4650 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4651 | u8 *uuid, u8 type, u64 subid) | |
4652 | { | |
4653 | struct btrfs_key key; | |
4654 | int ret = 0; | |
4655 | struct btrfs_root *subvol_root; | |
4656 | ||
4657 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4658 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4659 | goto out; | |
4660 | ||
4661 | key.objectid = subid; | |
4662 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4663 | key.offset = (u64)-1; | |
4664 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4665 | if (IS_ERR(subvol_root)) { | |
4666 | ret = PTR_ERR(subvol_root); | |
4667 | if (ret == -ENOENT) | |
4668 | ret = 1; | |
4669 | goto out; | |
4670 | } | |
4671 | ||
4672 | switch (type) { | |
4673 | case BTRFS_UUID_KEY_SUBVOL: | |
4674 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4675 | ret = 1; | |
4676 | break; | |
4677 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4678 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4679 | BTRFS_UUID_SIZE)) | |
4680 | ret = 1; | |
4681 | break; | |
4682 | } | |
4683 | ||
4684 | out: | |
4685 | return ret; | |
4686 | } | |
4687 | ||
4688 | static int btrfs_uuid_rescan_kthread(void *data) | |
4689 | { | |
4690 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4691 | int ret; | |
4692 | ||
4693 | /* | |
4694 | * 1st step is to iterate through the existing UUID tree and | |
4695 | * to delete all entries that contain outdated data. | |
4696 | * 2nd step is to add all missing entries to the UUID tree. | |
4697 | */ | |
4698 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4699 | if (ret < 0) { | |
efe120a0 | 4700 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4701 | up(&fs_info->uuid_tree_rescan_sem); |
4702 | return ret; | |
4703 | } | |
4704 | return btrfs_uuid_scan_kthread(data); | |
4705 | } | |
4706 | ||
f7a81ea4 SB |
4707 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4708 | { | |
4709 | struct btrfs_trans_handle *trans; | |
4710 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4711 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4712 | struct task_struct *task; |
4713 | int ret; | |
f7a81ea4 SB |
4714 | |
4715 | /* | |
4716 | * 1 - root node | |
4717 | * 1 - root item | |
4718 | */ | |
4719 | trans = btrfs_start_transaction(tree_root, 2); | |
4720 | if (IS_ERR(trans)) | |
4721 | return PTR_ERR(trans); | |
4722 | ||
4723 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4724 | BTRFS_UUID_TREE_OBJECTID); | |
4725 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4726 | ret = PTR_ERR(uuid_root); |
66642832 | 4727 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4728 | btrfs_end_transaction(trans); |
6d13f549 | 4729 | return ret; |
f7a81ea4 SB |
4730 | } |
4731 | ||
4732 | fs_info->uuid_root = uuid_root; | |
4733 | ||
3a45bb20 | 4734 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4735 | if (ret) |
4736 | return ret; | |
4737 | ||
4738 | down(&fs_info->uuid_tree_rescan_sem); | |
4739 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4740 | if (IS_ERR(task)) { | |
70f80175 | 4741 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4742 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4743 | up(&fs_info->uuid_tree_rescan_sem); |
4744 | return PTR_ERR(task); | |
4745 | } | |
4746 | ||
4747 | return 0; | |
f7a81ea4 | 4748 | } |
803b2f54 | 4749 | |
70f80175 SB |
4750 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4751 | { | |
4752 | struct task_struct *task; | |
4753 | ||
4754 | down(&fs_info->uuid_tree_rescan_sem); | |
4755 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4756 | if (IS_ERR(task)) { | |
4757 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4758 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4759 | up(&fs_info->uuid_tree_rescan_sem); |
4760 | return PTR_ERR(task); | |
4761 | } | |
4762 | ||
4763 | return 0; | |
4764 | } | |
4765 | ||
8f18cf13 CM |
4766 | /* |
4767 | * shrinking a device means finding all of the device extents past | |
4768 | * the new size, and then following the back refs to the chunks. | |
4769 | * The chunk relocation code actually frees the device extent | |
4770 | */ | |
4771 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4772 | { | |
0b246afa JM |
4773 | struct btrfs_fs_info *fs_info = device->fs_info; |
4774 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4775 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4776 | struct btrfs_dev_extent *dev_extent = NULL; |
4777 | struct btrfs_path *path; | |
4778 | u64 length; | |
8f18cf13 CM |
4779 | u64 chunk_offset; |
4780 | int ret; | |
4781 | int slot; | |
ba1bf481 JB |
4782 | int failed = 0; |
4783 | bool retried = false; | |
53e489bc | 4784 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4785 | struct extent_buffer *l; |
4786 | struct btrfs_key key; | |
0b246afa | 4787 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4788 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4789 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4790 | u64 diff; |
4791 | ||
4792 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4793 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4794 | |
401e29c1 | 4795 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4796 | return -EINVAL; |
4797 | ||
8f18cf13 CM |
4798 | path = btrfs_alloc_path(); |
4799 | if (!path) | |
4800 | return -ENOMEM; | |
4801 | ||
0338dff6 | 4802 | path->reada = READA_BACK; |
8f18cf13 | 4803 | |
34441361 | 4804 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4805 | |
7cc8e58d | 4806 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4807 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4808 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4809 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4810 | } |
34441361 | 4811 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4812 | |
ba1bf481 | 4813 | again: |
8f18cf13 CM |
4814 | key.objectid = device->devid; |
4815 | key.offset = (u64)-1; | |
4816 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4817 | ||
213e64da | 4818 | do { |
0b246afa | 4819 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4820 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4821 | if (ret < 0) { |
0b246afa | 4822 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4823 | goto done; |
67c5e7d4 | 4824 | } |
8f18cf13 CM |
4825 | |
4826 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4827 | if (ret) |
0b246afa | 4828 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4829 | if (ret < 0) |
4830 | goto done; | |
4831 | if (ret) { | |
4832 | ret = 0; | |
b3b4aa74 | 4833 | btrfs_release_path(path); |
bf1fb512 | 4834 | break; |
8f18cf13 CM |
4835 | } |
4836 | ||
4837 | l = path->nodes[0]; | |
4838 | slot = path->slots[0]; | |
4839 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4840 | ||
ba1bf481 | 4841 | if (key.objectid != device->devid) { |
0b246afa | 4842 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4843 | btrfs_release_path(path); |
bf1fb512 | 4844 | break; |
ba1bf481 | 4845 | } |
8f18cf13 CM |
4846 | |
4847 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4848 | length = btrfs_dev_extent_length(l, dev_extent); | |
4849 | ||
ba1bf481 | 4850 | if (key.offset + length <= new_size) { |
0b246afa | 4851 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4852 | btrfs_release_path(path); |
d6397bae | 4853 | break; |
ba1bf481 | 4854 | } |
8f18cf13 | 4855 | |
8f18cf13 | 4856 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4857 | btrfs_release_path(path); |
8f18cf13 | 4858 | |
a6f93c71 LB |
4859 | /* |
4860 | * We may be relocating the only data chunk we have, | |
4861 | * which could potentially end up with losing data's | |
4862 | * raid profile, so lets allocate an empty one in | |
4863 | * advance. | |
4864 | */ | |
4865 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4866 | if (ret < 0) { | |
4867 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4868 | goto done; | |
4869 | } | |
4870 | ||
0b246afa JM |
4871 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4872 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4873 | if (ret == -ENOSPC) { |
ba1bf481 | 4874 | failed++; |
eede2bf3 OS |
4875 | } else if (ret) { |
4876 | if (ret == -ETXTBSY) { | |
4877 | btrfs_warn(fs_info, | |
4878 | "could not shrink block group %llu due to active swapfile", | |
4879 | chunk_offset); | |
4880 | } | |
4881 | goto done; | |
4882 | } | |
213e64da | 4883 | } while (key.offset-- > 0); |
ba1bf481 JB |
4884 | |
4885 | if (failed && !retried) { | |
4886 | failed = 0; | |
4887 | retried = true; | |
4888 | goto again; | |
4889 | } else if (failed && retried) { | |
4890 | ret = -ENOSPC; | |
ba1bf481 | 4891 | goto done; |
8f18cf13 CM |
4892 | } |
4893 | ||
d6397bae | 4894 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4895 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4896 | if (IS_ERR(trans)) { |
4897 | ret = PTR_ERR(trans); | |
4898 | goto done; | |
4899 | } | |
4900 | ||
34441361 | 4901 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4902 | |
4903 | /* | |
4904 | * We checked in the above loop all device extents that were already in | |
4905 | * the device tree. However before we have updated the device's | |
4906 | * total_bytes to the new size, we might have had chunk allocations that | |
4907 | * have not complete yet (new block groups attached to transaction | |
4908 | * handles), and therefore their device extents were not yet in the | |
4909 | * device tree and we missed them in the loop above. So if we have any | |
4910 | * pending chunk using a device extent that overlaps the device range | |
4911 | * that we can not use anymore, commit the current transaction and | |
4912 | * repeat the search on the device tree - this way we guarantee we will | |
4913 | * not have chunks using device extents that end beyond 'new_size'. | |
4914 | */ | |
4915 | if (!checked_pending_chunks) { | |
4916 | u64 start = new_size; | |
4917 | u64 len = old_size - new_size; | |
4918 | ||
499f377f JM |
4919 | if (contains_pending_extent(trans->transaction, device, |
4920 | &start, len)) { | |
34441361 | 4921 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4922 | checked_pending_chunks = true; |
4923 | failed = 0; | |
4924 | retried = false; | |
3a45bb20 | 4925 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4926 | if (ret) |
4927 | goto done; | |
4928 | goto again; | |
4929 | } | |
4930 | } | |
4931 | ||
7cc8e58d | 4932 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4933 | if (list_empty(&device->resized_list)) |
4934 | list_add_tail(&device->resized_list, | |
0b246afa | 4935 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4936 | |
d6397bae | 4937 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4938 | btrfs_set_super_total_bytes(super_copy, |
4939 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4940 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4941 | |
4942 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4943 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4944 | if (ret < 0) { |
4945 | btrfs_abort_transaction(trans, ret); | |
4946 | btrfs_end_transaction(trans); | |
4947 | } else { | |
4948 | ret = btrfs_commit_transaction(trans); | |
4949 | } | |
8f18cf13 CM |
4950 | done: |
4951 | btrfs_free_path(path); | |
53e489bc | 4952 | if (ret) { |
34441361 | 4953 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4954 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4955 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4956 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4957 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4958 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4959 | } |
8f18cf13 CM |
4960 | return ret; |
4961 | } | |
4962 | ||
2ff7e61e | 4963 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4964 | struct btrfs_key *key, |
4965 | struct btrfs_chunk *chunk, int item_size) | |
4966 | { | |
0b246afa | 4967 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4968 | struct btrfs_disk_key disk_key; |
4969 | u32 array_size; | |
4970 | u8 *ptr; | |
4971 | ||
34441361 | 4972 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4973 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4974 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4975 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4976 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4977 | return -EFBIG; |
fe48a5c0 | 4978 | } |
0b86a832 CM |
4979 | |
4980 | ptr = super_copy->sys_chunk_array + array_size; | |
4981 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4982 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4983 | ptr += sizeof(disk_key); | |
4984 | memcpy(ptr, chunk, item_size); | |
4985 | item_size += sizeof(disk_key); | |
4986 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4987 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4988 | |
0b86a832 CM |
4989 | return 0; |
4990 | } | |
4991 | ||
73c5de00 AJ |
4992 | /* |
4993 | * sort the devices in descending order by max_avail, total_avail | |
4994 | */ | |
4995 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4996 | { |
73c5de00 AJ |
4997 | const struct btrfs_device_info *di_a = a; |
4998 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4999 | |
73c5de00 | 5000 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 5001 | return -1; |
73c5de00 | 5002 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 5003 | return 1; |
73c5de00 AJ |
5004 | if (di_a->total_avail > di_b->total_avail) |
5005 | return -1; | |
5006 | if (di_a->total_avail < di_b->total_avail) | |
5007 | return 1; | |
5008 | return 0; | |
b2117a39 | 5009 | } |
0b86a832 | 5010 | |
53b381b3 DW |
5011 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5012 | { | |
ffe2d203 | 5013 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
5014 | return; |
5015 | ||
ceda0864 | 5016 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
5017 | } |
5018 | ||
062d4d1f | 5019 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
5020 | - sizeof(struct btrfs_chunk)) \ |
5021 | / sizeof(struct btrfs_stripe) + 1) | |
5022 | ||
5023 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
5024 | - 2 * sizeof(struct btrfs_disk_key) \ | |
5025 | - 2 * sizeof(struct btrfs_chunk)) \ | |
5026 | / sizeof(struct btrfs_stripe) + 1) | |
5027 | ||
73c5de00 | 5028 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 5029 | u64 start, u64 type) |
b2117a39 | 5030 | { |
2ff7e61e | 5031 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 5032 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 5033 | struct btrfs_device *device; |
73c5de00 AJ |
5034 | struct map_lookup *map = NULL; |
5035 | struct extent_map_tree *em_tree; | |
5036 | struct extent_map *em; | |
5037 | struct btrfs_device_info *devices_info = NULL; | |
5038 | u64 total_avail; | |
5039 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
5040 | int data_stripes; /* number of stripes that count for |
5041 | block group size */ | |
73c5de00 AJ |
5042 | int sub_stripes; /* sub_stripes info for map */ |
5043 | int dev_stripes; /* stripes per dev */ | |
5044 | int devs_max; /* max devs to use */ | |
5045 | int devs_min; /* min devs needed */ | |
5046 | int devs_increment; /* ndevs has to be a multiple of this */ | |
5047 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
5048 | int nparity; /* number of stripes worth of bytes to |
5049 | store parity information */ | |
73c5de00 AJ |
5050 | int ret; |
5051 | u64 max_stripe_size; | |
5052 | u64 max_chunk_size; | |
5053 | u64 stripe_size; | |
23f0ff1e | 5054 | u64 chunk_size; |
73c5de00 AJ |
5055 | int ndevs; |
5056 | int i; | |
5057 | int j; | |
31e50229 | 5058 | int index; |
593060d7 | 5059 | |
0c460c0d | 5060 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 5061 | |
4117f207 QW |
5062 | if (list_empty(&fs_devices->alloc_list)) { |
5063 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5064 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 5065 | return -ENOSPC; |
4117f207 | 5066 | } |
b2117a39 | 5067 | |
3e72ee88 | 5068 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 5069 | |
31e50229 LB |
5070 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
5071 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5072 | devs_max = btrfs_raid_array[index].devs_max; | |
5073 | devs_min = btrfs_raid_array[index].devs_min; | |
5074 | devs_increment = btrfs_raid_array[index].devs_increment; | |
5075 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 5076 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 5077 | |
9b3f68b9 | 5078 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 5079 | max_stripe_size = SZ_1G; |
fce466ea | 5080 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 5081 | if (!devs_max) |
062d4d1f | 5082 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 5083 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 5084 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
5085 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
5086 | max_stripe_size = SZ_1G; | |
1100373f | 5087 | else |
ee22184b | 5088 | max_stripe_size = SZ_256M; |
73c5de00 | 5089 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 5090 | if (!devs_max) |
062d4d1f | 5091 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 5092 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 5093 | max_stripe_size = SZ_32M; |
73c5de00 | 5094 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
5095 | if (!devs_max) |
5096 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 5097 | } else { |
351fd353 | 5098 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
5099 | type); |
5100 | BUG_ON(1); | |
9b3f68b9 CM |
5101 | } |
5102 | ||
2b82032c YZ |
5103 | /* we don't want a chunk larger than 10% of writeable space */ |
5104 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
5105 | max_chunk_size); | |
9b3f68b9 | 5106 | |
31e818fe | 5107 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
5108 | GFP_NOFS); |
5109 | if (!devices_info) | |
5110 | return -ENOMEM; | |
0cad8a11 | 5111 | |
9f680ce0 | 5112 | /* |
73c5de00 AJ |
5113 | * in the first pass through the devices list, we gather information |
5114 | * about the available holes on each device. | |
9f680ce0 | 5115 | */ |
73c5de00 | 5116 | ndevs = 0; |
ebcc9301 | 5117 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
5118 | u64 max_avail; |
5119 | u64 dev_offset; | |
b2117a39 | 5120 | |
ebbede42 | 5121 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5122 | WARN(1, KERN_ERR |
efe120a0 | 5123 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5124 | continue; |
5125 | } | |
b2117a39 | 5126 | |
e12c9621 AJ |
5127 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5128 | &device->dev_state) || | |
401e29c1 | 5129 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5130 | continue; |
b2117a39 | 5131 | |
73c5de00 AJ |
5132 | if (device->total_bytes > device->bytes_used) |
5133 | total_avail = device->total_bytes - device->bytes_used; | |
5134 | else | |
5135 | total_avail = 0; | |
38c01b96 | 5136 | |
5137 | /* If there is no space on this device, skip it. */ | |
5138 | if (total_avail == 0) | |
5139 | continue; | |
b2117a39 | 5140 | |
6df9a95e | 5141 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
5142 | max_stripe_size * dev_stripes, |
5143 | &dev_offset, &max_avail); | |
5144 | if (ret && ret != -ENOSPC) | |
5145 | goto error; | |
b2117a39 | 5146 | |
73c5de00 AJ |
5147 | if (ret == 0) |
5148 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5149 | |
4117f207 QW |
5150 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5151 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5152 | btrfs_debug(info, | |
5153 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5154 | __func__, device->devid, max_avail, | |
5155 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5156 | continue; |
4117f207 | 5157 | } |
b2117a39 | 5158 | |
063d006f ES |
5159 | if (ndevs == fs_devices->rw_devices) { |
5160 | WARN(1, "%s: found more than %llu devices\n", | |
5161 | __func__, fs_devices->rw_devices); | |
5162 | break; | |
5163 | } | |
73c5de00 AJ |
5164 | devices_info[ndevs].dev_offset = dev_offset; |
5165 | devices_info[ndevs].max_avail = max_avail; | |
5166 | devices_info[ndevs].total_avail = total_avail; | |
5167 | devices_info[ndevs].dev = device; | |
5168 | ++ndevs; | |
5169 | } | |
b2117a39 | 5170 | |
73c5de00 AJ |
5171 | /* |
5172 | * now sort the devices by hole size / available space | |
5173 | */ | |
5174 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5175 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5176 | |
73c5de00 | 5177 | /* round down to number of usable stripes */ |
e5600fd6 | 5178 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5179 | |
ba89b802 | 5180 | if (ndevs < devs_min) { |
73c5de00 | 5181 | ret = -ENOSPC; |
4117f207 QW |
5182 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5183 | btrfs_debug(info, | |
5184 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5185 | __func__, ndevs, devs_min); |
4117f207 | 5186 | } |
73c5de00 | 5187 | goto error; |
b2117a39 | 5188 | } |
9f680ce0 | 5189 | |
f148ef4d NB |
5190 | ndevs = min(ndevs, devs_max); |
5191 | ||
73c5de00 | 5192 | /* |
92e222df HK |
5193 | * The primary goal is to maximize the number of stripes, so use as |
5194 | * many devices as possible, even if the stripes are not maximum sized. | |
5195 | * | |
5196 | * The DUP profile stores more than one stripe per device, the | |
5197 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5198 | */ |
92e222df | 5199 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5200 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5201 | |
53b381b3 DW |
5202 | /* |
5203 | * this will have to be fixed for RAID1 and RAID10 over | |
5204 | * more drives | |
5205 | */ | |
b50836ed | 5206 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5207 | |
5208 | /* | |
5209 | * Use the number of data stripes to figure out how big this chunk | |
5210 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5211 | * and compare that answer with the max chunk size. If it's higher, |
5212 | * we try to reduce stripe_size. | |
86db2578 CM |
5213 | */ |
5214 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5215 | /* |
baf92114 HK |
5216 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5217 | * then use it, unless it ends up being even bigger than the | |
5218 | * previous value we had already. | |
86db2578 | 5219 | */ |
baf92114 HK |
5220 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5221 | data_stripes), SZ_16M), | |
793ff2c8 | 5222 | stripe_size); |
86db2578 CM |
5223 | } |
5224 | ||
37db63a4 | 5225 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5226 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5227 | |
5228 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5229 | if (!map) { | |
5230 | ret = -ENOMEM; | |
5231 | goto error; | |
5232 | } | |
5233 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5234 | |
73c5de00 AJ |
5235 | for (i = 0; i < ndevs; ++i) { |
5236 | for (j = 0; j < dev_stripes; ++j) { | |
5237 | int s = i * dev_stripes + j; | |
5238 | map->stripes[s].dev = devices_info[i].dev; | |
5239 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5240 | j * stripe_size; | |
6324fbf3 | 5241 | } |
6324fbf3 | 5242 | } |
500ceed8 NB |
5243 | map->stripe_len = BTRFS_STRIPE_LEN; |
5244 | map->io_align = BTRFS_STRIPE_LEN; | |
5245 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5246 | map->type = type; |
2b82032c | 5247 | map->sub_stripes = sub_stripes; |
0b86a832 | 5248 | |
23f0ff1e | 5249 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5250 | |
23f0ff1e | 5251 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5252 | |
172ddd60 | 5253 | em = alloc_extent_map(); |
2b82032c | 5254 | if (!em) { |
298a8f9c | 5255 | kfree(map); |
b2117a39 MX |
5256 | ret = -ENOMEM; |
5257 | goto error; | |
593060d7 | 5258 | } |
298a8f9c | 5259 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5260 | em->map_lookup = map; |
2b82032c | 5261 | em->start = start; |
23f0ff1e | 5262 | em->len = chunk_size; |
2b82032c YZ |
5263 | em->block_start = 0; |
5264 | em->block_len = em->len; | |
6df9a95e | 5265 | em->orig_block_len = stripe_size; |
593060d7 | 5266 | |
0b246afa | 5267 | em_tree = &info->mapping_tree.map_tree; |
890871be | 5268 | write_lock(&em_tree->lock); |
09a2a8f9 | 5269 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5270 | if (ret) { |
1efb72a3 | 5271 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5272 | free_extent_map(em); |
1dd4602f | 5273 | goto error; |
0f5d42b2 | 5274 | } |
0b86a832 | 5275 | |
1efb72a3 NB |
5276 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
5277 | refcount_inc(&em->refs); | |
5278 | write_unlock(&em_tree->lock); | |
5279 | ||
23f0ff1e | 5280 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5281 | if (ret) |
5282 | goto error_del_extent; | |
2b82032c | 5283 | |
2f29df4f HK |
5284 | for (i = 0; i < map->num_stripes; i++) |
5285 | btrfs_device_set_bytes_used(map->stripes[i].dev, | |
5286 | map->stripes[i].dev->bytes_used + stripe_size); | |
43530c46 | 5287 | |
a5ed45f8 | 5288 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5289 | |
0f5d42b2 | 5290 | free_extent_map(em); |
0b246afa | 5291 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5292 | |
b2117a39 | 5293 | kfree(devices_info); |
2b82032c | 5294 | return 0; |
b2117a39 | 5295 | |
6df9a95e | 5296 | error_del_extent: |
0f5d42b2 JB |
5297 | write_lock(&em_tree->lock); |
5298 | remove_extent_mapping(em_tree, em); | |
5299 | write_unlock(&em_tree->lock); | |
5300 | ||
5301 | /* One for our allocation */ | |
5302 | free_extent_map(em); | |
5303 | /* One for the tree reference */ | |
5304 | free_extent_map(em); | |
495e64f4 FM |
5305 | /* One for the pending_chunks list reference */ |
5306 | free_extent_map(em); | |
b2117a39 | 5307 | error: |
b2117a39 MX |
5308 | kfree(devices_info); |
5309 | return ret; | |
2b82032c YZ |
5310 | } |
5311 | ||
6df9a95e | 5312 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5313 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5314 | { |
97aff912 | 5315 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5316 | struct btrfs_root *extent_root = fs_info->extent_root; |
5317 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5318 | struct btrfs_key key; |
2b82032c YZ |
5319 | struct btrfs_device *device; |
5320 | struct btrfs_chunk *chunk; | |
5321 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5322 | struct extent_map *em; |
5323 | struct map_lookup *map; | |
5324 | size_t item_size; | |
5325 | u64 dev_offset; | |
5326 | u64 stripe_size; | |
5327 | int i = 0; | |
140e639f | 5328 | int ret = 0; |
2b82032c | 5329 | |
60ca842e | 5330 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5331 | if (IS_ERR(em)) |
5332 | return PTR_ERR(em); | |
6df9a95e | 5333 | |
95617d69 | 5334 | map = em->map_lookup; |
6df9a95e JB |
5335 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5336 | stripe_size = em->orig_block_len; | |
5337 | ||
2b82032c | 5338 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5339 | if (!chunk) { |
5340 | ret = -ENOMEM; | |
5341 | goto out; | |
5342 | } | |
5343 | ||
50460e37 FM |
5344 | /* |
5345 | * Take the device list mutex to prevent races with the final phase of | |
5346 | * a device replace operation that replaces the device object associated | |
5347 | * with the map's stripes, because the device object's id can change | |
5348 | * at any time during that final phase of the device replace operation | |
5349 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5350 | */ | |
0b246afa | 5351 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5352 | for (i = 0; i < map->num_stripes; i++) { |
5353 | device = map->stripes[i].dev; | |
5354 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5355 | |
0b86a832 | 5356 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5357 | if (ret) |
50460e37 | 5358 | break; |
b5d9071c NB |
5359 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5360 | dev_offset, stripe_size); | |
6df9a95e | 5361 | if (ret) |
50460e37 FM |
5362 | break; |
5363 | } | |
5364 | if (ret) { | |
0b246afa | 5365 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5366 | goto out; |
2b82032c YZ |
5367 | } |
5368 | ||
2b82032c | 5369 | stripe = &chunk->stripe; |
6df9a95e JB |
5370 | for (i = 0; i < map->num_stripes; i++) { |
5371 | device = map->stripes[i].dev; | |
5372 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5373 | |
e17cade2 CM |
5374 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5375 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5376 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5377 | stripe++; |
0b86a832 | 5378 | } |
0b246afa | 5379 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5380 | |
2b82032c | 5381 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5382 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5383 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5384 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5385 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5386 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5387 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5388 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5389 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5390 | |
2b82032c YZ |
5391 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5392 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5393 | key.offset = chunk_offset; | |
0b86a832 | 5394 | |
2b82032c | 5395 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5396 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5397 | /* | |
5398 | * TODO: Cleanup of inserted chunk root in case of | |
5399 | * failure. | |
5400 | */ | |
2ff7e61e | 5401 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5402 | } |
1abe9b8a | 5403 | |
6df9a95e | 5404 | out: |
0b86a832 | 5405 | kfree(chunk); |
6df9a95e | 5406 | free_extent_map(em); |
4ed1d16e | 5407 | return ret; |
2b82032c | 5408 | } |
0b86a832 | 5409 | |
2b82032c YZ |
5410 | /* |
5411 | * Chunk allocation falls into two parts. The first part does works | |
5412 | * that make the new allocated chunk useable, but not do any operation | |
5413 | * that modifies the chunk tree. The second part does the works that | |
5414 | * require modifying the chunk tree. This division is important for the | |
5415 | * bootstrap process of adding storage to a seed btrfs. | |
5416 | */ | |
c216b203 | 5417 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5418 | { |
5419 | u64 chunk_offset; | |
2b82032c | 5420 | |
c216b203 NB |
5421 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5422 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5423 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5424 | } |
5425 | ||
d397712b | 5426 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5427 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5428 | { |
5429 | u64 chunk_offset; | |
5430 | u64 sys_chunk_offset; | |
2b82032c | 5431 | u64 alloc_profile; |
2b82032c YZ |
5432 | int ret; |
5433 | ||
6df9a95e | 5434 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5435 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5436 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5437 | if (ret) |
5438 | return ret; | |
2b82032c | 5439 | |
0b246afa | 5440 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5441 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5442 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5443 | return ret; |
2b82032c YZ |
5444 | } |
5445 | ||
d20983b4 MX |
5446 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5447 | { | |
5448 | int max_errors; | |
5449 | ||
5450 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5451 | BTRFS_BLOCK_GROUP_RAID10 | | |
5452 | BTRFS_BLOCK_GROUP_RAID5 | | |
5453 | BTRFS_BLOCK_GROUP_DUP)) { | |
5454 | max_errors = 1; | |
5455 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5456 | max_errors = 2; | |
5457 | } else { | |
5458 | max_errors = 0; | |
005d6427 | 5459 | } |
2b82032c | 5460 | |
d20983b4 | 5461 | return max_errors; |
2b82032c YZ |
5462 | } |
5463 | ||
2ff7e61e | 5464 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5465 | { |
5466 | struct extent_map *em; | |
5467 | struct map_lookup *map; | |
2b82032c | 5468 | int readonly = 0; |
d20983b4 | 5469 | int miss_ndevs = 0; |
2b82032c YZ |
5470 | int i; |
5471 | ||
60ca842e | 5472 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5473 | if (IS_ERR(em)) |
2b82032c YZ |
5474 | return 1; |
5475 | ||
95617d69 | 5476 | map = em->map_lookup; |
2b82032c | 5477 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5478 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5479 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5480 | miss_ndevs++; |
5481 | continue; | |
5482 | } | |
ebbede42 AJ |
5483 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5484 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5485 | readonly = 1; |
d20983b4 | 5486 | goto end; |
2b82032c YZ |
5487 | } |
5488 | } | |
d20983b4 MX |
5489 | |
5490 | /* | |
5491 | * If the number of missing devices is larger than max errors, | |
5492 | * we can not write the data into that chunk successfully, so | |
5493 | * set it readonly. | |
5494 | */ | |
5495 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5496 | readonly = 1; | |
5497 | end: | |
0b86a832 | 5498 | free_extent_map(em); |
2b82032c | 5499 | return readonly; |
0b86a832 CM |
5500 | } |
5501 | ||
5502 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5503 | { | |
a8067e02 | 5504 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5505 | } |
5506 | ||
5507 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5508 | { | |
5509 | struct extent_map *em; | |
5510 | ||
d397712b | 5511 | while (1) { |
890871be | 5512 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5513 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5514 | if (em) | |
5515 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5516 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5517 | if (!em) |
5518 | break; | |
0b86a832 CM |
5519 | /* once for us */ |
5520 | free_extent_map(em); | |
5521 | /* once for the tree */ | |
5522 | free_extent_map(em); | |
5523 | } | |
5524 | } | |
5525 | ||
5d964051 | 5526 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5527 | { |
5528 | struct extent_map *em; | |
5529 | struct map_lookup *map; | |
f188591e CM |
5530 | int ret; |
5531 | ||
60ca842e | 5532 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5533 | if (IS_ERR(em)) |
5534 | /* | |
5535 | * We could return errors for these cases, but that could get | |
5536 | * ugly and we'd probably do the same thing which is just not do | |
5537 | * anything else and exit, so return 1 so the callers don't try | |
5538 | * to use other copies. | |
5539 | */ | |
fb7669b5 | 5540 | return 1; |
fb7669b5 | 5541 | |
95617d69 | 5542 | map = em->map_lookup; |
f188591e CM |
5543 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5544 | ret = map->num_stripes; | |
321aecc6 CM |
5545 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5546 | ret = map->sub_stripes; | |
53b381b3 DW |
5547 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5548 | ret = 2; | |
5549 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5550 | /* |
5551 | * There could be two corrupted data stripes, we need | |
5552 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5553 | * |
8810f751 LB |
5554 | * Fail a stripe at a time on every retry except the |
5555 | * stripe under reconstruction. | |
5556 | */ | |
5557 | ret = map->num_stripes; | |
f188591e CM |
5558 | else |
5559 | ret = 1; | |
5560 | free_extent_map(em); | |
ad6d620e | 5561 | |
cb5583dd | 5562 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5563 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5564 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5565 | ret++; |
cb5583dd | 5566 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5567 | |
f188591e CM |
5568 | return ret; |
5569 | } | |
5570 | ||
2ff7e61e | 5571 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5572 | u64 logical) |
5573 | { | |
5574 | struct extent_map *em; | |
5575 | struct map_lookup *map; | |
0b246afa | 5576 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5577 | |
60ca842e | 5578 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5579 | |
69f03f13 NB |
5580 | if (!WARN_ON(IS_ERR(em))) { |
5581 | map = em->map_lookup; | |
5582 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5583 | len = map->stripe_len * nr_data_stripes(map); | |
5584 | free_extent_map(em); | |
5585 | } | |
53b381b3 DW |
5586 | return len; |
5587 | } | |
5588 | ||
e4ff5fb5 | 5589 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5590 | { |
5591 | struct extent_map *em; | |
5592 | struct map_lookup *map; | |
53b381b3 DW |
5593 | int ret = 0; |
5594 | ||
60ca842e | 5595 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5596 | |
69f03f13 NB |
5597 | if(!WARN_ON(IS_ERR(em))) { |
5598 | map = em->map_lookup; | |
5599 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5600 | ret = 1; | |
5601 | free_extent_map(em); | |
5602 | } | |
53b381b3 DW |
5603 | return ret; |
5604 | } | |
5605 | ||
30d9861f | 5606 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5607 | struct map_lookup *map, int first, |
8ba0ae78 | 5608 | int dev_replace_is_ongoing) |
dfe25020 CM |
5609 | { |
5610 | int i; | |
99f92a7c | 5611 | int num_stripes; |
8ba0ae78 | 5612 | int preferred_mirror; |
30d9861f SB |
5613 | int tolerance; |
5614 | struct btrfs_device *srcdev; | |
5615 | ||
99f92a7c AJ |
5616 | ASSERT((map->type & |
5617 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5618 | ||
5619 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5620 | num_stripes = map->sub_stripes; | |
5621 | else | |
5622 | num_stripes = map->num_stripes; | |
5623 | ||
8ba0ae78 AJ |
5624 | preferred_mirror = first + current->pid % num_stripes; |
5625 | ||
30d9861f SB |
5626 | if (dev_replace_is_ongoing && |
5627 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5628 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5629 | srcdev = fs_info->dev_replace.srcdev; | |
5630 | else | |
5631 | srcdev = NULL; | |
5632 | ||
5633 | /* | |
5634 | * try to avoid the drive that is the source drive for a | |
5635 | * dev-replace procedure, only choose it if no other non-missing | |
5636 | * mirror is available | |
5637 | */ | |
5638 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5639 | if (map->stripes[preferred_mirror].dev->bdev && |
5640 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5641 | return preferred_mirror; | |
99f92a7c | 5642 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5643 | if (map->stripes[i].dev->bdev && |
5644 | (tolerance || map->stripes[i].dev != srcdev)) | |
5645 | return i; | |
5646 | } | |
dfe25020 | 5647 | } |
30d9861f | 5648 | |
dfe25020 CM |
5649 | /* we couldn't find one that doesn't fail. Just return something |
5650 | * and the io error handling code will clean up eventually | |
5651 | */ | |
8ba0ae78 | 5652 | return preferred_mirror; |
dfe25020 CM |
5653 | } |
5654 | ||
53b381b3 DW |
5655 | static inline int parity_smaller(u64 a, u64 b) |
5656 | { | |
5657 | return a > b; | |
5658 | } | |
5659 | ||
5660 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5661 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5662 | { |
5663 | struct btrfs_bio_stripe s; | |
5664 | int i; | |
5665 | u64 l; | |
5666 | int again = 1; | |
5667 | ||
5668 | while (again) { | |
5669 | again = 0; | |
cc7539ed | 5670 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5671 | if (parity_smaller(bbio->raid_map[i], |
5672 | bbio->raid_map[i+1])) { | |
53b381b3 | 5673 | s = bbio->stripes[i]; |
8e5cfb55 | 5674 | l = bbio->raid_map[i]; |
53b381b3 | 5675 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5676 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5677 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5678 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5679 | |
53b381b3 DW |
5680 | again = 1; |
5681 | } | |
5682 | } | |
5683 | } | |
5684 | } | |
5685 | ||
6e9606d2 ZL |
5686 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5687 | { | |
5688 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5689 | /* the size of the btrfs_bio */ |
6e9606d2 | 5690 | sizeof(struct btrfs_bio) + |
e57cf21e | 5691 | /* plus the variable array for the stripes */ |
6e9606d2 | 5692 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5693 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5694 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5695 | /* |
5696 | * plus the raid_map, which includes both the tgt dev | |
5697 | * and the stripes | |
5698 | */ | |
5699 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5700 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5701 | |
5702 | atomic_set(&bbio->error, 0); | |
140475ae | 5703 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5704 | |
5705 | return bbio; | |
5706 | } | |
5707 | ||
5708 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5709 | { | |
140475ae ER |
5710 | WARN_ON(!refcount_read(&bbio->refs)); |
5711 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5712 | } |
5713 | ||
5714 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5715 | { | |
5716 | if (!bbio) | |
5717 | return; | |
140475ae | 5718 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5719 | kfree(bbio); |
5720 | } | |
5721 | ||
0b3d4cd3 LB |
5722 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5723 | /* | |
5724 | * Please note that, discard won't be sent to target device of device | |
5725 | * replace. | |
5726 | */ | |
5727 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5728 | u64 logical, u64 length, | |
5729 | struct btrfs_bio **bbio_ret) | |
5730 | { | |
5731 | struct extent_map *em; | |
5732 | struct map_lookup *map; | |
5733 | struct btrfs_bio *bbio; | |
5734 | u64 offset; | |
5735 | u64 stripe_nr; | |
5736 | u64 stripe_nr_end; | |
5737 | u64 stripe_end_offset; | |
5738 | u64 stripe_cnt; | |
5739 | u64 stripe_len; | |
5740 | u64 stripe_offset; | |
5741 | u64 num_stripes; | |
5742 | u32 stripe_index; | |
5743 | u32 factor = 0; | |
5744 | u32 sub_stripes = 0; | |
5745 | u64 stripes_per_dev = 0; | |
5746 | u32 remaining_stripes = 0; | |
5747 | u32 last_stripe = 0; | |
5748 | int ret = 0; | |
5749 | int i; | |
5750 | ||
5751 | /* discard always return a bbio */ | |
5752 | ASSERT(bbio_ret); | |
5753 | ||
60ca842e | 5754 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5755 | if (IS_ERR(em)) |
5756 | return PTR_ERR(em); | |
5757 | ||
5758 | map = em->map_lookup; | |
5759 | /* we don't discard raid56 yet */ | |
5760 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5761 | ret = -EOPNOTSUPP; | |
5762 | goto out; | |
5763 | } | |
5764 | ||
5765 | offset = logical - em->start; | |
5766 | length = min_t(u64, em->len - offset, length); | |
5767 | ||
5768 | stripe_len = map->stripe_len; | |
5769 | /* | |
5770 | * stripe_nr counts the total number of stripes we have to stride | |
5771 | * to get to this block | |
5772 | */ | |
5773 | stripe_nr = div64_u64(offset, stripe_len); | |
5774 | ||
5775 | /* stripe_offset is the offset of this block in its stripe */ | |
5776 | stripe_offset = offset - stripe_nr * stripe_len; | |
5777 | ||
5778 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5779 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5780 | stripe_cnt = stripe_nr_end - stripe_nr; |
5781 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5782 | (offset + length); | |
5783 | /* | |
5784 | * after this, stripe_nr is the number of stripes on this | |
5785 | * device we have to walk to find the data, and stripe_index is | |
5786 | * the number of our device in the stripe array | |
5787 | */ | |
5788 | num_stripes = 1; | |
5789 | stripe_index = 0; | |
5790 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5791 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5792 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5793 | sub_stripes = 1; | |
5794 | else | |
5795 | sub_stripes = map->sub_stripes; | |
5796 | ||
5797 | factor = map->num_stripes / sub_stripes; | |
5798 | num_stripes = min_t(u64, map->num_stripes, | |
5799 | sub_stripes * stripe_cnt); | |
5800 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5801 | stripe_index *= sub_stripes; | |
5802 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5803 | &remaining_stripes); | |
5804 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5805 | last_stripe *= sub_stripes; | |
5806 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5807 | BTRFS_BLOCK_GROUP_DUP)) { | |
5808 | num_stripes = map->num_stripes; | |
5809 | } else { | |
5810 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5811 | &stripe_index); | |
5812 | } | |
5813 | ||
5814 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5815 | if (!bbio) { | |
5816 | ret = -ENOMEM; | |
5817 | goto out; | |
5818 | } | |
5819 | ||
5820 | for (i = 0; i < num_stripes; i++) { | |
5821 | bbio->stripes[i].physical = | |
5822 | map->stripes[stripe_index].physical + | |
5823 | stripe_offset + stripe_nr * map->stripe_len; | |
5824 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5825 | ||
5826 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5827 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5828 | bbio->stripes[i].length = stripes_per_dev * | |
5829 | map->stripe_len; | |
5830 | ||
5831 | if (i / sub_stripes < remaining_stripes) | |
5832 | bbio->stripes[i].length += | |
5833 | map->stripe_len; | |
5834 | ||
5835 | /* | |
5836 | * Special for the first stripe and | |
5837 | * the last stripe: | |
5838 | * | |
5839 | * |-------|...|-------| | |
5840 | * |----------| | |
5841 | * off end_off | |
5842 | */ | |
5843 | if (i < sub_stripes) | |
5844 | bbio->stripes[i].length -= | |
5845 | stripe_offset; | |
5846 | ||
5847 | if (stripe_index >= last_stripe && | |
5848 | stripe_index <= (last_stripe + | |
5849 | sub_stripes - 1)) | |
5850 | bbio->stripes[i].length -= | |
5851 | stripe_end_offset; | |
5852 | ||
5853 | if (i == sub_stripes - 1) | |
5854 | stripe_offset = 0; | |
5855 | } else { | |
5856 | bbio->stripes[i].length = length; | |
5857 | } | |
5858 | ||
5859 | stripe_index++; | |
5860 | if (stripe_index == map->num_stripes) { | |
5861 | stripe_index = 0; | |
5862 | stripe_nr++; | |
5863 | } | |
5864 | } | |
5865 | ||
5866 | *bbio_ret = bbio; | |
5867 | bbio->map_type = map->type; | |
5868 | bbio->num_stripes = num_stripes; | |
5869 | out: | |
5870 | free_extent_map(em); | |
5871 | return ret; | |
5872 | } | |
5873 | ||
5ab56090 LB |
5874 | /* |
5875 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5876 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5877 | * left cursor can also be read from the target drive. | |
5878 | * | |
5879 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5880 | * array of stripes. | |
5881 | * For READ, it also needs to be supported using the same mirror number. | |
5882 | * | |
5883 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5884 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5885 | * case. | |
5886 | */ | |
5887 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5888 | u64 logical, u64 length, | |
5889 | u64 srcdev_devid, int *mirror_num, | |
5890 | u64 *physical) | |
5891 | { | |
5892 | struct btrfs_bio *bbio = NULL; | |
5893 | int num_stripes; | |
5894 | int index_srcdev = 0; | |
5895 | int found = 0; | |
5896 | u64 physical_of_found = 0; | |
5897 | int i; | |
5898 | int ret = 0; | |
5899 | ||
5900 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5901 | logical, &length, &bbio, 0, 0); | |
5902 | if (ret) { | |
5903 | ASSERT(bbio == NULL); | |
5904 | return ret; | |
5905 | } | |
5906 | ||
5907 | num_stripes = bbio->num_stripes; | |
5908 | if (*mirror_num > num_stripes) { | |
5909 | /* | |
5910 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5911 | * that means that the requested area is not left of the left | |
5912 | * cursor | |
5913 | */ | |
5914 | btrfs_put_bbio(bbio); | |
5915 | return -EIO; | |
5916 | } | |
5917 | ||
5918 | /* | |
5919 | * process the rest of the function using the mirror_num of the source | |
5920 | * drive. Therefore look it up first. At the end, patch the device | |
5921 | * pointer to the one of the target drive. | |
5922 | */ | |
5923 | for (i = 0; i < num_stripes; i++) { | |
5924 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5925 | continue; | |
5926 | ||
5927 | /* | |
5928 | * In case of DUP, in order to keep it simple, only add the | |
5929 | * mirror with the lowest physical address | |
5930 | */ | |
5931 | if (found && | |
5932 | physical_of_found <= bbio->stripes[i].physical) | |
5933 | continue; | |
5934 | ||
5935 | index_srcdev = i; | |
5936 | found = 1; | |
5937 | physical_of_found = bbio->stripes[i].physical; | |
5938 | } | |
5939 | ||
5940 | btrfs_put_bbio(bbio); | |
5941 | ||
5942 | ASSERT(found); | |
5943 | if (!found) | |
5944 | return -EIO; | |
5945 | ||
5946 | *mirror_num = index_srcdev + 1; | |
5947 | *physical = physical_of_found; | |
5948 | return ret; | |
5949 | } | |
5950 | ||
73c0f228 LB |
5951 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5952 | struct btrfs_bio **bbio_ret, | |
5953 | struct btrfs_dev_replace *dev_replace, | |
5954 | int *num_stripes_ret, int *max_errors_ret) | |
5955 | { | |
5956 | struct btrfs_bio *bbio = *bbio_ret; | |
5957 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5958 | int tgtdev_indexes = 0; | |
5959 | int num_stripes = *num_stripes_ret; | |
5960 | int max_errors = *max_errors_ret; | |
5961 | int i; | |
5962 | ||
5963 | if (op == BTRFS_MAP_WRITE) { | |
5964 | int index_where_to_add; | |
5965 | ||
5966 | /* | |
5967 | * duplicate the write operations while the dev replace | |
5968 | * procedure is running. Since the copying of the old disk to | |
5969 | * the new disk takes place at run time while the filesystem is | |
5970 | * mounted writable, the regular write operations to the old | |
5971 | * disk have to be duplicated to go to the new disk as well. | |
5972 | * | |
5973 | * Note that device->missing is handled by the caller, and that | |
5974 | * the write to the old disk is already set up in the stripes | |
5975 | * array. | |
5976 | */ | |
5977 | index_where_to_add = num_stripes; | |
5978 | for (i = 0; i < num_stripes; i++) { | |
5979 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5980 | /* write to new disk, too */ | |
5981 | struct btrfs_bio_stripe *new = | |
5982 | bbio->stripes + index_where_to_add; | |
5983 | struct btrfs_bio_stripe *old = | |
5984 | bbio->stripes + i; | |
5985 | ||
5986 | new->physical = old->physical; | |
5987 | new->length = old->length; | |
5988 | new->dev = dev_replace->tgtdev; | |
5989 | bbio->tgtdev_map[i] = index_where_to_add; | |
5990 | index_where_to_add++; | |
5991 | max_errors++; | |
5992 | tgtdev_indexes++; | |
5993 | } | |
5994 | } | |
5995 | num_stripes = index_where_to_add; | |
5996 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5997 | int index_srcdev = 0; | |
5998 | int found = 0; | |
5999 | u64 physical_of_found = 0; | |
6000 | ||
6001 | /* | |
6002 | * During the dev-replace procedure, the target drive can also | |
6003 | * be used to read data in case it is needed to repair a corrupt | |
6004 | * block elsewhere. This is possible if the requested area is | |
6005 | * left of the left cursor. In this area, the target drive is a | |
6006 | * full copy of the source drive. | |
6007 | */ | |
6008 | for (i = 0; i < num_stripes; i++) { | |
6009 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
6010 | /* | |
6011 | * In case of DUP, in order to keep it simple, | |
6012 | * only add the mirror with the lowest physical | |
6013 | * address | |
6014 | */ | |
6015 | if (found && | |
6016 | physical_of_found <= | |
6017 | bbio->stripes[i].physical) | |
6018 | continue; | |
6019 | index_srcdev = i; | |
6020 | found = 1; | |
6021 | physical_of_found = bbio->stripes[i].physical; | |
6022 | } | |
6023 | } | |
6024 | if (found) { | |
6025 | struct btrfs_bio_stripe *tgtdev_stripe = | |
6026 | bbio->stripes + num_stripes; | |
6027 | ||
6028 | tgtdev_stripe->physical = physical_of_found; | |
6029 | tgtdev_stripe->length = | |
6030 | bbio->stripes[index_srcdev].length; | |
6031 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
6032 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
6033 | ||
6034 | tgtdev_indexes++; | |
6035 | num_stripes++; | |
6036 | } | |
6037 | } | |
6038 | ||
6039 | *num_stripes_ret = num_stripes; | |
6040 | *max_errors_ret = max_errors; | |
6041 | bbio->num_tgtdevs = tgtdev_indexes; | |
6042 | *bbio_ret = bbio; | |
6043 | } | |
6044 | ||
2b19a1fe LB |
6045 | static bool need_full_stripe(enum btrfs_map_op op) |
6046 | { | |
6047 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
6048 | } | |
6049 | ||
cf8cddd3 CH |
6050 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
6051 | enum btrfs_map_op op, | |
f2d8d74d | 6052 | u64 logical, u64 *length, |
a1d3c478 | 6053 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6054 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6055 | { |
6056 | struct extent_map *em; | |
6057 | struct map_lookup *map; | |
0b86a832 | 6058 | u64 offset; |
593060d7 CM |
6059 | u64 stripe_offset; |
6060 | u64 stripe_nr; | |
53b381b3 | 6061 | u64 stripe_len; |
9d644a62 | 6062 | u32 stripe_index; |
cea9e445 | 6063 | int i; |
de11cc12 | 6064 | int ret = 0; |
f2d8d74d | 6065 | int num_stripes; |
a236aed1 | 6066 | int max_errors = 0; |
2c8cdd6e | 6067 | int tgtdev_indexes = 0; |
a1d3c478 | 6068 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6069 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6070 | int dev_replace_is_ongoing = 0; | |
6071 | int num_alloc_stripes; | |
ad6d620e SB |
6072 | int patch_the_first_stripe_for_dev_replace = 0; |
6073 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6074 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 6075 | |
0b3d4cd3 LB |
6076 | if (op == BTRFS_MAP_DISCARD) |
6077 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6078 | *length, bbio_ret); | |
6079 | ||
60ca842e | 6080 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
6081 | if (IS_ERR(em)) |
6082 | return PTR_ERR(em); | |
0b86a832 | 6083 | |
95617d69 | 6084 | map = em->map_lookup; |
0b86a832 | 6085 | offset = logical - em->start; |
593060d7 | 6086 | |
53b381b3 | 6087 | stripe_len = map->stripe_len; |
593060d7 CM |
6088 | stripe_nr = offset; |
6089 | /* | |
6090 | * stripe_nr counts the total number of stripes we have to stride | |
6091 | * to get to this block | |
6092 | */ | |
47c5713f | 6093 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 6094 | |
53b381b3 | 6095 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 6096 | if (offset < stripe_offset) { |
5d163e0e JM |
6097 | btrfs_crit(fs_info, |
6098 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
6099 | stripe_offset, offset, em->start, logical, |
6100 | stripe_len); | |
6101 | free_extent_map(em); | |
6102 | return -EINVAL; | |
6103 | } | |
593060d7 CM |
6104 | |
6105 | /* stripe_offset is the offset of this block in its stripe*/ | |
6106 | stripe_offset = offset - stripe_offset; | |
6107 | ||
53b381b3 | 6108 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 6109 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
6110 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
6111 | raid56_full_stripe_start = offset; | |
6112 | ||
6113 | /* allow a write of a full stripe, but make sure we don't | |
6114 | * allow straddling of stripes | |
6115 | */ | |
47c5713f DS |
6116 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
6117 | full_stripe_len); | |
53b381b3 DW |
6118 | raid56_full_stripe_start *= full_stripe_len; |
6119 | } | |
6120 | ||
0b3d4cd3 | 6121 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
6122 | u64 max_len; |
6123 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
6124 | For other RAID types and for RAID[56] reads, just allow a single | |
6125 | stripe (on a single disk). */ | |
ffe2d203 | 6126 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 6127 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
6128 | max_len = stripe_len * nr_data_stripes(map) - |
6129 | (offset - raid56_full_stripe_start); | |
6130 | } else { | |
6131 | /* we limit the length of each bio to what fits in a stripe */ | |
6132 | max_len = stripe_len - stripe_offset; | |
6133 | } | |
6134 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
6135 | } else { |
6136 | *length = em->len - offset; | |
6137 | } | |
f2d8d74d | 6138 | |
da12fe54 NB |
6139 | /* |
6140 | * This is for when we're called from btrfs_bio_fits_in_stripe and all | |
6141 | * it cares about is the length | |
6142 | */ | |
a1d3c478 | 6143 | if (!bbio_ret) |
cea9e445 CM |
6144 | goto out; |
6145 | ||
cb5583dd | 6146 | down_read(&dev_replace->rwsem); |
472262f3 | 6147 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6148 | /* |
6149 | * Hold the semaphore for read during the whole operation, write is | |
6150 | * requested at commit time but must wait. | |
6151 | */ | |
472262f3 | 6152 | if (!dev_replace_is_ongoing) |
cb5583dd | 6153 | up_read(&dev_replace->rwsem); |
472262f3 | 6154 | |
ad6d620e | 6155 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6156 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6157 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6158 | dev_replace->srcdev->devid, | |
6159 | &mirror_num, | |
6160 | &physical_to_patch_in_first_stripe); | |
6161 | if (ret) | |
ad6d620e | 6162 | goto out; |
5ab56090 LB |
6163 | else |
6164 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6165 | } else if (mirror_num > map->num_stripes) { |
6166 | mirror_num = 0; | |
6167 | } | |
6168 | ||
f2d8d74d | 6169 | num_stripes = 1; |
cea9e445 | 6170 | stripe_index = 0; |
fce3bb9a | 6171 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6172 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6173 | &stripe_index); | |
de483734 | 6174 | if (!need_full_stripe(op)) |
28e1cc7d | 6175 | mirror_num = 1; |
fce3bb9a | 6176 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 6177 | if (need_full_stripe(op)) |
f2d8d74d | 6178 | num_stripes = map->num_stripes; |
2fff734f | 6179 | else if (mirror_num) |
f188591e | 6180 | stripe_index = mirror_num - 1; |
dfe25020 | 6181 | else { |
30d9861f | 6182 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6183 | dev_replace_is_ongoing); |
a1d3c478 | 6184 | mirror_num = stripe_index + 1; |
dfe25020 | 6185 | } |
2fff734f | 6186 | |
611f0e00 | 6187 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6188 | if (need_full_stripe(op)) { |
f2d8d74d | 6189 | num_stripes = map->num_stripes; |
a1d3c478 | 6190 | } else if (mirror_num) { |
f188591e | 6191 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6192 | } else { |
6193 | mirror_num = 1; | |
6194 | } | |
2fff734f | 6195 | |
321aecc6 | 6196 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6197 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6198 | |
47c5713f | 6199 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6200 | stripe_index *= map->sub_stripes; |
6201 | ||
de483734 | 6202 | if (need_full_stripe(op)) |
f2d8d74d | 6203 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6204 | else if (mirror_num) |
6205 | stripe_index += mirror_num - 1; | |
dfe25020 | 6206 | else { |
3e74317a | 6207 | int old_stripe_index = stripe_index; |
30d9861f SB |
6208 | stripe_index = find_live_mirror(fs_info, map, |
6209 | stripe_index, | |
30d9861f | 6210 | dev_replace_is_ongoing); |
3e74317a | 6211 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6212 | } |
53b381b3 | 6213 | |
ffe2d203 | 6214 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6215 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6216 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6217 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 6218 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
6219 | |
6220 | /* RAID[56] write or recovery. Return all stripes */ | |
6221 | num_stripes = map->num_stripes; | |
6222 | max_errors = nr_parity_stripes(map); | |
6223 | ||
53b381b3 DW |
6224 | *length = map->stripe_len; |
6225 | stripe_index = 0; | |
6226 | stripe_offset = 0; | |
6227 | } else { | |
6228 | /* | |
6229 | * Mirror #0 or #1 means the original data block. | |
6230 | * Mirror #2 is RAID5 parity block. | |
6231 | * Mirror #3 is RAID6 Q block. | |
6232 | */ | |
47c5713f DS |
6233 | stripe_nr = div_u64_rem(stripe_nr, |
6234 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
6235 | if (mirror_num > 1) |
6236 | stripe_index = nr_data_stripes(map) + | |
6237 | mirror_num - 2; | |
6238 | ||
6239 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6240 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6241 | &stripe_index); | |
de483734 | 6242 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6243 | mirror_num = 1; |
53b381b3 | 6244 | } |
8790d502 CM |
6245 | } else { |
6246 | /* | |
47c5713f DS |
6247 | * after this, stripe_nr is the number of stripes on this |
6248 | * device we have to walk to find the data, and stripe_index is | |
6249 | * the number of our device in the stripe array | |
8790d502 | 6250 | */ |
47c5713f DS |
6251 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6252 | &stripe_index); | |
a1d3c478 | 6253 | mirror_num = stripe_index + 1; |
8790d502 | 6254 | } |
e042d1ec | 6255 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6256 | btrfs_crit(fs_info, |
6257 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6258 | stripe_index, map->num_stripes); |
6259 | ret = -EINVAL; | |
6260 | goto out; | |
6261 | } | |
cea9e445 | 6262 | |
472262f3 | 6263 | num_alloc_stripes = num_stripes; |
6fad823f | 6264 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6265 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6266 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6267 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6268 | num_alloc_stripes++; |
2c8cdd6e | 6269 | tgtdev_indexes = num_stripes; |
ad6d620e | 6270 | } |
2c8cdd6e | 6271 | |
6e9606d2 | 6272 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6273 | if (!bbio) { |
6274 | ret = -ENOMEM; | |
6275 | goto out; | |
6276 | } | |
6fad823f | 6277 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6278 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6279 | |
8e5cfb55 | 6280 | /* build raid_map */ |
2b19a1fe LB |
6281 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6282 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6283 | u64 tmp; |
9d644a62 | 6284 | unsigned rot; |
8e5cfb55 ZL |
6285 | |
6286 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6287 | sizeof(struct btrfs_bio_stripe) * | |
6288 | num_alloc_stripes + | |
6289 | sizeof(int) * tgtdev_indexes); | |
6290 | ||
6291 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6292 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6293 | |
6294 | /* Fill in the logical address of each stripe */ | |
6295 | tmp = stripe_nr * nr_data_stripes(map); | |
6296 | for (i = 0; i < nr_data_stripes(map); i++) | |
6297 | bbio->raid_map[(i+rot) % num_stripes] = | |
6298 | em->start + (tmp + i) * map->stripe_len; | |
6299 | ||
6300 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6301 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6302 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6303 | RAID6_Q_STRIPE; | |
6304 | } | |
6305 | ||
b89203f7 | 6306 | |
0b3d4cd3 LB |
6307 | for (i = 0; i < num_stripes; i++) { |
6308 | bbio->stripes[i].physical = | |
6309 | map->stripes[stripe_index].physical + | |
6310 | stripe_offset + | |
6311 | stripe_nr * map->stripe_len; | |
6312 | bbio->stripes[i].dev = | |
6313 | map->stripes[stripe_index].dev; | |
6314 | stripe_index++; | |
593060d7 | 6315 | } |
de11cc12 | 6316 | |
2b19a1fe | 6317 | if (need_full_stripe(op)) |
d20983b4 | 6318 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6319 | |
8e5cfb55 ZL |
6320 | if (bbio->raid_map) |
6321 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6322 | |
73c0f228 | 6323 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6324 | need_full_stripe(op)) { |
73c0f228 LB |
6325 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6326 | &max_errors); | |
472262f3 SB |
6327 | } |
6328 | ||
de11cc12 | 6329 | *bbio_ret = bbio; |
10f11900 | 6330 | bbio->map_type = map->type; |
de11cc12 LZ |
6331 | bbio->num_stripes = num_stripes; |
6332 | bbio->max_errors = max_errors; | |
6333 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6334 | |
6335 | /* | |
6336 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6337 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6338 | * available as a mirror | |
6339 | */ | |
6340 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6341 | WARN_ON(num_stripes > 1); | |
6342 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6343 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6344 | bbio->mirror_num = map->num_stripes + 1; | |
6345 | } | |
cea9e445 | 6346 | out: |
73beece9 | 6347 | if (dev_replace_is_ongoing) { |
53176dde DS |
6348 | lockdep_assert_held(&dev_replace->rwsem); |
6349 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6350 | up_read(&dev_replace->rwsem); |
73beece9 | 6351 | } |
0b86a832 | 6352 | free_extent_map(em); |
de11cc12 | 6353 | return ret; |
0b86a832 CM |
6354 | } |
6355 | ||
cf8cddd3 | 6356 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6357 | u64 logical, u64 *length, |
a1d3c478 | 6358 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6359 | { |
b3d3fa51 | 6360 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6361 | mirror_num, 0); |
f2d8d74d CM |
6362 | } |
6363 | ||
af8e2d1d | 6364 | /* For Scrub/replace */ |
cf8cddd3 | 6365 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6366 | u64 logical, u64 *length, |
825ad4c9 | 6367 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6368 | { |
825ad4c9 | 6369 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6370 | } |
6371 | ||
63a9c7b9 NB |
6372 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6373 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6374 | { |
a512bbf8 YZ |
6375 | struct extent_map *em; |
6376 | struct map_lookup *map; | |
6377 | u64 *buf; | |
6378 | u64 bytenr; | |
6379 | u64 length; | |
6380 | u64 stripe_nr; | |
53b381b3 | 6381 | u64 rmap_len; |
a512bbf8 YZ |
6382 | int i, j, nr = 0; |
6383 | ||
60ca842e | 6384 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6385 | if (IS_ERR(em)) |
835d974f | 6386 | return -EIO; |
835d974f | 6387 | |
95617d69 | 6388 | map = em->map_lookup; |
a512bbf8 | 6389 | length = em->len; |
53b381b3 DW |
6390 | rmap_len = map->stripe_len; |
6391 | ||
a512bbf8 | 6392 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6393 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6394 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6395 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6396 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6397 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6398 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6399 | } | |
a512bbf8 | 6400 | |
31e818fe | 6401 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6402 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6403 | |
6404 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6405 | if (map->stripes[i].physical > physical || |
6406 | map->stripes[i].physical + length <= physical) | |
6407 | continue; | |
6408 | ||
6409 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6410 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6411 | |
6412 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6413 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6414 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6415 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6416 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6417 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6418 | * Alternatively, just use rmap_len below instead of | |
6419 | * map->stripe_len */ | |
6420 | ||
6421 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6422 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6423 | for (j = 0; j < nr; j++) { |
6424 | if (buf[j] == bytenr) | |
6425 | break; | |
6426 | } | |
934d375b CM |
6427 | if (j == nr) { |
6428 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6429 | buf[nr++] = bytenr; |
934d375b | 6430 | } |
a512bbf8 YZ |
6431 | } |
6432 | ||
a512bbf8 YZ |
6433 | *logical = buf; |
6434 | *naddrs = nr; | |
53b381b3 | 6435 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6436 | |
6437 | free_extent_map(em); | |
6438 | return 0; | |
f2d8d74d CM |
6439 | } |
6440 | ||
4246a0b6 | 6441 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6442 | { |
326e1dbb MS |
6443 | bio->bi_private = bbio->private; |
6444 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6445 | bio_endio(bio); |
326e1dbb | 6446 | |
6e9606d2 | 6447 | btrfs_put_bbio(bbio); |
8408c716 MX |
6448 | } |
6449 | ||
4246a0b6 | 6450 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6451 | { |
9be3395b | 6452 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6453 | int is_orig_bio = 0; |
8790d502 | 6454 | |
4e4cbee9 | 6455 | if (bio->bi_status) { |
a1d3c478 | 6456 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6457 | if (bio->bi_status == BLK_STS_IOERR || |
6458 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6459 | unsigned int stripe_index = |
9be3395b | 6460 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6461 | struct btrfs_device *dev; |
442a4f63 SB |
6462 | |
6463 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6464 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6465 | if (dev->bdev) { |
37226b21 | 6466 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6467 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6468 | BTRFS_DEV_STAT_WRITE_ERRS); |
6469 | else | |
1cb34c8e | 6470 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6471 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6472 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6473 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6474 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6475 | } |
442a4f63 SB |
6476 | } |
6477 | } | |
8790d502 | 6478 | |
a1d3c478 | 6479 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6480 | is_orig_bio = 1; |
6481 | ||
c404e0dc MX |
6482 | btrfs_bio_counter_dec(bbio->fs_info); |
6483 | ||
a1d3c478 | 6484 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6485 | if (!is_orig_bio) { |
6486 | bio_put(bio); | |
a1d3c478 | 6487 | bio = bbio->orig_bio; |
7d2b4daa | 6488 | } |
c7b22bb1 | 6489 | |
9be3395b | 6490 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6491 | /* only send an error to the higher layers if it is |
53b381b3 | 6492 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6493 | */ |
a1d3c478 | 6494 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6495 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6496 | } else { |
1259ab75 CM |
6497 | /* |
6498 | * this bio is actually up to date, we didn't | |
6499 | * go over the max number of errors | |
6500 | */ | |
2dbe0c77 | 6501 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6502 | } |
c55f1396 | 6503 | |
4246a0b6 | 6504 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6505 | } else if (!is_orig_bio) { |
8790d502 CM |
6506 | bio_put(bio); |
6507 | } | |
8790d502 CM |
6508 | } |
6509 | ||
8b712842 CM |
6510 | /* |
6511 | * see run_scheduled_bios for a description of why bios are collected for | |
6512 | * async submit. | |
6513 | * | |
6514 | * This will add one bio to the pending list for a device and make sure | |
6515 | * the work struct is scheduled. | |
6516 | */ | |
2ff7e61e | 6517 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6518 | struct bio *bio) |
8b712842 | 6519 | { |
0b246afa | 6520 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6521 | int should_queue = 1; |
ffbd517d | 6522 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6523 | |
6524 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6525 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6526 | btrfsic_submit_bio(bio); |
143bede5 | 6527 | return; |
8b712842 CM |
6528 | } |
6529 | ||
492bb6de | 6530 | WARN_ON(bio->bi_next); |
8b712842 | 6531 | bio->bi_next = NULL; |
8b712842 CM |
6532 | |
6533 | spin_lock(&device->io_lock); | |
67f055c7 | 6534 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6535 | pending_bios = &device->pending_sync_bios; |
6536 | else | |
6537 | pending_bios = &device->pending_bios; | |
8b712842 | 6538 | |
ffbd517d CM |
6539 | if (pending_bios->tail) |
6540 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6541 | |
ffbd517d CM |
6542 | pending_bios->tail = bio; |
6543 | if (!pending_bios->head) | |
6544 | pending_bios->head = bio; | |
8b712842 CM |
6545 | if (device->running_pending) |
6546 | should_queue = 0; | |
6547 | ||
6548 | spin_unlock(&device->io_lock); | |
6549 | ||
6550 | if (should_queue) | |
0b246afa | 6551 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6552 | } |
6553 | ||
2ff7e61e JM |
6554 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6555 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6556 | { |
6557 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6558 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6559 | |
6560 | bio->bi_private = bbio; | |
9be3395b | 6561 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6562 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6563 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6564 | btrfs_debug_in_rcu(fs_info, |
6565 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6566 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6567 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6568 | bio->bi_iter.bi_size); | |
74d46992 | 6569 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6570 | |
2ff7e61e | 6571 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6572 | |
de1ee92a | 6573 | if (async) |
2ff7e61e | 6574 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6575 | else |
4e49ea4a | 6576 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6577 | } |
6578 | ||
de1ee92a JB |
6579 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6580 | { | |
6581 | atomic_inc(&bbio->error); | |
6582 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6583 | /* Should be the original bio. */ |
8408c716 MX |
6584 | WARN_ON(bio != bbio->orig_bio); |
6585 | ||
9be3395b | 6586 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6587 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6588 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6589 | bio->bi_status = BLK_STS_IOERR; | |
6590 | else | |
6591 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6592 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6593 | } |
6594 | } | |
6595 | ||
58efbc9f OS |
6596 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6597 | int mirror_num, int async_submit) | |
0b86a832 | 6598 | { |
0b86a832 | 6599 | struct btrfs_device *dev; |
8790d502 | 6600 | struct bio *first_bio = bio; |
4f024f37 | 6601 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6602 | u64 length = 0; |
6603 | u64 map_length; | |
0b86a832 | 6604 | int ret; |
08da757d ZL |
6605 | int dev_nr; |
6606 | int total_devs; | |
a1d3c478 | 6607 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6608 | |
4f024f37 | 6609 | length = bio->bi_iter.bi_size; |
0b86a832 | 6610 | map_length = length; |
cea9e445 | 6611 | |
0b246afa | 6612 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6613 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6614 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6615 | if (ret) { |
0b246afa | 6616 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6617 | return errno_to_blk_status(ret); |
c404e0dc | 6618 | } |
cea9e445 | 6619 | |
a1d3c478 | 6620 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6621 | bbio->orig_bio = first_bio; |
6622 | bbio->private = first_bio->bi_private; | |
6623 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6624 | bbio->fs_info = fs_info; |
53b381b3 DW |
6625 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6626 | ||
ad1ba2a0 | 6627 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6628 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6629 | /* In this case, map_length has been set to the length of |
6630 | a single stripe; not the whole write */ | |
37226b21 | 6631 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6632 | ret = raid56_parity_write(fs_info, bio, bbio, |
6633 | map_length); | |
53b381b3 | 6634 | } else { |
2ff7e61e JM |
6635 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6636 | map_length, mirror_num, 1); | |
53b381b3 | 6637 | } |
4245215d | 6638 | |
0b246afa | 6639 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6640 | return errno_to_blk_status(ret); |
53b381b3 DW |
6641 | } |
6642 | ||
cea9e445 | 6643 | if (map_length < length) { |
0b246afa | 6644 | btrfs_crit(fs_info, |
5d163e0e JM |
6645 | "mapping failed logical %llu bio len %llu len %llu", |
6646 | logical, length, map_length); | |
cea9e445 CM |
6647 | BUG(); |
6648 | } | |
a1d3c478 | 6649 | |
08da757d | 6650 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6651 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6652 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6653 | &dev->dev_state) || | |
ebbede42 AJ |
6654 | (bio_op(first_bio) == REQ_OP_WRITE && |
6655 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6656 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6657 | continue; |
6658 | } | |
6659 | ||
3aa8e074 | 6660 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6661 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6662 | else |
a1d3c478 | 6663 | bio = first_bio; |
de1ee92a | 6664 | |
2ff7e61e JM |
6665 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6666 | dev_nr, async_submit); | |
8790d502 | 6667 | } |
0b246afa | 6668 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6669 | return BLK_STS_OK; |
0b86a832 CM |
6670 | } |
6671 | ||
aa1b8cd4 | 6672 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6673 | u8 *uuid, u8 *fsid) |
0b86a832 | 6674 | { |
2b82032c YZ |
6675 | struct btrfs_device *device; |
6676 | struct btrfs_fs_devices *cur_devices; | |
6677 | ||
aa1b8cd4 | 6678 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6679 | while (cur_devices) { |
6680 | if (!fsid || | |
7239ff4b | 6681 | !memcmp(cur_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
35c70103 | 6682 | device = find_device(cur_devices, devid, uuid); |
2b82032c YZ |
6683 | if (device) |
6684 | return device; | |
6685 | } | |
6686 | cur_devices = cur_devices->seed; | |
6687 | } | |
6688 | return NULL; | |
0b86a832 CM |
6689 | } |
6690 | ||
2ff7e61e | 6691 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6692 | u64 devid, u8 *dev_uuid) |
6693 | { | |
6694 | struct btrfs_device *device; | |
dfe25020 | 6695 | |
12bd2fc0 ID |
6696 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6697 | if (IS_ERR(device)) | |
adfb69af | 6698 | return device; |
12bd2fc0 ID |
6699 | |
6700 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6701 | device->fs_devices = fs_devices; |
dfe25020 | 6702 | fs_devices->num_devices++; |
12bd2fc0 | 6703 | |
e6e674bd | 6704 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6705 | fs_devices->missing_devices++; |
12bd2fc0 | 6706 | |
dfe25020 CM |
6707 | return device; |
6708 | } | |
6709 | ||
12bd2fc0 ID |
6710 | /** |
6711 | * btrfs_alloc_device - allocate struct btrfs_device | |
6712 | * @fs_info: used only for generating a new devid, can be NULL if | |
6713 | * devid is provided (i.e. @devid != NULL). | |
6714 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6715 | * is generated. | |
6716 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6717 | * is generated. | |
6718 | * | |
6719 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6720 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6721 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6722 | */ |
6723 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6724 | const u64 *devid, | |
6725 | const u8 *uuid) | |
6726 | { | |
6727 | struct btrfs_device *dev; | |
6728 | u64 tmp; | |
6729 | ||
fae7f21c | 6730 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6731 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6732 | |
6733 | dev = __alloc_device(); | |
6734 | if (IS_ERR(dev)) | |
6735 | return dev; | |
6736 | ||
6737 | if (devid) | |
6738 | tmp = *devid; | |
6739 | else { | |
6740 | int ret; | |
6741 | ||
6742 | ret = find_next_devid(fs_info, &tmp); | |
6743 | if (ret) { | |
a425f9d4 | 6744 | btrfs_free_device(dev); |
12bd2fc0 ID |
6745 | return ERR_PTR(ret); |
6746 | } | |
6747 | } | |
6748 | dev->devid = tmp; | |
6749 | ||
6750 | if (uuid) | |
6751 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6752 | else | |
6753 | generate_random_uuid(dev->uuid); | |
6754 | ||
9e0af237 LB |
6755 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6756 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6757 | |
6758 | return dev; | |
6759 | } | |
6760 | ||
e06cd3dd | 6761 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6762 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6763 | struct extent_buffer *leaf, |
6764 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6765 | { |
0b86a832 | 6766 | u64 length; |
f04b772b | 6767 | u64 stripe_len; |
e06cd3dd LB |
6768 | u16 num_stripes; |
6769 | u16 sub_stripes; | |
6770 | u64 type; | |
315409b0 GJ |
6771 | u64 features; |
6772 | bool mixed = false; | |
0b86a832 | 6773 | |
e17cade2 | 6774 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6775 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6776 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6777 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6778 | type = btrfs_chunk_type(leaf, chunk); | |
6779 | ||
f04b772b | 6780 | if (!num_stripes) { |
0b246afa | 6781 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6782 | num_stripes); |
6783 | return -EIO; | |
6784 | } | |
0b246afa JM |
6785 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6786 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6787 | return -EIO; |
6788 | } | |
0b246afa JM |
6789 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6790 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6791 | btrfs_chunk_sector_size(leaf, chunk)); |
6792 | return -EIO; | |
6793 | } | |
0b246afa JM |
6794 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6795 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6796 | return -EIO; |
6797 | } | |
3d8da678 | 6798 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6799 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6800 | stripe_len); |
6801 | return -EIO; | |
6802 | } | |
6803 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6804 | type) { |
0b246afa | 6805 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6806 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6807 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6808 | btrfs_chunk_type(leaf, chunk)); | |
6809 | return -EIO; | |
6810 | } | |
315409b0 GJ |
6811 | |
6812 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6813 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6814 | return -EIO; | |
6815 | } | |
6816 | ||
6817 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6818 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6819 | btrfs_err(fs_info, | |
6820 | "system chunk with data or metadata type: 0x%llx", type); | |
6821 | return -EIO; | |
6822 | } | |
6823 | ||
6824 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6825 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6826 | mixed = true; | |
6827 | ||
6828 | if (!mixed) { | |
6829 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6830 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6831 | btrfs_err(fs_info, | |
6832 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6833 | return -EIO; | |
6834 | } | |
6835 | } | |
6836 | ||
e06cd3dd LB |
6837 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6838 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6839 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6840 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6841 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6842 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6843 | num_stripes != 1)) { | |
0b246afa | 6844 | btrfs_err(fs_info, |
e06cd3dd LB |
6845 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6846 | num_stripes, sub_stripes, | |
6847 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6848 | return -EIO; | |
6849 | } | |
6850 | ||
6851 | return 0; | |
6852 | } | |
6853 | ||
5a2b8e60 | 6854 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6855 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6856 | { |
2b902dfc AJ |
6857 | if (error) |
6858 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6859 | devid, uuid); | |
6860 | else | |
6861 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6862 | devid, uuid); | |
5a2b8e60 AJ |
6863 | } |
6864 | ||
2ff7e61e | 6865 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6866 | struct extent_buffer *leaf, |
6867 | struct btrfs_chunk *chunk) | |
6868 | { | |
0b246afa | 6869 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6870 | struct map_lookup *map; |
6871 | struct extent_map *em; | |
6872 | u64 logical; | |
6873 | u64 length; | |
e06cd3dd LB |
6874 | u64 devid; |
6875 | u8 uuid[BTRFS_UUID_SIZE]; | |
6876 | int num_stripes; | |
6877 | int ret; | |
6878 | int i; | |
6879 | ||
6880 | logical = key->offset; | |
6881 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6882 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6883 | ||
2ff7e61e | 6884 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6885 | if (ret) |
6886 | return ret; | |
a061fc8d | 6887 | |
890871be | 6888 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6889 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6890 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6891 | |
6892 | /* already mapped? */ | |
6893 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6894 | free_extent_map(em); | |
0b86a832 CM |
6895 | return 0; |
6896 | } else if (em) { | |
6897 | free_extent_map(em); | |
6898 | } | |
0b86a832 | 6899 | |
172ddd60 | 6900 | em = alloc_extent_map(); |
0b86a832 CM |
6901 | if (!em) |
6902 | return -ENOMEM; | |
593060d7 | 6903 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6904 | if (!map) { |
6905 | free_extent_map(em); | |
6906 | return -ENOMEM; | |
6907 | } | |
6908 | ||
298a8f9c | 6909 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6910 | em->map_lookup = map; |
0b86a832 CM |
6911 | em->start = logical; |
6912 | em->len = length; | |
70c8a91c | 6913 | em->orig_start = 0; |
0b86a832 | 6914 | em->block_start = 0; |
c8b97818 | 6915 | em->block_len = em->len; |
0b86a832 | 6916 | |
593060d7 CM |
6917 | map->num_stripes = num_stripes; |
6918 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6919 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6920 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6921 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6922 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6923 | map->verified_stripes = 0; |
593060d7 CM |
6924 | for (i = 0; i < num_stripes; i++) { |
6925 | map->stripes[i].physical = | |
6926 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6927 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6928 | read_extent_buffer(leaf, uuid, (unsigned long) |
6929 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6930 | BTRFS_UUID_SIZE); | |
0b246afa | 6931 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6932 | uuid, NULL); |
3cdde224 | 6933 | if (!map->stripes[i].dev && |
0b246afa | 6934 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6935 | free_extent_map(em); |
2b902dfc | 6936 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6937 | return -ENOENT; |
593060d7 | 6938 | } |
dfe25020 CM |
6939 | if (!map->stripes[i].dev) { |
6940 | map->stripes[i].dev = | |
2ff7e61e JM |
6941 | add_missing_dev(fs_info->fs_devices, devid, |
6942 | uuid); | |
adfb69af | 6943 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6944 | free_extent_map(em); |
adfb69af AJ |
6945 | btrfs_err(fs_info, |
6946 | "failed to init missing dev %llu: %ld", | |
6947 | devid, PTR_ERR(map->stripes[i].dev)); | |
6948 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6949 | } |
2b902dfc | 6950 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6951 | } |
e12c9621 AJ |
6952 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6953 | &(map->stripes[i].dev->dev_state)); | |
6954 | ||
0b86a832 CM |
6955 | } |
6956 | ||
890871be | 6957 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6958 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6959 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6960 | if (ret < 0) { |
6961 | btrfs_err(fs_info, | |
6962 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6963 | em->start, em->len, ret); | |
6964 | } | |
0b86a832 CM |
6965 | free_extent_map(em); |
6966 | ||
64f64f43 | 6967 | return ret; |
0b86a832 CM |
6968 | } |
6969 | ||
143bede5 | 6970 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6971 | struct btrfs_dev_item *dev_item, |
6972 | struct btrfs_device *device) | |
6973 | { | |
6974 | unsigned long ptr; | |
0b86a832 CM |
6975 | |
6976 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6977 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6978 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6979 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6980 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6981 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6982 | device->type = btrfs_device_type(leaf, dev_item); |
6983 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6984 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6985 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6986 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6987 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6988 | |
410ba3a2 | 6989 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6990 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6991 | } |
6992 | ||
2ff7e61e | 6993 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6994 | u8 *fsid) |
2b82032c YZ |
6995 | { |
6996 | struct btrfs_fs_devices *fs_devices; | |
6997 | int ret; | |
6998 | ||
a32bf9a3 | 6999 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 7000 | ASSERT(fsid); |
2b82032c | 7001 | |
0b246afa | 7002 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 7003 | while (fs_devices) { |
44880fdc | 7004 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
7005 | return fs_devices; |
7006 | ||
2b82032c YZ |
7007 | fs_devices = fs_devices->seed; |
7008 | } | |
7009 | ||
7239ff4b | 7010 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 7011 | if (!fs_devices) { |
0b246afa | 7012 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
7013 | return ERR_PTR(-ENOENT); |
7014 | ||
7239ff4b | 7015 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
7016 | if (IS_ERR(fs_devices)) |
7017 | return fs_devices; | |
7018 | ||
7019 | fs_devices->seeding = 1; | |
7020 | fs_devices->opened = 1; | |
7021 | return fs_devices; | |
2b82032c | 7022 | } |
e4404d6e YZ |
7023 | |
7024 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
7025 | if (IS_ERR(fs_devices)) |
7026 | return fs_devices; | |
2b82032c | 7027 | |
897fb573 | 7028 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
7029 | if (ret) { |
7030 | free_fs_devices(fs_devices); | |
5f375835 | 7031 | fs_devices = ERR_PTR(ret); |
2b82032c | 7032 | goto out; |
48d28232 | 7033 | } |
2b82032c YZ |
7034 | |
7035 | if (!fs_devices->seeding) { | |
0226e0eb | 7036 | close_fs_devices(fs_devices); |
e4404d6e | 7037 | free_fs_devices(fs_devices); |
5f375835 | 7038 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
7039 | goto out; |
7040 | } | |
7041 | ||
0b246afa JM |
7042 | fs_devices->seed = fs_info->fs_devices->seed; |
7043 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 7044 | out: |
5f375835 | 7045 | return fs_devices; |
2b82032c YZ |
7046 | } |
7047 | ||
2ff7e61e | 7048 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
7049 | struct extent_buffer *leaf, |
7050 | struct btrfs_dev_item *dev_item) | |
7051 | { | |
0b246afa | 7052 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
7053 | struct btrfs_device *device; |
7054 | u64 devid; | |
7055 | int ret; | |
44880fdc | 7056 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
7057 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
7058 | ||
0b86a832 | 7059 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 7060 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 7061 | BTRFS_UUID_SIZE); |
1473b24e | 7062 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 7063 | BTRFS_FSID_SIZE); |
2b82032c | 7064 | |
de37aa51 | 7065 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 7066 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
7067 | if (IS_ERR(fs_devices)) |
7068 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
7069 | } |
7070 | ||
0b246afa | 7071 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 7072 | if (!device) { |
c5502451 | 7073 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
7074 | btrfs_report_missing_device(fs_info, devid, |
7075 | dev_uuid, true); | |
45dbdbc9 | 7076 | return -ENOENT; |
c5502451 | 7077 | } |
2b82032c | 7078 | |
2ff7e61e | 7079 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
7080 | if (IS_ERR(device)) { |
7081 | btrfs_err(fs_info, | |
7082 | "failed to add missing dev %llu: %ld", | |
7083 | devid, PTR_ERR(device)); | |
7084 | return PTR_ERR(device); | |
7085 | } | |
2b902dfc | 7086 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 7087 | } else { |
c5502451 | 7088 | if (!device->bdev) { |
2b902dfc AJ |
7089 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
7090 | btrfs_report_missing_device(fs_info, | |
7091 | devid, dev_uuid, true); | |
45dbdbc9 | 7092 | return -ENOENT; |
2b902dfc AJ |
7093 | } |
7094 | btrfs_report_missing_device(fs_info, devid, | |
7095 | dev_uuid, false); | |
c5502451 | 7096 | } |
5f375835 | 7097 | |
e6e674bd AJ |
7098 | if (!device->bdev && |
7099 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7100 | /* |
7101 | * this happens when a device that was properly setup | |
7102 | * in the device info lists suddenly goes bad. | |
7103 | * device->bdev is NULL, and so we have to set | |
7104 | * device->missing to one here | |
7105 | */ | |
5f375835 | 7106 | device->fs_devices->missing_devices++; |
e6e674bd | 7107 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7108 | } |
5f375835 MX |
7109 | |
7110 | /* Move the device to its own fs_devices */ | |
7111 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7112 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7113 | &device->dev_state)); | |
5f375835 MX |
7114 | |
7115 | list_move(&device->dev_list, &fs_devices->devices); | |
7116 | device->fs_devices->num_devices--; | |
7117 | fs_devices->num_devices++; | |
7118 | ||
7119 | device->fs_devices->missing_devices--; | |
7120 | fs_devices->missing_devices++; | |
7121 | ||
7122 | device->fs_devices = fs_devices; | |
7123 | } | |
2b82032c YZ |
7124 | } |
7125 | ||
0b246afa | 7126 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7127 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7128 | if (device->generation != |
7129 | btrfs_device_generation(leaf, dev_item)) | |
7130 | return -EINVAL; | |
6324fbf3 | 7131 | } |
0b86a832 CM |
7132 | |
7133 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 7134 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7135 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7136 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7137 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7138 | atomic64_add(device->total_bytes - device->bytes_used, |
7139 | &fs_info->free_chunk_space); | |
2bf64758 | 7140 | } |
0b86a832 | 7141 | ret = 0; |
0b86a832 CM |
7142 | return ret; |
7143 | } | |
7144 | ||
6bccf3ab | 7145 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7146 | { |
6bccf3ab | 7147 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 7148 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7149 | struct extent_buffer *sb; |
0b86a832 | 7150 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7151 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7152 | u8 *array_ptr; |
7153 | unsigned long sb_array_offset; | |
84eed90f | 7154 | int ret = 0; |
0b86a832 CM |
7155 | u32 num_stripes; |
7156 | u32 array_size; | |
7157 | u32 len = 0; | |
1ffb22cf | 7158 | u32 cur_offset; |
e06cd3dd | 7159 | u64 type; |
84eed90f | 7160 | struct btrfs_key key; |
0b86a832 | 7161 | |
0b246afa | 7162 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7163 | /* |
7164 | * This will create extent buffer of nodesize, superblock size is | |
7165 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7166 | * overallocate but we can keep it as-is, only the first page is used. | |
7167 | */ | |
2ff7e61e | 7168 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7169 | if (IS_ERR(sb)) |
7170 | return PTR_ERR(sb); | |
4db8c528 | 7171 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7172 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7173 | /* |
01327610 | 7174 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7175 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7176 | * pages up-to-date when the page is larger: extent does not cover the |
7177 | * whole page and consequently check_page_uptodate does not find all | |
7178 | * the page's extents up-to-date (the hole beyond sb), | |
7179 | * write_extent_buffer then triggers a WARN_ON. | |
7180 | * | |
7181 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7182 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7183 | * to silence the warning eg. on PowerPC 64. | |
7184 | */ | |
09cbfeaf | 7185 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7186 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7187 | |
a061fc8d | 7188 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7189 | array_size = btrfs_super_sys_array_size(super_copy); |
7190 | ||
1ffb22cf DS |
7191 | array_ptr = super_copy->sys_chunk_array; |
7192 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7193 | cur_offset = 0; | |
0b86a832 | 7194 | |
1ffb22cf DS |
7195 | while (cur_offset < array_size) { |
7196 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7197 | len = sizeof(*disk_key); |
7198 | if (cur_offset + len > array_size) | |
7199 | goto out_short_read; | |
7200 | ||
0b86a832 CM |
7201 | btrfs_disk_key_to_cpu(&key, disk_key); |
7202 | ||
1ffb22cf DS |
7203 | array_ptr += len; |
7204 | sb_array_offset += len; | |
7205 | cur_offset += len; | |
0b86a832 | 7206 | |
0d81ba5d | 7207 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7208 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7209 | /* |
7210 | * At least one btrfs_chunk with one stripe must be | |
7211 | * present, exact stripe count check comes afterwards | |
7212 | */ | |
7213 | len = btrfs_chunk_item_size(1); | |
7214 | if (cur_offset + len > array_size) | |
7215 | goto out_short_read; | |
7216 | ||
7217 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7218 | if (!num_stripes) { |
ab8d0fc4 JM |
7219 | btrfs_err(fs_info, |
7220 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7221 | num_stripes, cur_offset); |
7222 | ret = -EIO; | |
7223 | break; | |
7224 | } | |
7225 | ||
e06cd3dd LB |
7226 | type = btrfs_chunk_type(sb, chunk); |
7227 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7228 | btrfs_err(fs_info, |
e06cd3dd LB |
7229 | "invalid chunk type %llu in sys_array at offset %u", |
7230 | type, cur_offset); | |
7231 | ret = -EIO; | |
7232 | break; | |
7233 | } | |
7234 | ||
e3540eab DS |
7235 | len = btrfs_chunk_item_size(num_stripes); |
7236 | if (cur_offset + len > array_size) | |
7237 | goto out_short_read; | |
7238 | ||
2ff7e61e | 7239 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
7240 | if (ret) |
7241 | break; | |
0b86a832 | 7242 | } else { |
ab8d0fc4 JM |
7243 | btrfs_err(fs_info, |
7244 | "unexpected item type %u in sys_array at offset %u", | |
7245 | (u32)key.type, cur_offset); | |
84eed90f CM |
7246 | ret = -EIO; |
7247 | break; | |
0b86a832 | 7248 | } |
1ffb22cf DS |
7249 | array_ptr += len; |
7250 | sb_array_offset += len; | |
7251 | cur_offset += len; | |
0b86a832 | 7252 | } |
d865177a | 7253 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7254 | free_extent_buffer_stale(sb); |
84eed90f | 7255 | return ret; |
e3540eab DS |
7256 | |
7257 | out_short_read: | |
ab8d0fc4 | 7258 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7259 | len, cur_offset); |
d865177a | 7260 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7261 | free_extent_buffer_stale(sb); |
e3540eab | 7262 | return -EIO; |
0b86a832 CM |
7263 | } |
7264 | ||
21634a19 QW |
7265 | /* |
7266 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7267 | * | |
6528b99d AJ |
7268 | * If the @failing_dev is specified, it's accounted as missing. |
7269 | * | |
21634a19 QW |
7270 | * Return true if all chunks meet the minimal RW mount requirements. |
7271 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7272 | */ | |
6528b99d AJ |
7273 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7274 | struct btrfs_device *failing_dev) | |
21634a19 QW |
7275 | { |
7276 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
7277 | struct extent_map *em; | |
7278 | u64 next_start = 0; | |
7279 | bool ret = true; | |
7280 | ||
7281 | read_lock(&map_tree->map_tree.lock); | |
7282 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
7283 | read_unlock(&map_tree->map_tree.lock); | |
7284 | /* No chunk at all? Return false anyway */ | |
7285 | if (!em) { | |
7286 | ret = false; | |
7287 | goto out; | |
7288 | } | |
7289 | while (em) { | |
7290 | struct map_lookup *map; | |
7291 | int missing = 0; | |
7292 | int max_tolerated; | |
7293 | int i; | |
7294 | ||
7295 | map = em->map_lookup; | |
7296 | max_tolerated = | |
7297 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7298 | map->type); | |
7299 | for (i = 0; i < map->num_stripes; i++) { | |
7300 | struct btrfs_device *dev = map->stripes[i].dev; | |
7301 | ||
e6e674bd AJ |
7302 | if (!dev || !dev->bdev || |
7303 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7304 | dev->last_flush_error) |
7305 | missing++; | |
6528b99d AJ |
7306 | else if (failing_dev && failing_dev == dev) |
7307 | missing++; | |
21634a19 QW |
7308 | } |
7309 | if (missing > max_tolerated) { | |
6528b99d AJ |
7310 | if (!failing_dev) |
7311 | btrfs_warn(fs_info, | |
21634a19 QW |
7312 | "chunk %llu missing %d devices, max tolerance is %d for writeable mount", |
7313 | em->start, missing, max_tolerated); | |
7314 | free_extent_map(em); | |
7315 | ret = false; | |
7316 | goto out; | |
7317 | } | |
7318 | next_start = extent_map_end(em); | |
7319 | free_extent_map(em); | |
7320 | ||
7321 | read_lock(&map_tree->map_tree.lock); | |
7322 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
7323 | (u64)(-1) - next_start); | |
7324 | read_unlock(&map_tree->map_tree.lock); | |
7325 | } | |
7326 | out: | |
7327 | return ret; | |
7328 | } | |
7329 | ||
5b4aacef | 7330 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7331 | { |
5b4aacef | 7332 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7333 | struct btrfs_path *path; |
7334 | struct extent_buffer *leaf; | |
7335 | struct btrfs_key key; | |
7336 | struct btrfs_key found_key; | |
7337 | int ret; | |
7338 | int slot; | |
99e3ecfc | 7339 | u64 total_dev = 0; |
0b86a832 | 7340 | |
0b86a832 CM |
7341 | path = btrfs_alloc_path(); |
7342 | if (!path) | |
7343 | return -ENOMEM; | |
7344 | ||
3dd0f7a3 AJ |
7345 | /* |
7346 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7347 | * otherwise we don't need it. | |
7348 | */ | |
b367e47f | 7349 | mutex_lock(&uuid_mutex); |
34441361 | 7350 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7351 | |
395927a9 FDBM |
7352 | /* |
7353 | * Read all device items, and then all the chunk items. All | |
7354 | * device items are found before any chunk item (their object id | |
7355 | * is smaller than the lowest possible object id for a chunk | |
7356 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7357 | */ |
7358 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7359 | key.offset = 0; | |
7360 | key.type = 0; | |
0b86a832 | 7361 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7362 | if (ret < 0) |
7363 | goto error; | |
d397712b | 7364 | while (1) { |
0b86a832 CM |
7365 | leaf = path->nodes[0]; |
7366 | slot = path->slots[0]; | |
7367 | if (slot >= btrfs_header_nritems(leaf)) { | |
7368 | ret = btrfs_next_leaf(root, path); | |
7369 | if (ret == 0) | |
7370 | continue; | |
7371 | if (ret < 0) | |
7372 | goto error; | |
7373 | break; | |
7374 | } | |
7375 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7376 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7377 | struct btrfs_dev_item *dev_item; | |
7378 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7379 | struct btrfs_dev_item); |
2ff7e61e | 7380 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7381 | if (ret) |
7382 | goto error; | |
99e3ecfc | 7383 | total_dev++; |
0b86a832 CM |
7384 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7385 | struct btrfs_chunk *chunk; | |
7386 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7387 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7388 | if (ret) |
7389 | goto error; | |
0b86a832 CM |
7390 | } |
7391 | path->slots[0]++; | |
7392 | } | |
99e3ecfc LB |
7393 | |
7394 | /* | |
7395 | * After loading chunk tree, we've got all device information, | |
7396 | * do another round of validation checks. | |
7397 | */ | |
0b246afa JM |
7398 | if (total_dev != fs_info->fs_devices->total_devices) { |
7399 | btrfs_err(fs_info, | |
99e3ecfc | 7400 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7401 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7402 | total_dev); |
7403 | ret = -EINVAL; | |
7404 | goto error; | |
7405 | } | |
0b246afa JM |
7406 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7407 | fs_info->fs_devices->total_rw_bytes) { | |
7408 | btrfs_err(fs_info, | |
99e3ecfc | 7409 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7410 | btrfs_super_total_bytes(fs_info->super_copy), |
7411 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7412 | ret = -EINVAL; |
7413 | goto error; | |
7414 | } | |
0b86a832 CM |
7415 | ret = 0; |
7416 | error: | |
34441361 | 7417 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7418 | mutex_unlock(&uuid_mutex); |
7419 | ||
2b82032c | 7420 | btrfs_free_path(path); |
0b86a832 CM |
7421 | return ret; |
7422 | } | |
442a4f63 | 7423 | |
cb517eab MX |
7424 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7425 | { | |
7426 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7427 | struct btrfs_device *device; | |
7428 | ||
29cc83f6 LB |
7429 | while (fs_devices) { |
7430 | mutex_lock(&fs_devices->device_list_mutex); | |
7431 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7432 | device->fs_info = fs_info; |
29cc83f6 LB |
7433 | mutex_unlock(&fs_devices->device_list_mutex); |
7434 | ||
7435 | fs_devices = fs_devices->seed; | |
7436 | } | |
cb517eab MX |
7437 | } |
7438 | ||
733f4fbb SB |
7439 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7440 | { | |
7441 | int i; | |
7442 | ||
7443 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7444 | btrfs_dev_stat_reset(dev, i); | |
7445 | } | |
7446 | ||
7447 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7448 | { | |
7449 | struct btrfs_key key; | |
7450 | struct btrfs_key found_key; | |
7451 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7452 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7453 | struct extent_buffer *eb; | |
7454 | int slot; | |
7455 | int ret = 0; | |
7456 | struct btrfs_device *device; | |
7457 | struct btrfs_path *path = NULL; | |
7458 | int i; | |
7459 | ||
7460 | path = btrfs_alloc_path(); | |
7461 | if (!path) { | |
7462 | ret = -ENOMEM; | |
7463 | goto out; | |
7464 | } | |
7465 | ||
7466 | mutex_lock(&fs_devices->device_list_mutex); | |
7467 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7468 | int item_size; | |
7469 | struct btrfs_dev_stats_item *ptr; | |
7470 | ||
242e2956 DS |
7471 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7472 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7473 | key.offset = device->devid; |
7474 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7475 | if (ret) { | |
733f4fbb SB |
7476 | __btrfs_reset_dev_stats(device); |
7477 | device->dev_stats_valid = 1; | |
7478 | btrfs_release_path(path); | |
7479 | continue; | |
7480 | } | |
7481 | slot = path->slots[0]; | |
7482 | eb = path->nodes[0]; | |
7483 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7484 | item_size = btrfs_item_size_nr(eb, slot); | |
7485 | ||
7486 | ptr = btrfs_item_ptr(eb, slot, | |
7487 | struct btrfs_dev_stats_item); | |
7488 | ||
7489 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7490 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7491 | btrfs_dev_stat_set(device, i, | |
7492 | btrfs_dev_stats_value(eb, ptr, i)); | |
7493 | else | |
7494 | btrfs_dev_stat_reset(device, i); | |
7495 | } | |
7496 | ||
7497 | device->dev_stats_valid = 1; | |
7498 | btrfs_dev_stat_print_on_load(device); | |
7499 | btrfs_release_path(path); | |
7500 | } | |
7501 | mutex_unlock(&fs_devices->device_list_mutex); | |
7502 | ||
7503 | out: | |
7504 | btrfs_free_path(path); | |
7505 | return ret < 0 ? ret : 0; | |
7506 | } | |
7507 | ||
7508 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7509 | struct btrfs_device *device) |
7510 | { | |
5495f195 | 7511 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7512 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7513 | struct btrfs_path *path; |
7514 | struct btrfs_key key; | |
7515 | struct extent_buffer *eb; | |
7516 | struct btrfs_dev_stats_item *ptr; | |
7517 | int ret; | |
7518 | int i; | |
7519 | ||
242e2956 DS |
7520 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7521 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7522 | key.offset = device->devid; |
7523 | ||
7524 | path = btrfs_alloc_path(); | |
fa252992 DS |
7525 | if (!path) |
7526 | return -ENOMEM; | |
733f4fbb SB |
7527 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7528 | if (ret < 0) { | |
0b246afa | 7529 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7530 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7531 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7532 | goto out; |
7533 | } | |
7534 | ||
7535 | if (ret == 0 && | |
7536 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7537 | /* need to delete old one and insert a new one */ | |
7538 | ret = btrfs_del_item(trans, dev_root, path); | |
7539 | if (ret != 0) { | |
0b246afa | 7540 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7541 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7542 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7543 | goto out; |
7544 | } | |
7545 | ret = 1; | |
7546 | } | |
7547 | ||
7548 | if (ret == 1) { | |
7549 | /* need to insert a new item */ | |
7550 | btrfs_release_path(path); | |
7551 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7552 | &key, sizeof(*ptr)); | |
7553 | if (ret < 0) { | |
0b246afa | 7554 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7555 | "insert dev_stats item for device %s failed %d", |
7556 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7557 | goto out; |
7558 | } | |
7559 | } | |
7560 | ||
7561 | eb = path->nodes[0]; | |
7562 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7563 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7564 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7565 | btrfs_dev_stat_read(device, i)); | |
7566 | btrfs_mark_buffer_dirty(eb); | |
7567 | ||
7568 | out: | |
7569 | btrfs_free_path(path); | |
7570 | return ret; | |
7571 | } | |
7572 | ||
7573 | /* | |
7574 | * called from commit_transaction. Writes all changed device stats to disk. | |
7575 | */ | |
7576 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7577 | struct btrfs_fs_info *fs_info) | |
7578 | { | |
733f4fbb SB |
7579 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7580 | struct btrfs_device *device; | |
addc3fa7 | 7581 | int stats_cnt; |
733f4fbb SB |
7582 | int ret = 0; |
7583 | ||
7584 | mutex_lock(&fs_devices->device_list_mutex); | |
7585 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7586 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7587 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7588 | continue; |
7589 | ||
9deae968 NB |
7590 | |
7591 | /* | |
7592 | * There is a LOAD-LOAD control dependency between the value of | |
7593 | * dev_stats_ccnt and updating the on-disk values which requires | |
7594 | * reading the in-memory counters. Such control dependencies | |
7595 | * require explicit read memory barriers. | |
7596 | * | |
7597 | * This memory barriers pairs with smp_mb__before_atomic in | |
7598 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7599 | * barrier implied by atomic_xchg in | |
7600 | * btrfs_dev_stats_read_and_reset | |
7601 | */ | |
7602 | smp_rmb(); | |
7603 | ||
5495f195 | 7604 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7605 | if (!ret) |
addc3fa7 | 7606 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7607 | } |
7608 | mutex_unlock(&fs_devices->device_list_mutex); | |
7609 | ||
7610 | return ret; | |
7611 | } | |
7612 | ||
442a4f63 SB |
7613 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7614 | { | |
7615 | btrfs_dev_stat_inc(dev, index); | |
7616 | btrfs_dev_stat_print_on_error(dev); | |
7617 | } | |
7618 | ||
48a3b636 | 7619 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7620 | { |
733f4fbb SB |
7621 | if (!dev->dev_stats_valid) |
7622 | return; | |
fb456252 | 7623 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7624 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7625 | rcu_str_deref(dev->name), |
442a4f63 SB |
7626 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7627 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7628 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7629 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7630 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7631 | } |
c11d2c23 | 7632 | |
733f4fbb SB |
7633 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7634 | { | |
a98cdb85 SB |
7635 | int i; |
7636 | ||
7637 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7638 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7639 | break; | |
7640 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7641 | return; /* all values == 0, suppress message */ | |
7642 | ||
fb456252 | 7643 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7644 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7645 | rcu_str_deref(dev->name), |
733f4fbb SB |
7646 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7647 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7648 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7649 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7650 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7651 | } | |
7652 | ||
2ff7e61e | 7653 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7654 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7655 | { |
7656 | struct btrfs_device *dev; | |
0b246afa | 7657 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7658 | int i; |
7659 | ||
7660 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7661 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7662 | mutex_unlock(&fs_devices->device_list_mutex); |
7663 | ||
7664 | if (!dev) { | |
0b246afa | 7665 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7666 | return -ENODEV; |
733f4fbb | 7667 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7668 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7669 | return -ENODEV; |
b27f7c0c | 7670 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7671 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7672 | if (stats->nr_items > i) | |
7673 | stats->values[i] = | |
7674 | btrfs_dev_stat_read_and_reset(dev, i); | |
7675 | else | |
7676 | btrfs_dev_stat_reset(dev, i); | |
7677 | } | |
7678 | } else { | |
7679 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7680 | if (stats->nr_items > i) | |
7681 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7682 | } | |
7683 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7684 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7685 | return 0; | |
7686 | } | |
a8a6dab7 | 7687 | |
da353f6b | 7688 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7689 | { |
7690 | struct buffer_head *bh; | |
7691 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7692 | int copy_num; |
a8a6dab7 | 7693 | |
12b1c263 AJ |
7694 | if (!bdev) |
7695 | return; | |
a8a6dab7 | 7696 | |
12b1c263 AJ |
7697 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7698 | copy_num++) { | |
a8a6dab7 | 7699 | |
12b1c263 AJ |
7700 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7701 | continue; | |
7702 | ||
7703 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7704 | ||
7705 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7706 | set_buffer_dirty(bh); | |
7707 | sync_dirty_buffer(bh); | |
7708 | brelse(bh); | |
7709 | } | |
7710 | ||
7711 | /* Notify udev that device has changed */ | |
7712 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7713 | ||
7714 | /* Update ctime/mtime for device path for libblkid */ | |
7715 | update_dev_time(device_path); | |
a8a6dab7 | 7716 | } |
935e5cc9 MX |
7717 | |
7718 | /* | |
7719 | * Update the size of all devices, which is used for writing out the | |
7720 | * super blocks. | |
7721 | */ | |
7722 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7723 | { | |
7724 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7725 | struct btrfs_device *curr, *next; | |
7726 | ||
7727 | if (list_empty(&fs_devices->resized_devices)) | |
7728 | return; | |
7729 | ||
7730 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7731 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7732 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7733 | resized_list) { | |
7734 | list_del_init(&curr->resized_list); | |
7735 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7736 | } | |
34441361 | 7737 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7738 | mutex_unlock(&fs_devices->device_list_mutex); |
7739 | } | |
ce7213c7 MX |
7740 | |
7741 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7742 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7743 | { |
e9b919b1 | 7744 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7745 | struct extent_map *em; |
7746 | struct map_lookup *map; | |
7747 | struct btrfs_device *dev; | |
7748 | int i; | |
7749 | ||
e9b919b1 | 7750 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7751 | return; |
7752 | ||
7753 | /* In order to kick the device replace finish process */ | |
34441361 | 7754 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7755 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7756 | map = em->map_lookup; |
ce7213c7 MX |
7757 | |
7758 | for (i = 0; i < map->num_stripes; i++) { | |
7759 | dev = map->stripes[i].dev; | |
7760 | dev->commit_bytes_used = dev->bytes_used; | |
7761 | } | |
7762 | } | |
34441361 | 7763 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7764 | } |
5a13f430 AJ |
7765 | |
7766 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7767 | { | |
7768 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7769 | while (fs_devices) { | |
7770 | fs_devices->fs_info = fs_info; | |
7771 | fs_devices = fs_devices->seed; | |
7772 | } | |
7773 | } | |
7774 | ||
7775 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7776 | { | |
7777 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7778 | while (fs_devices) { | |
7779 | fs_devices->fs_info = NULL; | |
7780 | fs_devices = fs_devices->seed; | |
7781 | } | |
7782 | } | |
46df06b8 DS |
7783 | |
7784 | /* | |
7785 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7786 | */ | |
7787 | int btrfs_bg_type_to_factor(u64 flags) | |
7788 | { | |
7789 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7790 | BTRFS_BLOCK_GROUP_RAID10)) | |
7791 | return 2; | |
7792 | return 1; | |
7793 | } | |
cf90d884 QW |
7794 | |
7795 | ||
7796 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7797 | { | |
7798 | int index = btrfs_bg_flags_to_raid_index(type); | |
7799 | int ncopies = btrfs_raid_array[index].ncopies; | |
7800 | int data_stripes; | |
7801 | ||
7802 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7803 | case BTRFS_BLOCK_GROUP_RAID5: | |
7804 | data_stripes = num_stripes - 1; | |
7805 | break; | |
7806 | case BTRFS_BLOCK_GROUP_RAID6: | |
7807 | data_stripes = num_stripes - 2; | |
7808 | break; | |
7809 | default: | |
7810 | data_stripes = num_stripes / ncopies; | |
7811 | break; | |
7812 | } | |
7813 | return div_u64(chunk_len, data_stripes); | |
7814 | } | |
7815 | ||
7816 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7817 | u64 chunk_offset, u64 devid, | |
7818 | u64 physical_offset, u64 physical_len) | |
7819 | { | |
7820 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7821 | struct extent_map *em; | |
7822 | struct map_lookup *map; | |
05a37c48 | 7823 | struct btrfs_device *dev; |
cf90d884 QW |
7824 | u64 stripe_len; |
7825 | bool found = false; | |
7826 | int ret = 0; | |
7827 | int i; | |
7828 | ||
7829 | read_lock(&em_tree->lock); | |
7830 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7831 | read_unlock(&em_tree->lock); | |
7832 | ||
7833 | if (!em) { | |
7834 | btrfs_err(fs_info, | |
7835 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7836 | physical_offset, devid); | |
7837 | ret = -EUCLEAN; | |
7838 | goto out; | |
7839 | } | |
7840 | ||
7841 | map = em->map_lookup; | |
7842 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7843 | if (physical_len != stripe_len) { | |
7844 | btrfs_err(fs_info, | |
7845 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7846 | physical_offset, devid, em->start, physical_len, | |
7847 | stripe_len); | |
7848 | ret = -EUCLEAN; | |
7849 | goto out; | |
7850 | } | |
7851 | ||
7852 | for (i = 0; i < map->num_stripes; i++) { | |
7853 | if (map->stripes[i].dev->devid == devid && | |
7854 | map->stripes[i].physical == physical_offset) { | |
7855 | found = true; | |
7856 | if (map->verified_stripes >= map->num_stripes) { | |
7857 | btrfs_err(fs_info, | |
7858 | "too many dev extents for chunk %llu found", | |
7859 | em->start); | |
7860 | ret = -EUCLEAN; | |
7861 | goto out; | |
7862 | } | |
7863 | map->verified_stripes++; | |
7864 | break; | |
7865 | } | |
7866 | } | |
7867 | if (!found) { | |
7868 | btrfs_err(fs_info, | |
7869 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7870 | physical_offset, devid); | |
7871 | ret = -EUCLEAN; | |
7872 | } | |
05a37c48 QW |
7873 | |
7874 | /* Make sure no dev extent is beyond device bondary */ | |
7875 | dev = btrfs_find_device(fs_info, devid, NULL, NULL); | |
7876 | if (!dev) { | |
7877 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7878 | ret = -EUCLEAN; | |
7879 | goto out; | |
7880 | } | |
7881 | if (physical_offset + physical_len > dev->disk_total_bytes) { | |
7882 | btrfs_err(fs_info, | |
7883 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7884 | devid, physical_offset, physical_len, | |
7885 | dev->disk_total_bytes); | |
7886 | ret = -EUCLEAN; | |
7887 | goto out; | |
7888 | } | |
cf90d884 QW |
7889 | out: |
7890 | free_extent_map(em); | |
7891 | return ret; | |
7892 | } | |
7893 | ||
7894 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7895 | { | |
7896 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7897 | struct extent_map *em; | |
7898 | struct rb_node *node; | |
7899 | int ret = 0; | |
7900 | ||
7901 | read_lock(&em_tree->lock); | |
07e1ce09 | 7902 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7903 | em = rb_entry(node, struct extent_map, rb_node); |
7904 | if (em->map_lookup->num_stripes != | |
7905 | em->map_lookup->verified_stripes) { | |
7906 | btrfs_err(fs_info, | |
7907 | "chunk %llu has missing dev extent, have %d expect %d", | |
7908 | em->start, em->map_lookup->verified_stripes, | |
7909 | em->map_lookup->num_stripes); | |
7910 | ret = -EUCLEAN; | |
7911 | goto out; | |
7912 | } | |
7913 | } | |
7914 | out: | |
7915 | read_unlock(&em_tree->lock); | |
7916 | return ret; | |
7917 | } | |
7918 | ||
7919 | /* | |
7920 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7921 | * later chunk allocation/free would cause unexpected behavior. | |
7922 | * | |
7923 | * NOTE: This will iterate through the whole device tree, which should be of | |
7924 | * the same size level as the chunk tree. This slightly increases mount time. | |
7925 | */ | |
7926 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7927 | { | |
7928 | struct btrfs_path *path; | |
7929 | struct btrfs_root *root = fs_info->dev_root; | |
7930 | struct btrfs_key key; | |
5eb19381 QW |
7931 | u64 prev_devid = 0; |
7932 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7933 | int ret = 0; |
7934 | ||
7935 | key.objectid = 1; | |
7936 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7937 | key.offset = 0; | |
7938 | ||
7939 | path = btrfs_alloc_path(); | |
7940 | if (!path) | |
7941 | return -ENOMEM; | |
7942 | ||
7943 | path->reada = READA_FORWARD; | |
7944 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7945 | if (ret < 0) | |
7946 | goto out; | |
7947 | ||
7948 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7949 | ret = btrfs_next_item(root, path); | |
7950 | if (ret < 0) | |
7951 | goto out; | |
7952 | /* No dev extents at all? Not good */ | |
7953 | if (ret > 0) { | |
7954 | ret = -EUCLEAN; | |
7955 | goto out; | |
7956 | } | |
7957 | } | |
7958 | while (1) { | |
7959 | struct extent_buffer *leaf = path->nodes[0]; | |
7960 | struct btrfs_dev_extent *dext; | |
7961 | int slot = path->slots[0]; | |
7962 | u64 chunk_offset; | |
7963 | u64 physical_offset; | |
7964 | u64 physical_len; | |
7965 | u64 devid; | |
7966 | ||
7967 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7968 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7969 | break; | |
7970 | devid = key.objectid; | |
7971 | physical_offset = key.offset; | |
7972 | ||
7973 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7974 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7975 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7976 | ||
5eb19381 QW |
7977 | /* Check if this dev extent overlaps with the previous one */ |
7978 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7979 | btrfs_err(fs_info, | |
7980 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7981 | devid, physical_offset, prev_dev_ext_end); | |
7982 | ret = -EUCLEAN; | |
7983 | goto out; | |
7984 | } | |
7985 | ||
cf90d884 QW |
7986 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7987 | physical_offset, physical_len); | |
7988 | if (ret < 0) | |
7989 | goto out; | |
5eb19381 QW |
7990 | prev_devid = devid; |
7991 | prev_dev_ext_end = physical_offset + physical_len; | |
7992 | ||
cf90d884 QW |
7993 | ret = btrfs_next_item(root, path); |
7994 | if (ret < 0) | |
7995 | goto out; | |
7996 | if (ret > 0) { | |
7997 | ret = 0; | |
7998 | break; | |
7999 | } | |
8000 | } | |
8001 | ||
8002 | /* Ensure all chunks have corresponding dev extents */ | |
8003 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
8004 | out: | |
8005 | btrfs_free_path(path); | |
8006 | return ret; | |
8007 | } | |
eede2bf3 OS |
8008 | |
8009 | /* | |
8010 | * Check whether the given block group or device is pinned by any inode being | |
8011 | * used as a swapfile. | |
8012 | */ | |
8013 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
8014 | { | |
8015 | struct btrfs_swapfile_pin *sp; | |
8016 | struct rb_node *node; | |
8017 | ||
8018 | spin_lock(&fs_info->swapfile_pins_lock); | |
8019 | node = fs_info->swapfile_pins.rb_node; | |
8020 | while (node) { | |
8021 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
8022 | if (ptr < sp->ptr) | |
8023 | node = node->rb_left; | |
8024 | else if (ptr > sp->ptr) | |
8025 | node = node->rb_right; | |
8026 | else | |
8027 | break; | |
8028 | } | |
8029 | spin_unlock(&fs_info->swapfile_pins_lock); | |
8030 | return node != NULL; | |
8031 | } |