]>
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, | |
52042d8e | 215 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
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 | ||
7239ff4b NB |
418 | static noinline struct btrfs_fs_devices *find_fsid( |
419 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 420 | { |
8a4b83cc CM |
421 | struct btrfs_fs_devices *fs_devices; |
422 | ||
7239ff4b NB |
423 | ASSERT(fsid); |
424 | ||
7a62d0f0 NB |
425 | if (metadata_fsid) { |
426 | /* | |
427 | * Handle scanned device having completed its fsid change but | |
428 | * belonging to a fs_devices that was created by first scanning | |
429 | * a device which didn't have its fsid/metadata_uuid changed | |
430 | * at all and the CHANGING_FSID_V2 flag set. | |
431 | */ | |
432 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
433 | if (fs_devices->fsid_change && | |
434 | memcmp(metadata_fsid, fs_devices->fsid, | |
435 | BTRFS_FSID_SIZE) == 0 && | |
436 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
437 | BTRFS_FSID_SIZE) == 0) { | |
438 | return fs_devices; | |
439 | } | |
440 | } | |
cc5de4e7 NB |
441 | /* |
442 | * Handle scanned device having completed its fsid change but | |
443 | * belonging to a fs_devices that was created by a device that | |
444 | * has an outdated pair of fsid/metadata_uuid and | |
445 | * CHANGING_FSID_V2 flag set. | |
446 | */ | |
447 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
448 | if (fs_devices->fsid_change && | |
449 | memcmp(fs_devices->metadata_uuid, | |
450 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
451 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
452 | BTRFS_FSID_SIZE) == 0) { | |
453 | return fs_devices; | |
454 | } | |
455 | } | |
7a62d0f0 NB |
456 | } |
457 | ||
458 | /* Handle non-split brain cases */ | |
c4babc5e | 459 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
460 | if (metadata_fsid) { |
461 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
462 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
463 | BTRFS_FSID_SIZE) == 0) | |
464 | return fs_devices; | |
465 | } else { | |
466 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
467 | return fs_devices; | |
468 | } | |
8a4b83cc CM |
469 | } |
470 | return NULL; | |
471 | } | |
472 | ||
beaf8ab3 SB |
473 | static int |
474 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
475 | int flush, struct block_device **bdev, | |
476 | struct buffer_head **bh) | |
477 | { | |
478 | int ret; | |
479 | ||
480 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
481 | ||
482 | if (IS_ERR(*bdev)) { | |
483 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
484 | goto error; |
485 | } | |
486 | ||
487 | if (flush) | |
488 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 489 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
490 | if (ret) { |
491 | blkdev_put(*bdev, flags); | |
492 | goto error; | |
493 | } | |
494 | invalidate_bdev(*bdev); | |
495 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
496 | if (IS_ERR(*bh)) { |
497 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
498 | blkdev_put(*bdev, flags); |
499 | goto error; | |
500 | } | |
501 | ||
502 | return 0; | |
503 | ||
504 | error: | |
505 | *bdev = NULL; | |
506 | *bh = NULL; | |
507 | return ret; | |
508 | } | |
509 | ||
ffbd517d CM |
510 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
511 | struct bio *head, struct bio *tail) | |
512 | { | |
513 | ||
514 | struct bio *old_head; | |
515 | ||
516 | old_head = pending_bios->head; | |
517 | pending_bios->head = head; | |
518 | if (pending_bios->tail) | |
519 | tail->bi_next = old_head; | |
520 | else | |
521 | pending_bios->tail = tail; | |
522 | } | |
523 | ||
8b712842 CM |
524 | /* |
525 | * we try to collect pending bios for a device so we don't get a large | |
526 | * number of procs sending bios down to the same device. This greatly | |
527 | * improves the schedulers ability to collect and merge the bios. | |
528 | * | |
529 | * But, it also turns into a long list of bios to process and that is sure | |
530 | * to eventually make the worker thread block. The solution here is to | |
531 | * make some progress and then put this work struct back at the end of | |
532 | * the list if the block device is congested. This way, multiple devices | |
533 | * can make progress from a single worker thread. | |
534 | */ | |
143bede5 | 535 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 536 | { |
0b246afa | 537 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
538 | struct bio *pending; |
539 | struct backing_dev_info *bdi; | |
ffbd517d | 540 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
541 | struct bio *tail; |
542 | struct bio *cur; | |
543 | int again = 0; | |
ffbd517d | 544 | unsigned long num_run; |
d644d8a1 | 545 | unsigned long batch_run = 0; |
b765ead5 | 546 | unsigned long last_waited = 0; |
d84275c9 | 547 | int force_reg = 0; |
0e588859 | 548 | int sync_pending = 0; |
211588ad CM |
549 | struct blk_plug plug; |
550 | ||
551 | /* | |
552 | * this function runs all the bios we've collected for | |
553 | * a particular device. We don't want to wander off to | |
554 | * another device without first sending all of these down. | |
555 | * So, setup a plug here and finish it off before we return | |
556 | */ | |
557 | blk_start_plug(&plug); | |
8b712842 | 558 | |
efa7c9f9 | 559 | bdi = device->bdev->bd_bdi; |
b64a2851 | 560 | |
8b712842 CM |
561 | loop: |
562 | spin_lock(&device->io_lock); | |
563 | ||
a6837051 | 564 | loop_lock: |
d84275c9 | 565 | num_run = 0; |
ffbd517d | 566 | |
8b712842 CM |
567 | /* take all the bios off the list at once and process them |
568 | * later on (without the lock held). But, remember the | |
569 | * tail and other pointers so the bios can be properly reinserted | |
570 | * into the list if we hit congestion | |
571 | */ | |
d84275c9 | 572 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 573 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
574 | force_reg = 1; |
575 | } else { | |
ffbd517d | 576 | pending_bios = &device->pending_bios; |
d84275c9 CM |
577 | force_reg = 0; |
578 | } | |
ffbd517d CM |
579 | |
580 | pending = pending_bios->head; | |
581 | tail = pending_bios->tail; | |
8b712842 | 582 | WARN_ON(pending && !tail); |
8b712842 CM |
583 | |
584 | /* | |
585 | * if pending was null this time around, no bios need processing | |
586 | * at all and we can stop. Otherwise it'll loop back up again | |
587 | * and do an additional check so no bios are missed. | |
588 | * | |
589 | * device->running_pending is used to synchronize with the | |
590 | * schedule_bio code. | |
591 | */ | |
ffbd517d CM |
592 | if (device->pending_sync_bios.head == NULL && |
593 | device->pending_bios.head == NULL) { | |
8b712842 CM |
594 | again = 0; |
595 | device->running_pending = 0; | |
ffbd517d CM |
596 | } else { |
597 | again = 1; | |
598 | device->running_pending = 1; | |
8b712842 | 599 | } |
ffbd517d CM |
600 | |
601 | pending_bios->head = NULL; | |
602 | pending_bios->tail = NULL; | |
603 | ||
8b712842 CM |
604 | spin_unlock(&device->io_lock); |
605 | ||
d397712b | 606 | while (pending) { |
ffbd517d CM |
607 | |
608 | rmb(); | |
d84275c9 CM |
609 | /* we want to work on both lists, but do more bios on the |
610 | * sync list than the regular list | |
611 | */ | |
612 | if ((num_run > 32 && | |
613 | pending_bios != &device->pending_sync_bios && | |
614 | device->pending_sync_bios.head) || | |
615 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
616 | device->pending_bios.head)) { | |
ffbd517d CM |
617 | spin_lock(&device->io_lock); |
618 | requeue_list(pending_bios, pending, tail); | |
619 | goto loop_lock; | |
620 | } | |
621 | ||
8b712842 CM |
622 | cur = pending; |
623 | pending = pending->bi_next; | |
624 | cur->bi_next = NULL; | |
b64a2851 | 625 | |
dac56212 | 626 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 627 | |
2ab1ba68 CM |
628 | /* |
629 | * if we're doing the sync list, record that our | |
630 | * plug has some sync requests on it | |
631 | * | |
632 | * If we're doing the regular list and there are | |
633 | * sync requests sitting around, unplug before | |
634 | * we add more | |
635 | */ | |
636 | if (pending_bios == &device->pending_sync_bios) { | |
637 | sync_pending = 1; | |
638 | } else if (sync_pending) { | |
639 | blk_finish_plug(&plug); | |
640 | blk_start_plug(&plug); | |
641 | sync_pending = 0; | |
642 | } | |
643 | ||
4e49ea4a | 644 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
645 | num_run++; |
646 | batch_run++; | |
853d8ec4 DS |
647 | |
648 | cond_resched(); | |
8b712842 CM |
649 | |
650 | /* | |
651 | * we made progress, there is more work to do and the bdi | |
652 | * is now congested. Back off and let other work structs | |
653 | * run instead | |
654 | */ | |
57fd5a5f | 655 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 656 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 657 | struct io_context *ioc; |
8b712842 | 658 | |
b765ead5 CM |
659 | ioc = current->io_context; |
660 | ||
661 | /* | |
662 | * the main goal here is that we don't want to | |
663 | * block if we're going to be able to submit | |
664 | * more requests without blocking. | |
665 | * | |
666 | * This code does two great things, it pokes into | |
667 | * the elevator code from a filesystem _and_ | |
668 | * it makes assumptions about how batching works. | |
669 | */ | |
670 | if (ioc && ioc->nr_batch_requests > 0 && | |
671 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
672 | (last_waited == 0 || | |
673 | ioc->last_waited == last_waited)) { | |
674 | /* | |
675 | * we want to go through our batch of | |
676 | * requests and stop. So, we copy out | |
677 | * the ioc->last_waited time and test | |
678 | * against it before looping | |
679 | */ | |
680 | last_waited = ioc->last_waited; | |
853d8ec4 | 681 | cond_resched(); |
b765ead5 CM |
682 | continue; |
683 | } | |
8b712842 | 684 | spin_lock(&device->io_lock); |
ffbd517d | 685 | requeue_list(pending_bios, pending, tail); |
a6837051 | 686 | device->running_pending = 1; |
8b712842 CM |
687 | |
688 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
689 | btrfs_queue_work(fs_info->submit_workers, |
690 | &device->work); | |
8b712842 CM |
691 | goto done; |
692 | } | |
693 | } | |
ffbd517d | 694 | |
51684082 CM |
695 | cond_resched(); |
696 | if (again) | |
697 | goto loop; | |
698 | ||
699 | spin_lock(&device->io_lock); | |
700 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
701 | goto loop_lock; | |
702 | spin_unlock(&device->io_lock); | |
703 | ||
8b712842 | 704 | done: |
211588ad | 705 | blk_finish_plug(&plug); |
8b712842 CM |
706 | } |
707 | ||
b2950863 | 708 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
709 | { |
710 | struct btrfs_device *device; | |
711 | ||
712 | device = container_of(work, struct btrfs_device, work); | |
713 | run_scheduled_bios(device); | |
714 | } | |
715 | ||
70bc7088 AJ |
716 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
717 | { | |
718 | int found; | |
719 | ||
720 | rcu_read_lock(); | |
721 | found = strcmp(rcu_str_deref(device->name), path); | |
722 | rcu_read_unlock(); | |
723 | ||
724 | return found == 0; | |
725 | } | |
726 | ||
d8367db3 AJ |
727 | /* |
728 | * Search and remove all stale (devices which are not mounted) devices. | |
729 | * When both inputs are NULL, it will search and release all stale devices. | |
730 | * path: Optional. When provided will it release all unmounted devices | |
731 | * matching this path only. | |
732 | * skip_dev: Optional. Will skip this device when searching for the stale | |
733 | * devices. | |
70bc7088 AJ |
734 | * Return: 0 for success or if @path is NULL. |
735 | * -EBUSY if @path is a mounted device. | |
736 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 737 | */ |
70bc7088 | 738 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 739 | struct btrfs_device *skip_device) |
4fde46f0 | 740 | { |
fa6d2ae5 AJ |
741 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
742 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
743 | int ret = 0; |
744 | ||
745 | if (path) | |
746 | ret = -ENOENT; | |
4fde46f0 | 747 | |
fa6d2ae5 | 748 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 749 | |
70bc7088 | 750 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
751 | list_for_each_entry_safe(device, tmp_device, |
752 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 753 | if (skip_device && skip_device == device) |
d8367db3 | 754 | continue; |
fa6d2ae5 | 755 | if (path && !device->name) |
4fde46f0 | 756 | continue; |
70bc7088 | 757 | if (path && !device_path_matched(path, device)) |
38cf665d | 758 | continue; |
70bc7088 AJ |
759 | if (fs_devices->opened) { |
760 | /* for an already deleted device return 0 */ | |
761 | if (path && ret != 0) | |
762 | ret = -EBUSY; | |
763 | break; | |
764 | } | |
4fde46f0 | 765 | |
4fde46f0 | 766 | /* delete the stale device */ |
7bcb8164 AJ |
767 | fs_devices->num_devices--; |
768 | list_del(&device->dev_list); | |
769 | btrfs_free_device(device); | |
770 | ||
70bc7088 | 771 | ret = 0; |
7bcb8164 | 772 | if (fs_devices->num_devices == 0) |
fd649f10 | 773 | break; |
7bcb8164 AJ |
774 | } |
775 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 776 | |
7bcb8164 AJ |
777 | if (fs_devices->num_devices == 0) { |
778 | btrfs_sysfs_remove_fsid(fs_devices); | |
779 | list_del(&fs_devices->fs_list); | |
780 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
781 | } |
782 | } | |
70bc7088 AJ |
783 | |
784 | return ret; | |
4fde46f0 AJ |
785 | } |
786 | ||
0fb08bcc AJ |
787 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
788 | struct btrfs_device *device, fmode_t flags, | |
789 | void *holder) | |
790 | { | |
791 | struct request_queue *q; | |
792 | struct block_device *bdev; | |
793 | struct buffer_head *bh; | |
794 | struct btrfs_super_block *disk_super; | |
795 | u64 devid; | |
796 | int ret; | |
797 | ||
798 | if (device->bdev) | |
799 | return -EINVAL; | |
800 | if (!device->name) | |
801 | return -EINVAL; | |
802 | ||
803 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
804 | &bdev, &bh); | |
805 | if (ret) | |
806 | return ret; | |
807 | ||
808 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
809 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
810 | if (devid != device->devid) | |
811 | goto error_brelse; | |
812 | ||
813 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
814 | goto error_brelse; | |
815 | ||
816 | device->generation = btrfs_super_generation(disk_super); | |
817 | ||
818 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
819 | if (btrfs_super_incompat_flags(disk_super) & |
820 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
821 | pr_err( | |
822 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
823 | goto error_brelse; | |
824 | } | |
825 | ||
ebbede42 | 826 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
827 | fs_devices->seeding = 1; |
828 | } else { | |
ebbede42 AJ |
829 | if (bdev_read_only(bdev)) |
830 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
831 | else | |
832 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
833 | } |
834 | ||
835 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
836 | if (!blk_queue_nonrot(q)) |
837 | fs_devices->rotating = 1; | |
838 | ||
839 | device->bdev = bdev; | |
e12c9621 | 840 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
841 | device->mode = flags; |
842 | ||
843 | fs_devices->open_devices++; | |
ebbede42 AJ |
844 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
845 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 846 | fs_devices->rw_devices++; |
b1b8e386 | 847 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
848 | } |
849 | brelse(bh); | |
850 | ||
851 | return 0; | |
852 | ||
853 | error_brelse: | |
854 | brelse(bh); | |
855 | blkdev_put(bdev, flags); | |
856 | ||
857 | return -EINVAL; | |
858 | } | |
859 | ||
7a62d0f0 NB |
860 | /* |
861 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
862 | * being created with a disk that has already completed its fsid change. | |
863 | */ | |
864 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
865 | struct btrfs_super_block *disk_super) | |
866 | { | |
867 | struct btrfs_fs_devices *fs_devices; | |
868 | ||
869 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
870 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
871 | BTRFS_FSID_SIZE) != 0 && | |
872 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
873 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
874 | return fs_devices; | |
875 | } | |
876 | } | |
877 | ||
878 | return NULL; | |
879 | } | |
880 | ||
cc5de4e7 NB |
881 | |
882 | static struct btrfs_fs_devices *find_fsid_changed( | |
883 | struct btrfs_super_block *disk_super) | |
884 | { | |
885 | struct btrfs_fs_devices *fs_devices; | |
886 | ||
887 | /* | |
888 | * Handles the case where scanned device is part of an fs that had | |
889 | * multiple successful changes of FSID but curently device didn't | |
890 | * observe it. Meaning our fsid will be different than theirs. | |
891 | */ | |
892 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
893 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
894 | BTRFS_FSID_SIZE) != 0 && | |
895 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
896 | BTRFS_FSID_SIZE) == 0 && | |
897 | memcmp(fs_devices->fsid, disk_super->fsid, | |
898 | BTRFS_FSID_SIZE) != 0) { | |
899 | return fs_devices; | |
900 | } | |
901 | } | |
902 | ||
903 | return NULL; | |
904 | } | |
60999ca4 DS |
905 | /* |
906 | * Add new device to list of registered devices | |
907 | * | |
908 | * Returns: | |
e124ece5 AJ |
909 | * device pointer which was just added or updated when successful |
910 | * error pointer when failed | |
60999ca4 | 911 | */ |
e124ece5 | 912 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
913 | struct btrfs_super_block *disk_super, |
914 | bool *new_device_added) | |
8a4b83cc CM |
915 | { |
916 | struct btrfs_device *device; | |
7a62d0f0 | 917 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 918 | struct rcu_string *name; |
8a4b83cc | 919 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 920 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
921 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
922 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
923 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
924 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 925 | |
cc5de4e7 NB |
926 | if (fsid_change_in_progress) { |
927 | if (!has_metadata_uuid) { | |
928 | /* | |
929 | * When we have an image which has CHANGING_FSID_V2 set | |
930 | * it might belong to either a filesystem which has | |
931 | * disks with completed fsid change or it might belong | |
932 | * to fs with no UUID changes in effect, handle both. | |
933 | */ | |
934 | fs_devices = find_fsid_inprogress(disk_super); | |
935 | if (!fs_devices) | |
936 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
937 | } else { | |
938 | fs_devices = find_fsid_changed(disk_super); | |
939 | } | |
7a62d0f0 NB |
940 | } else if (has_metadata_uuid) { |
941 | fs_devices = find_fsid(disk_super->fsid, | |
942 | disk_super->metadata_uuid); | |
943 | } else { | |
7239ff4b | 944 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
945 | } |
946 | ||
8a4b83cc | 947 | |
8a4b83cc | 948 | if (!fs_devices) { |
7239ff4b NB |
949 | if (has_metadata_uuid) |
950 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
951 | disk_super->metadata_uuid); | |
952 | else | |
953 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
954 | ||
2208a378 | 955 | if (IS_ERR(fs_devices)) |
e124ece5 | 956 | return ERR_CAST(fs_devices); |
2208a378 | 957 | |
92900e51 AV |
958 | fs_devices->fsid_change = fsid_change_in_progress; |
959 | ||
9c6d173e | 960 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 961 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 962 | |
8a4b83cc CM |
963 | device = NULL; |
964 | } else { | |
9c6d173e | 965 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
966 | device = btrfs_find_device(fs_devices, devid, |
967 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
968 | |
969 | /* | |
970 | * If this disk has been pulled into an fs devices created by | |
971 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
972 | * metadata_uuid/fsid values of the fs_devices. | |
973 | */ | |
974 | if (has_metadata_uuid && fs_devices->fsid_change && | |
975 | found_transid > fs_devices->latest_generation) { | |
976 | memcpy(fs_devices->fsid, disk_super->fsid, | |
977 | BTRFS_FSID_SIZE); | |
978 | memcpy(fs_devices->metadata_uuid, | |
979 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
980 | ||
981 | fs_devices->fsid_change = false; | |
982 | } | |
8a4b83cc | 983 | } |
443f24fe | 984 | |
8a4b83cc | 985 | if (!device) { |
9c6d173e AJ |
986 | if (fs_devices->opened) { |
987 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 988 | return ERR_PTR(-EBUSY); |
9c6d173e | 989 | } |
2b82032c | 990 | |
12bd2fc0 ID |
991 | device = btrfs_alloc_device(NULL, &devid, |
992 | disk_super->dev_item.uuid); | |
993 | if (IS_ERR(device)) { | |
9c6d173e | 994 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 995 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 996 | return device; |
8a4b83cc | 997 | } |
606686ee JB |
998 | |
999 | name = rcu_string_strdup(path, GFP_NOFS); | |
1000 | if (!name) { | |
a425f9d4 | 1001 | btrfs_free_device(device); |
9c6d173e | 1002 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1003 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 1004 | } |
606686ee | 1005 | rcu_assign_pointer(device->name, name); |
90519d66 | 1006 | |
1f78160c | 1007 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 1008 | fs_devices->num_devices++; |
e5e9a520 | 1009 | |
2b82032c | 1010 | device->fs_devices = fs_devices; |
4306a974 | 1011 | *new_device_added = true; |
327f18cc AJ |
1012 | |
1013 | if (disk_super->label[0]) | |
1014 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
1015 | disk_super->label, devid, found_transid, path); | |
1016 | else | |
1017 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
1018 | disk_super->fsid, devid, found_transid, path); | |
1019 | ||
606686ee | 1020 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
1021 | /* |
1022 | * When FS is already mounted. | |
1023 | * 1. If you are here and if the device->name is NULL that | |
1024 | * means this device was missing at time of FS mount. | |
1025 | * 2. If you are here and if the device->name is different | |
1026 | * from 'path' that means either | |
1027 | * a. The same device disappeared and reappeared with | |
1028 | * different name. or | |
1029 | * b. The missing-disk-which-was-replaced, has | |
1030 | * reappeared now. | |
1031 | * | |
1032 | * We must allow 1 and 2a above. But 2b would be a spurious | |
1033 | * and unintentional. | |
1034 | * | |
1035 | * Further in case of 1 and 2a above, the disk at 'path' | |
1036 | * would have missed some transaction when it was away and | |
1037 | * in case of 2a the stale bdev has to be updated as well. | |
1038 | * 2b must not be allowed at all time. | |
1039 | */ | |
1040 | ||
1041 | /* | |
0f23ae74 CM |
1042 | * For now, we do allow update to btrfs_fs_device through the |
1043 | * btrfs dev scan cli after FS has been mounted. We're still | |
1044 | * tracking a problem where systems fail mount by subvolume id | |
1045 | * when we reject replacement on a mounted FS. | |
b96de000 | 1046 | */ |
0f23ae74 | 1047 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
1048 | /* |
1049 | * That is if the FS is _not_ mounted and if you | |
1050 | * are here, that means there is more than one | |
1051 | * disk with same uuid and devid.We keep the one | |
1052 | * with larger generation number or the last-in if | |
1053 | * generation are equal. | |
1054 | */ | |
9c6d173e | 1055 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1056 | return ERR_PTR(-EEXIST); |
77bdae4d | 1057 | } |
b96de000 | 1058 | |
a9261d41 AJ |
1059 | /* |
1060 | * We are going to replace the device path for a given devid, | |
1061 | * make sure it's the same device if the device is mounted | |
1062 | */ | |
1063 | if (device->bdev) { | |
1064 | struct block_device *path_bdev; | |
1065 | ||
1066 | path_bdev = lookup_bdev(path); | |
1067 | if (IS_ERR(path_bdev)) { | |
1068 | mutex_unlock(&fs_devices->device_list_mutex); | |
1069 | return ERR_CAST(path_bdev); | |
1070 | } | |
1071 | ||
1072 | if (device->bdev != path_bdev) { | |
1073 | bdput(path_bdev); | |
1074 | mutex_unlock(&fs_devices->device_list_mutex); | |
1075 | btrfs_warn_in_rcu(device->fs_info, | |
1076 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1077 | disk_super->fsid, devid, | |
1078 | rcu_str_deref(device->name), path); | |
1079 | return ERR_PTR(-EEXIST); | |
1080 | } | |
1081 | bdput(path_bdev); | |
1082 | btrfs_info_in_rcu(device->fs_info, | |
1083 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1084 | disk_super->fsid, devid, | |
1085 | rcu_str_deref(device->name), path); | |
1086 | } | |
1087 | ||
606686ee | 1088 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1089 | if (!name) { |
1090 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1091 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1092 | } |
606686ee JB |
1093 | rcu_string_free(device->name); |
1094 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1095 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1096 | fs_devices->missing_devices--; |
e6e674bd | 1097 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1098 | } |
8a4b83cc CM |
1099 | } |
1100 | ||
77bdae4d AJ |
1101 | /* |
1102 | * Unmount does not free the btrfs_device struct but would zero | |
1103 | * generation along with most of the other members. So just update | |
1104 | * it back. We need it to pick the disk with largest generation | |
1105 | * (as above). | |
1106 | */ | |
d1a63002 | 1107 | if (!fs_devices->opened) { |
77bdae4d | 1108 | device->generation = found_transid; |
d1a63002 NB |
1109 | fs_devices->latest_generation = max_t(u64, found_transid, |
1110 | fs_devices->latest_generation); | |
1111 | } | |
77bdae4d | 1112 | |
f2788d2f AJ |
1113 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1114 | ||
9c6d173e | 1115 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1116 | return device; |
8a4b83cc CM |
1117 | } |
1118 | ||
e4404d6e YZ |
1119 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1120 | { | |
1121 | struct btrfs_fs_devices *fs_devices; | |
1122 | struct btrfs_device *device; | |
1123 | struct btrfs_device *orig_dev; | |
1124 | ||
7239ff4b | 1125 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1126 | if (IS_ERR(fs_devices)) |
1127 | return fs_devices; | |
e4404d6e | 1128 | |
adbbb863 | 1129 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1130 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 1131 | |
46224705 | 1132 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 1133 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
1134 | struct rcu_string *name; |
1135 | ||
12bd2fc0 ID |
1136 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1137 | orig_dev->uuid); | |
1138 | if (IS_ERR(device)) | |
e4404d6e YZ |
1139 | goto error; |
1140 | ||
606686ee JB |
1141 | /* |
1142 | * This is ok to do without rcu read locked because we hold the | |
1143 | * uuid mutex so nothing we touch in here is going to disappear. | |
1144 | */ | |
e755f780 | 1145 | if (orig_dev->name) { |
78f2c9e6 DS |
1146 | name = rcu_string_strdup(orig_dev->name->str, |
1147 | GFP_KERNEL); | |
e755f780 | 1148 | if (!name) { |
a425f9d4 | 1149 | btrfs_free_device(device); |
e755f780 AJ |
1150 | goto error; |
1151 | } | |
1152 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1153 | } |
e4404d6e | 1154 | |
e4404d6e YZ |
1155 | list_add(&device->dev_list, &fs_devices->devices); |
1156 | device->fs_devices = fs_devices; | |
1157 | fs_devices->num_devices++; | |
1158 | } | |
adbbb863 | 1159 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1160 | return fs_devices; |
1161 | error: | |
adbbb863 | 1162 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1163 | free_fs_devices(fs_devices); |
1164 | return ERR_PTR(-ENOMEM); | |
1165 | } | |
1166 | ||
9b99b115 AJ |
1167 | /* |
1168 | * After we have read the system tree and know devids belonging to | |
1169 | * this filesystem, remove the device which does not belong there. | |
1170 | */ | |
1171 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1172 | { |
c6e30871 | 1173 | struct btrfs_device *device, *next; |
443f24fe | 1174 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1175 | |
dfe25020 CM |
1176 | mutex_lock(&uuid_mutex); |
1177 | again: | |
46224705 | 1178 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1179 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1180 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1181 | &device->dev_state)) { | |
401e29c1 AJ |
1182 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1183 | &device->dev_state) && | |
1184 | (!latest_dev || | |
1185 | device->generation > latest_dev->generation)) { | |
443f24fe | 1186 | latest_dev = device; |
a6b0d5c8 | 1187 | } |
2b82032c | 1188 | continue; |
a6b0d5c8 | 1189 | } |
2b82032c | 1190 | |
8dabb742 SB |
1191 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1192 | /* | |
1193 | * In the first step, keep the device which has | |
1194 | * the correct fsid and the devid that is used | |
1195 | * for the dev_replace procedure. | |
1196 | * In the second step, the dev_replace state is | |
1197 | * read from the device tree and it is known | |
1198 | * whether the procedure is really active or | |
1199 | * not, which means whether this device is | |
1200 | * used or whether it should be removed. | |
1201 | */ | |
401e29c1 AJ |
1202 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1203 | &device->dev_state)) { | |
8dabb742 SB |
1204 | continue; |
1205 | } | |
1206 | } | |
2b82032c | 1207 | if (device->bdev) { |
d4d77629 | 1208 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1209 | device->bdev = NULL; |
1210 | fs_devices->open_devices--; | |
1211 | } | |
ebbede42 | 1212 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1213 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1214 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1215 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1216 | &device->dev_state)) | |
8dabb742 | 1217 | fs_devices->rw_devices--; |
2b82032c | 1218 | } |
e4404d6e YZ |
1219 | list_del_init(&device->dev_list); |
1220 | fs_devices->num_devices--; | |
a425f9d4 | 1221 | btrfs_free_device(device); |
dfe25020 | 1222 | } |
2b82032c YZ |
1223 | |
1224 | if (fs_devices->seed) { | |
1225 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1226 | goto again; |
1227 | } | |
1228 | ||
443f24fe | 1229 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1230 | |
dfe25020 | 1231 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1232 | } |
a0af469b | 1233 | |
f06c5965 | 1234 | static void free_device_rcu(struct rcu_head *head) |
1f78160c XG |
1235 | { |
1236 | struct btrfs_device *device; | |
1237 | ||
9f5316c1 | 1238 | device = container_of(head, struct btrfs_device, rcu); |
a425f9d4 | 1239 | btrfs_free_device(device); |
1f78160c XG |
1240 | } |
1241 | ||
14238819 AJ |
1242 | static void btrfs_close_bdev(struct btrfs_device *device) |
1243 | { | |
08ffcae8 DS |
1244 | if (!device->bdev) |
1245 | return; | |
1246 | ||
ebbede42 | 1247 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1248 | sync_blockdev(device->bdev); |
1249 | invalidate_bdev(device->bdev); | |
1250 | } | |
1251 | ||
08ffcae8 | 1252 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1253 | } |
1254 | ||
959b1c04 | 1255 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1256 | { |
1257 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1258 | struct btrfs_device *new_device; | |
1259 | struct rcu_string *name; | |
1260 | ||
1261 | if (device->bdev) | |
1262 | fs_devices->open_devices--; | |
1263 | ||
ebbede42 | 1264 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1265 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1266 | list_del_init(&device->dev_alloc_list); | |
1267 | fs_devices->rw_devices--; | |
1268 | } | |
1269 | ||
e6e674bd | 1270 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1271 | fs_devices->missing_devices--; |
1272 | ||
959b1c04 NB |
1273 | btrfs_close_bdev(device); |
1274 | ||
f448341a AJ |
1275 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1276 | device->uuid); | |
1277 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1278 | ||
1279 | /* Safe because we are under uuid_mutex */ | |
1280 | if (device->name) { | |
1281 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1282 | BUG_ON(!name); /* -ENOMEM */ | |
1283 | rcu_assign_pointer(new_device->name, name); | |
1284 | } | |
1285 | ||
1286 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1287 | new_device->fs_devices = device->fs_devices; | |
959b1c04 NB |
1288 | |
1289 | call_rcu(&device->rcu, free_device_rcu); | |
f448341a AJ |
1290 | } |
1291 | ||
0226e0eb | 1292 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1293 | { |
2037a093 | 1294 | struct btrfs_device *device, *tmp; |
e4404d6e | 1295 | |
2b82032c YZ |
1296 | if (--fs_devices->opened > 0) |
1297 | return 0; | |
8a4b83cc | 1298 | |
c9513edb | 1299 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1300 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1301 | btrfs_close_one_device(device); |
8a4b83cc | 1302 | } |
c9513edb XG |
1303 | mutex_unlock(&fs_devices->device_list_mutex); |
1304 | ||
e4404d6e YZ |
1305 | WARN_ON(fs_devices->open_devices); |
1306 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1307 | fs_devices->opened = 0; |
1308 | fs_devices->seeding = 0; | |
2b82032c | 1309 | |
8a4b83cc CM |
1310 | return 0; |
1311 | } | |
1312 | ||
2b82032c YZ |
1313 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1314 | { | |
e4404d6e | 1315 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1316 | int ret; |
1317 | ||
1318 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1319 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1320 | if (!fs_devices->opened) { |
1321 | seed_devices = fs_devices->seed; | |
1322 | fs_devices->seed = NULL; | |
1323 | } | |
2b82032c | 1324 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1325 | |
1326 | while (seed_devices) { | |
1327 | fs_devices = seed_devices; | |
1328 | seed_devices = fs_devices->seed; | |
0226e0eb | 1329 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1330 | free_fs_devices(fs_devices); |
1331 | } | |
2b82032c YZ |
1332 | return ret; |
1333 | } | |
1334 | ||
897fb573 | 1335 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1336 | fmode_t flags, void *holder) |
8a4b83cc | 1337 | { |
8a4b83cc | 1338 | struct btrfs_device *device; |
443f24fe | 1339 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1340 | int ret = 0; |
8a4b83cc | 1341 | |
d4d77629 TH |
1342 | flags |= FMODE_EXCL; |
1343 | ||
f117e290 | 1344 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1345 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1346 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1347 | continue; |
a0af469b | 1348 | |
9f050db4 AJ |
1349 | if (!latest_dev || |
1350 | device->generation > latest_dev->generation) | |
1351 | latest_dev = device; | |
8a4b83cc | 1352 | } |
a0af469b | 1353 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1354 | ret = -EINVAL; |
a0af469b CM |
1355 | goto out; |
1356 | } | |
2b82032c | 1357 | fs_devices->opened = 1; |
443f24fe | 1358 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1359 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1360 | out: |
2b82032c YZ |
1361 | return ret; |
1362 | } | |
1363 | ||
f8e10cd3 AJ |
1364 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1365 | { | |
1366 | struct btrfs_device *dev1, *dev2; | |
1367 | ||
1368 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1369 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1370 | ||
1371 | if (dev1->devid < dev2->devid) | |
1372 | return -1; | |
1373 | else if (dev1->devid > dev2->devid) | |
1374 | return 1; | |
1375 | return 0; | |
1376 | } | |
1377 | ||
2b82032c | 1378 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1379 | fmode_t flags, void *holder) |
2b82032c YZ |
1380 | { |
1381 | int ret; | |
1382 | ||
f5194e34 DS |
1383 | lockdep_assert_held(&uuid_mutex); |
1384 | ||
542c5908 | 1385 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1386 | if (fs_devices->opened) { |
e4404d6e YZ |
1387 | fs_devices->opened++; |
1388 | ret = 0; | |
2b82032c | 1389 | } else { |
f8e10cd3 | 1390 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1391 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1392 | } |
542c5908 AJ |
1393 | mutex_unlock(&fs_devices->device_list_mutex); |
1394 | ||
8a4b83cc CM |
1395 | return ret; |
1396 | } | |
1397 | ||
c9162bdf | 1398 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1399 | { |
1400 | kunmap(page); | |
1401 | put_page(page); | |
1402 | } | |
1403 | ||
c9162bdf OS |
1404 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1405 | struct page **page, | |
1406 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1407 | { |
1408 | void *p; | |
1409 | pgoff_t index; | |
1410 | ||
1411 | /* make sure our super fits in the device */ | |
1412 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1413 | return 1; | |
1414 | ||
1415 | /* make sure our super fits in the page */ | |
1416 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1417 | return 1; | |
1418 | ||
1419 | /* make sure our super doesn't straddle pages on disk */ | |
1420 | index = bytenr >> PAGE_SHIFT; | |
1421 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1422 | return 1; | |
1423 | ||
1424 | /* pull in the page with our super */ | |
1425 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1426 | index, GFP_KERNEL); | |
1427 | ||
1428 | if (IS_ERR_OR_NULL(*page)) | |
1429 | return 1; | |
1430 | ||
1431 | p = kmap(*page); | |
1432 | ||
1433 | /* align our pointer to the offset of the super block */ | |
7073017a | 1434 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1435 | |
1436 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1437 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1438 | btrfs_release_disk_super(*page); | |
1439 | return 1; | |
1440 | } | |
1441 | ||
1442 | if ((*disk_super)->label[0] && | |
1443 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1444 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1445 | ||
1446 | return 0; | |
1447 | } | |
1448 | ||
6f60cbd3 DS |
1449 | /* |
1450 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1451 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1452 | * is read via pagecache | |
1453 | */ | |
36350e95 GJ |
1454 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1455 | void *holder) | |
8a4b83cc CM |
1456 | { |
1457 | struct btrfs_super_block *disk_super; | |
4306a974 | 1458 | bool new_device_added = false; |
36350e95 | 1459 | struct btrfs_device *device = NULL; |
8a4b83cc | 1460 | struct block_device *bdev; |
6f60cbd3 | 1461 | struct page *page; |
6f60cbd3 | 1462 | u64 bytenr; |
8a4b83cc | 1463 | |
899f9307 DS |
1464 | lockdep_assert_held(&uuid_mutex); |
1465 | ||
6f60cbd3 DS |
1466 | /* |
1467 | * we would like to check all the supers, but that would make | |
1468 | * a btrfs mount succeed after a mkfs from a different FS. | |
1469 | * So, we need to add a special mount option to scan for | |
1470 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1471 | */ | |
1472 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1473 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1474 | |
1475 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1476 | if (IS_ERR(bdev)) |
36350e95 | 1477 | return ERR_CAST(bdev); |
6f60cbd3 | 1478 | |
05a5c55d | 1479 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1480 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1481 | goto error_bdev_put; |
05a5c55d | 1482 | } |
6f60cbd3 | 1483 | |
4306a974 | 1484 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1485 | if (!IS_ERR(device)) { |
4306a974 AJ |
1486 | if (new_device_added) |
1487 | btrfs_free_stale_devices(path, device); | |
1488 | } | |
6f60cbd3 | 1489 | |
6cf86a00 | 1490 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1491 | |
1492 | error_bdev_put: | |
d4d77629 | 1493 | blkdev_put(bdev, flags); |
b6ed73bc | 1494 | |
36350e95 | 1495 | return device; |
8a4b83cc | 1496 | } |
0b86a832 | 1497 | |
499f377f | 1498 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1499 | struct btrfs_device *device, |
1500 | u64 *start, u64 len) | |
1501 | { | |
fb456252 | 1502 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1503 | struct extent_map *em; |
499f377f | 1504 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1505 | int ret = 0; |
1b984508 | 1506 | u64 physical_start = *start; |
6df9a95e | 1507 | |
499f377f JM |
1508 | if (transaction) |
1509 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1510 | again: |
1511 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1512 | struct map_lookup *map; |
1513 | int i; | |
1514 | ||
95617d69 | 1515 | map = em->map_lookup; |
6df9a95e | 1516 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1517 | u64 end; |
1518 | ||
6df9a95e JB |
1519 | if (map->stripes[i].dev != device) |
1520 | continue; | |
1b984508 | 1521 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1522 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1523 | physical_start) |
6df9a95e | 1524 | continue; |
c152b63e FM |
1525 | /* |
1526 | * Make sure that while processing the pinned list we do | |
1527 | * not override our *start with a lower value, because | |
1528 | * we can have pinned chunks that fall within this | |
1529 | * device hole and that have lower physical addresses | |
1530 | * than the pending chunks we processed before. If we | |
1531 | * do not take this special care we can end up getting | |
1532 | * 2 pending chunks that start at the same physical | |
1533 | * device offsets because the end offset of a pinned | |
1534 | * chunk can be equal to the start offset of some | |
1535 | * pending chunk. | |
1536 | */ | |
1537 | end = map->stripes[i].physical + em->orig_block_len; | |
1538 | if (end > *start) { | |
1539 | *start = end; | |
1540 | ret = 1; | |
1541 | } | |
6df9a95e JB |
1542 | } |
1543 | } | |
499f377f JM |
1544 | if (search_list != &fs_info->pinned_chunks) { |
1545 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1546 | goto again; |
1547 | } | |
6df9a95e JB |
1548 | |
1549 | return ret; | |
1550 | } | |
1551 | ||
1552 | ||
0b86a832 | 1553 | /* |
499f377f JM |
1554 | * find_free_dev_extent_start - find free space in the specified device |
1555 | * @device: the device which we search the free space in | |
1556 | * @num_bytes: the size of the free space that we need | |
1557 | * @search_start: the position from which to begin the search | |
1558 | * @start: store the start of the free space. | |
1559 | * @len: the size of the free space. that we find, or the size | |
1560 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1561 | * |
0b86a832 CM |
1562 | * this uses a pretty simple search, the expectation is that it is |
1563 | * called very infrequently and that a given device has a small number | |
1564 | * of extents | |
7bfc837d MX |
1565 | * |
1566 | * @start is used to store the start of the free space if we find. But if we | |
1567 | * don't find suitable free space, it will be used to store the start position | |
1568 | * of the max free space. | |
1569 | * | |
1570 | * @len is used to store the size of the free space that we find. | |
1571 | * But if we don't find suitable free space, it is used to store the size of | |
1572 | * the max free space. | |
0b86a832 | 1573 | */ |
499f377f JM |
1574 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1575 | struct btrfs_device *device, u64 num_bytes, | |
1576 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1577 | { |
0b246afa JM |
1578 | struct btrfs_fs_info *fs_info = device->fs_info; |
1579 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1580 | struct btrfs_key key; |
7bfc837d | 1581 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1582 | struct btrfs_path *path; |
7bfc837d MX |
1583 | u64 hole_size; |
1584 | u64 max_hole_start; | |
1585 | u64 max_hole_size; | |
1586 | u64 extent_end; | |
0b86a832 CM |
1587 | u64 search_end = device->total_bytes; |
1588 | int ret; | |
7bfc837d | 1589 | int slot; |
0b86a832 | 1590 | struct extent_buffer *l; |
8cdc7c5b FM |
1591 | |
1592 | /* | |
1593 | * We don't want to overwrite the superblock on the drive nor any area | |
1594 | * used by the boot loader (grub for example), so we make sure to start | |
1595 | * at an offset of at least 1MB. | |
1596 | */ | |
0d0c71b3 | 1597 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1598 | |
6df9a95e JB |
1599 | path = btrfs_alloc_path(); |
1600 | if (!path) | |
1601 | return -ENOMEM; | |
f2ab7618 | 1602 | |
7bfc837d MX |
1603 | max_hole_start = search_start; |
1604 | max_hole_size = 0; | |
1605 | ||
f2ab7618 | 1606 | again: |
401e29c1 AJ |
1607 | if (search_start >= search_end || |
1608 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1609 | ret = -ENOSPC; |
6df9a95e | 1610 | goto out; |
7bfc837d MX |
1611 | } |
1612 | ||
e4058b54 | 1613 | path->reada = READA_FORWARD; |
6df9a95e JB |
1614 | path->search_commit_root = 1; |
1615 | path->skip_locking = 1; | |
7bfc837d | 1616 | |
0b86a832 CM |
1617 | key.objectid = device->devid; |
1618 | key.offset = search_start; | |
1619 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1620 | |
125ccb0a | 1621 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1622 | if (ret < 0) |
7bfc837d | 1623 | goto out; |
1fcbac58 YZ |
1624 | if (ret > 0) { |
1625 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1626 | if (ret < 0) | |
7bfc837d | 1627 | goto out; |
1fcbac58 | 1628 | } |
7bfc837d | 1629 | |
0b86a832 CM |
1630 | while (1) { |
1631 | l = path->nodes[0]; | |
1632 | slot = path->slots[0]; | |
1633 | if (slot >= btrfs_header_nritems(l)) { | |
1634 | ret = btrfs_next_leaf(root, path); | |
1635 | if (ret == 0) | |
1636 | continue; | |
1637 | if (ret < 0) | |
7bfc837d MX |
1638 | goto out; |
1639 | ||
1640 | break; | |
0b86a832 CM |
1641 | } |
1642 | btrfs_item_key_to_cpu(l, &key, slot); | |
1643 | ||
1644 | if (key.objectid < device->devid) | |
1645 | goto next; | |
1646 | ||
1647 | if (key.objectid > device->devid) | |
7bfc837d | 1648 | break; |
0b86a832 | 1649 | |
962a298f | 1650 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1651 | goto next; |
9779b72f | 1652 | |
7bfc837d MX |
1653 | if (key.offset > search_start) { |
1654 | hole_size = key.offset - search_start; | |
9779b72f | 1655 | |
6df9a95e JB |
1656 | /* |
1657 | * Have to check before we set max_hole_start, otherwise | |
1658 | * we could end up sending back this offset anyway. | |
1659 | */ | |
499f377f | 1660 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1661 | &search_start, |
1b984508 FL |
1662 | hole_size)) { |
1663 | if (key.offset >= search_start) { | |
1664 | hole_size = key.offset - search_start; | |
1665 | } else { | |
1666 | WARN_ON_ONCE(1); | |
1667 | hole_size = 0; | |
1668 | } | |
1669 | } | |
6df9a95e | 1670 | |
7bfc837d MX |
1671 | if (hole_size > max_hole_size) { |
1672 | max_hole_start = search_start; | |
1673 | max_hole_size = hole_size; | |
1674 | } | |
9779b72f | 1675 | |
7bfc837d MX |
1676 | /* |
1677 | * If this free space is greater than which we need, | |
1678 | * it must be the max free space that we have found | |
1679 | * until now, so max_hole_start must point to the start | |
1680 | * of this free space and the length of this free space | |
1681 | * is stored in max_hole_size. Thus, we return | |
1682 | * max_hole_start and max_hole_size and go back to the | |
1683 | * caller. | |
1684 | */ | |
1685 | if (hole_size >= num_bytes) { | |
1686 | ret = 0; | |
1687 | goto out; | |
0b86a832 CM |
1688 | } |
1689 | } | |
0b86a832 | 1690 | |
0b86a832 | 1691 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1692 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1693 | dev_extent); | |
1694 | if (extent_end > search_start) | |
1695 | search_start = extent_end; | |
0b86a832 CM |
1696 | next: |
1697 | path->slots[0]++; | |
1698 | cond_resched(); | |
1699 | } | |
0b86a832 | 1700 | |
38c01b96 | 1701 | /* |
1702 | * At this point, search_start should be the end of | |
1703 | * allocated dev extents, and when shrinking the device, | |
1704 | * search_end may be smaller than search_start. | |
1705 | */ | |
f2ab7618 | 1706 | if (search_end > search_start) { |
38c01b96 | 1707 | hole_size = search_end - search_start; |
1708 | ||
499f377f | 1709 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1710 | hole_size)) { |
1711 | btrfs_release_path(path); | |
1712 | goto again; | |
1713 | } | |
0b86a832 | 1714 | |
f2ab7618 ZL |
1715 | if (hole_size > max_hole_size) { |
1716 | max_hole_start = search_start; | |
1717 | max_hole_size = hole_size; | |
1718 | } | |
6df9a95e JB |
1719 | } |
1720 | ||
7bfc837d | 1721 | /* See above. */ |
f2ab7618 | 1722 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1723 | ret = -ENOSPC; |
1724 | else | |
1725 | ret = 0; | |
1726 | ||
1727 | out: | |
2b82032c | 1728 | btrfs_free_path(path); |
7bfc837d | 1729 | *start = max_hole_start; |
b2117a39 | 1730 | if (len) |
7bfc837d | 1731 | *len = max_hole_size; |
0b86a832 CM |
1732 | return ret; |
1733 | } | |
1734 | ||
499f377f JM |
1735 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1736 | struct btrfs_device *device, u64 num_bytes, | |
1737 | u64 *start, u64 *len) | |
1738 | { | |
499f377f | 1739 | /* FIXME use last free of some kind */ |
499f377f | 1740 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1741 | num_bytes, 0, start, len); |
499f377f JM |
1742 | } |
1743 | ||
b2950863 | 1744 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1745 | struct btrfs_device *device, |
2196d6e8 | 1746 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1747 | { |
0b246afa JM |
1748 | struct btrfs_fs_info *fs_info = device->fs_info; |
1749 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1750 | int ret; |
1751 | struct btrfs_path *path; | |
8f18cf13 | 1752 | struct btrfs_key key; |
a061fc8d CM |
1753 | struct btrfs_key found_key; |
1754 | struct extent_buffer *leaf = NULL; | |
1755 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1756 | |
1757 | path = btrfs_alloc_path(); | |
1758 | if (!path) | |
1759 | return -ENOMEM; | |
1760 | ||
1761 | key.objectid = device->devid; | |
1762 | key.offset = start; | |
1763 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1764 | again: |
8f18cf13 | 1765 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1766 | if (ret > 0) { |
1767 | ret = btrfs_previous_item(root, path, key.objectid, | |
1768 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1769 | if (ret) |
1770 | goto out; | |
a061fc8d CM |
1771 | leaf = path->nodes[0]; |
1772 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1773 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1774 | struct btrfs_dev_extent); | |
1775 | BUG_ON(found_key.offset > start || found_key.offset + | |
1776 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1777 | key = found_key; |
1778 | btrfs_release_path(path); | |
1779 | goto again; | |
a061fc8d CM |
1780 | } else if (ret == 0) { |
1781 | leaf = path->nodes[0]; | |
1782 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1783 | struct btrfs_dev_extent); | |
79787eaa | 1784 | } else { |
0b246afa | 1785 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1786 | goto out; |
a061fc8d | 1787 | } |
8f18cf13 | 1788 | |
2196d6e8 MX |
1789 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1790 | ||
8f18cf13 | 1791 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1792 | if (ret) { |
0b246afa JM |
1793 | btrfs_handle_fs_error(fs_info, ret, |
1794 | "Failed to remove dev extent item"); | |
13212b54 | 1795 | } else { |
3204d33c | 1796 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1797 | } |
b0b802d7 | 1798 | out: |
8f18cf13 CM |
1799 | btrfs_free_path(path); |
1800 | return ret; | |
1801 | } | |
1802 | ||
48a3b636 ES |
1803 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1804 | struct btrfs_device *device, | |
48a3b636 | 1805 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1806 | { |
1807 | int ret; | |
1808 | struct btrfs_path *path; | |
0b246afa JM |
1809 | struct btrfs_fs_info *fs_info = device->fs_info; |
1810 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1811 | struct btrfs_dev_extent *extent; |
1812 | struct extent_buffer *leaf; | |
1813 | struct btrfs_key key; | |
1814 | ||
e12c9621 | 1815 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1816 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1817 | path = btrfs_alloc_path(); |
1818 | if (!path) | |
1819 | return -ENOMEM; | |
1820 | ||
0b86a832 | 1821 | key.objectid = device->devid; |
2b82032c | 1822 | key.offset = start; |
0b86a832 CM |
1823 | key.type = BTRFS_DEV_EXTENT_KEY; |
1824 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1825 | sizeof(*extent)); | |
2cdcecbc MF |
1826 | if (ret) |
1827 | goto out; | |
0b86a832 CM |
1828 | |
1829 | leaf = path->nodes[0]; | |
1830 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1831 | struct btrfs_dev_extent); | |
b5d9071c NB |
1832 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1833 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1834 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1835 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1836 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1837 | ||
0b86a832 CM |
1838 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1839 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1840 | out: |
0b86a832 CM |
1841 | btrfs_free_path(path); |
1842 | return ret; | |
1843 | } | |
1844 | ||
6df9a95e | 1845 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1846 | { |
6df9a95e JB |
1847 | struct extent_map_tree *em_tree; |
1848 | struct extent_map *em; | |
1849 | struct rb_node *n; | |
1850 | u64 ret = 0; | |
0b86a832 | 1851 | |
6df9a95e JB |
1852 | em_tree = &fs_info->mapping_tree.map_tree; |
1853 | read_lock(&em_tree->lock); | |
07e1ce09 | 1854 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1855 | if (n) { |
1856 | em = rb_entry(n, struct extent_map, rb_node); | |
1857 | ret = em->start + em->len; | |
0b86a832 | 1858 | } |
6df9a95e JB |
1859 | read_unlock(&em_tree->lock); |
1860 | ||
0b86a832 CM |
1861 | return ret; |
1862 | } | |
1863 | ||
53f10659 ID |
1864 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1865 | u64 *devid_ret) | |
0b86a832 CM |
1866 | { |
1867 | int ret; | |
1868 | struct btrfs_key key; | |
1869 | struct btrfs_key found_key; | |
2b82032c YZ |
1870 | struct btrfs_path *path; |
1871 | ||
2b82032c YZ |
1872 | path = btrfs_alloc_path(); |
1873 | if (!path) | |
1874 | return -ENOMEM; | |
0b86a832 CM |
1875 | |
1876 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1877 | key.type = BTRFS_DEV_ITEM_KEY; | |
1878 | key.offset = (u64)-1; | |
1879 | ||
53f10659 | 1880 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1881 | if (ret < 0) |
1882 | goto error; | |
1883 | ||
79787eaa | 1884 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1885 | |
53f10659 ID |
1886 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1887 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1888 | BTRFS_DEV_ITEM_KEY); |
1889 | if (ret) { | |
53f10659 | 1890 | *devid_ret = 1; |
0b86a832 CM |
1891 | } else { |
1892 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1893 | path->slots[0]); | |
53f10659 | 1894 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1895 | } |
1896 | ret = 0; | |
1897 | error: | |
2b82032c | 1898 | btrfs_free_path(path); |
0b86a832 CM |
1899 | return ret; |
1900 | } | |
1901 | ||
1902 | /* | |
1903 | * the device information is stored in the chunk root | |
1904 | * the btrfs_device struct should be fully filled in | |
1905 | */ | |
c74a0b02 | 1906 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1907 | struct btrfs_device *device) |
0b86a832 CM |
1908 | { |
1909 | int ret; | |
1910 | struct btrfs_path *path; | |
1911 | struct btrfs_dev_item *dev_item; | |
1912 | struct extent_buffer *leaf; | |
1913 | struct btrfs_key key; | |
1914 | unsigned long ptr; | |
0b86a832 | 1915 | |
0b86a832 CM |
1916 | path = btrfs_alloc_path(); |
1917 | if (!path) | |
1918 | return -ENOMEM; | |
1919 | ||
0b86a832 CM |
1920 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1921 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1922 | key.offset = device->devid; |
0b86a832 | 1923 | |
8e87e856 NB |
1924 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1925 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1926 | if (ret) |
1927 | goto out; | |
1928 | ||
1929 | leaf = path->nodes[0]; | |
1930 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1931 | ||
1932 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1933 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1934 | btrfs_set_device_type(leaf, dev_item, device->type); |
1935 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1936 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1937 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1938 | btrfs_set_device_total_bytes(leaf, dev_item, |
1939 | btrfs_device_get_disk_total_bytes(device)); | |
1940 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1941 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1942 | btrfs_set_device_group(leaf, dev_item, 0); |
1943 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1944 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1945 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1946 | |
410ba3a2 | 1947 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1948 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1949 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1950 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1951 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1952 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1953 | |
2b82032c | 1954 | ret = 0; |
0b86a832 CM |
1955 | out: |
1956 | btrfs_free_path(path); | |
1957 | return ret; | |
1958 | } | |
8f18cf13 | 1959 | |
5a1972bd QW |
1960 | /* |
1961 | * Function to update ctime/mtime for a given device path. | |
1962 | * Mainly used for ctime/mtime based probe like libblkid. | |
1963 | */ | |
da353f6b | 1964 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1965 | { |
1966 | struct file *filp; | |
1967 | ||
1968 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1969 | if (IS_ERR(filp)) |
5a1972bd QW |
1970 | return; |
1971 | file_update_time(filp); | |
1972 | filp_close(filp, NULL); | |
5a1972bd QW |
1973 | } |
1974 | ||
5b4aacef | 1975 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1976 | struct btrfs_device *device) |
1977 | { | |
5b4aacef | 1978 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1979 | int ret; |
1980 | struct btrfs_path *path; | |
a061fc8d | 1981 | struct btrfs_key key; |
a061fc8d CM |
1982 | struct btrfs_trans_handle *trans; |
1983 | ||
a061fc8d CM |
1984 | path = btrfs_alloc_path(); |
1985 | if (!path) | |
1986 | return -ENOMEM; | |
1987 | ||
a22285a6 | 1988 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1989 | if (IS_ERR(trans)) { |
1990 | btrfs_free_path(path); | |
1991 | return PTR_ERR(trans); | |
1992 | } | |
a061fc8d CM |
1993 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1994 | key.type = BTRFS_DEV_ITEM_KEY; | |
1995 | key.offset = device->devid; | |
1996 | ||
1997 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1998 | if (ret) { |
1999 | if (ret > 0) | |
2000 | ret = -ENOENT; | |
2001 | btrfs_abort_transaction(trans, ret); | |
2002 | btrfs_end_transaction(trans); | |
a061fc8d CM |
2003 | goto out; |
2004 | } | |
2005 | ||
2006 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
2007 | if (ret) { |
2008 | btrfs_abort_transaction(trans, ret); | |
2009 | btrfs_end_transaction(trans); | |
2010 | } | |
2011 | ||
a061fc8d CM |
2012 | out: |
2013 | btrfs_free_path(path); | |
5e9f2ad5 NB |
2014 | if (!ret) |
2015 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
2016 | return ret; |
2017 | } | |
2018 | ||
3cc31a0d DS |
2019 | /* |
2020 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
2021 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
2022 | * replace. | |
2023 | */ | |
2024 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
2025 | u64 num_devices) | |
a061fc8d | 2026 | { |
a061fc8d | 2027 | u64 all_avail; |
de98ced9 | 2028 | unsigned seq; |
418775a2 | 2029 | int i; |
a061fc8d | 2030 | |
de98ced9 | 2031 | do { |
bd45ffbc | 2032 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2033 | |
bd45ffbc AJ |
2034 | all_avail = fs_info->avail_data_alloc_bits | |
2035 | fs_info->avail_system_alloc_bits | | |
2036 | fs_info->avail_metadata_alloc_bits; | |
2037 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2038 | |
418775a2 | 2039 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2040 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2041 | continue; |
a061fc8d | 2042 | |
418775a2 | 2043 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2044 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2045 | |
418775a2 DS |
2046 | if (ret) |
2047 | return ret; | |
2048 | } | |
53b381b3 DW |
2049 | } |
2050 | ||
bd45ffbc | 2051 | return 0; |
f1fa7f26 AJ |
2052 | } |
2053 | ||
c9162bdf OS |
2054 | static struct btrfs_device * btrfs_find_next_active_device( |
2055 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2056 | { |
2b82032c | 2057 | struct btrfs_device *next_device; |
88acff64 AJ |
2058 | |
2059 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2060 | if (next_device != device && | |
e6e674bd AJ |
2061 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2062 | && next_device->bdev) | |
88acff64 AJ |
2063 | return next_device; |
2064 | } | |
2065 | ||
2066 | return NULL; | |
2067 | } | |
2068 | ||
2069 | /* | |
2070 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2071 | * and replace it with the provided or the next active device, in the context | |
2072 | * where this function called, there should be always be another device (or | |
2073 | * this_dev) which is active. | |
2074 | */ | |
d6507cf1 NB |
2075 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2076 | struct btrfs_device *this_dev) | |
88acff64 | 2077 | { |
d6507cf1 | 2078 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2079 | struct btrfs_device *next_device; |
2080 | ||
2081 | if (this_dev) | |
2082 | next_device = this_dev; | |
2083 | else | |
2084 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2085 | device); | |
2086 | ASSERT(next_device); | |
2087 | ||
2088 | if (fs_info->sb->s_bdev && | |
2089 | (fs_info->sb->s_bdev == device->bdev)) | |
2090 | fs_info->sb->s_bdev = next_device->bdev; | |
2091 | ||
2092 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2093 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2094 | } | |
2095 | ||
1da73967 AJ |
2096 | /* |
2097 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2098 | * currently replaced. | |
2099 | */ | |
2100 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2101 | { | |
2102 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2103 | ||
cb5583dd | 2104 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2105 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2106 | ASSERT(num_devices > 1); | |
2107 | num_devices--; | |
2108 | } | |
cb5583dd | 2109 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2110 | |
2111 | return num_devices; | |
2112 | } | |
2113 | ||
da353f6b DS |
2114 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2115 | u64 devid) | |
f1fa7f26 AJ |
2116 | { |
2117 | struct btrfs_device *device; | |
1f78160c | 2118 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2119 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2120 | u64 num_devices; |
a061fc8d CM |
2121 | int ret = 0; |
2122 | ||
a061fc8d CM |
2123 | mutex_lock(&uuid_mutex); |
2124 | ||
1da73967 | 2125 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2126 | |
0b246afa | 2127 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2128 | if (ret) |
a061fc8d | 2129 | goto out; |
a061fc8d | 2130 | |
a27a94c2 NB |
2131 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2132 | ||
2133 | if (IS_ERR(device)) { | |
2134 | if (PTR_ERR(device) == -ENOENT && | |
2135 | strcmp(device_path, "missing") == 0) | |
2136 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2137 | else | |
2138 | ret = PTR_ERR(device); | |
53b381b3 | 2139 | goto out; |
a27a94c2 | 2140 | } |
dfe25020 | 2141 | |
eede2bf3 OS |
2142 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2143 | btrfs_warn_in_rcu(fs_info, | |
2144 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2145 | rcu_str_deref(device->name), device->devid); | |
2146 | ret = -ETXTBSY; | |
2147 | goto out; | |
2148 | } | |
2149 | ||
401e29c1 | 2150 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2151 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2152 | goto out; |
63a212ab SB |
2153 | } |
2154 | ||
ebbede42 AJ |
2155 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2156 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2157 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2158 | goto out; |
2b82032c YZ |
2159 | } |
2160 | ||
ebbede42 | 2161 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2162 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2163 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2164 | device->fs_devices->rw_devices--; |
34441361 | 2165 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2166 | } |
a061fc8d | 2167 | |
d7901554 | 2168 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2169 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2170 | mutex_lock(&uuid_mutex); |
a061fc8d | 2171 | if (ret) |
9b3517e9 | 2172 | goto error_undo; |
a061fc8d | 2173 | |
63a212ab SB |
2174 | /* |
2175 | * TODO: the superblock still includes this device in its num_devices | |
2176 | * counter although write_all_supers() is not locked out. This | |
2177 | * could give a filesystem state which requires a degraded mount. | |
2178 | */ | |
0b246afa | 2179 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 2180 | if (ret) |
9b3517e9 | 2181 | goto error_undo; |
a061fc8d | 2182 | |
e12c9621 | 2183 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 2184 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
2185 | |
2186 | /* | |
2187 | * the device list mutex makes sure that we don't change | |
2188 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2189 | * the device supers. Whoever is writing all supers, should |
2190 | * lock the device list mutex before getting the number of | |
2191 | * devices in the super block (super_copy). Conversely, | |
2192 | * whoever updates the number of devices in the super block | |
2193 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2194 | */ |
1f78160c | 2195 | |
41a52a0f AJ |
2196 | /* |
2197 | * In normal cases the cur_devices == fs_devices. But in case | |
2198 | * of deleting a seed device, the cur_devices should point to | |
2199 | * its own fs_devices listed under the fs_devices->seed. | |
2200 | */ | |
1f78160c | 2201 | cur_devices = device->fs_devices; |
b5185197 | 2202 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2203 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2204 | |
41a52a0f AJ |
2205 | cur_devices->num_devices--; |
2206 | cur_devices->total_devices--; | |
b4993e64 AJ |
2207 | /* Update total_devices of the parent fs_devices if it's seed */ |
2208 | if (cur_devices != fs_devices) | |
2209 | fs_devices->total_devices--; | |
2b82032c | 2210 | |
e6e674bd | 2211 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2212 | cur_devices->missing_devices--; |
cd02dca5 | 2213 | |
d6507cf1 | 2214 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2215 | |
0bfaa9c5 | 2216 | if (device->bdev) { |
41a52a0f | 2217 | cur_devices->open_devices--; |
0bfaa9c5 | 2218 | /* remove sysfs entry */ |
b5185197 | 2219 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2220 | } |
99994cde | 2221 | |
0b246afa JM |
2222 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2223 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2224 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2225 | |
cea67ab9 JM |
2226 | /* |
2227 | * at this point, the device is zero sized and detached from | |
2228 | * the devices list. All that's left is to zero out the old | |
2229 | * supers and free the device. | |
2230 | */ | |
ebbede42 | 2231 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2232 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2233 | ||
2234 | btrfs_close_bdev(device); | |
f06c5965 | 2235 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2236 | |
1f78160c | 2237 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2238 | while (fs_devices) { |
8321cf25 RS |
2239 | if (fs_devices->seed == cur_devices) { |
2240 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2241 | break; |
8321cf25 | 2242 | } |
e4404d6e | 2243 | fs_devices = fs_devices->seed; |
2b82032c | 2244 | } |
1f78160c | 2245 | cur_devices->seed = NULL; |
0226e0eb | 2246 | close_fs_devices(cur_devices); |
1f78160c | 2247 | free_fs_devices(cur_devices); |
2b82032c YZ |
2248 | } |
2249 | ||
a061fc8d CM |
2250 | out: |
2251 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2252 | return ret; |
24fc572f | 2253 | |
9b3517e9 | 2254 | error_undo: |
ebbede42 | 2255 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2256 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2257 | list_add(&device->dev_alloc_list, |
b5185197 | 2258 | &fs_devices->alloc_list); |
c3929c36 | 2259 | device->fs_devices->rw_devices++; |
34441361 | 2260 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2261 | } |
24fc572f | 2262 | goto out; |
a061fc8d CM |
2263 | } |
2264 | ||
68a9db5f | 2265 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2266 | { |
d51908ce AJ |
2267 | struct btrfs_fs_devices *fs_devices; |
2268 | ||
68a9db5f | 2269 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2270 | |
25e8e911 AJ |
2271 | /* |
2272 | * in case of fs with no seed, srcdev->fs_devices will point | |
2273 | * to fs_devices of fs_info. However when the dev being replaced is | |
2274 | * a seed dev it will point to the seed's local fs_devices. In short | |
2275 | * srcdev will have its correct fs_devices in both the cases. | |
2276 | */ | |
2277 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2278 | |
e93c89c1 | 2279 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2280 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2281 | fs_devices->num_devices--; |
e6e674bd | 2282 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2283 | fs_devices->missing_devices--; |
e93c89c1 | 2284 | |
ebbede42 | 2285 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2286 | fs_devices->rw_devices--; |
1357272f | 2287 | |
82372bc8 | 2288 | if (srcdev->bdev) |
d51908ce | 2289 | fs_devices->open_devices--; |
084b6e7c QW |
2290 | } |
2291 | ||
2292 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2293 | struct btrfs_device *srcdev) | |
2294 | { | |
2295 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2296 | |
ebbede42 | 2297 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2298 | /* zero out the old super if it is writable */ |
2299 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2300 | } | |
14238819 AJ |
2301 | |
2302 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2303 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2304 | |
94d5f0c2 AJ |
2305 | /* if this is no devs we rather delete the fs_devices */ |
2306 | if (!fs_devices->num_devices) { | |
2307 | struct btrfs_fs_devices *tmp_fs_devices; | |
2308 | ||
6dd38f81 AJ |
2309 | /* |
2310 | * On a mounted FS, num_devices can't be zero unless it's a | |
2311 | * seed. In case of a seed device being replaced, the replace | |
2312 | * target added to the sprout FS, so there will be no more | |
2313 | * device left under the seed FS. | |
2314 | */ | |
2315 | ASSERT(fs_devices->seeding); | |
2316 | ||
94d5f0c2 AJ |
2317 | tmp_fs_devices = fs_info->fs_devices; |
2318 | while (tmp_fs_devices) { | |
2319 | if (tmp_fs_devices->seed == fs_devices) { | |
2320 | tmp_fs_devices->seed = fs_devices->seed; | |
2321 | break; | |
2322 | } | |
2323 | tmp_fs_devices = tmp_fs_devices->seed; | |
2324 | } | |
2325 | fs_devices->seed = NULL; | |
0226e0eb | 2326 | close_fs_devices(fs_devices); |
8bef8401 | 2327 | free_fs_devices(fs_devices); |
94d5f0c2 | 2328 | } |
e93c89c1 SB |
2329 | } |
2330 | ||
4f5ad7bd | 2331 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2332 | { |
4f5ad7bd | 2333 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2334 | |
e93c89c1 | 2335 | WARN_ON(!tgtdev); |
d9a071f0 | 2336 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2337 | |
d9a071f0 | 2338 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2339 | |
779bf3fe | 2340 | if (tgtdev->bdev) |
d9a071f0 | 2341 | fs_devices->open_devices--; |
779bf3fe | 2342 | |
d9a071f0 | 2343 | fs_devices->num_devices--; |
e93c89c1 | 2344 | |
d6507cf1 | 2345 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2346 | |
e93c89c1 | 2347 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2348 | |
d9a071f0 | 2349 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2350 | |
2351 | /* | |
2352 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2353 | * may lead to a call to btrfs_show_devname() which will try | |
2354 | * to hold device_list_mutex. And here this device | |
2355 | * is already out of device list, so we don't have to hold | |
2356 | * the device_list_mutex lock. | |
2357 | */ | |
2358 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2359 | |
2360 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2361 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2362 | } |
2363 | ||
b444ad46 NB |
2364 | static struct btrfs_device *btrfs_find_device_by_path( |
2365 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2366 | { |
2367 | int ret = 0; | |
2368 | struct btrfs_super_block *disk_super; | |
2369 | u64 devid; | |
2370 | u8 *dev_uuid; | |
2371 | struct block_device *bdev; | |
2372 | struct buffer_head *bh; | |
b444ad46 | 2373 | struct btrfs_device *device; |
7ba15b7d | 2374 | |
7ba15b7d | 2375 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2376 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2377 | if (ret) |
b444ad46 | 2378 | return ERR_PTR(ret); |
7ba15b7d SB |
2379 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2380 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2381 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2382 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2383 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2384 | disk_super->metadata_uuid, true); |
7239ff4b | 2385 | else |
e4319cd9 | 2386 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2387 | disk_super->fsid, true); |
7239ff4b | 2388 | |
7ba15b7d | 2389 | brelse(bh); |
b444ad46 NB |
2390 | if (!device) |
2391 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2392 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2393 | return device; |
7ba15b7d SB |
2394 | } |
2395 | ||
5c5c0df0 DS |
2396 | /* |
2397 | * Lookup a device given by device id, or the path if the id is 0. | |
2398 | */ | |
a27a94c2 | 2399 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2400 | struct btrfs_fs_info *fs_info, u64 devid, |
2401 | const char *device_path) | |
24e0474b | 2402 | { |
a27a94c2 | 2403 | struct btrfs_device *device; |
24e0474b | 2404 | |
5c5c0df0 | 2405 | if (devid) { |
e4319cd9 | 2406 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2407 | NULL, true); |
a27a94c2 NB |
2408 | if (!device) |
2409 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2410 | return device; |
2411 | } | |
2412 | ||
2413 | if (!device_path || !device_path[0]) | |
2414 | return ERR_PTR(-EINVAL); | |
2415 | ||
2416 | if (strcmp(device_path, "missing") == 0) { | |
2417 | /* Find first missing device */ | |
2418 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2419 | dev_list) { | |
2420 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2421 | &device->dev_state) && !device->bdev) | |
2422 | return device; | |
d95a830c | 2423 | } |
6e927ceb | 2424 | return ERR_PTR(-ENOENT); |
24e0474b | 2425 | } |
6e927ceb AJ |
2426 | |
2427 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2428 | } |
2429 | ||
2b82032c YZ |
2430 | /* |
2431 | * does all the dirty work required for changing file system's UUID. | |
2432 | */ | |
2ff7e61e | 2433 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2434 | { |
0b246afa | 2435 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2436 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2437 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2438 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2439 | struct btrfs_device *device; |
2440 | u64 super_flags; | |
2441 | ||
a32bf9a3 | 2442 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2443 | if (!fs_devices->seeding) |
2b82032c YZ |
2444 | return -EINVAL; |
2445 | ||
7239ff4b | 2446 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2447 | if (IS_ERR(seed_devices)) |
2448 | return PTR_ERR(seed_devices); | |
2b82032c | 2449 | |
e4404d6e YZ |
2450 | old_devices = clone_fs_devices(fs_devices); |
2451 | if (IS_ERR(old_devices)) { | |
2452 | kfree(seed_devices); | |
2453 | return PTR_ERR(old_devices); | |
2b82032c | 2454 | } |
e4404d6e | 2455 | |
c4babc5e | 2456 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2457 | |
e4404d6e YZ |
2458 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2459 | seed_devices->opened = 1; | |
2460 | INIT_LIST_HEAD(&seed_devices->devices); | |
2461 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2462 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2463 | |
321a4bf7 | 2464 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2465 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2466 | synchronize_rcu); | |
2196d6e8 MX |
2467 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2468 | device->fs_devices = seed_devices; | |
c9513edb | 2469 | |
34441361 | 2470 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2471 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2472 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2473 | |
2b82032c YZ |
2474 | fs_devices->seeding = 0; |
2475 | fs_devices->num_devices = 0; | |
2476 | fs_devices->open_devices = 0; | |
69611ac8 | 2477 | fs_devices->missing_devices = 0; |
69611ac8 | 2478 | fs_devices->rotating = 0; |
e4404d6e | 2479 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2480 | |
2481 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2482 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2483 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2484 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2485 | |
2b82032c YZ |
2486 | super_flags = btrfs_super_flags(disk_super) & |
2487 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2488 | btrfs_set_super_flags(disk_super, super_flags); | |
2489 | ||
2490 | return 0; | |
2491 | } | |
2492 | ||
2493 | /* | |
01327610 | 2494 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2495 | */ |
2496 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2497 | struct btrfs_fs_info *fs_info) |
2b82032c | 2498 | { |
5b4aacef | 2499 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2500 | struct btrfs_path *path; |
2501 | struct extent_buffer *leaf; | |
2502 | struct btrfs_dev_item *dev_item; | |
2503 | struct btrfs_device *device; | |
2504 | struct btrfs_key key; | |
44880fdc | 2505 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2506 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2507 | u64 devid; | |
2508 | int ret; | |
2509 | ||
2510 | path = btrfs_alloc_path(); | |
2511 | if (!path) | |
2512 | return -ENOMEM; | |
2513 | ||
2b82032c YZ |
2514 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2515 | key.offset = 0; | |
2516 | key.type = BTRFS_DEV_ITEM_KEY; | |
2517 | ||
2518 | while (1) { | |
2519 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2520 | if (ret < 0) | |
2521 | goto error; | |
2522 | ||
2523 | leaf = path->nodes[0]; | |
2524 | next_slot: | |
2525 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2526 | ret = btrfs_next_leaf(root, path); | |
2527 | if (ret > 0) | |
2528 | break; | |
2529 | if (ret < 0) | |
2530 | goto error; | |
2531 | leaf = path->nodes[0]; | |
2532 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2533 | btrfs_release_path(path); |
2b82032c YZ |
2534 | continue; |
2535 | } | |
2536 | ||
2537 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2538 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2539 | key.type != BTRFS_DEV_ITEM_KEY) | |
2540 | break; | |
2541 | ||
2542 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2543 | struct btrfs_dev_item); | |
2544 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2545 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2546 | BTRFS_UUID_SIZE); |
1473b24e | 2547 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2548 | BTRFS_FSID_SIZE); |
e4319cd9 | 2549 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2550 | fs_uuid, true); |
79787eaa | 2551 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2552 | |
2553 | if (device->fs_devices->seeding) { | |
2554 | btrfs_set_device_generation(leaf, dev_item, | |
2555 | device->generation); | |
2556 | btrfs_mark_buffer_dirty(leaf); | |
2557 | } | |
2558 | ||
2559 | path->slots[0]++; | |
2560 | goto next_slot; | |
2561 | } | |
2562 | ret = 0; | |
2563 | error: | |
2564 | btrfs_free_path(path); | |
2565 | return ret; | |
2566 | } | |
2567 | ||
da353f6b | 2568 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2569 | { |
5112febb | 2570 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2571 | struct request_queue *q; |
788f20eb CM |
2572 | struct btrfs_trans_handle *trans; |
2573 | struct btrfs_device *device; | |
2574 | struct block_device *bdev; | |
0b246afa | 2575 | struct super_block *sb = fs_info->sb; |
606686ee | 2576 | struct rcu_string *name; |
5da54bc1 | 2577 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2578 | u64 orig_super_total_bytes; |
2579 | u64 orig_super_num_devices; | |
2b82032c | 2580 | int seeding_dev = 0; |
788f20eb | 2581 | int ret = 0; |
7132a262 | 2582 | bool unlocked = false; |
788f20eb | 2583 | |
5da54bc1 | 2584 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2585 | return -EROFS; |
788f20eb | 2586 | |
a5d16333 | 2587 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2588 | fs_info->bdev_holder); |
7f59203a JB |
2589 | if (IS_ERR(bdev)) |
2590 | return PTR_ERR(bdev); | |
a2135011 | 2591 | |
5da54bc1 | 2592 | if (fs_devices->seeding) { |
2b82032c YZ |
2593 | seeding_dev = 1; |
2594 | down_write(&sb->s_umount); | |
2595 | mutex_lock(&uuid_mutex); | |
2596 | } | |
2597 | ||
8c8bee1d | 2598 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2599 | |
5da54bc1 | 2600 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2601 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2602 | if (device->bdev == bdev) { |
2603 | ret = -EEXIST; | |
d25628bd | 2604 | mutex_unlock( |
5da54bc1 | 2605 | &fs_devices->device_list_mutex); |
2b82032c | 2606 | goto error; |
788f20eb CM |
2607 | } |
2608 | } | |
5da54bc1 | 2609 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2610 | |
0b246afa | 2611 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2612 | if (IS_ERR(device)) { |
788f20eb | 2613 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2614 | ret = PTR_ERR(device); |
2b82032c | 2615 | goto error; |
788f20eb CM |
2616 | } |
2617 | ||
78f2c9e6 | 2618 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2619 | if (!name) { |
2b82032c | 2620 | ret = -ENOMEM; |
5c4cf6c9 | 2621 | goto error_free_device; |
788f20eb | 2622 | } |
606686ee | 2623 | rcu_assign_pointer(device->name, name); |
2b82032c | 2624 | |
a22285a6 | 2625 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2626 | if (IS_ERR(trans)) { |
98d5dc13 | 2627 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2628 | goto error_free_device; |
98d5dc13 TI |
2629 | } |
2630 | ||
d5e2003c | 2631 | q = bdev_get_queue(bdev); |
ebbede42 | 2632 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2633 | device->generation = trans->transid; |
0b246afa JM |
2634 | device->io_width = fs_info->sectorsize; |
2635 | device->io_align = fs_info->sectorsize; | |
2636 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2637 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2638 | fs_info->sectorsize); | |
2cc3c559 | 2639 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2640 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2641 | device->fs_info = fs_info; |
788f20eb | 2642 | device->bdev = bdev; |
e12c9621 | 2643 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2644 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2645 | device->mode = FMODE_EXCL; |
27087f37 | 2646 | device->dev_stats_valid = 1; |
9f6d2510 | 2647 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2648 | |
2b82032c | 2649 | if (seeding_dev) { |
1751e8a6 | 2650 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2651 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2652 | if (ret) { |
2653 | btrfs_abort_transaction(trans, ret); | |
2654 | goto error_trans; | |
2655 | } | |
2b82032c | 2656 | } |
788f20eb | 2657 | |
5da54bc1 | 2658 | device->fs_devices = fs_devices; |
e5e9a520 | 2659 | |
5da54bc1 | 2660 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2661 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2662 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2663 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2664 | fs_devices->num_devices++; | |
2665 | fs_devices->open_devices++; | |
2666 | fs_devices->rw_devices++; | |
2667 | fs_devices->total_devices++; | |
2668 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2669 | |
a5ed45f8 | 2670 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2671 | |
e884f4f0 | 2672 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2673 | fs_devices->rotating = 1; |
c289811c | 2674 | |
39379faa | 2675 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2676 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2677 | round_down(orig_super_total_bytes + device->total_bytes, |
2678 | fs_info->sectorsize)); | |
788f20eb | 2679 | |
39379faa NA |
2680 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2681 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2682 | orig_super_num_devices + 1); | |
0d39376a AJ |
2683 | |
2684 | /* add sysfs device entry */ | |
5da54bc1 | 2685 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2686 | |
2196d6e8 MX |
2687 | /* |
2688 | * we've got more storage, clear any full flags on the space | |
2689 | * infos | |
2690 | */ | |
0b246afa | 2691 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2692 | |
34441361 | 2693 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2694 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2695 | |
2b82032c | 2696 | if (seeding_dev) { |
34441361 | 2697 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2698 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2699 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2700 | if (ret) { |
66642832 | 2701 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2702 | goto error_sysfs; |
005d6427 | 2703 | } |
2196d6e8 MX |
2704 | } |
2705 | ||
8e87e856 | 2706 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2707 | if (ret) { |
66642832 | 2708 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2709 | goto error_sysfs; |
2196d6e8 MX |
2710 | } |
2711 | ||
2712 | if (seeding_dev) { | |
2713 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2714 | ||
0b246afa | 2715 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2716 | if (ret) { |
66642832 | 2717 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2718 | goto error_sysfs; |
005d6427 | 2719 | } |
b2373f25 AJ |
2720 | |
2721 | /* Sprouting would change fsid of the mounted root, | |
2722 | * so rename the fsid on the sysfs | |
2723 | */ | |
2724 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
de37aa51 | 2725 | fs_info->fs_devices->fsid); |
5da54bc1 | 2726 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2727 | btrfs_warn(fs_info, |
2728 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2729 | } |
2730 | ||
3a45bb20 | 2731 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2732 | |
2b82032c YZ |
2733 | if (seeding_dev) { |
2734 | mutex_unlock(&uuid_mutex); | |
2735 | up_write(&sb->s_umount); | |
7132a262 | 2736 | unlocked = true; |
788f20eb | 2737 | |
79787eaa JM |
2738 | if (ret) /* transaction commit */ |
2739 | return ret; | |
2740 | ||
2ff7e61e | 2741 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2742 | if (ret < 0) |
0b246afa | 2743 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2744 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2745 | trans = btrfs_attach_transaction(root); |
2746 | if (IS_ERR(trans)) { | |
2747 | if (PTR_ERR(trans) == -ENOENT) | |
2748 | return 0; | |
7132a262 AJ |
2749 | ret = PTR_ERR(trans); |
2750 | trans = NULL; | |
2751 | goto error_sysfs; | |
671415b7 | 2752 | } |
3a45bb20 | 2753 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2754 | } |
c9e9f97b | 2755 | |
5a1972bd QW |
2756 | /* Update ctime/mtime for libblkid */ |
2757 | update_dev_time(device_path); | |
2b82032c | 2758 | return ret; |
79787eaa | 2759 | |
d31c32f6 | 2760 | error_sysfs: |
5da54bc1 | 2761 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2762 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2763 | mutex_lock(&fs_info->chunk_mutex); | |
2764 | list_del_rcu(&device->dev_list); | |
2765 | list_del(&device->dev_alloc_list); | |
2766 | fs_info->fs_devices->num_devices--; | |
2767 | fs_info->fs_devices->open_devices--; | |
2768 | fs_info->fs_devices->rw_devices--; | |
2769 | fs_info->fs_devices->total_devices--; | |
2770 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2771 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2772 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2773 | orig_super_total_bytes); | |
2774 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2775 | orig_super_num_devices); | |
2776 | mutex_unlock(&fs_info->chunk_mutex); | |
2777 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2778 | error_trans: |
0af2c4bf | 2779 | if (seeding_dev) |
1751e8a6 | 2780 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2781 | if (trans) |
2782 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2783 | error_free_device: |
a425f9d4 | 2784 | btrfs_free_device(device); |
2b82032c | 2785 | error: |
e525fd89 | 2786 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2787 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2788 | mutex_unlock(&uuid_mutex); |
2789 | up_write(&sb->s_umount); | |
2790 | } | |
c9e9f97b | 2791 | return ret; |
788f20eb CM |
2792 | } |
2793 | ||
d397712b CM |
2794 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2795 | struct btrfs_device *device) | |
0b86a832 CM |
2796 | { |
2797 | int ret; | |
2798 | struct btrfs_path *path; | |
0b246afa | 2799 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2800 | struct btrfs_dev_item *dev_item; |
2801 | struct extent_buffer *leaf; | |
2802 | struct btrfs_key key; | |
2803 | ||
0b86a832 CM |
2804 | path = btrfs_alloc_path(); |
2805 | if (!path) | |
2806 | return -ENOMEM; | |
2807 | ||
2808 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2809 | key.type = BTRFS_DEV_ITEM_KEY; | |
2810 | key.offset = device->devid; | |
2811 | ||
2812 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2813 | if (ret < 0) | |
2814 | goto out; | |
2815 | ||
2816 | if (ret > 0) { | |
2817 | ret = -ENOENT; | |
2818 | goto out; | |
2819 | } | |
2820 | ||
2821 | leaf = path->nodes[0]; | |
2822 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2823 | ||
2824 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2825 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2826 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2827 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2828 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2829 | btrfs_set_device_total_bytes(leaf, dev_item, |
2830 | btrfs_device_get_disk_total_bytes(device)); | |
2831 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2832 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2833 | btrfs_mark_buffer_dirty(leaf); |
2834 | ||
2835 | out: | |
2836 | btrfs_free_path(path); | |
2837 | return ret; | |
2838 | } | |
2839 | ||
2196d6e8 | 2840 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2841 | struct btrfs_device *device, u64 new_size) |
2842 | { | |
0b246afa JM |
2843 | struct btrfs_fs_info *fs_info = device->fs_info; |
2844 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2845 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2846 | u64 old_total; |
2847 | u64 diff; | |
8f18cf13 | 2848 | |
ebbede42 | 2849 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2850 | return -EACCES; |
2196d6e8 | 2851 | |
7dfb8be1 NB |
2852 | new_size = round_down(new_size, fs_info->sectorsize); |
2853 | ||
34441361 | 2854 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2855 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2856 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2857 | |
63a212ab | 2858 | if (new_size <= device->total_bytes || |
401e29c1 | 2859 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2860 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2861 | return -EINVAL; |
2196d6e8 | 2862 | } |
2b82032c | 2863 | |
0b246afa | 2864 | fs_devices = fs_info->fs_devices; |
2b82032c | 2865 | |
7dfb8be1 NB |
2866 | btrfs_set_super_total_bytes(super_copy, |
2867 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2868 | device->fs_devices->total_rw_bytes += diff; |
2869 | ||
7cc8e58d MX |
2870 | btrfs_device_set_total_bytes(device, new_size); |
2871 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2872 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2873 | if (list_empty(&device->resized_list)) |
2874 | list_add_tail(&device->resized_list, | |
2875 | &fs_devices->resized_devices); | |
34441361 | 2876 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2877 | |
8f18cf13 CM |
2878 | return btrfs_update_device(trans, device); |
2879 | } | |
2880 | ||
f4208794 | 2881 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2882 | { |
f4208794 | 2883 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2884 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2885 | int ret; |
2886 | struct btrfs_path *path; | |
2887 | struct btrfs_key key; | |
2888 | ||
8f18cf13 CM |
2889 | path = btrfs_alloc_path(); |
2890 | if (!path) | |
2891 | return -ENOMEM; | |
2892 | ||
408fbf19 | 2893 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2894 | key.offset = chunk_offset; |
2895 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2896 | ||
2897 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2898 | if (ret < 0) |
2899 | goto out; | |
2900 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2901 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2902 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2903 | ret = -ENOENT; |
2904 | goto out; | |
2905 | } | |
8f18cf13 CM |
2906 | |
2907 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2908 | if (ret < 0) |
0b246afa JM |
2909 | btrfs_handle_fs_error(fs_info, ret, |
2910 | "Failed to delete chunk item."); | |
79787eaa | 2911 | out: |
8f18cf13 | 2912 | btrfs_free_path(path); |
65a246c5 | 2913 | return ret; |
8f18cf13 CM |
2914 | } |
2915 | ||
408fbf19 | 2916 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2917 | { |
0b246afa | 2918 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2919 | struct btrfs_disk_key *disk_key; |
2920 | struct btrfs_chunk *chunk; | |
2921 | u8 *ptr; | |
2922 | int ret = 0; | |
2923 | u32 num_stripes; | |
2924 | u32 array_size; | |
2925 | u32 len = 0; | |
2926 | u32 cur; | |
2927 | struct btrfs_key key; | |
2928 | ||
34441361 | 2929 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2930 | array_size = btrfs_super_sys_array_size(super_copy); |
2931 | ||
2932 | ptr = super_copy->sys_chunk_array; | |
2933 | cur = 0; | |
2934 | ||
2935 | while (cur < array_size) { | |
2936 | disk_key = (struct btrfs_disk_key *)ptr; | |
2937 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2938 | ||
2939 | len = sizeof(*disk_key); | |
2940 | ||
2941 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2942 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2943 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2944 | len += btrfs_chunk_item_size(num_stripes); | |
2945 | } else { | |
2946 | ret = -EIO; | |
2947 | break; | |
2948 | } | |
408fbf19 | 2949 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2950 | key.offset == chunk_offset) { |
2951 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2952 | array_size -= len; | |
2953 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2954 | } else { | |
2955 | ptr += len; | |
2956 | cur += len; | |
2957 | } | |
2958 | } | |
34441361 | 2959 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2960 | return ret; |
2961 | } | |
2962 | ||
60ca842e OS |
2963 | /* |
2964 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2965 | * @logical: Logical block offset in bytes. | |
2966 | * @length: Length of extent in bytes. | |
2967 | * | |
2968 | * Return: Chunk mapping or ERR_PTR. | |
2969 | */ | |
2970 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2971 | u64 logical, u64 length) | |
592d92ee LB |
2972 | { |
2973 | struct extent_map_tree *em_tree; | |
2974 | struct extent_map *em; | |
2975 | ||
2976 | em_tree = &fs_info->mapping_tree.map_tree; | |
2977 | read_lock(&em_tree->lock); | |
2978 | em = lookup_extent_mapping(em_tree, logical, length); | |
2979 | read_unlock(&em_tree->lock); | |
2980 | ||
2981 | if (!em) { | |
2982 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2983 | logical, length); | |
2984 | return ERR_PTR(-EINVAL); | |
2985 | } | |
2986 | ||
2987 | if (em->start > logical || em->start + em->len < logical) { | |
2988 | btrfs_crit(fs_info, | |
2989 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2990 | logical, length, em->start, em->start + em->len); | |
2991 | free_extent_map(em); | |
2992 | return ERR_PTR(-EINVAL); | |
2993 | } | |
2994 | ||
2995 | /* callers are responsible for dropping em's ref. */ | |
2996 | return em; | |
2997 | } | |
2998 | ||
97aff912 | 2999 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3000 | { |
97aff912 | 3001 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3002 | struct extent_map *em; |
3003 | struct map_lookup *map; | |
2196d6e8 | 3004 | u64 dev_extent_len = 0; |
47ab2a6c | 3005 | int i, ret = 0; |
0b246afa | 3006 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3007 | |
60ca842e | 3008 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3009 | if (IS_ERR(em)) { |
47ab2a6c JB |
3010 | /* |
3011 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3012 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3013 | * do anything we still error out. |
3014 | */ | |
3015 | ASSERT(0); | |
592d92ee | 3016 | return PTR_ERR(em); |
47ab2a6c | 3017 | } |
95617d69 | 3018 | map = em->map_lookup; |
34441361 | 3019 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 3020 | check_system_chunk(trans, map->type); |
34441361 | 3021 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 3022 | |
57ba4cb8 FM |
3023 | /* |
3024 | * Take the device list mutex to prevent races with the final phase of | |
3025 | * a device replace operation that replaces the device object associated | |
3026 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
3027 | */ | |
3028 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3029 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3030 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3031 | ret = btrfs_free_dev_extent(trans, device, |
3032 | map->stripes[i].physical, | |
3033 | &dev_extent_len); | |
47ab2a6c | 3034 | if (ret) { |
57ba4cb8 | 3035 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3036 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3037 | goto out; |
3038 | } | |
a061fc8d | 3039 | |
2196d6e8 | 3040 | if (device->bytes_used > 0) { |
34441361 | 3041 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3042 | btrfs_device_set_bytes_used(device, |
3043 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3044 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3045 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3046 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3047 | } |
a061fc8d | 3048 | |
64bc6c2a NB |
3049 | ret = btrfs_update_device(trans, device); |
3050 | if (ret) { | |
3051 | mutex_unlock(&fs_devices->device_list_mutex); | |
3052 | btrfs_abort_transaction(trans, ret); | |
3053 | goto out; | |
dfe25020 | 3054 | } |
8f18cf13 | 3055 | } |
57ba4cb8 FM |
3056 | mutex_unlock(&fs_devices->device_list_mutex); |
3057 | ||
f4208794 | 3058 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3059 | if (ret) { |
66642832 | 3060 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3061 | goto out; |
3062 | } | |
8f18cf13 | 3063 | |
6bccf3ab | 3064 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3065 | |
8f18cf13 | 3066 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3067 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3068 | if (ret) { |
66642832 | 3069 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3070 | goto out; |
3071 | } | |
8f18cf13 CM |
3072 | } |
3073 | ||
5a98ec01 | 3074 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3075 | if (ret) { |
66642832 | 3076 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3077 | goto out; |
3078 | } | |
2b82032c | 3079 | |
47ab2a6c | 3080 | out: |
2b82032c YZ |
3081 | /* once for us */ |
3082 | free_extent_map(em); | |
47ab2a6c JB |
3083 | return ret; |
3084 | } | |
2b82032c | 3085 | |
5b4aacef | 3086 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3087 | { |
5b4aacef | 3088 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3089 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3090 | int ret; |
2b82032c | 3091 | |
67c5e7d4 FM |
3092 | /* |
3093 | * Prevent races with automatic removal of unused block groups. | |
3094 | * After we relocate and before we remove the chunk with offset | |
3095 | * chunk_offset, automatic removal of the block group can kick in, | |
3096 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3097 | * | |
3098 | * Make sure to acquire this mutex before doing a tree search (dev | |
3099 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3100 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3101 | * we release the path used to search the chunk/dev tree and before | |
3102 | * the current task acquires this mutex and calls us. | |
3103 | */ | |
a32bf9a3 | 3104 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3105 | |
0b246afa | 3106 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
3107 | if (ret) |
3108 | return -ENOSPC; | |
3109 | ||
3110 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3111 | btrfs_scrub_pause(fs_info); |
0b246afa | 3112 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3113 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3114 | if (ret) |
3115 | return ret; | |
3116 | ||
75cb379d JM |
3117 | /* |
3118 | * We add the kobjects here (and after forcing data chunk creation) | |
3119 | * since relocation is the only place we'll create chunks of a new | |
3120 | * type at runtime. The only place where we'll remove the last | |
3121 | * chunk of a type is the call immediately below this one. Even | |
3122 | * so, we're protected against races with the cleaner thread since | |
3123 | * we're covered by the delete_unused_bgs_mutex. | |
3124 | */ | |
3125 | btrfs_add_raid_kobjects(fs_info); | |
3126 | ||
19c4d2f9 CM |
3127 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3128 | chunk_offset); | |
3129 | if (IS_ERR(trans)) { | |
3130 | ret = PTR_ERR(trans); | |
3131 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3132 | return ret; | |
3133 | } | |
3134 | ||
47ab2a6c | 3135 | /* |
19c4d2f9 CM |
3136 | * step two, delete the device extents and the |
3137 | * chunk tree entries | |
47ab2a6c | 3138 | */ |
97aff912 | 3139 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3140 | btrfs_end_transaction(trans); |
19c4d2f9 | 3141 | return ret; |
2b82032c YZ |
3142 | } |
3143 | ||
2ff7e61e | 3144 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3145 | { |
0b246afa | 3146 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3147 | struct btrfs_path *path; |
3148 | struct extent_buffer *leaf; | |
3149 | struct btrfs_chunk *chunk; | |
3150 | struct btrfs_key key; | |
3151 | struct btrfs_key found_key; | |
2b82032c | 3152 | u64 chunk_type; |
ba1bf481 JB |
3153 | bool retried = false; |
3154 | int failed = 0; | |
2b82032c YZ |
3155 | int ret; |
3156 | ||
3157 | path = btrfs_alloc_path(); | |
3158 | if (!path) | |
3159 | return -ENOMEM; | |
3160 | ||
ba1bf481 | 3161 | again: |
2b82032c YZ |
3162 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3163 | key.offset = (u64)-1; | |
3164 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3165 | ||
3166 | while (1) { | |
0b246afa | 3167 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3168 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3169 | if (ret < 0) { |
0b246afa | 3170 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3171 | goto error; |
67c5e7d4 | 3172 | } |
79787eaa | 3173 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3174 | |
3175 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3176 | key.type); | |
67c5e7d4 | 3177 | if (ret) |
0b246afa | 3178 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3179 | if (ret < 0) |
3180 | goto error; | |
3181 | if (ret > 0) | |
3182 | break; | |
1a40e23b | 3183 | |
2b82032c YZ |
3184 | leaf = path->nodes[0]; |
3185 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3186 | |
2b82032c YZ |
3187 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3188 | struct btrfs_chunk); | |
3189 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3190 | btrfs_release_path(path); |
8f18cf13 | 3191 | |
2b82032c | 3192 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3193 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3194 | if (ret == -ENOSPC) |
3195 | failed++; | |
14586651 HS |
3196 | else |
3197 | BUG_ON(ret); | |
2b82032c | 3198 | } |
0b246afa | 3199 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3200 | |
2b82032c YZ |
3201 | if (found_key.offset == 0) |
3202 | break; | |
3203 | key.offset = found_key.offset - 1; | |
3204 | } | |
3205 | ret = 0; | |
ba1bf481 JB |
3206 | if (failed && !retried) { |
3207 | failed = 0; | |
3208 | retried = true; | |
3209 | goto again; | |
fae7f21c | 3210 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3211 | ret = -ENOSPC; |
3212 | } | |
2b82032c YZ |
3213 | error: |
3214 | btrfs_free_path(path); | |
3215 | return ret; | |
8f18cf13 CM |
3216 | } |
3217 | ||
a6f93c71 LB |
3218 | /* |
3219 | * return 1 : allocate a data chunk successfully, | |
3220 | * return <0: errors during allocating a data chunk, | |
3221 | * return 0 : no need to allocate a data chunk. | |
3222 | */ | |
3223 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3224 | u64 chunk_offset) | |
3225 | { | |
3226 | struct btrfs_block_group_cache *cache; | |
3227 | u64 bytes_used; | |
3228 | u64 chunk_type; | |
3229 | ||
3230 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3231 | ASSERT(cache); | |
3232 | chunk_type = cache->flags; | |
3233 | btrfs_put_block_group(cache); | |
3234 | ||
3235 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3236 | spin_lock(&fs_info->data_sinfo->lock); | |
3237 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3238 | spin_unlock(&fs_info->data_sinfo->lock); | |
3239 | ||
3240 | if (!bytes_used) { | |
3241 | struct btrfs_trans_handle *trans; | |
3242 | int ret; | |
3243 | ||
3244 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3245 | if (IS_ERR(trans)) | |
3246 | return PTR_ERR(trans); | |
3247 | ||
43a7e99d | 3248 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3249 | BTRFS_BLOCK_GROUP_DATA); |
3250 | btrfs_end_transaction(trans); | |
3251 | if (ret < 0) | |
3252 | return ret; | |
3253 | ||
75cb379d JM |
3254 | btrfs_add_raid_kobjects(fs_info); |
3255 | ||
a6f93c71 LB |
3256 | return 1; |
3257 | } | |
3258 | } | |
3259 | return 0; | |
3260 | } | |
3261 | ||
6bccf3ab | 3262 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3263 | struct btrfs_balance_control *bctl) |
3264 | { | |
6bccf3ab | 3265 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3266 | struct btrfs_trans_handle *trans; |
3267 | struct btrfs_balance_item *item; | |
3268 | struct btrfs_disk_balance_args disk_bargs; | |
3269 | struct btrfs_path *path; | |
3270 | struct extent_buffer *leaf; | |
3271 | struct btrfs_key key; | |
3272 | int ret, err; | |
3273 | ||
3274 | path = btrfs_alloc_path(); | |
3275 | if (!path) | |
3276 | return -ENOMEM; | |
3277 | ||
3278 | trans = btrfs_start_transaction(root, 0); | |
3279 | if (IS_ERR(trans)) { | |
3280 | btrfs_free_path(path); | |
3281 | return PTR_ERR(trans); | |
3282 | } | |
3283 | ||
3284 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3285 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3286 | key.offset = 0; |
3287 | ||
3288 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3289 | sizeof(*item)); | |
3290 | if (ret) | |
3291 | goto out; | |
3292 | ||
3293 | leaf = path->nodes[0]; | |
3294 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3295 | ||
b159fa28 | 3296 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3297 | |
3298 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3299 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3300 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3301 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3302 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3303 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3304 | ||
3305 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3306 | ||
3307 | btrfs_mark_buffer_dirty(leaf); | |
3308 | out: | |
3309 | btrfs_free_path(path); | |
3a45bb20 | 3310 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3311 | if (err && !ret) |
3312 | ret = err; | |
3313 | return ret; | |
3314 | } | |
3315 | ||
6bccf3ab | 3316 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3317 | { |
6bccf3ab | 3318 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3319 | struct btrfs_trans_handle *trans; |
3320 | struct btrfs_path *path; | |
3321 | struct btrfs_key key; | |
3322 | int ret, err; | |
3323 | ||
3324 | path = btrfs_alloc_path(); | |
3325 | if (!path) | |
3326 | return -ENOMEM; | |
3327 | ||
3328 | trans = btrfs_start_transaction(root, 0); | |
3329 | if (IS_ERR(trans)) { | |
3330 | btrfs_free_path(path); | |
3331 | return PTR_ERR(trans); | |
3332 | } | |
3333 | ||
3334 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3335 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3336 | key.offset = 0; |
3337 | ||
3338 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3339 | if (ret < 0) | |
3340 | goto out; | |
3341 | if (ret > 0) { | |
3342 | ret = -ENOENT; | |
3343 | goto out; | |
3344 | } | |
3345 | ||
3346 | ret = btrfs_del_item(trans, root, path); | |
3347 | out: | |
3348 | btrfs_free_path(path); | |
3a45bb20 | 3349 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3350 | if (err && !ret) |
3351 | ret = err; | |
3352 | return ret; | |
3353 | } | |
3354 | ||
59641015 ID |
3355 | /* |
3356 | * This is a heuristic used to reduce the number of chunks balanced on | |
3357 | * resume after balance was interrupted. | |
3358 | */ | |
3359 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3360 | { | |
3361 | /* | |
3362 | * Turn on soft mode for chunk types that were being converted. | |
3363 | */ | |
3364 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3365 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3366 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3367 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3368 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3369 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3370 | ||
3371 | /* | |
3372 | * Turn on usage filter if is not already used. The idea is | |
3373 | * that chunks that we have already balanced should be | |
3374 | * reasonably full. Don't do it for chunks that are being | |
3375 | * converted - that will keep us from relocating unconverted | |
3376 | * (albeit full) chunks. | |
3377 | */ | |
3378 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3379 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3380 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3381 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3382 | bctl->data.usage = 90; | |
3383 | } | |
3384 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3385 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3386 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3387 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3388 | bctl->sys.usage = 90; | |
3389 | } | |
3390 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3391 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3392 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3393 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3394 | bctl->meta.usage = 90; | |
3395 | } | |
3396 | } | |
3397 | ||
149196a2 DS |
3398 | /* |
3399 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3400 | */ | |
3401 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3402 | { |
3403 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3404 | int ret; |
c9e9f97b ID |
3405 | |
3406 | BUG_ON(!fs_info->balance_ctl); | |
3407 | ||
3408 | spin_lock(&fs_info->balance_lock); | |
3409 | fs_info->balance_ctl = NULL; | |
3410 | spin_unlock(&fs_info->balance_lock); | |
3411 | ||
3412 | kfree(bctl); | |
149196a2 DS |
3413 | ret = del_balance_item(fs_info); |
3414 | if (ret) | |
3415 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3416 | } |
3417 | ||
ed25e9b2 ID |
3418 | /* |
3419 | * Balance filters. Return 1 if chunk should be filtered out | |
3420 | * (should not be balanced). | |
3421 | */ | |
899c81ea | 3422 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3423 | struct btrfs_balance_args *bargs) |
3424 | { | |
899c81ea ID |
3425 | chunk_type = chunk_to_extended(chunk_type) & |
3426 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3427 | |
899c81ea | 3428 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3429 | return 0; |
3430 | ||
3431 | return 1; | |
3432 | } | |
3433 | ||
dba72cb3 | 3434 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3435 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3436 | { |
3437 | struct btrfs_block_group_cache *cache; | |
3438 | u64 chunk_used; | |
3439 | u64 user_thresh_min; | |
3440 | u64 user_thresh_max; | |
3441 | int ret = 1; | |
3442 | ||
3443 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3444 | chunk_used = btrfs_block_group_used(&cache->item); | |
3445 | ||
3446 | if (bargs->usage_min == 0) | |
3447 | user_thresh_min = 0; | |
3448 | else | |
3449 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3450 | bargs->usage_min); | |
3451 | ||
3452 | if (bargs->usage_max == 0) | |
3453 | user_thresh_max = 1; | |
3454 | else if (bargs->usage_max > 100) | |
3455 | user_thresh_max = cache->key.offset; | |
3456 | else | |
3457 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3458 | bargs->usage_max); | |
3459 | ||
3460 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3461 | ret = 0; | |
3462 | ||
3463 | btrfs_put_block_group(cache); | |
3464 | return ret; | |
3465 | } | |
3466 | ||
dba72cb3 | 3467 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3468 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3469 | { |
3470 | struct btrfs_block_group_cache *cache; | |
3471 | u64 chunk_used, user_thresh; | |
3472 | int ret = 1; | |
3473 | ||
3474 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3475 | chunk_used = btrfs_block_group_used(&cache->item); | |
3476 | ||
bc309467 | 3477 | if (bargs->usage_min == 0) |
3e39cea6 | 3478 | user_thresh = 1; |
a105bb88 ID |
3479 | else if (bargs->usage > 100) |
3480 | user_thresh = cache->key.offset; | |
3481 | else | |
3482 | user_thresh = div_factor_fine(cache->key.offset, | |
3483 | bargs->usage); | |
3484 | ||
5ce5b3c0 ID |
3485 | if (chunk_used < user_thresh) |
3486 | ret = 0; | |
3487 | ||
3488 | btrfs_put_block_group(cache); | |
3489 | return ret; | |
3490 | } | |
3491 | ||
409d404b ID |
3492 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3493 | struct btrfs_chunk *chunk, | |
3494 | struct btrfs_balance_args *bargs) | |
3495 | { | |
3496 | struct btrfs_stripe *stripe; | |
3497 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3498 | int i; | |
3499 | ||
3500 | for (i = 0; i < num_stripes; i++) { | |
3501 | stripe = btrfs_stripe_nr(chunk, i); | |
3502 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3503 | return 0; | |
3504 | } | |
3505 | ||
3506 | return 1; | |
3507 | } | |
3508 | ||
94e60d5a ID |
3509 | /* [pstart, pend) */ |
3510 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3511 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3512 | struct btrfs_balance_args *bargs) |
3513 | { | |
3514 | struct btrfs_stripe *stripe; | |
3515 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3516 | u64 stripe_offset; | |
3517 | u64 stripe_length; | |
3518 | int factor; | |
3519 | int i; | |
3520 | ||
3521 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3522 | return 0; | |
3523 | ||
3524 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3525 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3526 | factor = num_stripes / 2; | |
3527 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3528 | factor = num_stripes - 1; | |
3529 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3530 | factor = num_stripes - 2; | |
3531 | } else { | |
3532 | factor = num_stripes; | |
3533 | } | |
94e60d5a ID |
3534 | |
3535 | for (i = 0; i < num_stripes; i++) { | |
3536 | stripe = btrfs_stripe_nr(chunk, i); | |
3537 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3538 | continue; | |
3539 | ||
3540 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3541 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3542 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3543 | |
3544 | if (stripe_offset < bargs->pend && | |
3545 | stripe_offset + stripe_length > bargs->pstart) | |
3546 | return 0; | |
3547 | } | |
3548 | ||
3549 | return 1; | |
3550 | } | |
3551 | ||
ea67176a ID |
3552 | /* [vstart, vend) */ |
3553 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3554 | struct btrfs_chunk *chunk, | |
3555 | u64 chunk_offset, | |
3556 | struct btrfs_balance_args *bargs) | |
3557 | { | |
3558 | if (chunk_offset < bargs->vend && | |
3559 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3560 | /* at least part of the chunk is inside this vrange */ | |
3561 | return 0; | |
3562 | ||
3563 | return 1; | |
3564 | } | |
3565 | ||
dee32d0a GAP |
3566 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3567 | struct btrfs_chunk *chunk, | |
3568 | struct btrfs_balance_args *bargs) | |
3569 | { | |
3570 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3571 | ||
3572 | if (bargs->stripes_min <= num_stripes | |
3573 | && num_stripes <= bargs->stripes_max) | |
3574 | return 0; | |
3575 | ||
3576 | return 1; | |
3577 | } | |
3578 | ||
899c81ea | 3579 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3580 | struct btrfs_balance_args *bargs) |
3581 | { | |
3582 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3583 | return 0; | |
3584 | ||
899c81ea ID |
3585 | chunk_type = chunk_to_extended(chunk_type) & |
3586 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3587 | |
899c81ea | 3588 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3589 | return 1; |
3590 | ||
3591 | return 0; | |
3592 | } | |
3593 | ||
2ff7e61e | 3594 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3595 | struct extent_buffer *leaf, |
3596 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3597 | { | |
0b246afa | 3598 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3599 | struct btrfs_balance_args *bargs = NULL; |
3600 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3601 | ||
3602 | /* type filter */ | |
3603 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3604 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3605 | return 0; | |
3606 | } | |
3607 | ||
3608 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3609 | bargs = &bctl->data; | |
3610 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3611 | bargs = &bctl->sys; | |
3612 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3613 | bargs = &bctl->meta; | |
3614 | ||
ed25e9b2 ID |
3615 | /* profiles filter */ |
3616 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3617 | chunk_profiles_filter(chunk_type, bargs)) { | |
3618 | return 0; | |
5ce5b3c0 ID |
3619 | } |
3620 | ||
3621 | /* usage filter */ | |
3622 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3623 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3624 | return 0; |
bc309467 | 3625 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3626 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3627 | return 0; |
409d404b ID |
3628 | } |
3629 | ||
3630 | /* devid filter */ | |
3631 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3632 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3633 | return 0; | |
94e60d5a ID |
3634 | } |
3635 | ||
3636 | /* drange filter, makes sense only with devid filter */ | |
3637 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3638 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3639 | return 0; |
ea67176a ID |
3640 | } |
3641 | ||
3642 | /* vrange filter */ | |
3643 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3644 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3645 | return 0; | |
ed25e9b2 ID |
3646 | } |
3647 | ||
dee32d0a GAP |
3648 | /* stripes filter */ |
3649 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3650 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3651 | return 0; | |
3652 | } | |
3653 | ||
cfa4c961 ID |
3654 | /* soft profile changing mode */ |
3655 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3656 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3657 | return 0; | |
3658 | } | |
3659 | ||
7d824b6f DS |
3660 | /* |
3661 | * limited by count, must be the last filter | |
3662 | */ | |
3663 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3664 | if (bargs->limit == 0) | |
3665 | return 0; | |
3666 | else | |
3667 | bargs->limit--; | |
12907fc7 DS |
3668 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3669 | /* | |
3670 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3671 | * determined here because we do not have the global information |
12907fc7 DS |
3672 | * about the count of all chunks that satisfy the filters. |
3673 | */ | |
3674 | if (bargs->limit_max == 0) | |
3675 | return 0; | |
3676 | else | |
3677 | bargs->limit_max--; | |
7d824b6f DS |
3678 | } |
3679 | ||
f43ffb60 ID |
3680 | return 1; |
3681 | } | |
3682 | ||
c9e9f97b | 3683 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3684 | { |
19a39dce | 3685 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3686 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3687 | u64 chunk_type; |
f43ffb60 | 3688 | struct btrfs_chunk *chunk; |
5a488b9d | 3689 | struct btrfs_path *path = NULL; |
ec44a35c | 3690 | struct btrfs_key key; |
ec44a35c | 3691 | struct btrfs_key found_key; |
f43ffb60 ID |
3692 | struct extent_buffer *leaf; |
3693 | int slot; | |
c9e9f97b ID |
3694 | int ret; |
3695 | int enospc_errors = 0; | |
19a39dce | 3696 | bool counting = true; |
12907fc7 | 3697 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3698 | u64 limit_data = bctl->data.limit; |
3699 | u64 limit_meta = bctl->meta.limit; | |
3700 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3701 | u32 count_data = 0; |
3702 | u32 count_meta = 0; | |
3703 | u32 count_sys = 0; | |
2c9fe835 | 3704 | int chunk_reserved = 0; |
ec44a35c | 3705 | |
ec44a35c | 3706 | path = btrfs_alloc_path(); |
17e9f796 MF |
3707 | if (!path) { |
3708 | ret = -ENOMEM; | |
3709 | goto error; | |
3710 | } | |
19a39dce ID |
3711 | |
3712 | /* zero out stat counters */ | |
3713 | spin_lock(&fs_info->balance_lock); | |
3714 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3715 | spin_unlock(&fs_info->balance_lock); | |
3716 | again: | |
7d824b6f | 3717 | if (!counting) { |
12907fc7 DS |
3718 | /* |
3719 | * The single value limit and min/max limits use the same bytes | |
3720 | * in the | |
3721 | */ | |
7d824b6f DS |
3722 | bctl->data.limit = limit_data; |
3723 | bctl->meta.limit = limit_meta; | |
3724 | bctl->sys.limit = limit_sys; | |
3725 | } | |
ec44a35c CM |
3726 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3727 | key.offset = (u64)-1; | |
3728 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3729 | ||
d397712b | 3730 | while (1) { |
19a39dce | 3731 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3732 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3733 | ret = -ECANCELED; |
3734 | goto error; | |
3735 | } | |
3736 | ||
67c5e7d4 | 3737 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3738 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3739 | if (ret < 0) { |
3740 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3741 | goto error; |
67c5e7d4 | 3742 | } |
ec44a35c CM |
3743 | |
3744 | /* | |
3745 | * this shouldn't happen, it means the last relocate | |
3746 | * failed | |
3747 | */ | |
3748 | if (ret == 0) | |
c9e9f97b | 3749 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3750 | |
3751 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3752 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3753 | if (ret) { |
67c5e7d4 | 3754 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3755 | ret = 0; |
ec44a35c | 3756 | break; |
c9e9f97b | 3757 | } |
7d9eb12c | 3758 | |
f43ffb60 ID |
3759 | leaf = path->nodes[0]; |
3760 | slot = path->slots[0]; | |
3761 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3762 | |
67c5e7d4 FM |
3763 | if (found_key.objectid != key.objectid) { |
3764 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3765 | break; |
67c5e7d4 | 3766 | } |
7d9eb12c | 3767 | |
f43ffb60 | 3768 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3769 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3770 | |
19a39dce ID |
3771 | if (!counting) { |
3772 | spin_lock(&fs_info->balance_lock); | |
3773 | bctl->stat.considered++; | |
3774 | spin_unlock(&fs_info->balance_lock); | |
3775 | } | |
3776 | ||
2ff7e61e | 3777 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3778 | found_key.offset); |
2c9fe835 | 3779 | |
b3b4aa74 | 3780 | btrfs_release_path(path); |
67c5e7d4 FM |
3781 | if (!ret) { |
3782 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3783 | goto loop; |
67c5e7d4 | 3784 | } |
f43ffb60 | 3785 | |
19a39dce | 3786 | if (counting) { |
67c5e7d4 | 3787 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3788 | spin_lock(&fs_info->balance_lock); |
3789 | bctl->stat.expected++; | |
3790 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3791 | |
3792 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3793 | count_data++; | |
3794 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3795 | count_sys++; | |
3796 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3797 | count_meta++; | |
3798 | ||
3799 | goto loop; | |
3800 | } | |
3801 | ||
3802 | /* | |
3803 | * Apply limit_min filter, no need to check if the LIMITS | |
3804 | * filter is used, limit_min is 0 by default | |
3805 | */ | |
3806 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3807 | count_data < bctl->data.limit_min) | |
3808 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3809 | count_meta < bctl->meta.limit_min) | |
3810 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3811 | count_sys < bctl->sys.limit_min)) { | |
3812 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3813 | goto loop; |
3814 | } | |
3815 | ||
a6f93c71 LB |
3816 | if (!chunk_reserved) { |
3817 | /* | |
3818 | * We may be relocating the only data chunk we have, | |
3819 | * which could potentially end up with losing data's | |
3820 | * raid profile, so lets allocate an empty one in | |
3821 | * advance. | |
3822 | */ | |
3823 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3824 | found_key.offset); | |
2c9fe835 ZL |
3825 | if (ret < 0) { |
3826 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3827 | goto error; | |
a6f93c71 LB |
3828 | } else if (ret == 1) { |
3829 | chunk_reserved = 1; | |
2c9fe835 | 3830 | } |
2c9fe835 ZL |
3831 | } |
3832 | ||
5b4aacef | 3833 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3834 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3835 | if (ret == -ENOSPC) { |
c9e9f97b | 3836 | enospc_errors++; |
eede2bf3 OS |
3837 | } else if (ret == -ETXTBSY) { |
3838 | btrfs_info(fs_info, | |
3839 | "skipping relocation of block group %llu due to active swapfile", | |
3840 | found_key.offset); | |
3841 | ret = 0; | |
3842 | } else if (ret) { | |
3843 | goto error; | |
19a39dce ID |
3844 | } else { |
3845 | spin_lock(&fs_info->balance_lock); | |
3846 | bctl->stat.completed++; | |
3847 | spin_unlock(&fs_info->balance_lock); | |
3848 | } | |
f43ffb60 | 3849 | loop: |
795a3321 ID |
3850 | if (found_key.offset == 0) |
3851 | break; | |
ba1bf481 | 3852 | key.offset = found_key.offset - 1; |
ec44a35c | 3853 | } |
c9e9f97b | 3854 | |
19a39dce ID |
3855 | if (counting) { |
3856 | btrfs_release_path(path); | |
3857 | counting = false; | |
3858 | goto again; | |
3859 | } | |
ec44a35c CM |
3860 | error: |
3861 | btrfs_free_path(path); | |
c9e9f97b | 3862 | if (enospc_errors) { |
efe120a0 | 3863 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3864 | enospc_errors); |
c9e9f97b ID |
3865 | if (!ret) |
3866 | ret = -ENOSPC; | |
3867 | } | |
3868 | ||
ec44a35c CM |
3869 | return ret; |
3870 | } | |
3871 | ||
0c460c0d ID |
3872 | /** |
3873 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3874 | * @flags: profile to validate | |
3875 | * @extended: if true @flags is treated as an extended profile | |
3876 | */ | |
3877 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3878 | { | |
3879 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3880 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3881 | ||
3882 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3883 | ||
3884 | /* 1) check that all other bits are zeroed */ | |
3885 | if (flags & ~mask) | |
3886 | return 0; | |
3887 | ||
3888 | /* 2) see if profile is reduced */ | |
3889 | if (flags == 0) | |
3890 | return !extended; /* "0" is valid for usual profiles */ | |
3891 | ||
3892 | /* true if exactly one bit set */ | |
818255fe | 3893 | return is_power_of_2(flags); |
0c460c0d ID |
3894 | } |
3895 | ||
837d5b6e ID |
3896 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3897 | { | |
a7e99c69 ID |
3898 | /* cancel requested || normal exit path */ |
3899 | return atomic_read(&fs_info->balance_cancel_req) || | |
3900 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3901 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3902 | } |
3903 | ||
bdcd3c97 AM |
3904 | /* Non-zero return value signifies invalidity */ |
3905 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3906 | u64 allowed) | |
3907 | { | |
3908 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3909 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3910 | (bctl_arg->target & ~allowed))); | |
3911 | } | |
3912 | ||
56fc37d9 AJ |
3913 | /* |
3914 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3915 | * @size_buf including the terminating null. The output may be trimmed if it | |
3916 | * does not fit into the provided buffer. | |
3917 | */ | |
3918 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3919 | u32 size_buf) | |
3920 | { | |
3921 | int ret; | |
3922 | u32 size_bp = size_buf; | |
3923 | char *bp = buf; | |
3924 | u64 flags = bargs->flags; | |
3925 | char tmp_buf[128] = {'\0'}; | |
3926 | ||
3927 | if (!flags) | |
3928 | return; | |
3929 | ||
3930 | #define CHECK_APPEND_NOARG(a) \ | |
3931 | do { \ | |
3932 | ret = snprintf(bp, size_bp, (a)); \ | |
3933 | if (ret < 0 || ret >= size_bp) \ | |
3934 | goto out_overflow; \ | |
3935 | size_bp -= ret; \ | |
3936 | bp += ret; \ | |
3937 | } while (0) | |
3938 | ||
3939 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3940 | do { \ | |
3941 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3942 | if (ret < 0 || ret >= size_bp) \ | |
3943 | goto out_overflow; \ | |
3944 | size_bp -= ret; \ | |
3945 | bp += ret; \ | |
3946 | } while (0) | |
3947 | ||
3948 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3949 | do { \ | |
3950 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3951 | if (ret < 0 || ret >= size_bp) \ | |
3952 | goto out_overflow; \ | |
3953 | size_bp -= ret; \ | |
3954 | bp += ret; \ | |
3955 | } while (0) | |
3956 | ||
3957 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3958 | int index = btrfs_bg_flags_to_raid_index(bargs->target); | |
3959 | ||
3960 | CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index)); | |
3961 | } | |
3962 | ||
3963 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3964 | CHECK_APPEND_NOARG("soft,"); | |
3965 | ||
3966 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3967 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3968 | sizeof(tmp_buf)); | |
3969 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3970 | } | |
3971 | ||
3972 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3973 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3974 | ||
3975 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3976 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3977 | bargs->usage_min, bargs->usage_max); | |
3978 | ||
3979 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3980 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3981 | ||
3982 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3983 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3984 | bargs->pstart, bargs->pend); | |
3985 | ||
3986 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3987 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3988 | bargs->vstart, bargs->vend); | |
3989 | ||
3990 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3991 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3992 | ||
3993 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3994 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3995 | bargs->limit_min, bargs->limit_max); | |
3996 | ||
3997 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3998 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3999 | bargs->stripes_min, bargs->stripes_max); | |
4000 | ||
4001 | #undef CHECK_APPEND_2ARG | |
4002 | #undef CHECK_APPEND_1ARG | |
4003 | #undef CHECK_APPEND_NOARG | |
4004 | ||
4005 | out_overflow: | |
4006 | ||
4007 | if (size_bp < size_buf) | |
4008 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4009 | else | |
4010 | buf[0] = '\0'; | |
4011 | } | |
4012 | ||
4013 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4014 | { | |
4015 | u32 size_buf = 1024; | |
4016 | char tmp_buf[192] = {'\0'}; | |
4017 | char *buf; | |
4018 | char *bp; | |
4019 | u32 size_bp = size_buf; | |
4020 | int ret; | |
4021 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4022 | ||
4023 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4024 | if (!buf) | |
4025 | return; | |
4026 | ||
4027 | bp = buf; | |
4028 | ||
4029 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4030 | do { \ | |
4031 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4032 | if (ret < 0 || ret >= size_bp) \ | |
4033 | goto out_overflow; \ | |
4034 | size_bp -= ret; \ | |
4035 | bp += ret; \ | |
4036 | } while (0) | |
4037 | ||
4038 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4039 | CHECK_APPEND_1ARG("%s", "-f "); | |
4040 | ||
4041 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4042 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4043 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4044 | } | |
4045 | ||
4046 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4047 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4048 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4049 | } | |
4050 | ||
4051 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4052 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4053 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4054 | } | |
4055 | ||
4056 | #undef CHECK_APPEND_1ARG | |
4057 | ||
4058 | out_overflow: | |
4059 | ||
4060 | if (size_bp < size_buf) | |
4061 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4062 | btrfs_info(fs_info, "balance: %s %s", | |
4063 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4064 | "resume" : "start", buf); | |
4065 | ||
4066 | kfree(buf); | |
4067 | } | |
4068 | ||
c9e9f97b | 4069 | /* |
dccdb07b | 4070 | * Should be called with balance mutexe held |
c9e9f97b | 4071 | */ |
6fcf6e2b DS |
4072 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4073 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4074 | struct btrfs_ioctl_balance_args *bargs) |
4075 | { | |
14506127 | 4076 | u64 meta_target, data_target; |
f43ffb60 | 4077 | u64 allowed; |
e4837f8f | 4078 | int mixed = 0; |
c9e9f97b | 4079 | int ret; |
8dabb742 | 4080 | u64 num_devices; |
de98ced9 | 4081 | unsigned seq; |
5a8067c0 | 4082 | bool reducing_integrity; |
c9e9f97b | 4083 | |
837d5b6e | 4084 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4085 | atomic_read(&fs_info->balance_pause_req) || |
4086 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4087 | ret = -EINVAL; |
4088 | goto out; | |
4089 | } | |
4090 | ||
e4837f8f ID |
4091 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4092 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4093 | mixed = 1; | |
4094 | ||
f43ffb60 ID |
4095 | /* |
4096 | * In case of mixed groups both data and meta should be picked, | |
4097 | * and identical options should be given for both of them. | |
4098 | */ | |
e4837f8f ID |
4099 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4100 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4101 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4102 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4103 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4104 | btrfs_err(fs_info, |
6dac13f8 | 4105 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4106 | ret = -EINVAL; |
4107 | goto out; | |
4108 | } | |
4109 | } | |
4110 | ||
1da73967 AJ |
4111 | num_devices = btrfs_num_devices(fs_info); |
4112 | ||
88be159c AH |
4113 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
4114 | if (num_devices > 1) | |
e4d8ec0f | 4115 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
4116 | if (num_devices > 2) |
4117 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
4118 | if (num_devices > 3) | |
4119 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
4120 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 4121 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
4122 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
4123 | ||
5d163e0e | 4124 | btrfs_err(fs_info, |
6dac13f8 AJ |
4125 | "balance: invalid convert data profile %s", |
4126 | get_raid_name(index)); | |
e4d8ec0f ID |
4127 | ret = -EINVAL; |
4128 | goto out; | |
4129 | } | |
bdcd3c97 | 4130 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
4131 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
4132 | ||
efe120a0 | 4133 | btrfs_err(fs_info, |
6dac13f8 AJ |
4134 | "balance: invalid convert metadata profile %s", |
4135 | get_raid_name(index)); | |
e4d8ec0f ID |
4136 | ret = -EINVAL; |
4137 | goto out; | |
4138 | } | |
bdcd3c97 | 4139 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
4140 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
4141 | ||
efe120a0 | 4142 | btrfs_err(fs_info, |
6dac13f8 AJ |
4143 | "balance: invalid convert system profile %s", |
4144 | get_raid_name(index)); | |
e4d8ec0f ID |
4145 | ret = -EINVAL; |
4146 | goto out; | |
4147 | } | |
4148 | ||
e4d8ec0f ID |
4149 | /* allow to reduce meta or sys integrity only if force set */ |
4150 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
4151 | BTRFS_BLOCK_GROUP_RAID10 | |
4152 | BTRFS_BLOCK_GROUP_RAID5 | | |
4153 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
4154 | do { |
4155 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4156 | ||
4157 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4158 | (fs_info->avail_system_alloc_bits & allowed) && | |
4159 | !(bctl->sys.target & allowed)) || | |
4160 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4161 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4162 | !(bctl->meta.target & allowed))) |
4163 | reducing_integrity = true; | |
4164 | else | |
4165 | reducing_integrity = false; | |
4166 | ||
4167 | /* if we're not converting, the target field is uninitialized */ | |
4168 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4169 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4170 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4171 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4172 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4173 | |
5a8067c0 FM |
4174 | if (reducing_integrity) { |
4175 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4176 | btrfs_info(fs_info, | |
4177 | "balance: force reducing metadata integrity"); | |
4178 | } else { | |
4179 | btrfs_err(fs_info, | |
4180 | "balance: reduces metadata integrity, use --force if you want this"); | |
4181 | ret = -EINVAL; | |
4182 | goto out; | |
4183 | } | |
4184 | } | |
4185 | ||
14506127 AB |
4186 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4187 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
4188 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
4189 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
4190 | ||
ee592d07 | 4191 | btrfs_warn(fs_info, |
6dac13f8 AJ |
4192 | "balance: metadata profile %s has lower redundancy than data profile %s", |
4193 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
4194 | } |
4195 | ||
6bccf3ab | 4196 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4197 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4198 | goto out; |
4199 | ||
59641015 ID |
4200 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4201 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4202 | BUG_ON(fs_info->balance_ctl); |
4203 | spin_lock(&fs_info->balance_lock); | |
4204 | fs_info->balance_ctl = bctl; | |
4205 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4206 | } else { |
4207 | BUG_ON(ret != -EEXIST); | |
4208 | spin_lock(&fs_info->balance_lock); | |
4209 | update_balance_args(bctl); | |
4210 | spin_unlock(&fs_info->balance_lock); | |
4211 | } | |
c9e9f97b | 4212 | |
3009a62f DS |
4213 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4214 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4215 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4216 | mutex_unlock(&fs_info->balance_mutex); |
4217 | ||
4218 | ret = __btrfs_balance(fs_info); | |
4219 | ||
4220 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4221 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4222 | btrfs_info(fs_info, "balance: paused"); | |
4223 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4224 | btrfs_info(fs_info, "balance: canceled"); | |
4225 | else | |
4226 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4227 | ||
3009a62f | 4228 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4229 | |
4230 | if (bargs) { | |
4231 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4232 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4233 | } |
4234 | ||
3a01aa7a ID |
4235 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4236 | balance_need_close(fs_info)) { | |
149196a2 | 4237 | reset_balance_state(fs_info); |
a17c95df | 4238 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4239 | } |
4240 | ||
837d5b6e | 4241 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4242 | |
4243 | return ret; | |
4244 | out: | |
59641015 | 4245 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4246 | reset_balance_state(fs_info); |
a17c95df | 4247 | else |
59641015 | 4248 | kfree(bctl); |
a17c95df DS |
4249 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4250 | ||
59641015 ID |
4251 | return ret; |
4252 | } | |
4253 | ||
4254 | static int balance_kthread(void *data) | |
4255 | { | |
2b6ba629 | 4256 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4257 | int ret = 0; |
59641015 | 4258 | |
59641015 | 4259 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4260 | if (fs_info->balance_ctl) |
6fcf6e2b | 4261 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4262 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4263 | |
59641015 ID |
4264 | return ret; |
4265 | } | |
4266 | ||
2b6ba629 ID |
4267 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4268 | { | |
4269 | struct task_struct *tsk; | |
4270 | ||
1354e1a1 | 4271 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4272 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4273 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4274 | return 0; |
4275 | } | |
1354e1a1 | 4276 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4277 | |
3cdde224 | 4278 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4279 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4280 | return 0; |
4281 | } | |
4282 | ||
02ee654d AJ |
4283 | /* |
4284 | * A ro->rw remount sequence should continue with the paused balance | |
4285 | * regardless of who pauses it, system or the user as of now, so set | |
4286 | * the resume flag. | |
4287 | */ | |
4288 | spin_lock(&fs_info->balance_lock); | |
4289 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4290 | spin_unlock(&fs_info->balance_lock); | |
4291 | ||
2b6ba629 | 4292 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4293 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4294 | } |
4295 | ||
68310a5e | 4296 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4297 | { |
59641015 ID |
4298 | struct btrfs_balance_control *bctl; |
4299 | struct btrfs_balance_item *item; | |
4300 | struct btrfs_disk_balance_args disk_bargs; | |
4301 | struct btrfs_path *path; | |
4302 | struct extent_buffer *leaf; | |
4303 | struct btrfs_key key; | |
4304 | int ret; | |
4305 | ||
4306 | path = btrfs_alloc_path(); | |
4307 | if (!path) | |
4308 | return -ENOMEM; | |
4309 | ||
59641015 | 4310 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4311 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4312 | key.offset = 0; |
4313 | ||
68310a5e | 4314 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4315 | if (ret < 0) |
68310a5e | 4316 | goto out; |
59641015 ID |
4317 | if (ret > 0) { /* ret = -ENOENT; */ |
4318 | ret = 0; | |
68310a5e ID |
4319 | goto out; |
4320 | } | |
4321 | ||
4322 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4323 | if (!bctl) { | |
4324 | ret = -ENOMEM; | |
4325 | goto out; | |
59641015 ID |
4326 | } |
4327 | ||
4328 | leaf = path->nodes[0]; | |
4329 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4330 | ||
68310a5e ID |
4331 | bctl->flags = btrfs_balance_flags(leaf, item); |
4332 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4333 | |
4334 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4335 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4336 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4337 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4338 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4339 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4340 | ||
eee95e3f DS |
4341 | /* |
4342 | * This should never happen, as the paused balance state is recovered | |
4343 | * during mount without any chance of other exclusive ops to collide. | |
4344 | * | |
4345 | * This gives the exclusive op status to balance and keeps in paused | |
4346 | * state until user intervention (cancel or umount). If the ownership | |
4347 | * cannot be assigned, show a message but do not fail. The balance | |
4348 | * is in a paused state and must have fs_info::balance_ctl properly | |
4349 | * set up. | |
4350 | */ | |
4351 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4352 | btrfs_warn(fs_info, | |
6dac13f8 | 4353 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4354 | |
68310a5e | 4355 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4356 | BUG_ON(fs_info->balance_ctl); |
4357 | spin_lock(&fs_info->balance_lock); | |
4358 | fs_info->balance_ctl = bctl; | |
4359 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4360 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4361 | out: |
4362 | btrfs_free_path(path); | |
ec44a35c CM |
4363 | return ret; |
4364 | } | |
4365 | ||
837d5b6e ID |
4366 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4367 | { | |
4368 | int ret = 0; | |
4369 | ||
4370 | mutex_lock(&fs_info->balance_mutex); | |
4371 | if (!fs_info->balance_ctl) { | |
4372 | mutex_unlock(&fs_info->balance_mutex); | |
4373 | return -ENOTCONN; | |
4374 | } | |
4375 | ||
3009a62f | 4376 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4377 | atomic_inc(&fs_info->balance_pause_req); |
4378 | mutex_unlock(&fs_info->balance_mutex); | |
4379 | ||
4380 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4381 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4382 | |
4383 | mutex_lock(&fs_info->balance_mutex); | |
4384 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4385 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4386 | atomic_dec(&fs_info->balance_pause_req); |
4387 | } else { | |
4388 | ret = -ENOTCONN; | |
4389 | } | |
4390 | ||
4391 | mutex_unlock(&fs_info->balance_mutex); | |
4392 | return ret; | |
4393 | } | |
4394 | ||
a7e99c69 ID |
4395 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4396 | { | |
4397 | mutex_lock(&fs_info->balance_mutex); | |
4398 | if (!fs_info->balance_ctl) { | |
4399 | mutex_unlock(&fs_info->balance_mutex); | |
4400 | return -ENOTCONN; | |
4401 | } | |
4402 | ||
cf7d20f4 DS |
4403 | /* |
4404 | * A paused balance with the item stored on disk can be resumed at | |
4405 | * mount time if the mount is read-write. Otherwise it's still paused | |
4406 | * and we must not allow cancelling as it deletes the item. | |
4407 | */ | |
4408 | if (sb_rdonly(fs_info->sb)) { | |
4409 | mutex_unlock(&fs_info->balance_mutex); | |
4410 | return -EROFS; | |
4411 | } | |
4412 | ||
a7e99c69 ID |
4413 | atomic_inc(&fs_info->balance_cancel_req); |
4414 | /* | |
4415 | * if we are running just wait and return, balance item is | |
4416 | * deleted in btrfs_balance in this case | |
4417 | */ | |
3009a62f | 4418 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4419 | mutex_unlock(&fs_info->balance_mutex); |
4420 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4421 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4422 | mutex_lock(&fs_info->balance_mutex); |
4423 | } else { | |
a7e99c69 | 4424 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4425 | /* |
4426 | * Lock released to allow other waiters to continue, we'll | |
4427 | * reexamine the status again. | |
4428 | */ | |
a7e99c69 ID |
4429 | mutex_lock(&fs_info->balance_mutex); |
4430 | ||
a17c95df | 4431 | if (fs_info->balance_ctl) { |
149196a2 | 4432 | reset_balance_state(fs_info); |
a17c95df | 4433 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4434 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4435 | } |
a7e99c69 ID |
4436 | } |
4437 | ||
3009a62f DS |
4438 | BUG_ON(fs_info->balance_ctl || |
4439 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4440 | atomic_dec(&fs_info->balance_cancel_req); |
4441 | mutex_unlock(&fs_info->balance_mutex); | |
4442 | return 0; | |
4443 | } | |
4444 | ||
803b2f54 SB |
4445 | static int btrfs_uuid_scan_kthread(void *data) |
4446 | { | |
4447 | struct btrfs_fs_info *fs_info = data; | |
4448 | struct btrfs_root *root = fs_info->tree_root; | |
4449 | struct btrfs_key key; | |
803b2f54 SB |
4450 | struct btrfs_path *path = NULL; |
4451 | int ret = 0; | |
4452 | struct extent_buffer *eb; | |
4453 | int slot; | |
4454 | struct btrfs_root_item root_item; | |
4455 | u32 item_size; | |
f45388f3 | 4456 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4457 | |
4458 | path = btrfs_alloc_path(); | |
4459 | if (!path) { | |
4460 | ret = -ENOMEM; | |
4461 | goto out; | |
4462 | } | |
4463 | ||
4464 | key.objectid = 0; | |
4465 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4466 | key.offset = 0; | |
4467 | ||
803b2f54 | 4468 | while (1) { |
7c829b72 AJ |
4469 | ret = btrfs_search_forward(root, &key, path, |
4470 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4471 | if (ret) { |
4472 | if (ret > 0) | |
4473 | ret = 0; | |
4474 | break; | |
4475 | } | |
4476 | ||
4477 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4478 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4479 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4480 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4481 | goto skip; | |
4482 | ||
4483 | eb = path->nodes[0]; | |
4484 | slot = path->slots[0]; | |
4485 | item_size = btrfs_item_size_nr(eb, slot); | |
4486 | if (item_size < sizeof(root_item)) | |
4487 | goto skip; | |
4488 | ||
803b2f54 SB |
4489 | read_extent_buffer(eb, &root_item, |
4490 | btrfs_item_ptr_offset(eb, slot), | |
4491 | (int)sizeof(root_item)); | |
4492 | if (btrfs_root_refs(&root_item) == 0) | |
4493 | goto skip; | |
f45388f3 FDBM |
4494 | |
4495 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4496 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4497 | if (trans) | |
4498 | goto update_tree; | |
4499 | ||
4500 | btrfs_release_path(path); | |
803b2f54 SB |
4501 | /* |
4502 | * 1 - subvol uuid item | |
4503 | * 1 - received_subvol uuid item | |
4504 | */ | |
4505 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4506 | if (IS_ERR(trans)) { | |
4507 | ret = PTR_ERR(trans); | |
4508 | break; | |
4509 | } | |
f45388f3 FDBM |
4510 | continue; |
4511 | } else { | |
4512 | goto skip; | |
4513 | } | |
4514 | update_tree: | |
4515 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4516 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4517 | BTRFS_UUID_KEY_SUBVOL, |
4518 | key.objectid); | |
4519 | if (ret < 0) { | |
efe120a0 | 4520 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4521 | ret); |
803b2f54 SB |
4522 | break; |
4523 | } | |
4524 | } | |
4525 | ||
4526 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4527 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4528 | root_item.received_uuid, |
4529 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4530 | key.objectid); | |
4531 | if (ret < 0) { | |
efe120a0 | 4532 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4533 | ret); |
803b2f54 SB |
4534 | break; |
4535 | } | |
4536 | } | |
4537 | ||
f45388f3 | 4538 | skip: |
803b2f54 | 4539 | if (trans) { |
3a45bb20 | 4540 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4541 | trans = NULL; |
803b2f54 SB |
4542 | if (ret) |
4543 | break; | |
4544 | } | |
4545 | ||
803b2f54 SB |
4546 | btrfs_release_path(path); |
4547 | if (key.offset < (u64)-1) { | |
4548 | key.offset++; | |
4549 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4550 | key.offset = 0; | |
4551 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4552 | } else if (key.objectid < (u64)-1) { | |
4553 | key.offset = 0; | |
4554 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4555 | key.objectid++; | |
4556 | } else { | |
4557 | break; | |
4558 | } | |
4559 | cond_resched(); | |
4560 | } | |
4561 | ||
4562 | out: | |
4563 | btrfs_free_path(path); | |
f45388f3 | 4564 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4565 | btrfs_end_transaction(trans); |
803b2f54 | 4566 | if (ret) |
efe120a0 | 4567 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4568 | else |
afcdd129 | 4569 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4570 | up(&fs_info->uuid_tree_rescan_sem); |
4571 | return 0; | |
4572 | } | |
4573 | ||
70f80175 SB |
4574 | /* |
4575 | * Callback for btrfs_uuid_tree_iterate(). | |
4576 | * returns: | |
4577 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4578 | * < 0 if an error occurred. |
70f80175 SB |
4579 | * > 0 if the check failed, which means the caller shall remove the entry. |
4580 | */ | |
4581 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4582 | u8 *uuid, u8 type, u64 subid) | |
4583 | { | |
4584 | struct btrfs_key key; | |
4585 | int ret = 0; | |
4586 | struct btrfs_root *subvol_root; | |
4587 | ||
4588 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4589 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4590 | goto out; | |
4591 | ||
4592 | key.objectid = subid; | |
4593 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4594 | key.offset = (u64)-1; | |
4595 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4596 | if (IS_ERR(subvol_root)) { | |
4597 | ret = PTR_ERR(subvol_root); | |
4598 | if (ret == -ENOENT) | |
4599 | ret = 1; | |
4600 | goto out; | |
4601 | } | |
4602 | ||
4603 | switch (type) { | |
4604 | case BTRFS_UUID_KEY_SUBVOL: | |
4605 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4606 | ret = 1; | |
4607 | break; | |
4608 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4609 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4610 | BTRFS_UUID_SIZE)) | |
4611 | ret = 1; | |
4612 | break; | |
4613 | } | |
4614 | ||
4615 | out: | |
4616 | return ret; | |
4617 | } | |
4618 | ||
4619 | static int btrfs_uuid_rescan_kthread(void *data) | |
4620 | { | |
4621 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4622 | int ret; | |
4623 | ||
4624 | /* | |
4625 | * 1st step is to iterate through the existing UUID tree and | |
4626 | * to delete all entries that contain outdated data. | |
4627 | * 2nd step is to add all missing entries to the UUID tree. | |
4628 | */ | |
4629 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4630 | if (ret < 0) { | |
efe120a0 | 4631 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4632 | up(&fs_info->uuid_tree_rescan_sem); |
4633 | return ret; | |
4634 | } | |
4635 | return btrfs_uuid_scan_kthread(data); | |
4636 | } | |
4637 | ||
f7a81ea4 SB |
4638 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4639 | { | |
4640 | struct btrfs_trans_handle *trans; | |
4641 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4642 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4643 | struct task_struct *task; |
4644 | int ret; | |
f7a81ea4 SB |
4645 | |
4646 | /* | |
4647 | * 1 - root node | |
4648 | * 1 - root item | |
4649 | */ | |
4650 | trans = btrfs_start_transaction(tree_root, 2); | |
4651 | if (IS_ERR(trans)) | |
4652 | return PTR_ERR(trans); | |
4653 | ||
4654 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4655 | BTRFS_UUID_TREE_OBJECTID); | |
4656 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4657 | ret = PTR_ERR(uuid_root); |
66642832 | 4658 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4659 | btrfs_end_transaction(trans); |
6d13f549 | 4660 | return ret; |
f7a81ea4 SB |
4661 | } |
4662 | ||
4663 | fs_info->uuid_root = uuid_root; | |
4664 | ||
3a45bb20 | 4665 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4666 | if (ret) |
4667 | return ret; | |
4668 | ||
4669 | down(&fs_info->uuid_tree_rescan_sem); | |
4670 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4671 | if (IS_ERR(task)) { | |
70f80175 | 4672 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4673 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4674 | up(&fs_info->uuid_tree_rescan_sem); |
4675 | return PTR_ERR(task); | |
4676 | } | |
4677 | ||
4678 | return 0; | |
f7a81ea4 | 4679 | } |
803b2f54 | 4680 | |
70f80175 SB |
4681 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4682 | { | |
4683 | struct task_struct *task; | |
4684 | ||
4685 | down(&fs_info->uuid_tree_rescan_sem); | |
4686 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4687 | if (IS_ERR(task)) { | |
4688 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4689 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4690 | up(&fs_info->uuid_tree_rescan_sem); |
4691 | return PTR_ERR(task); | |
4692 | } | |
4693 | ||
4694 | return 0; | |
4695 | } | |
4696 | ||
8f18cf13 CM |
4697 | /* |
4698 | * shrinking a device means finding all of the device extents past | |
4699 | * the new size, and then following the back refs to the chunks. | |
4700 | * The chunk relocation code actually frees the device extent | |
4701 | */ | |
4702 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4703 | { | |
0b246afa JM |
4704 | struct btrfs_fs_info *fs_info = device->fs_info; |
4705 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4706 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4707 | struct btrfs_dev_extent *dev_extent = NULL; |
4708 | struct btrfs_path *path; | |
4709 | u64 length; | |
8f18cf13 CM |
4710 | u64 chunk_offset; |
4711 | int ret; | |
4712 | int slot; | |
ba1bf481 JB |
4713 | int failed = 0; |
4714 | bool retried = false; | |
53e489bc | 4715 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4716 | struct extent_buffer *l; |
4717 | struct btrfs_key key; | |
0b246afa | 4718 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4719 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4720 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4721 | u64 diff; |
4722 | ||
4723 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4724 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4725 | |
401e29c1 | 4726 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4727 | return -EINVAL; |
4728 | ||
8f18cf13 CM |
4729 | path = btrfs_alloc_path(); |
4730 | if (!path) | |
4731 | return -ENOMEM; | |
4732 | ||
0338dff6 | 4733 | path->reada = READA_BACK; |
8f18cf13 | 4734 | |
34441361 | 4735 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4736 | |
7cc8e58d | 4737 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4738 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4739 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4740 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4741 | } |
34441361 | 4742 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4743 | |
ba1bf481 | 4744 | again: |
8f18cf13 CM |
4745 | key.objectid = device->devid; |
4746 | key.offset = (u64)-1; | |
4747 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4748 | ||
213e64da | 4749 | do { |
0b246afa | 4750 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4751 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4752 | if (ret < 0) { |
0b246afa | 4753 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4754 | goto done; |
67c5e7d4 | 4755 | } |
8f18cf13 CM |
4756 | |
4757 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4758 | if (ret) |
0b246afa | 4759 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4760 | if (ret < 0) |
4761 | goto done; | |
4762 | if (ret) { | |
4763 | ret = 0; | |
b3b4aa74 | 4764 | btrfs_release_path(path); |
bf1fb512 | 4765 | break; |
8f18cf13 CM |
4766 | } |
4767 | ||
4768 | l = path->nodes[0]; | |
4769 | slot = path->slots[0]; | |
4770 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4771 | ||
ba1bf481 | 4772 | if (key.objectid != device->devid) { |
0b246afa | 4773 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4774 | btrfs_release_path(path); |
bf1fb512 | 4775 | break; |
ba1bf481 | 4776 | } |
8f18cf13 CM |
4777 | |
4778 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4779 | length = btrfs_dev_extent_length(l, dev_extent); | |
4780 | ||
ba1bf481 | 4781 | if (key.offset + length <= new_size) { |
0b246afa | 4782 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4783 | btrfs_release_path(path); |
d6397bae | 4784 | break; |
ba1bf481 | 4785 | } |
8f18cf13 | 4786 | |
8f18cf13 | 4787 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4788 | btrfs_release_path(path); |
8f18cf13 | 4789 | |
a6f93c71 LB |
4790 | /* |
4791 | * We may be relocating the only data chunk we have, | |
4792 | * which could potentially end up with losing data's | |
4793 | * raid profile, so lets allocate an empty one in | |
4794 | * advance. | |
4795 | */ | |
4796 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4797 | if (ret < 0) { | |
4798 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4799 | goto done; | |
4800 | } | |
4801 | ||
0b246afa JM |
4802 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4803 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4804 | if (ret == -ENOSPC) { |
ba1bf481 | 4805 | failed++; |
eede2bf3 OS |
4806 | } else if (ret) { |
4807 | if (ret == -ETXTBSY) { | |
4808 | btrfs_warn(fs_info, | |
4809 | "could not shrink block group %llu due to active swapfile", | |
4810 | chunk_offset); | |
4811 | } | |
4812 | goto done; | |
4813 | } | |
213e64da | 4814 | } while (key.offset-- > 0); |
ba1bf481 JB |
4815 | |
4816 | if (failed && !retried) { | |
4817 | failed = 0; | |
4818 | retried = true; | |
4819 | goto again; | |
4820 | } else if (failed && retried) { | |
4821 | ret = -ENOSPC; | |
ba1bf481 | 4822 | goto done; |
8f18cf13 CM |
4823 | } |
4824 | ||
d6397bae | 4825 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4826 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4827 | if (IS_ERR(trans)) { |
4828 | ret = PTR_ERR(trans); | |
4829 | goto done; | |
4830 | } | |
4831 | ||
34441361 | 4832 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4833 | |
4834 | /* | |
4835 | * We checked in the above loop all device extents that were already in | |
4836 | * the device tree. However before we have updated the device's | |
4837 | * total_bytes to the new size, we might have had chunk allocations that | |
4838 | * have not complete yet (new block groups attached to transaction | |
4839 | * handles), and therefore their device extents were not yet in the | |
4840 | * device tree and we missed them in the loop above. So if we have any | |
4841 | * pending chunk using a device extent that overlaps the device range | |
4842 | * that we can not use anymore, commit the current transaction and | |
4843 | * repeat the search on the device tree - this way we guarantee we will | |
4844 | * not have chunks using device extents that end beyond 'new_size'. | |
4845 | */ | |
4846 | if (!checked_pending_chunks) { | |
4847 | u64 start = new_size; | |
4848 | u64 len = old_size - new_size; | |
4849 | ||
499f377f JM |
4850 | if (contains_pending_extent(trans->transaction, device, |
4851 | &start, len)) { | |
34441361 | 4852 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4853 | checked_pending_chunks = true; |
4854 | failed = 0; | |
4855 | retried = false; | |
3a45bb20 | 4856 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4857 | if (ret) |
4858 | goto done; | |
4859 | goto again; | |
4860 | } | |
4861 | } | |
4862 | ||
7cc8e58d | 4863 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4864 | if (list_empty(&device->resized_list)) |
4865 | list_add_tail(&device->resized_list, | |
0b246afa | 4866 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4867 | |
d6397bae | 4868 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4869 | btrfs_set_super_total_bytes(super_copy, |
4870 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4871 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4872 | |
4873 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4874 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4875 | if (ret < 0) { |
4876 | btrfs_abort_transaction(trans, ret); | |
4877 | btrfs_end_transaction(trans); | |
4878 | } else { | |
4879 | ret = btrfs_commit_transaction(trans); | |
4880 | } | |
8f18cf13 CM |
4881 | done: |
4882 | btrfs_free_path(path); | |
53e489bc | 4883 | if (ret) { |
34441361 | 4884 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4885 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4886 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4887 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4888 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4889 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4890 | } |
8f18cf13 CM |
4891 | return ret; |
4892 | } | |
4893 | ||
2ff7e61e | 4894 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4895 | struct btrfs_key *key, |
4896 | struct btrfs_chunk *chunk, int item_size) | |
4897 | { | |
0b246afa | 4898 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4899 | struct btrfs_disk_key disk_key; |
4900 | u32 array_size; | |
4901 | u8 *ptr; | |
4902 | ||
34441361 | 4903 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4904 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4905 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4906 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4907 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4908 | return -EFBIG; |
fe48a5c0 | 4909 | } |
0b86a832 CM |
4910 | |
4911 | ptr = super_copy->sys_chunk_array + array_size; | |
4912 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4913 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4914 | ptr += sizeof(disk_key); | |
4915 | memcpy(ptr, chunk, item_size); | |
4916 | item_size += sizeof(disk_key); | |
4917 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4918 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4919 | |
0b86a832 CM |
4920 | return 0; |
4921 | } | |
4922 | ||
73c5de00 AJ |
4923 | /* |
4924 | * sort the devices in descending order by max_avail, total_avail | |
4925 | */ | |
4926 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4927 | { |
73c5de00 AJ |
4928 | const struct btrfs_device_info *di_a = a; |
4929 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4930 | |
73c5de00 | 4931 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4932 | return -1; |
73c5de00 | 4933 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4934 | return 1; |
73c5de00 AJ |
4935 | if (di_a->total_avail > di_b->total_avail) |
4936 | return -1; | |
4937 | if (di_a->total_avail < di_b->total_avail) | |
4938 | return 1; | |
4939 | return 0; | |
b2117a39 | 4940 | } |
0b86a832 | 4941 | |
53b381b3 DW |
4942 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4943 | { | |
ffe2d203 | 4944 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4945 | return; |
4946 | ||
ceda0864 | 4947 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4948 | } |
4949 | ||
062d4d1f | 4950 | #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \ |
23f8f9b7 GH |
4951 | - sizeof(struct btrfs_chunk)) \ |
4952 | / sizeof(struct btrfs_stripe) + 1) | |
4953 | ||
4954 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4955 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4956 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4957 | / sizeof(struct btrfs_stripe) + 1) | |
4958 | ||
73c5de00 | 4959 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4960 | u64 start, u64 type) |
b2117a39 | 4961 | { |
2ff7e61e | 4962 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4963 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4964 | struct btrfs_device *device; |
73c5de00 AJ |
4965 | struct map_lookup *map = NULL; |
4966 | struct extent_map_tree *em_tree; | |
4967 | struct extent_map *em; | |
4968 | struct btrfs_device_info *devices_info = NULL; | |
4969 | u64 total_avail; | |
4970 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4971 | int data_stripes; /* number of stripes that count for |
4972 | block group size */ | |
73c5de00 AJ |
4973 | int sub_stripes; /* sub_stripes info for map */ |
4974 | int dev_stripes; /* stripes per dev */ | |
4975 | int devs_max; /* max devs to use */ | |
4976 | int devs_min; /* min devs needed */ | |
4977 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4978 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4979 | int nparity; /* number of stripes worth of bytes to |
4980 | store parity information */ | |
73c5de00 AJ |
4981 | int ret; |
4982 | u64 max_stripe_size; | |
4983 | u64 max_chunk_size; | |
4984 | u64 stripe_size; | |
23f0ff1e | 4985 | u64 chunk_size; |
73c5de00 AJ |
4986 | int ndevs; |
4987 | int i; | |
4988 | int j; | |
31e50229 | 4989 | int index; |
593060d7 | 4990 | |
0c460c0d | 4991 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4992 | |
4117f207 QW |
4993 | if (list_empty(&fs_devices->alloc_list)) { |
4994 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4995 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4996 | return -ENOSPC; |
4117f207 | 4997 | } |
b2117a39 | 4998 | |
3e72ee88 | 4999 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 5000 | |
31e50229 LB |
5001 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
5002 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5003 | devs_max = btrfs_raid_array[index].devs_max; | |
5004 | devs_min = btrfs_raid_array[index].devs_min; | |
5005 | devs_increment = btrfs_raid_array[index].devs_increment; | |
5006 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 5007 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 5008 | |
9b3f68b9 | 5009 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 5010 | max_stripe_size = SZ_1G; |
fce466ea | 5011 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 5012 | if (!devs_max) |
062d4d1f | 5013 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 5014 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 5015 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
5016 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
5017 | max_stripe_size = SZ_1G; | |
1100373f | 5018 | else |
ee22184b | 5019 | max_stripe_size = SZ_256M; |
73c5de00 | 5020 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 5021 | if (!devs_max) |
062d4d1f | 5022 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 5023 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 5024 | max_stripe_size = SZ_32M; |
73c5de00 | 5025 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
5026 | if (!devs_max) |
5027 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 5028 | } else { |
351fd353 | 5029 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
5030 | type); |
5031 | BUG_ON(1); | |
9b3f68b9 CM |
5032 | } |
5033 | ||
52042d8e | 5034 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
5035 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
5036 | max_chunk_size); | |
9b3f68b9 | 5037 | |
31e818fe | 5038 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
5039 | GFP_NOFS); |
5040 | if (!devices_info) | |
5041 | return -ENOMEM; | |
0cad8a11 | 5042 | |
9f680ce0 | 5043 | /* |
73c5de00 AJ |
5044 | * in the first pass through the devices list, we gather information |
5045 | * about the available holes on each device. | |
9f680ce0 | 5046 | */ |
73c5de00 | 5047 | ndevs = 0; |
ebcc9301 | 5048 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
5049 | u64 max_avail; |
5050 | u64 dev_offset; | |
b2117a39 | 5051 | |
ebbede42 | 5052 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5053 | WARN(1, KERN_ERR |
efe120a0 | 5054 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5055 | continue; |
5056 | } | |
b2117a39 | 5057 | |
e12c9621 AJ |
5058 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5059 | &device->dev_state) || | |
401e29c1 | 5060 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5061 | continue; |
b2117a39 | 5062 | |
73c5de00 AJ |
5063 | if (device->total_bytes > device->bytes_used) |
5064 | total_avail = device->total_bytes - device->bytes_used; | |
5065 | else | |
5066 | total_avail = 0; | |
38c01b96 | 5067 | |
5068 | /* If there is no space on this device, skip it. */ | |
5069 | if (total_avail == 0) | |
5070 | continue; | |
b2117a39 | 5071 | |
6df9a95e | 5072 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
5073 | max_stripe_size * dev_stripes, |
5074 | &dev_offset, &max_avail); | |
5075 | if (ret && ret != -ENOSPC) | |
5076 | goto error; | |
b2117a39 | 5077 | |
73c5de00 AJ |
5078 | if (ret == 0) |
5079 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5080 | |
4117f207 QW |
5081 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5082 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5083 | btrfs_debug(info, | |
5084 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5085 | __func__, device->devid, max_avail, | |
5086 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5087 | continue; |
4117f207 | 5088 | } |
b2117a39 | 5089 | |
063d006f ES |
5090 | if (ndevs == fs_devices->rw_devices) { |
5091 | WARN(1, "%s: found more than %llu devices\n", | |
5092 | __func__, fs_devices->rw_devices); | |
5093 | break; | |
5094 | } | |
73c5de00 AJ |
5095 | devices_info[ndevs].dev_offset = dev_offset; |
5096 | devices_info[ndevs].max_avail = max_avail; | |
5097 | devices_info[ndevs].total_avail = total_avail; | |
5098 | devices_info[ndevs].dev = device; | |
5099 | ++ndevs; | |
5100 | } | |
b2117a39 | 5101 | |
73c5de00 AJ |
5102 | /* |
5103 | * now sort the devices by hole size / available space | |
5104 | */ | |
5105 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5106 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5107 | |
73c5de00 | 5108 | /* round down to number of usable stripes */ |
e5600fd6 | 5109 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5110 | |
ba89b802 | 5111 | if (ndevs < devs_min) { |
73c5de00 | 5112 | ret = -ENOSPC; |
4117f207 QW |
5113 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5114 | btrfs_debug(info, | |
5115 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5116 | __func__, ndevs, devs_min); |
4117f207 | 5117 | } |
73c5de00 | 5118 | goto error; |
b2117a39 | 5119 | } |
9f680ce0 | 5120 | |
f148ef4d NB |
5121 | ndevs = min(ndevs, devs_max); |
5122 | ||
73c5de00 | 5123 | /* |
92e222df HK |
5124 | * The primary goal is to maximize the number of stripes, so use as |
5125 | * many devices as possible, even if the stripes are not maximum sized. | |
5126 | * | |
5127 | * The DUP profile stores more than one stripe per device, the | |
5128 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5129 | */ |
92e222df | 5130 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5131 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5132 | |
53b381b3 DW |
5133 | /* |
5134 | * this will have to be fixed for RAID1 and RAID10 over | |
5135 | * more drives | |
5136 | */ | |
b50836ed | 5137 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5138 | |
5139 | /* | |
5140 | * Use the number of data stripes to figure out how big this chunk | |
5141 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5142 | * and compare that answer with the max chunk size. If it's higher, |
5143 | * we try to reduce stripe_size. | |
86db2578 CM |
5144 | */ |
5145 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5146 | /* |
baf92114 HK |
5147 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5148 | * then use it, unless it ends up being even bigger than the | |
5149 | * previous value we had already. | |
86db2578 | 5150 | */ |
baf92114 HK |
5151 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5152 | data_stripes), SZ_16M), | |
793ff2c8 | 5153 | stripe_size); |
86db2578 CM |
5154 | } |
5155 | ||
37db63a4 | 5156 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5157 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5158 | |
5159 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5160 | if (!map) { | |
5161 | ret = -ENOMEM; | |
5162 | goto error; | |
5163 | } | |
5164 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5165 | |
73c5de00 AJ |
5166 | for (i = 0; i < ndevs; ++i) { |
5167 | for (j = 0; j < dev_stripes; ++j) { | |
5168 | int s = i * dev_stripes + j; | |
5169 | map->stripes[s].dev = devices_info[i].dev; | |
5170 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5171 | j * stripe_size; | |
6324fbf3 | 5172 | } |
6324fbf3 | 5173 | } |
500ceed8 NB |
5174 | map->stripe_len = BTRFS_STRIPE_LEN; |
5175 | map->io_align = BTRFS_STRIPE_LEN; | |
5176 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5177 | map->type = type; |
2b82032c | 5178 | map->sub_stripes = sub_stripes; |
0b86a832 | 5179 | |
23f0ff1e | 5180 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5181 | |
23f0ff1e | 5182 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5183 | |
172ddd60 | 5184 | em = alloc_extent_map(); |
2b82032c | 5185 | if (!em) { |
298a8f9c | 5186 | kfree(map); |
b2117a39 MX |
5187 | ret = -ENOMEM; |
5188 | goto error; | |
593060d7 | 5189 | } |
298a8f9c | 5190 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5191 | em->map_lookup = map; |
2b82032c | 5192 | em->start = start; |
23f0ff1e | 5193 | em->len = chunk_size; |
2b82032c YZ |
5194 | em->block_start = 0; |
5195 | em->block_len = em->len; | |
6df9a95e | 5196 | em->orig_block_len = stripe_size; |
593060d7 | 5197 | |
0b246afa | 5198 | em_tree = &info->mapping_tree.map_tree; |
890871be | 5199 | write_lock(&em_tree->lock); |
09a2a8f9 | 5200 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5201 | if (ret) { |
1efb72a3 | 5202 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5203 | free_extent_map(em); |
1dd4602f | 5204 | goto error; |
0f5d42b2 | 5205 | } |
0b86a832 | 5206 | |
1efb72a3 NB |
5207 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
5208 | refcount_inc(&em->refs); | |
5209 | write_unlock(&em_tree->lock); | |
5210 | ||
23f0ff1e | 5211 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5212 | if (ret) |
5213 | goto error_del_extent; | |
2b82032c | 5214 | |
2f29df4f HK |
5215 | for (i = 0; i < map->num_stripes; i++) |
5216 | btrfs_device_set_bytes_used(map->stripes[i].dev, | |
5217 | map->stripes[i].dev->bytes_used + stripe_size); | |
43530c46 | 5218 | |
a5ed45f8 | 5219 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5220 | |
0f5d42b2 | 5221 | free_extent_map(em); |
0b246afa | 5222 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5223 | |
b2117a39 | 5224 | kfree(devices_info); |
2b82032c | 5225 | return 0; |
b2117a39 | 5226 | |
6df9a95e | 5227 | error_del_extent: |
0f5d42b2 JB |
5228 | write_lock(&em_tree->lock); |
5229 | remove_extent_mapping(em_tree, em); | |
5230 | write_unlock(&em_tree->lock); | |
5231 | ||
5232 | /* One for our allocation */ | |
5233 | free_extent_map(em); | |
5234 | /* One for the tree reference */ | |
5235 | free_extent_map(em); | |
495e64f4 FM |
5236 | /* One for the pending_chunks list reference */ |
5237 | free_extent_map(em); | |
b2117a39 | 5238 | error: |
b2117a39 MX |
5239 | kfree(devices_info); |
5240 | return ret; | |
2b82032c YZ |
5241 | } |
5242 | ||
6df9a95e | 5243 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5244 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5245 | { |
97aff912 | 5246 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5247 | struct btrfs_root *extent_root = fs_info->extent_root; |
5248 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5249 | struct btrfs_key key; |
2b82032c YZ |
5250 | struct btrfs_device *device; |
5251 | struct btrfs_chunk *chunk; | |
5252 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5253 | struct extent_map *em; |
5254 | struct map_lookup *map; | |
5255 | size_t item_size; | |
5256 | u64 dev_offset; | |
5257 | u64 stripe_size; | |
5258 | int i = 0; | |
140e639f | 5259 | int ret = 0; |
2b82032c | 5260 | |
60ca842e | 5261 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5262 | if (IS_ERR(em)) |
5263 | return PTR_ERR(em); | |
6df9a95e | 5264 | |
95617d69 | 5265 | map = em->map_lookup; |
6df9a95e JB |
5266 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5267 | stripe_size = em->orig_block_len; | |
5268 | ||
2b82032c | 5269 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5270 | if (!chunk) { |
5271 | ret = -ENOMEM; | |
5272 | goto out; | |
5273 | } | |
5274 | ||
50460e37 FM |
5275 | /* |
5276 | * Take the device list mutex to prevent races with the final phase of | |
5277 | * a device replace operation that replaces the device object associated | |
5278 | * with the map's stripes, because the device object's id can change | |
5279 | * at any time during that final phase of the device replace operation | |
5280 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5281 | */ | |
0b246afa | 5282 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5283 | for (i = 0; i < map->num_stripes; i++) { |
5284 | device = map->stripes[i].dev; | |
5285 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5286 | |
0b86a832 | 5287 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5288 | if (ret) |
50460e37 | 5289 | break; |
b5d9071c NB |
5290 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5291 | dev_offset, stripe_size); | |
6df9a95e | 5292 | if (ret) |
50460e37 FM |
5293 | break; |
5294 | } | |
5295 | if (ret) { | |
0b246afa | 5296 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5297 | goto out; |
2b82032c YZ |
5298 | } |
5299 | ||
2b82032c | 5300 | stripe = &chunk->stripe; |
6df9a95e JB |
5301 | for (i = 0; i < map->num_stripes; i++) { |
5302 | device = map->stripes[i].dev; | |
5303 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5304 | |
e17cade2 CM |
5305 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5306 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5307 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5308 | stripe++; |
0b86a832 | 5309 | } |
0b246afa | 5310 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5311 | |
2b82032c | 5312 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5313 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5314 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5315 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5316 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5317 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5318 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5319 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5320 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5321 | |
2b82032c YZ |
5322 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5323 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5324 | key.offset = chunk_offset; | |
0b86a832 | 5325 | |
2b82032c | 5326 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5327 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5328 | /* | |
5329 | * TODO: Cleanup of inserted chunk root in case of | |
5330 | * failure. | |
5331 | */ | |
2ff7e61e | 5332 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5333 | } |
1abe9b8a | 5334 | |
6df9a95e | 5335 | out: |
0b86a832 | 5336 | kfree(chunk); |
6df9a95e | 5337 | free_extent_map(em); |
4ed1d16e | 5338 | return ret; |
2b82032c | 5339 | } |
0b86a832 | 5340 | |
2b82032c | 5341 | /* |
52042d8e AG |
5342 | * Chunk allocation falls into two parts. The first part does work |
5343 | * that makes the new allocated chunk usable, but does not do any operation | |
5344 | * that modifies the chunk tree. The second part does the work that | |
5345 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5346 | * bootstrap process of adding storage to a seed btrfs. |
5347 | */ | |
c216b203 | 5348 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5349 | { |
5350 | u64 chunk_offset; | |
2b82032c | 5351 | |
c216b203 NB |
5352 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5353 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5354 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5355 | } |
5356 | ||
d397712b | 5357 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5358 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5359 | { |
5360 | u64 chunk_offset; | |
5361 | u64 sys_chunk_offset; | |
2b82032c | 5362 | u64 alloc_profile; |
2b82032c YZ |
5363 | int ret; |
5364 | ||
6df9a95e | 5365 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5366 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5367 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5368 | if (ret) |
5369 | return ret; | |
2b82032c | 5370 | |
0b246afa | 5371 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5372 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5373 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5374 | return ret; |
2b82032c YZ |
5375 | } |
5376 | ||
d20983b4 MX |
5377 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5378 | { | |
5379 | int max_errors; | |
5380 | ||
5381 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5382 | BTRFS_BLOCK_GROUP_RAID10 | | |
5383 | BTRFS_BLOCK_GROUP_RAID5 | | |
5384 | BTRFS_BLOCK_GROUP_DUP)) { | |
5385 | max_errors = 1; | |
5386 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5387 | max_errors = 2; | |
5388 | } else { | |
5389 | max_errors = 0; | |
005d6427 | 5390 | } |
2b82032c | 5391 | |
d20983b4 | 5392 | return max_errors; |
2b82032c YZ |
5393 | } |
5394 | ||
2ff7e61e | 5395 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5396 | { |
5397 | struct extent_map *em; | |
5398 | struct map_lookup *map; | |
2b82032c | 5399 | int readonly = 0; |
d20983b4 | 5400 | int miss_ndevs = 0; |
2b82032c YZ |
5401 | int i; |
5402 | ||
60ca842e | 5403 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5404 | if (IS_ERR(em)) |
2b82032c YZ |
5405 | return 1; |
5406 | ||
95617d69 | 5407 | map = em->map_lookup; |
2b82032c | 5408 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5409 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5410 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5411 | miss_ndevs++; |
5412 | continue; | |
5413 | } | |
ebbede42 AJ |
5414 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5415 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5416 | readonly = 1; |
d20983b4 | 5417 | goto end; |
2b82032c YZ |
5418 | } |
5419 | } | |
d20983b4 MX |
5420 | |
5421 | /* | |
5422 | * If the number of missing devices is larger than max errors, | |
5423 | * we can not write the data into that chunk successfully, so | |
5424 | * set it readonly. | |
5425 | */ | |
5426 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5427 | readonly = 1; | |
5428 | end: | |
0b86a832 | 5429 | free_extent_map(em); |
2b82032c | 5430 | return readonly; |
0b86a832 CM |
5431 | } |
5432 | ||
5433 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5434 | { | |
a8067e02 | 5435 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5436 | } |
5437 | ||
5438 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5439 | { | |
5440 | struct extent_map *em; | |
5441 | ||
d397712b | 5442 | while (1) { |
890871be | 5443 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5444 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5445 | if (em) | |
5446 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5447 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5448 | if (!em) |
5449 | break; | |
0b86a832 CM |
5450 | /* once for us */ |
5451 | free_extent_map(em); | |
5452 | /* once for the tree */ | |
5453 | free_extent_map(em); | |
5454 | } | |
5455 | } | |
5456 | ||
5d964051 | 5457 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5458 | { |
5459 | struct extent_map *em; | |
5460 | struct map_lookup *map; | |
f188591e CM |
5461 | int ret; |
5462 | ||
60ca842e | 5463 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5464 | if (IS_ERR(em)) |
5465 | /* | |
5466 | * We could return errors for these cases, but that could get | |
5467 | * ugly and we'd probably do the same thing which is just not do | |
5468 | * anything else and exit, so return 1 so the callers don't try | |
5469 | * to use other copies. | |
5470 | */ | |
fb7669b5 | 5471 | return 1; |
fb7669b5 | 5472 | |
95617d69 | 5473 | map = em->map_lookup; |
f188591e CM |
5474 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5475 | ret = map->num_stripes; | |
321aecc6 CM |
5476 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5477 | ret = map->sub_stripes; | |
53b381b3 DW |
5478 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5479 | ret = 2; | |
5480 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5481 | /* |
5482 | * There could be two corrupted data stripes, we need | |
5483 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5484 | * |
8810f751 LB |
5485 | * Fail a stripe at a time on every retry except the |
5486 | * stripe under reconstruction. | |
5487 | */ | |
5488 | ret = map->num_stripes; | |
f188591e CM |
5489 | else |
5490 | ret = 1; | |
5491 | free_extent_map(em); | |
ad6d620e | 5492 | |
cb5583dd | 5493 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5494 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5495 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5496 | ret++; |
cb5583dd | 5497 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5498 | |
f188591e CM |
5499 | return ret; |
5500 | } | |
5501 | ||
2ff7e61e | 5502 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5503 | u64 logical) |
5504 | { | |
5505 | struct extent_map *em; | |
5506 | struct map_lookup *map; | |
0b246afa | 5507 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5508 | |
60ca842e | 5509 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5510 | |
69f03f13 NB |
5511 | if (!WARN_ON(IS_ERR(em))) { |
5512 | map = em->map_lookup; | |
5513 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5514 | len = map->stripe_len * nr_data_stripes(map); | |
5515 | free_extent_map(em); | |
5516 | } | |
53b381b3 DW |
5517 | return len; |
5518 | } | |
5519 | ||
e4ff5fb5 | 5520 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5521 | { |
5522 | struct extent_map *em; | |
5523 | struct map_lookup *map; | |
53b381b3 DW |
5524 | int ret = 0; |
5525 | ||
60ca842e | 5526 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5527 | |
69f03f13 NB |
5528 | if(!WARN_ON(IS_ERR(em))) { |
5529 | map = em->map_lookup; | |
5530 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5531 | ret = 1; | |
5532 | free_extent_map(em); | |
5533 | } | |
53b381b3 DW |
5534 | return ret; |
5535 | } | |
5536 | ||
30d9861f | 5537 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5538 | struct map_lookup *map, int first, |
8ba0ae78 | 5539 | int dev_replace_is_ongoing) |
dfe25020 CM |
5540 | { |
5541 | int i; | |
99f92a7c | 5542 | int num_stripes; |
8ba0ae78 | 5543 | int preferred_mirror; |
30d9861f SB |
5544 | int tolerance; |
5545 | struct btrfs_device *srcdev; | |
5546 | ||
99f92a7c AJ |
5547 | ASSERT((map->type & |
5548 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5549 | ||
5550 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5551 | num_stripes = map->sub_stripes; | |
5552 | else | |
5553 | num_stripes = map->num_stripes; | |
5554 | ||
8ba0ae78 AJ |
5555 | preferred_mirror = first + current->pid % num_stripes; |
5556 | ||
30d9861f SB |
5557 | if (dev_replace_is_ongoing && |
5558 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5559 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5560 | srcdev = fs_info->dev_replace.srcdev; | |
5561 | else | |
5562 | srcdev = NULL; | |
5563 | ||
5564 | /* | |
5565 | * try to avoid the drive that is the source drive for a | |
5566 | * dev-replace procedure, only choose it if no other non-missing | |
5567 | * mirror is available | |
5568 | */ | |
5569 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5570 | if (map->stripes[preferred_mirror].dev->bdev && |
5571 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5572 | return preferred_mirror; | |
99f92a7c | 5573 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5574 | if (map->stripes[i].dev->bdev && |
5575 | (tolerance || map->stripes[i].dev != srcdev)) | |
5576 | return i; | |
5577 | } | |
dfe25020 | 5578 | } |
30d9861f | 5579 | |
dfe25020 CM |
5580 | /* we couldn't find one that doesn't fail. Just return something |
5581 | * and the io error handling code will clean up eventually | |
5582 | */ | |
8ba0ae78 | 5583 | return preferred_mirror; |
dfe25020 CM |
5584 | } |
5585 | ||
53b381b3 DW |
5586 | static inline int parity_smaller(u64 a, u64 b) |
5587 | { | |
5588 | return a > b; | |
5589 | } | |
5590 | ||
5591 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5592 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5593 | { |
5594 | struct btrfs_bio_stripe s; | |
5595 | int i; | |
5596 | u64 l; | |
5597 | int again = 1; | |
5598 | ||
5599 | while (again) { | |
5600 | again = 0; | |
cc7539ed | 5601 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5602 | if (parity_smaller(bbio->raid_map[i], |
5603 | bbio->raid_map[i+1])) { | |
53b381b3 | 5604 | s = bbio->stripes[i]; |
8e5cfb55 | 5605 | l = bbio->raid_map[i]; |
53b381b3 | 5606 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5607 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5608 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5609 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5610 | |
53b381b3 DW |
5611 | again = 1; |
5612 | } | |
5613 | } | |
5614 | } | |
5615 | } | |
5616 | ||
6e9606d2 ZL |
5617 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5618 | { | |
5619 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5620 | /* the size of the btrfs_bio */ |
6e9606d2 | 5621 | sizeof(struct btrfs_bio) + |
e57cf21e | 5622 | /* plus the variable array for the stripes */ |
6e9606d2 | 5623 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5624 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5625 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5626 | /* |
5627 | * plus the raid_map, which includes both the tgt dev | |
5628 | * and the stripes | |
5629 | */ | |
5630 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5631 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5632 | |
5633 | atomic_set(&bbio->error, 0); | |
140475ae | 5634 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5635 | |
5636 | return bbio; | |
5637 | } | |
5638 | ||
5639 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5640 | { | |
140475ae ER |
5641 | WARN_ON(!refcount_read(&bbio->refs)); |
5642 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5643 | } |
5644 | ||
5645 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5646 | { | |
5647 | if (!bbio) | |
5648 | return; | |
140475ae | 5649 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5650 | kfree(bbio); |
5651 | } | |
5652 | ||
0b3d4cd3 LB |
5653 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5654 | /* | |
5655 | * Please note that, discard won't be sent to target device of device | |
5656 | * replace. | |
5657 | */ | |
5658 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5659 | u64 logical, u64 length, | |
5660 | struct btrfs_bio **bbio_ret) | |
5661 | { | |
5662 | struct extent_map *em; | |
5663 | struct map_lookup *map; | |
5664 | struct btrfs_bio *bbio; | |
5665 | u64 offset; | |
5666 | u64 stripe_nr; | |
5667 | u64 stripe_nr_end; | |
5668 | u64 stripe_end_offset; | |
5669 | u64 stripe_cnt; | |
5670 | u64 stripe_len; | |
5671 | u64 stripe_offset; | |
5672 | u64 num_stripes; | |
5673 | u32 stripe_index; | |
5674 | u32 factor = 0; | |
5675 | u32 sub_stripes = 0; | |
5676 | u64 stripes_per_dev = 0; | |
5677 | u32 remaining_stripes = 0; | |
5678 | u32 last_stripe = 0; | |
5679 | int ret = 0; | |
5680 | int i; | |
5681 | ||
5682 | /* discard always return a bbio */ | |
5683 | ASSERT(bbio_ret); | |
5684 | ||
60ca842e | 5685 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5686 | if (IS_ERR(em)) |
5687 | return PTR_ERR(em); | |
5688 | ||
5689 | map = em->map_lookup; | |
5690 | /* we don't discard raid56 yet */ | |
5691 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5692 | ret = -EOPNOTSUPP; | |
5693 | goto out; | |
5694 | } | |
5695 | ||
5696 | offset = logical - em->start; | |
5697 | length = min_t(u64, em->len - offset, length); | |
5698 | ||
5699 | stripe_len = map->stripe_len; | |
5700 | /* | |
5701 | * stripe_nr counts the total number of stripes we have to stride | |
5702 | * to get to this block | |
5703 | */ | |
5704 | stripe_nr = div64_u64(offset, stripe_len); | |
5705 | ||
5706 | /* stripe_offset is the offset of this block in its stripe */ | |
5707 | stripe_offset = offset - stripe_nr * stripe_len; | |
5708 | ||
5709 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5710 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5711 | stripe_cnt = stripe_nr_end - stripe_nr; |
5712 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5713 | (offset + length); | |
5714 | /* | |
5715 | * after this, stripe_nr is the number of stripes on this | |
5716 | * device we have to walk to find the data, and stripe_index is | |
5717 | * the number of our device in the stripe array | |
5718 | */ | |
5719 | num_stripes = 1; | |
5720 | stripe_index = 0; | |
5721 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5722 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5723 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5724 | sub_stripes = 1; | |
5725 | else | |
5726 | sub_stripes = map->sub_stripes; | |
5727 | ||
5728 | factor = map->num_stripes / sub_stripes; | |
5729 | num_stripes = min_t(u64, map->num_stripes, | |
5730 | sub_stripes * stripe_cnt); | |
5731 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5732 | stripe_index *= sub_stripes; | |
5733 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5734 | &remaining_stripes); | |
5735 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5736 | last_stripe *= sub_stripes; | |
5737 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5738 | BTRFS_BLOCK_GROUP_DUP)) { | |
5739 | num_stripes = map->num_stripes; | |
5740 | } else { | |
5741 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5742 | &stripe_index); | |
5743 | } | |
5744 | ||
5745 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5746 | if (!bbio) { | |
5747 | ret = -ENOMEM; | |
5748 | goto out; | |
5749 | } | |
5750 | ||
5751 | for (i = 0; i < num_stripes; i++) { | |
5752 | bbio->stripes[i].physical = | |
5753 | map->stripes[stripe_index].physical + | |
5754 | stripe_offset + stripe_nr * map->stripe_len; | |
5755 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5756 | ||
5757 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5758 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5759 | bbio->stripes[i].length = stripes_per_dev * | |
5760 | map->stripe_len; | |
5761 | ||
5762 | if (i / sub_stripes < remaining_stripes) | |
5763 | bbio->stripes[i].length += | |
5764 | map->stripe_len; | |
5765 | ||
5766 | /* | |
5767 | * Special for the first stripe and | |
5768 | * the last stripe: | |
5769 | * | |
5770 | * |-------|...|-------| | |
5771 | * |----------| | |
5772 | * off end_off | |
5773 | */ | |
5774 | if (i < sub_stripes) | |
5775 | bbio->stripes[i].length -= | |
5776 | stripe_offset; | |
5777 | ||
5778 | if (stripe_index >= last_stripe && | |
5779 | stripe_index <= (last_stripe + | |
5780 | sub_stripes - 1)) | |
5781 | bbio->stripes[i].length -= | |
5782 | stripe_end_offset; | |
5783 | ||
5784 | if (i == sub_stripes - 1) | |
5785 | stripe_offset = 0; | |
5786 | } else { | |
5787 | bbio->stripes[i].length = length; | |
5788 | } | |
5789 | ||
5790 | stripe_index++; | |
5791 | if (stripe_index == map->num_stripes) { | |
5792 | stripe_index = 0; | |
5793 | stripe_nr++; | |
5794 | } | |
5795 | } | |
5796 | ||
5797 | *bbio_ret = bbio; | |
5798 | bbio->map_type = map->type; | |
5799 | bbio->num_stripes = num_stripes; | |
5800 | out: | |
5801 | free_extent_map(em); | |
5802 | return ret; | |
5803 | } | |
5804 | ||
5ab56090 LB |
5805 | /* |
5806 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5807 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5808 | * left cursor can also be read from the target drive. | |
5809 | * | |
5810 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5811 | * array of stripes. | |
5812 | * For READ, it also needs to be supported using the same mirror number. | |
5813 | * | |
5814 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5815 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5816 | * case. | |
5817 | */ | |
5818 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5819 | u64 logical, u64 length, | |
5820 | u64 srcdev_devid, int *mirror_num, | |
5821 | u64 *physical) | |
5822 | { | |
5823 | struct btrfs_bio *bbio = NULL; | |
5824 | int num_stripes; | |
5825 | int index_srcdev = 0; | |
5826 | int found = 0; | |
5827 | u64 physical_of_found = 0; | |
5828 | int i; | |
5829 | int ret = 0; | |
5830 | ||
5831 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5832 | logical, &length, &bbio, 0, 0); | |
5833 | if (ret) { | |
5834 | ASSERT(bbio == NULL); | |
5835 | return ret; | |
5836 | } | |
5837 | ||
5838 | num_stripes = bbio->num_stripes; | |
5839 | if (*mirror_num > num_stripes) { | |
5840 | /* | |
5841 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5842 | * that means that the requested area is not left of the left | |
5843 | * cursor | |
5844 | */ | |
5845 | btrfs_put_bbio(bbio); | |
5846 | return -EIO; | |
5847 | } | |
5848 | ||
5849 | /* | |
5850 | * process the rest of the function using the mirror_num of the source | |
5851 | * drive. Therefore look it up first. At the end, patch the device | |
5852 | * pointer to the one of the target drive. | |
5853 | */ | |
5854 | for (i = 0; i < num_stripes; i++) { | |
5855 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5856 | continue; | |
5857 | ||
5858 | /* | |
5859 | * In case of DUP, in order to keep it simple, only add the | |
5860 | * mirror with the lowest physical address | |
5861 | */ | |
5862 | if (found && | |
5863 | physical_of_found <= bbio->stripes[i].physical) | |
5864 | continue; | |
5865 | ||
5866 | index_srcdev = i; | |
5867 | found = 1; | |
5868 | physical_of_found = bbio->stripes[i].physical; | |
5869 | } | |
5870 | ||
5871 | btrfs_put_bbio(bbio); | |
5872 | ||
5873 | ASSERT(found); | |
5874 | if (!found) | |
5875 | return -EIO; | |
5876 | ||
5877 | *mirror_num = index_srcdev + 1; | |
5878 | *physical = physical_of_found; | |
5879 | return ret; | |
5880 | } | |
5881 | ||
73c0f228 LB |
5882 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5883 | struct btrfs_bio **bbio_ret, | |
5884 | struct btrfs_dev_replace *dev_replace, | |
5885 | int *num_stripes_ret, int *max_errors_ret) | |
5886 | { | |
5887 | struct btrfs_bio *bbio = *bbio_ret; | |
5888 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5889 | int tgtdev_indexes = 0; | |
5890 | int num_stripes = *num_stripes_ret; | |
5891 | int max_errors = *max_errors_ret; | |
5892 | int i; | |
5893 | ||
5894 | if (op == BTRFS_MAP_WRITE) { | |
5895 | int index_where_to_add; | |
5896 | ||
5897 | /* | |
5898 | * duplicate the write operations while the dev replace | |
5899 | * procedure is running. Since the copying of the old disk to | |
5900 | * the new disk takes place at run time while the filesystem is | |
5901 | * mounted writable, the regular write operations to the old | |
5902 | * disk have to be duplicated to go to the new disk as well. | |
5903 | * | |
5904 | * Note that device->missing is handled by the caller, and that | |
5905 | * the write to the old disk is already set up in the stripes | |
5906 | * array. | |
5907 | */ | |
5908 | index_where_to_add = num_stripes; | |
5909 | for (i = 0; i < num_stripes; i++) { | |
5910 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5911 | /* write to new disk, too */ | |
5912 | struct btrfs_bio_stripe *new = | |
5913 | bbio->stripes + index_where_to_add; | |
5914 | struct btrfs_bio_stripe *old = | |
5915 | bbio->stripes + i; | |
5916 | ||
5917 | new->physical = old->physical; | |
5918 | new->length = old->length; | |
5919 | new->dev = dev_replace->tgtdev; | |
5920 | bbio->tgtdev_map[i] = index_where_to_add; | |
5921 | index_where_to_add++; | |
5922 | max_errors++; | |
5923 | tgtdev_indexes++; | |
5924 | } | |
5925 | } | |
5926 | num_stripes = index_where_to_add; | |
5927 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5928 | int index_srcdev = 0; | |
5929 | int found = 0; | |
5930 | u64 physical_of_found = 0; | |
5931 | ||
5932 | /* | |
5933 | * During the dev-replace procedure, the target drive can also | |
5934 | * be used to read data in case it is needed to repair a corrupt | |
5935 | * block elsewhere. This is possible if the requested area is | |
5936 | * left of the left cursor. In this area, the target drive is a | |
5937 | * full copy of the source drive. | |
5938 | */ | |
5939 | for (i = 0; i < num_stripes; i++) { | |
5940 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5941 | /* | |
5942 | * In case of DUP, in order to keep it simple, | |
5943 | * only add the mirror with the lowest physical | |
5944 | * address | |
5945 | */ | |
5946 | if (found && | |
5947 | physical_of_found <= | |
5948 | bbio->stripes[i].physical) | |
5949 | continue; | |
5950 | index_srcdev = i; | |
5951 | found = 1; | |
5952 | physical_of_found = bbio->stripes[i].physical; | |
5953 | } | |
5954 | } | |
5955 | if (found) { | |
5956 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5957 | bbio->stripes + num_stripes; | |
5958 | ||
5959 | tgtdev_stripe->physical = physical_of_found; | |
5960 | tgtdev_stripe->length = | |
5961 | bbio->stripes[index_srcdev].length; | |
5962 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5963 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5964 | ||
5965 | tgtdev_indexes++; | |
5966 | num_stripes++; | |
5967 | } | |
5968 | } | |
5969 | ||
5970 | *num_stripes_ret = num_stripes; | |
5971 | *max_errors_ret = max_errors; | |
5972 | bbio->num_tgtdevs = tgtdev_indexes; | |
5973 | *bbio_ret = bbio; | |
5974 | } | |
5975 | ||
2b19a1fe LB |
5976 | static bool need_full_stripe(enum btrfs_map_op op) |
5977 | { | |
5978 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5979 | } | |
5980 | ||
cf8cddd3 CH |
5981 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5982 | enum btrfs_map_op op, | |
f2d8d74d | 5983 | u64 logical, u64 *length, |
a1d3c478 | 5984 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5985 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5986 | { |
5987 | struct extent_map *em; | |
5988 | struct map_lookup *map; | |
0b86a832 | 5989 | u64 offset; |
593060d7 CM |
5990 | u64 stripe_offset; |
5991 | u64 stripe_nr; | |
53b381b3 | 5992 | u64 stripe_len; |
9d644a62 | 5993 | u32 stripe_index; |
cea9e445 | 5994 | int i; |
de11cc12 | 5995 | int ret = 0; |
f2d8d74d | 5996 | int num_stripes; |
a236aed1 | 5997 | int max_errors = 0; |
2c8cdd6e | 5998 | int tgtdev_indexes = 0; |
a1d3c478 | 5999 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6000 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6001 | int dev_replace_is_ongoing = 0; | |
6002 | int num_alloc_stripes; | |
ad6d620e SB |
6003 | int patch_the_first_stripe_for_dev_replace = 0; |
6004 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6005 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 6006 | |
0b3d4cd3 LB |
6007 | if (op == BTRFS_MAP_DISCARD) |
6008 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6009 | *length, bbio_ret); | |
6010 | ||
60ca842e | 6011 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
6012 | if (IS_ERR(em)) |
6013 | return PTR_ERR(em); | |
0b86a832 | 6014 | |
95617d69 | 6015 | map = em->map_lookup; |
0b86a832 | 6016 | offset = logical - em->start; |
593060d7 | 6017 | |
53b381b3 | 6018 | stripe_len = map->stripe_len; |
593060d7 CM |
6019 | stripe_nr = offset; |
6020 | /* | |
6021 | * stripe_nr counts the total number of stripes we have to stride | |
6022 | * to get to this block | |
6023 | */ | |
47c5713f | 6024 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 6025 | |
53b381b3 | 6026 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 6027 | if (offset < stripe_offset) { |
5d163e0e JM |
6028 | btrfs_crit(fs_info, |
6029 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
6030 | stripe_offset, offset, em->start, logical, |
6031 | stripe_len); | |
6032 | free_extent_map(em); | |
6033 | return -EINVAL; | |
6034 | } | |
593060d7 CM |
6035 | |
6036 | /* stripe_offset is the offset of this block in its stripe*/ | |
6037 | stripe_offset = offset - stripe_offset; | |
6038 | ||
53b381b3 | 6039 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 6040 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
6041 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
6042 | raid56_full_stripe_start = offset; | |
6043 | ||
6044 | /* allow a write of a full stripe, but make sure we don't | |
6045 | * allow straddling of stripes | |
6046 | */ | |
47c5713f DS |
6047 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
6048 | full_stripe_len); | |
53b381b3 DW |
6049 | raid56_full_stripe_start *= full_stripe_len; |
6050 | } | |
6051 | ||
0b3d4cd3 | 6052 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
6053 | u64 max_len; |
6054 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
6055 | For other RAID types and for RAID[56] reads, just allow a single | |
6056 | stripe (on a single disk). */ | |
ffe2d203 | 6057 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 6058 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
6059 | max_len = stripe_len * nr_data_stripes(map) - |
6060 | (offset - raid56_full_stripe_start); | |
6061 | } else { | |
6062 | /* we limit the length of each bio to what fits in a stripe */ | |
6063 | max_len = stripe_len - stripe_offset; | |
6064 | } | |
6065 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
6066 | } else { |
6067 | *length = em->len - offset; | |
6068 | } | |
f2d8d74d | 6069 | |
da12fe54 NB |
6070 | /* |
6071 | * This is for when we're called from btrfs_bio_fits_in_stripe and all | |
6072 | * it cares about is the length | |
6073 | */ | |
a1d3c478 | 6074 | if (!bbio_ret) |
cea9e445 CM |
6075 | goto out; |
6076 | ||
cb5583dd | 6077 | down_read(&dev_replace->rwsem); |
472262f3 | 6078 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6079 | /* |
6080 | * Hold the semaphore for read during the whole operation, write is | |
6081 | * requested at commit time but must wait. | |
6082 | */ | |
472262f3 | 6083 | if (!dev_replace_is_ongoing) |
cb5583dd | 6084 | up_read(&dev_replace->rwsem); |
472262f3 | 6085 | |
ad6d620e | 6086 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6087 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6088 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6089 | dev_replace->srcdev->devid, | |
6090 | &mirror_num, | |
6091 | &physical_to_patch_in_first_stripe); | |
6092 | if (ret) | |
ad6d620e | 6093 | goto out; |
5ab56090 LB |
6094 | else |
6095 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6096 | } else if (mirror_num > map->num_stripes) { |
6097 | mirror_num = 0; | |
6098 | } | |
6099 | ||
f2d8d74d | 6100 | num_stripes = 1; |
cea9e445 | 6101 | stripe_index = 0; |
fce3bb9a | 6102 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6103 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6104 | &stripe_index); | |
de483734 | 6105 | if (!need_full_stripe(op)) |
28e1cc7d | 6106 | mirror_num = 1; |
fce3bb9a | 6107 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 6108 | if (need_full_stripe(op)) |
f2d8d74d | 6109 | num_stripes = map->num_stripes; |
2fff734f | 6110 | else if (mirror_num) |
f188591e | 6111 | stripe_index = mirror_num - 1; |
dfe25020 | 6112 | else { |
30d9861f | 6113 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6114 | dev_replace_is_ongoing); |
a1d3c478 | 6115 | mirror_num = stripe_index + 1; |
dfe25020 | 6116 | } |
2fff734f | 6117 | |
611f0e00 | 6118 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6119 | if (need_full_stripe(op)) { |
f2d8d74d | 6120 | num_stripes = map->num_stripes; |
a1d3c478 | 6121 | } else if (mirror_num) { |
f188591e | 6122 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6123 | } else { |
6124 | mirror_num = 1; | |
6125 | } | |
2fff734f | 6126 | |
321aecc6 | 6127 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6128 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6129 | |
47c5713f | 6130 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6131 | stripe_index *= map->sub_stripes; |
6132 | ||
de483734 | 6133 | if (need_full_stripe(op)) |
f2d8d74d | 6134 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6135 | else if (mirror_num) |
6136 | stripe_index += mirror_num - 1; | |
dfe25020 | 6137 | else { |
3e74317a | 6138 | int old_stripe_index = stripe_index; |
30d9861f SB |
6139 | stripe_index = find_live_mirror(fs_info, map, |
6140 | stripe_index, | |
30d9861f | 6141 | dev_replace_is_ongoing); |
3e74317a | 6142 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6143 | } |
53b381b3 | 6144 | |
ffe2d203 | 6145 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6146 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6147 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6148 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 6149 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
6150 | |
6151 | /* RAID[56] write or recovery. Return all stripes */ | |
6152 | num_stripes = map->num_stripes; | |
6153 | max_errors = nr_parity_stripes(map); | |
6154 | ||
53b381b3 DW |
6155 | *length = map->stripe_len; |
6156 | stripe_index = 0; | |
6157 | stripe_offset = 0; | |
6158 | } else { | |
6159 | /* | |
6160 | * Mirror #0 or #1 means the original data block. | |
6161 | * Mirror #2 is RAID5 parity block. | |
6162 | * Mirror #3 is RAID6 Q block. | |
6163 | */ | |
47c5713f DS |
6164 | stripe_nr = div_u64_rem(stripe_nr, |
6165 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
6166 | if (mirror_num > 1) |
6167 | stripe_index = nr_data_stripes(map) + | |
6168 | mirror_num - 2; | |
6169 | ||
6170 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6171 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6172 | &stripe_index); | |
de483734 | 6173 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6174 | mirror_num = 1; |
53b381b3 | 6175 | } |
8790d502 CM |
6176 | } else { |
6177 | /* | |
47c5713f DS |
6178 | * after this, stripe_nr is the number of stripes on this |
6179 | * device we have to walk to find the data, and stripe_index is | |
6180 | * the number of our device in the stripe array | |
8790d502 | 6181 | */ |
47c5713f DS |
6182 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6183 | &stripe_index); | |
a1d3c478 | 6184 | mirror_num = stripe_index + 1; |
8790d502 | 6185 | } |
e042d1ec | 6186 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6187 | btrfs_crit(fs_info, |
6188 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6189 | stripe_index, map->num_stripes); |
6190 | ret = -EINVAL; | |
6191 | goto out; | |
6192 | } | |
cea9e445 | 6193 | |
472262f3 | 6194 | num_alloc_stripes = num_stripes; |
6fad823f | 6195 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6196 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6197 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6198 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6199 | num_alloc_stripes++; |
2c8cdd6e | 6200 | tgtdev_indexes = num_stripes; |
ad6d620e | 6201 | } |
2c8cdd6e | 6202 | |
6e9606d2 | 6203 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6204 | if (!bbio) { |
6205 | ret = -ENOMEM; | |
6206 | goto out; | |
6207 | } | |
6fad823f | 6208 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6209 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6210 | |
8e5cfb55 | 6211 | /* build raid_map */ |
2b19a1fe LB |
6212 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6213 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6214 | u64 tmp; |
9d644a62 | 6215 | unsigned rot; |
8e5cfb55 ZL |
6216 | |
6217 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6218 | sizeof(struct btrfs_bio_stripe) * | |
6219 | num_alloc_stripes + | |
6220 | sizeof(int) * tgtdev_indexes); | |
6221 | ||
6222 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6223 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6224 | |
6225 | /* Fill in the logical address of each stripe */ | |
6226 | tmp = stripe_nr * nr_data_stripes(map); | |
6227 | for (i = 0; i < nr_data_stripes(map); i++) | |
6228 | bbio->raid_map[(i+rot) % num_stripes] = | |
6229 | em->start + (tmp + i) * map->stripe_len; | |
6230 | ||
6231 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6232 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6233 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6234 | RAID6_Q_STRIPE; | |
6235 | } | |
6236 | ||
b89203f7 | 6237 | |
0b3d4cd3 LB |
6238 | for (i = 0; i < num_stripes; i++) { |
6239 | bbio->stripes[i].physical = | |
6240 | map->stripes[stripe_index].physical + | |
6241 | stripe_offset + | |
6242 | stripe_nr * map->stripe_len; | |
6243 | bbio->stripes[i].dev = | |
6244 | map->stripes[stripe_index].dev; | |
6245 | stripe_index++; | |
593060d7 | 6246 | } |
de11cc12 | 6247 | |
2b19a1fe | 6248 | if (need_full_stripe(op)) |
d20983b4 | 6249 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6250 | |
8e5cfb55 ZL |
6251 | if (bbio->raid_map) |
6252 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6253 | |
73c0f228 | 6254 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6255 | need_full_stripe(op)) { |
73c0f228 LB |
6256 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6257 | &max_errors); | |
472262f3 SB |
6258 | } |
6259 | ||
de11cc12 | 6260 | *bbio_ret = bbio; |
10f11900 | 6261 | bbio->map_type = map->type; |
de11cc12 LZ |
6262 | bbio->num_stripes = num_stripes; |
6263 | bbio->max_errors = max_errors; | |
6264 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6265 | |
6266 | /* | |
6267 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6268 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6269 | * available as a mirror | |
6270 | */ | |
6271 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6272 | WARN_ON(num_stripes > 1); | |
6273 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6274 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6275 | bbio->mirror_num = map->num_stripes + 1; | |
6276 | } | |
cea9e445 | 6277 | out: |
73beece9 | 6278 | if (dev_replace_is_ongoing) { |
53176dde DS |
6279 | lockdep_assert_held(&dev_replace->rwsem); |
6280 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6281 | up_read(&dev_replace->rwsem); |
73beece9 | 6282 | } |
0b86a832 | 6283 | free_extent_map(em); |
de11cc12 | 6284 | return ret; |
0b86a832 CM |
6285 | } |
6286 | ||
cf8cddd3 | 6287 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6288 | u64 logical, u64 *length, |
a1d3c478 | 6289 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6290 | { |
b3d3fa51 | 6291 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6292 | mirror_num, 0); |
f2d8d74d CM |
6293 | } |
6294 | ||
af8e2d1d | 6295 | /* For Scrub/replace */ |
cf8cddd3 | 6296 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6297 | u64 logical, u64 *length, |
825ad4c9 | 6298 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6299 | { |
825ad4c9 | 6300 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6301 | } |
6302 | ||
63a9c7b9 NB |
6303 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6304 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6305 | { |
a512bbf8 YZ |
6306 | struct extent_map *em; |
6307 | struct map_lookup *map; | |
6308 | u64 *buf; | |
6309 | u64 bytenr; | |
6310 | u64 length; | |
6311 | u64 stripe_nr; | |
53b381b3 | 6312 | u64 rmap_len; |
a512bbf8 YZ |
6313 | int i, j, nr = 0; |
6314 | ||
60ca842e | 6315 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6316 | if (IS_ERR(em)) |
835d974f | 6317 | return -EIO; |
835d974f | 6318 | |
95617d69 | 6319 | map = em->map_lookup; |
a512bbf8 | 6320 | length = em->len; |
53b381b3 DW |
6321 | rmap_len = map->stripe_len; |
6322 | ||
a512bbf8 | 6323 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6324 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6325 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6326 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6327 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6328 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6329 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6330 | } | |
a512bbf8 | 6331 | |
31e818fe | 6332 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6333 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6334 | |
6335 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6336 | if (map->stripes[i].physical > physical || |
6337 | map->stripes[i].physical + length <= physical) | |
6338 | continue; | |
6339 | ||
6340 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6341 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6342 | |
6343 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6344 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6345 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6346 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6347 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6348 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6349 | * Alternatively, just use rmap_len below instead of | |
6350 | * map->stripe_len */ | |
6351 | ||
6352 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6353 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6354 | for (j = 0; j < nr; j++) { |
6355 | if (buf[j] == bytenr) | |
6356 | break; | |
6357 | } | |
934d375b CM |
6358 | if (j == nr) { |
6359 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6360 | buf[nr++] = bytenr; |
934d375b | 6361 | } |
a512bbf8 YZ |
6362 | } |
6363 | ||
a512bbf8 YZ |
6364 | *logical = buf; |
6365 | *naddrs = nr; | |
53b381b3 | 6366 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6367 | |
6368 | free_extent_map(em); | |
6369 | return 0; | |
f2d8d74d CM |
6370 | } |
6371 | ||
4246a0b6 | 6372 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6373 | { |
326e1dbb MS |
6374 | bio->bi_private = bbio->private; |
6375 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6376 | bio_endio(bio); |
326e1dbb | 6377 | |
6e9606d2 | 6378 | btrfs_put_bbio(bbio); |
8408c716 MX |
6379 | } |
6380 | ||
4246a0b6 | 6381 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6382 | { |
9be3395b | 6383 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6384 | int is_orig_bio = 0; |
8790d502 | 6385 | |
4e4cbee9 | 6386 | if (bio->bi_status) { |
a1d3c478 | 6387 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6388 | if (bio->bi_status == BLK_STS_IOERR || |
6389 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6390 | unsigned int stripe_index = |
9be3395b | 6391 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6392 | struct btrfs_device *dev; |
442a4f63 SB |
6393 | |
6394 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6395 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6396 | if (dev->bdev) { |
37226b21 | 6397 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6398 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa SB |
6399 | BTRFS_DEV_STAT_WRITE_ERRS); |
6400 | else | |
1cb34c8e | 6401 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6402 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6403 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6404 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6405 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6406 | } |
442a4f63 SB |
6407 | } |
6408 | } | |
8790d502 | 6409 | |
a1d3c478 | 6410 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6411 | is_orig_bio = 1; |
6412 | ||
c404e0dc MX |
6413 | btrfs_bio_counter_dec(bbio->fs_info); |
6414 | ||
a1d3c478 | 6415 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6416 | if (!is_orig_bio) { |
6417 | bio_put(bio); | |
a1d3c478 | 6418 | bio = bbio->orig_bio; |
7d2b4daa | 6419 | } |
c7b22bb1 | 6420 | |
9be3395b | 6421 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6422 | /* only send an error to the higher layers if it is |
53b381b3 | 6423 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6424 | */ |
a1d3c478 | 6425 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6426 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6427 | } else { |
1259ab75 CM |
6428 | /* |
6429 | * this bio is actually up to date, we didn't | |
6430 | * go over the max number of errors | |
6431 | */ | |
2dbe0c77 | 6432 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6433 | } |
c55f1396 | 6434 | |
4246a0b6 | 6435 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6436 | } else if (!is_orig_bio) { |
8790d502 CM |
6437 | bio_put(bio); |
6438 | } | |
8790d502 CM |
6439 | } |
6440 | ||
8b712842 CM |
6441 | /* |
6442 | * see run_scheduled_bios for a description of why bios are collected for | |
6443 | * async submit. | |
6444 | * | |
6445 | * This will add one bio to the pending list for a device and make sure | |
6446 | * the work struct is scheduled. | |
6447 | */ | |
2ff7e61e | 6448 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6449 | struct bio *bio) |
8b712842 | 6450 | { |
0b246afa | 6451 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6452 | int should_queue = 1; |
ffbd517d | 6453 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6454 | |
6455 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6456 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6457 | btrfsic_submit_bio(bio); |
143bede5 | 6458 | return; |
8b712842 CM |
6459 | } |
6460 | ||
492bb6de | 6461 | WARN_ON(bio->bi_next); |
8b712842 | 6462 | bio->bi_next = NULL; |
8b712842 CM |
6463 | |
6464 | spin_lock(&device->io_lock); | |
67f055c7 | 6465 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6466 | pending_bios = &device->pending_sync_bios; |
6467 | else | |
6468 | pending_bios = &device->pending_bios; | |
8b712842 | 6469 | |
ffbd517d CM |
6470 | if (pending_bios->tail) |
6471 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6472 | |
ffbd517d CM |
6473 | pending_bios->tail = bio; |
6474 | if (!pending_bios->head) | |
6475 | pending_bios->head = bio; | |
8b712842 CM |
6476 | if (device->running_pending) |
6477 | should_queue = 0; | |
6478 | ||
6479 | spin_unlock(&device->io_lock); | |
6480 | ||
6481 | if (should_queue) | |
0b246afa | 6482 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6483 | } |
6484 | ||
2ff7e61e JM |
6485 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6486 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6487 | { |
6488 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6489 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6490 | |
6491 | bio->bi_private = bbio; | |
9be3395b | 6492 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6493 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6494 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6495 | btrfs_debug_in_rcu(fs_info, |
6496 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6497 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6498 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6499 | bio->bi_iter.bi_size); | |
74d46992 | 6500 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6501 | |
2ff7e61e | 6502 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6503 | |
de1ee92a | 6504 | if (async) |
2ff7e61e | 6505 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6506 | else |
4e49ea4a | 6507 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6508 | } |
6509 | ||
de1ee92a JB |
6510 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6511 | { | |
6512 | atomic_inc(&bbio->error); | |
6513 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6514 | /* Should be the original bio. */ |
8408c716 MX |
6515 | WARN_ON(bio != bbio->orig_bio); |
6516 | ||
9be3395b | 6517 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6518 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6519 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6520 | bio->bi_status = BLK_STS_IOERR; | |
6521 | else | |
6522 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6523 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6524 | } |
6525 | } | |
6526 | ||
58efbc9f OS |
6527 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6528 | int mirror_num, int async_submit) | |
0b86a832 | 6529 | { |
0b86a832 | 6530 | struct btrfs_device *dev; |
8790d502 | 6531 | struct bio *first_bio = bio; |
4f024f37 | 6532 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6533 | u64 length = 0; |
6534 | u64 map_length; | |
0b86a832 | 6535 | int ret; |
08da757d ZL |
6536 | int dev_nr; |
6537 | int total_devs; | |
a1d3c478 | 6538 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6539 | |
4f024f37 | 6540 | length = bio->bi_iter.bi_size; |
0b86a832 | 6541 | map_length = length; |
cea9e445 | 6542 | |
0b246afa | 6543 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6544 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6545 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6546 | if (ret) { |
0b246afa | 6547 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6548 | return errno_to_blk_status(ret); |
c404e0dc | 6549 | } |
cea9e445 | 6550 | |
a1d3c478 | 6551 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6552 | bbio->orig_bio = first_bio; |
6553 | bbio->private = first_bio->bi_private; | |
6554 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6555 | bbio->fs_info = fs_info; |
53b381b3 DW |
6556 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6557 | ||
ad1ba2a0 | 6558 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6559 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6560 | /* In this case, map_length has been set to the length of |
6561 | a single stripe; not the whole write */ | |
37226b21 | 6562 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6563 | ret = raid56_parity_write(fs_info, bio, bbio, |
6564 | map_length); | |
53b381b3 | 6565 | } else { |
2ff7e61e JM |
6566 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6567 | map_length, mirror_num, 1); | |
53b381b3 | 6568 | } |
4245215d | 6569 | |
0b246afa | 6570 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6571 | return errno_to_blk_status(ret); |
53b381b3 DW |
6572 | } |
6573 | ||
cea9e445 | 6574 | if (map_length < length) { |
0b246afa | 6575 | btrfs_crit(fs_info, |
5d163e0e JM |
6576 | "mapping failed logical %llu bio len %llu len %llu", |
6577 | logical, length, map_length); | |
cea9e445 CM |
6578 | BUG(); |
6579 | } | |
a1d3c478 | 6580 | |
08da757d | 6581 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6582 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6583 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6584 | &dev->dev_state) || | |
ebbede42 AJ |
6585 | (bio_op(first_bio) == REQ_OP_WRITE && |
6586 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6587 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6588 | continue; |
6589 | } | |
6590 | ||
3aa8e074 | 6591 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6592 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6593 | else |
a1d3c478 | 6594 | bio = first_bio; |
de1ee92a | 6595 | |
2ff7e61e JM |
6596 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6597 | dev_nr, async_submit); | |
8790d502 | 6598 | } |
0b246afa | 6599 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6600 | return BLK_STS_OK; |
0b86a832 CM |
6601 | } |
6602 | ||
09ba3bc9 AJ |
6603 | /* |
6604 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6605 | * return NULL. | |
6606 | * | |
6607 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6608 | * only devid is used. | |
6609 | * | |
6610 | * If @seed is true, traverse through the seed devices. | |
6611 | */ | |
e4319cd9 | 6612 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6613 | u64 devid, u8 *uuid, u8 *fsid, |
6614 | bool seed) | |
0b86a832 | 6615 | { |
2b82032c | 6616 | struct btrfs_device *device; |
2b82032c | 6617 | |
e4319cd9 | 6618 | while (fs_devices) { |
2b82032c | 6619 | if (!fsid || |
e4319cd9 | 6620 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6621 | list_for_each_entry(device, &fs_devices->devices, |
6622 | dev_list) { | |
6623 | if (device->devid == devid && | |
6624 | (!uuid || memcmp(device->uuid, uuid, | |
6625 | BTRFS_UUID_SIZE) == 0)) | |
6626 | return device; | |
6627 | } | |
2b82032c | 6628 | } |
09ba3bc9 AJ |
6629 | if (seed) |
6630 | fs_devices = fs_devices->seed; | |
6631 | else | |
6632 | return NULL; | |
2b82032c YZ |
6633 | } |
6634 | return NULL; | |
0b86a832 CM |
6635 | } |
6636 | ||
2ff7e61e | 6637 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6638 | u64 devid, u8 *dev_uuid) |
6639 | { | |
6640 | struct btrfs_device *device; | |
dfe25020 | 6641 | |
12bd2fc0 ID |
6642 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6643 | if (IS_ERR(device)) | |
adfb69af | 6644 | return device; |
12bd2fc0 ID |
6645 | |
6646 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6647 | device->fs_devices = fs_devices; |
dfe25020 | 6648 | fs_devices->num_devices++; |
12bd2fc0 | 6649 | |
e6e674bd | 6650 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6651 | fs_devices->missing_devices++; |
12bd2fc0 | 6652 | |
dfe25020 CM |
6653 | return device; |
6654 | } | |
6655 | ||
12bd2fc0 ID |
6656 | /** |
6657 | * btrfs_alloc_device - allocate struct btrfs_device | |
6658 | * @fs_info: used only for generating a new devid, can be NULL if | |
6659 | * devid is provided (i.e. @devid != NULL). | |
6660 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6661 | * is generated. | |
6662 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6663 | * is generated. | |
6664 | * | |
6665 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6666 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6667 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6668 | */ |
6669 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6670 | const u64 *devid, | |
6671 | const u8 *uuid) | |
6672 | { | |
6673 | struct btrfs_device *dev; | |
6674 | u64 tmp; | |
6675 | ||
fae7f21c | 6676 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6677 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6678 | |
6679 | dev = __alloc_device(); | |
6680 | if (IS_ERR(dev)) | |
6681 | return dev; | |
6682 | ||
6683 | if (devid) | |
6684 | tmp = *devid; | |
6685 | else { | |
6686 | int ret; | |
6687 | ||
6688 | ret = find_next_devid(fs_info, &tmp); | |
6689 | if (ret) { | |
a425f9d4 | 6690 | btrfs_free_device(dev); |
12bd2fc0 ID |
6691 | return ERR_PTR(ret); |
6692 | } | |
6693 | } | |
6694 | dev->devid = tmp; | |
6695 | ||
6696 | if (uuid) | |
6697 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6698 | else | |
6699 | generate_random_uuid(dev->uuid); | |
6700 | ||
9e0af237 LB |
6701 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6702 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6703 | |
6704 | return dev; | |
6705 | } | |
6706 | ||
e06cd3dd | 6707 | /* Return -EIO if any error, otherwise return 0. */ |
2ff7e61e | 6708 | static int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info, |
e06cd3dd LB |
6709 | struct extent_buffer *leaf, |
6710 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6711 | { |
0b86a832 | 6712 | u64 length; |
f04b772b | 6713 | u64 stripe_len; |
e06cd3dd LB |
6714 | u16 num_stripes; |
6715 | u16 sub_stripes; | |
6716 | u64 type; | |
315409b0 GJ |
6717 | u64 features; |
6718 | bool mixed = false; | |
0b86a832 | 6719 | |
e17cade2 | 6720 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6721 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6722 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6723 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6724 | type = btrfs_chunk_type(leaf, chunk); | |
6725 | ||
f04b772b | 6726 | if (!num_stripes) { |
0b246afa | 6727 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6728 | num_stripes); |
6729 | return -EIO; | |
6730 | } | |
0b246afa JM |
6731 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6732 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6733 | return -EIO; |
6734 | } | |
0b246afa JM |
6735 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6736 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6737 | btrfs_chunk_sector_size(leaf, chunk)); |
6738 | return -EIO; | |
6739 | } | |
0b246afa JM |
6740 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6741 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6742 | return -EIO; |
6743 | } | |
3d8da678 | 6744 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6745 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6746 | stripe_len); |
6747 | return -EIO; | |
6748 | } | |
6749 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6750 | type) { |
0b246afa | 6751 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6752 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6753 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6754 | btrfs_chunk_type(leaf, chunk)); | |
6755 | return -EIO; | |
6756 | } | |
315409b0 GJ |
6757 | |
6758 | if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) { | |
6759 | btrfs_err(fs_info, "missing chunk type flag: 0x%llx", type); | |
6760 | return -EIO; | |
6761 | } | |
6762 | ||
6763 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
6764 | (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) { | |
6765 | btrfs_err(fs_info, | |
6766 | "system chunk with data or metadata type: 0x%llx", type); | |
6767 | return -EIO; | |
6768 | } | |
6769 | ||
6770 | features = btrfs_super_incompat_flags(fs_info->super_copy); | |
6771 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
6772 | mixed = true; | |
6773 | ||
6774 | if (!mixed) { | |
6775 | if ((type & BTRFS_BLOCK_GROUP_METADATA) && | |
6776 | (type & BTRFS_BLOCK_GROUP_DATA)) { | |
6777 | btrfs_err(fs_info, | |
6778 | "mixed chunk type in non-mixed mode: 0x%llx", type); | |
6779 | return -EIO; | |
6780 | } | |
6781 | } | |
6782 | ||
e06cd3dd LB |
6783 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6784 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6785 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6786 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6787 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6788 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6789 | num_stripes != 1)) { | |
0b246afa | 6790 | btrfs_err(fs_info, |
e06cd3dd LB |
6791 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6792 | num_stripes, sub_stripes, | |
6793 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6794 | return -EIO; | |
6795 | } | |
6796 | ||
6797 | return 0; | |
6798 | } | |
6799 | ||
5a2b8e60 | 6800 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6801 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6802 | { |
2b902dfc AJ |
6803 | if (error) |
6804 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6805 | devid, uuid); | |
6806 | else | |
6807 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6808 | devid, uuid); | |
5a2b8e60 AJ |
6809 | } |
6810 | ||
2ff7e61e | 6811 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6812 | struct extent_buffer *leaf, |
6813 | struct btrfs_chunk *chunk) | |
6814 | { | |
0b246afa | 6815 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6816 | struct map_lookup *map; |
6817 | struct extent_map *em; | |
6818 | u64 logical; | |
6819 | u64 length; | |
e06cd3dd LB |
6820 | u64 devid; |
6821 | u8 uuid[BTRFS_UUID_SIZE]; | |
6822 | int num_stripes; | |
6823 | int ret; | |
6824 | int i; | |
6825 | ||
6826 | logical = key->offset; | |
6827 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6828 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6829 | ||
2ff7e61e | 6830 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); |
e06cd3dd LB |
6831 | if (ret) |
6832 | return ret; | |
a061fc8d | 6833 | |
890871be | 6834 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6835 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6836 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6837 | |
6838 | /* already mapped? */ | |
6839 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6840 | free_extent_map(em); | |
0b86a832 CM |
6841 | return 0; |
6842 | } else if (em) { | |
6843 | free_extent_map(em); | |
6844 | } | |
0b86a832 | 6845 | |
172ddd60 | 6846 | em = alloc_extent_map(); |
0b86a832 CM |
6847 | if (!em) |
6848 | return -ENOMEM; | |
593060d7 | 6849 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6850 | if (!map) { |
6851 | free_extent_map(em); | |
6852 | return -ENOMEM; | |
6853 | } | |
6854 | ||
298a8f9c | 6855 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6856 | em->map_lookup = map; |
0b86a832 CM |
6857 | em->start = logical; |
6858 | em->len = length; | |
70c8a91c | 6859 | em->orig_start = 0; |
0b86a832 | 6860 | em->block_start = 0; |
c8b97818 | 6861 | em->block_len = em->len; |
0b86a832 | 6862 | |
593060d7 CM |
6863 | map->num_stripes = num_stripes; |
6864 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6865 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6866 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6867 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6868 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6869 | map->verified_stripes = 0; |
593060d7 CM |
6870 | for (i = 0; i < num_stripes; i++) { |
6871 | map->stripes[i].physical = | |
6872 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6873 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6874 | read_extent_buffer(leaf, uuid, (unsigned long) |
6875 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6876 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6877 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6878 | devid, uuid, NULL, true); |
3cdde224 | 6879 | if (!map->stripes[i].dev && |
0b246afa | 6880 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6881 | free_extent_map(em); |
2b902dfc | 6882 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6883 | return -ENOENT; |
593060d7 | 6884 | } |
dfe25020 CM |
6885 | if (!map->stripes[i].dev) { |
6886 | map->stripes[i].dev = | |
2ff7e61e JM |
6887 | add_missing_dev(fs_info->fs_devices, devid, |
6888 | uuid); | |
adfb69af | 6889 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6890 | free_extent_map(em); |
adfb69af AJ |
6891 | btrfs_err(fs_info, |
6892 | "failed to init missing dev %llu: %ld", | |
6893 | devid, PTR_ERR(map->stripes[i].dev)); | |
6894 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6895 | } |
2b902dfc | 6896 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6897 | } |
e12c9621 AJ |
6898 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6899 | &(map->stripes[i].dev->dev_state)); | |
6900 | ||
0b86a832 CM |
6901 | } |
6902 | ||
890871be | 6903 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6904 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6905 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6906 | if (ret < 0) { |
6907 | btrfs_err(fs_info, | |
6908 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6909 | em->start, em->len, ret); | |
6910 | } | |
0b86a832 CM |
6911 | free_extent_map(em); |
6912 | ||
64f64f43 | 6913 | return ret; |
0b86a832 CM |
6914 | } |
6915 | ||
143bede5 | 6916 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6917 | struct btrfs_dev_item *dev_item, |
6918 | struct btrfs_device *device) | |
6919 | { | |
6920 | unsigned long ptr; | |
0b86a832 CM |
6921 | |
6922 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6923 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6924 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6925 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6926 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6927 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6928 | device->type = btrfs_device_type(leaf, dev_item); |
6929 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6930 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6931 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6932 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6933 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6934 | |
410ba3a2 | 6935 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6936 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6937 | } |
6938 | ||
2ff7e61e | 6939 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6940 | u8 *fsid) |
2b82032c YZ |
6941 | { |
6942 | struct btrfs_fs_devices *fs_devices; | |
6943 | int ret; | |
6944 | ||
a32bf9a3 | 6945 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6946 | ASSERT(fsid); |
2b82032c | 6947 | |
0b246afa | 6948 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6949 | while (fs_devices) { |
44880fdc | 6950 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6951 | return fs_devices; |
6952 | ||
2b82032c YZ |
6953 | fs_devices = fs_devices->seed; |
6954 | } | |
6955 | ||
7239ff4b | 6956 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6957 | if (!fs_devices) { |
0b246afa | 6958 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6959 | return ERR_PTR(-ENOENT); |
6960 | ||
7239ff4b | 6961 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6962 | if (IS_ERR(fs_devices)) |
6963 | return fs_devices; | |
6964 | ||
6965 | fs_devices->seeding = 1; | |
6966 | fs_devices->opened = 1; | |
6967 | return fs_devices; | |
2b82032c | 6968 | } |
e4404d6e YZ |
6969 | |
6970 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6971 | if (IS_ERR(fs_devices)) |
6972 | return fs_devices; | |
2b82032c | 6973 | |
897fb573 | 6974 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6975 | if (ret) { |
6976 | free_fs_devices(fs_devices); | |
5f375835 | 6977 | fs_devices = ERR_PTR(ret); |
2b82032c | 6978 | goto out; |
48d28232 | 6979 | } |
2b82032c YZ |
6980 | |
6981 | if (!fs_devices->seeding) { | |
0226e0eb | 6982 | close_fs_devices(fs_devices); |
e4404d6e | 6983 | free_fs_devices(fs_devices); |
5f375835 | 6984 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6985 | goto out; |
6986 | } | |
6987 | ||
0b246afa JM |
6988 | fs_devices->seed = fs_info->fs_devices->seed; |
6989 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6990 | out: |
5f375835 | 6991 | return fs_devices; |
2b82032c YZ |
6992 | } |
6993 | ||
2ff7e61e | 6994 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6995 | struct extent_buffer *leaf, |
6996 | struct btrfs_dev_item *dev_item) | |
6997 | { | |
0b246afa | 6998 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6999 | struct btrfs_device *device; |
7000 | u64 devid; | |
7001 | int ret; | |
44880fdc | 7002 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
7003 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
7004 | ||
0b86a832 | 7005 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 7006 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 7007 | BTRFS_UUID_SIZE); |
1473b24e | 7008 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 7009 | BTRFS_FSID_SIZE); |
2b82032c | 7010 | |
de37aa51 | 7011 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 7012 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
7013 | if (IS_ERR(fs_devices)) |
7014 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
7015 | } |
7016 | ||
e4319cd9 | 7017 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 7018 | fs_uuid, true); |
5f375835 | 7019 | if (!device) { |
c5502451 | 7020 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
7021 | btrfs_report_missing_device(fs_info, devid, |
7022 | dev_uuid, true); | |
45dbdbc9 | 7023 | return -ENOENT; |
c5502451 | 7024 | } |
2b82032c | 7025 | |
2ff7e61e | 7026 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
7027 | if (IS_ERR(device)) { |
7028 | btrfs_err(fs_info, | |
7029 | "failed to add missing dev %llu: %ld", | |
7030 | devid, PTR_ERR(device)); | |
7031 | return PTR_ERR(device); | |
7032 | } | |
2b902dfc | 7033 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 7034 | } else { |
c5502451 | 7035 | if (!device->bdev) { |
2b902dfc AJ |
7036 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
7037 | btrfs_report_missing_device(fs_info, | |
7038 | devid, dev_uuid, true); | |
45dbdbc9 | 7039 | return -ENOENT; |
2b902dfc AJ |
7040 | } |
7041 | btrfs_report_missing_device(fs_info, devid, | |
7042 | dev_uuid, false); | |
c5502451 | 7043 | } |
5f375835 | 7044 | |
e6e674bd AJ |
7045 | if (!device->bdev && |
7046 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7047 | /* |
7048 | * this happens when a device that was properly setup | |
7049 | * in the device info lists suddenly goes bad. | |
7050 | * device->bdev is NULL, and so we have to set | |
7051 | * device->missing to one here | |
7052 | */ | |
5f375835 | 7053 | device->fs_devices->missing_devices++; |
e6e674bd | 7054 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7055 | } |
5f375835 MX |
7056 | |
7057 | /* Move the device to its own fs_devices */ | |
7058 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7059 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7060 | &device->dev_state)); | |
5f375835 MX |
7061 | |
7062 | list_move(&device->dev_list, &fs_devices->devices); | |
7063 | device->fs_devices->num_devices--; | |
7064 | fs_devices->num_devices++; | |
7065 | ||
7066 | device->fs_devices->missing_devices--; | |
7067 | fs_devices->missing_devices++; | |
7068 | ||
7069 | device->fs_devices = fs_devices; | |
7070 | } | |
2b82032c YZ |
7071 | } |
7072 | ||
0b246afa | 7073 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7074 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7075 | if (device->generation != |
7076 | btrfs_device_generation(leaf, dev_item)) | |
7077 | return -EINVAL; | |
6324fbf3 | 7078 | } |
0b86a832 CM |
7079 | |
7080 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 7081 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7082 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7083 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7084 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7085 | atomic64_add(device->total_bytes - device->bytes_used, |
7086 | &fs_info->free_chunk_space); | |
2bf64758 | 7087 | } |
0b86a832 | 7088 | ret = 0; |
0b86a832 CM |
7089 | return ret; |
7090 | } | |
7091 | ||
6bccf3ab | 7092 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7093 | { |
6bccf3ab | 7094 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 7095 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7096 | struct extent_buffer *sb; |
0b86a832 | 7097 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7098 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7099 | u8 *array_ptr; |
7100 | unsigned long sb_array_offset; | |
84eed90f | 7101 | int ret = 0; |
0b86a832 CM |
7102 | u32 num_stripes; |
7103 | u32 array_size; | |
7104 | u32 len = 0; | |
1ffb22cf | 7105 | u32 cur_offset; |
e06cd3dd | 7106 | u64 type; |
84eed90f | 7107 | struct btrfs_key key; |
0b86a832 | 7108 | |
0b246afa | 7109 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7110 | /* |
7111 | * This will create extent buffer of nodesize, superblock size is | |
7112 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7113 | * overallocate but we can keep it as-is, only the first page is used. | |
7114 | */ | |
2ff7e61e | 7115 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7116 | if (IS_ERR(sb)) |
7117 | return PTR_ERR(sb); | |
4db8c528 | 7118 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7119 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7120 | /* |
01327610 | 7121 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7122 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7123 | * pages up-to-date when the page is larger: extent does not cover the |
7124 | * whole page and consequently check_page_uptodate does not find all | |
7125 | * the page's extents up-to-date (the hole beyond sb), | |
7126 | * write_extent_buffer then triggers a WARN_ON. | |
7127 | * | |
7128 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7129 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7130 | * to silence the warning eg. on PowerPC 64. | |
7131 | */ | |
09cbfeaf | 7132 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7133 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7134 | |
a061fc8d | 7135 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7136 | array_size = btrfs_super_sys_array_size(super_copy); |
7137 | ||
1ffb22cf DS |
7138 | array_ptr = super_copy->sys_chunk_array; |
7139 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7140 | cur_offset = 0; | |
0b86a832 | 7141 | |
1ffb22cf DS |
7142 | while (cur_offset < array_size) { |
7143 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7144 | len = sizeof(*disk_key); |
7145 | if (cur_offset + len > array_size) | |
7146 | goto out_short_read; | |
7147 | ||
0b86a832 CM |
7148 | btrfs_disk_key_to_cpu(&key, disk_key); |
7149 | ||
1ffb22cf DS |
7150 | array_ptr += len; |
7151 | sb_array_offset += len; | |
7152 | cur_offset += len; | |
0b86a832 | 7153 | |
0d81ba5d | 7154 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7155 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7156 | /* |
7157 | * At least one btrfs_chunk with one stripe must be | |
7158 | * present, exact stripe count check comes afterwards | |
7159 | */ | |
7160 | len = btrfs_chunk_item_size(1); | |
7161 | if (cur_offset + len > array_size) | |
7162 | goto out_short_read; | |
7163 | ||
7164 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7165 | if (!num_stripes) { |
ab8d0fc4 JM |
7166 | btrfs_err(fs_info, |
7167 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7168 | num_stripes, cur_offset); |
7169 | ret = -EIO; | |
7170 | break; | |
7171 | } | |
7172 | ||
e06cd3dd LB |
7173 | type = btrfs_chunk_type(sb, chunk); |
7174 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7175 | btrfs_err(fs_info, |
e06cd3dd LB |
7176 | "invalid chunk type %llu in sys_array at offset %u", |
7177 | type, cur_offset); | |
7178 | ret = -EIO; | |
7179 | break; | |
7180 | } | |
7181 | ||
e3540eab DS |
7182 | len = btrfs_chunk_item_size(num_stripes); |
7183 | if (cur_offset + len > array_size) | |
7184 | goto out_short_read; | |
7185 | ||
2ff7e61e | 7186 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
7187 | if (ret) |
7188 | break; | |
0b86a832 | 7189 | } else { |
ab8d0fc4 JM |
7190 | btrfs_err(fs_info, |
7191 | "unexpected item type %u in sys_array at offset %u", | |
7192 | (u32)key.type, cur_offset); | |
84eed90f CM |
7193 | ret = -EIO; |
7194 | break; | |
0b86a832 | 7195 | } |
1ffb22cf DS |
7196 | array_ptr += len; |
7197 | sb_array_offset += len; | |
7198 | cur_offset += len; | |
0b86a832 | 7199 | } |
d865177a | 7200 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7201 | free_extent_buffer_stale(sb); |
84eed90f | 7202 | return ret; |
e3540eab DS |
7203 | |
7204 | out_short_read: | |
ab8d0fc4 | 7205 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7206 | len, cur_offset); |
d865177a | 7207 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7208 | free_extent_buffer_stale(sb); |
e3540eab | 7209 | return -EIO; |
0b86a832 CM |
7210 | } |
7211 | ||
21634a19 QW |
7212 | /* |
7213 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7214 | * | |
6528b99d AJ |
7215 | * If the @failing_dev is specified, it's accounted as missing. |
7216 | * | |
21634a19 QW |
7217 | * Return true if all chunks meet the minimal RW mount requirements. |
7218 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7219 | */ | |
6528b99d AJ |
7220 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7221 | struct btrfs_device *failing_dev) | |
21634a19 QW |
7222 | { |
7223 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
7224 | struct extent_map *em; | |
7225 | u64 next_start = 0; | |
7226 | bool ret = true; | |
7227 | ||
7228 | read_lock(&map_tree->map_tree.lock); | |
7229 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
7230 | read_unlock(&map_tree->map_tree.lock); | |
7231 | /* No chunk at all? Return false anyway */ | |
7232 | if (!em) { | |
7233 | ret = false; | |
7234 | goto out; | |
7235 | } | |
7236 | while (em) { | |
7237 | struct map_lookup *map; | |
7238 | int missing = 0; | |
7239 | int max_tolerated; | |
7240 | int i; | |
7241 | ||
7242 | map = em->map_lookup; | |
7243 | max_tolerated = | |
7244 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7245 | map->type); | |
7246 | for (i = 0; i < map->num_stripes; i++) { | |
7247 | struct btrfs_device *dev = map->stripes[i].dev; | |
7248 | ||
e6e674bd AJ |
7249 | if (!dev || !dev->bdev || |
7250 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7251 | dev->last_flush_error) |
7252 | missing++; | |
6528b99d AJ |
7253 | else if (failing_dev && failing_dev == dev) |
7254 | missing++; | |
21634a19 QW |
7255 | } |
7256 | if (missing > max_tolerated) { | |
6528b99d AJ |
7257 | if (!failing_dev) |
7258 | btrfs_warn(fs_info, | |
52042d8e | 7259 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7260 | em->start, missing, max_tolerated); |
7261 | free_extent_map(em); | |
7262 | ret = false; | |
7263 | goto out; | |
7264 | } | |
7265 | next_start = extent_map_end(em); | |
7266 | free_extent_map(em); | |
7267 | ||
7268 | read_lock(&map_tree->map_tree.lock); | |
7269 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
7270 | (u64)(-1) - next_start); | |
7271 | read_unlock(&map_tree->map_tree.lock); | |
7272 | } | |
7273 | out: | |
7274 | return ret; | |
7275 | } | |
7276 | ||
5b4aacef | 7277 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7278 | { |
5b4aacef | 7279 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7280 | struct btrfs_path *path; |
7281 | struct extent_buffer *leaf; | |
7282 | struct btrfs_key key; | |
7283 | struct btrfs_key found_key; | |
7284 | int ret; | |
7285 | int slot; | |
99e3ecfc | 7286 | u64 total_dev = 0; |
0b86a832 | 7287 | |
0b86a832 CM |
7288 | path = btrfs_alloc_path(); |
7289 | if (!path) | |
7290 | return -ENOMEM; | |
7291 | ||
3dd0f7a3 AJ |
7292 | /* |
7293 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7294 | * otherwise we don't need it. | |
7295 | */ | |
b367e47f | 7296 | mutex_lock(&uuid_mutex); |
34441361 | 7297 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7298 | |
395927a9 FDBM |
7299 | /* |
7300 | * Read all device items, and then all the chunk items. All | |
7301 | * device items are found before any chunk item (their object id | |
7302 | * is smaller than the lowest possible object id for a chunk | |
7303 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7304 | */ |
7305 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7306 | key.offset = 0; | |
7307 | key.type = 0; | |
0b86a832 | 7308 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7309 | if (ret < 0) |
7310 | goto error; | |
d397712b | 7311 | while (1) { |
0b86a832 CM |
7312 | leaf = path->nodes[0]; |
7313 | slot = path->slots[0]; | |
7314 | if (slot >= btrfs_header_nritems(leaf)) { | |
7315 | ret = btrfs_next_leaf(root, path); | |
7316 | if (ret == 0) | |
7317 | continue; | |
7318 | if (ret < 0) | |
7319 | goto error; | |
7320 | break; | |
7321 | } | |
7322 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7323 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7324 | struct btrfs_dev_item *dev_item; | |
7325 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7326 | struct btrfs_dev_item); |
2ff7e61e | 7327 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7328 | if (ret) |
7329 | goto error; | |
99e3ecfc | 7330 | total_dev++; |
0b86a832 CM |
7331 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7332 | struct btrfs_chunk *chunk; | |
7333 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7334 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7335 | if (ret) |
7336 | goto error; | |
0b86a832 CM |
7337 | } |
7338 | path->slots[0]++; | |
7339 | } | |
99e3ecfc LB |
7340 | |
7341 | /* | |
7342 | * After loading chunk tree, we've got all device information, | |
7343 | * do another round of validation checks. | |
7344 | */ | |
0b246afa JM |
7345 | if (total_dev != fs_info->fs_devices->total_devices) { |
7346 | btrfs_err(fs_info, | |
99e3ecfc | 7347 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7348 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7349 | total_dev); |
7350 | ret = -EINVAL; | |
7351 | goto error; | |
7352 | } | |
0b246afa JM |
7353 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7354 | fs_info->fs_devices->total_rw_bytes) { | |
7355 | btrfs_err(fs_info, | |
99e3ecfc | 7356 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7357 | btrfs_super_total_bytes(fs_info->super_copy), |
7358 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7359 | ret = -EINVAL; |
7360 | goto error; | |
7361 | } | |
0b86a832 CM |
7362 | ret = 0; |
7363 | error: | |
34441361 | 7364 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7365 | mutex_unlock(&uuid_mutex); |
7366 | ||
2b82032c | 7367 | btrfs_free_path(path); |
0b86a832 CM |
7368 | return ret; |
7369 | } | |
442a4f63 | 7370 | |
cb517eab MX |
7371 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7372 | { | |
7373 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7374 | struct btrfs_device *device; | |
7375 | ||
29cc83f6 LB |
7376 | while (fs_devices) { |
7377 | mutex_lock(&fs_devices->device_list_mutex); | |
7378 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7379 | device->fs_info = fs_info; |
29cc83f6 LB |
7380 | mutex_unlock(&fs_devices->device_list_mutex); |
7381 | ||
7382 | fs_devices = fs_devices->seed; | |
7383 | } | |
cb517eab MX |
7384 | } |
7385 | ||
733f4fbb SB |
7386 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7387 | { | |
7388 | int i; | |
7389 | ||
7390 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7391 | btrfs_dev_stat_reset(dev, i); | |
7392 | } | |
7393 | ||
7394 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7395 | { | |
7396 | struct btrfs_key key; | |
7397 | struct btrfs_key found_key; | |
7398 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7399 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7400 | struct extent_buffer *eb; | |
7401 | int slot; | |
7402 | int ret = 0; | |
7403 | struct btrfs_device *device; | |
7404 | struct btrfs_path *path = NULL; | |
7405 | int i; | |
7406 | ||
7407 | path = btrfs_alloc_path(); | |
7408 | if (!path) { | |
7409 | ret = -ENOMEM; | |
7410 | goto out; | |
7411 | } | |
7412 | ||
7413 | mutex_lock(&fs_devices->device_list_mutex); | |
7414 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7415 | int item_size; | |
7416 | struct btrfs_dev_stats_item *ptr; | |
7417 | ||
242e2956 DS |
7418 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7419 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7420 | key.offset = device->devid; |
7421 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7422 | if (ret) { | |
733f4fbb SB |
7423 | __btrfs_reset_dev_stats(device); |
7424 | device->dev_stats_valid = 1; | |
7425 | btrfs_release_path(path); | |
7426 | continue; | |
7427 | } | |
7428 | slot = path->slots[0]; | |
7429 | eb = path->nodes[0]; | |
7430 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7431 | item_size = btrfs_item_size_nr(eb, slot); | |
7432 | ||
7433 | ptr = btrfs_item_ptr(eb, slot, | |
7434 | struct btrfs_dev_stats_item); | |
7435 | ||
7436 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7437 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7438 | btrfs_dev_stat_set(device, i, | |
7439 | btrfs_dev_stats_value(eb, ptr, i)); | |
7440 | else | |
7441 | btrfs_dev_stat_reset(device, i); | |
7442 | } | |
7443 | ||
7444 | device->dev_stats_valid = 1; | |
7445 | btrfs_dev_stat_print_on_load(device); | |
7446 | btrfs_release_path(path); | |
7447 | } | |
7448 | mutex_unlock(&fs_devices->device_list_mutex); | |
7449 | ||
7450 | out: | |
7451 | btrfs_free_path(path); | |
7452 | return ret < 0 ? ret : 0; | |
7453 | } | |
7454 | ||
7455 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7456 | struct btrfs_device *device) |
7457 | { | |
5495f195 | 7458 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7459 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7460 | struct btrfs_path *path; |
7461 | struct btrfs_key key; | |
7462 | struct extent_buffer *eb; | |
7463 | struct btrfs_dev_stats_item *ptr; | |
7464 | int ret; | |
7465 | int i; | |
7466 | ||
242e2956 DS |
7467 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7468 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7469 | key.offset = device->devid; |
7470 | ||
7471 | path = btrfs_alloc_path(); | |
fa252992 DS |
7472 | if (!path) |
7473 | return -ENOMEM; | |
733f4fbb SB |
7474 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7475 | if (ret < 0) { | |
0b246afa | 7476 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7477 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7478 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7479 | goto out; |
7480 | } | |
7481 | ||
7482 | if (ret == 0 && | |
7483 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7484 | /* need to delete old one and insert a new one */ | |
7485 | ret = btrfs_del_item(trans, dev_root, path); | |
7486 | if (ret != 0) { | |
0b246afa | 7487 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7488 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7489 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7490 | goto out; |
7491 | } | |
7492 | ret = 1; | |
7493 | } | |
7494 | ||
7495 | if (ret == 1) { | |
7496 | /* need to insert a new item */ | |
7497 | btrfs_release_path(path); | |
7498 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7499 | &key, sizeof(*ptr)); | |
7500 | if (ret < 0) { | |
0b246afa | 7501 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7502 | "insert dev_stats item for device %s failed %d", |
7503 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7504 | goto out; |
7505 | } | |
7506 | } | |
7507 | ||
7508 | eb = path->nodes[0]; | |
7509 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7510 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7511 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7512 | btrfs_dev_stat_read(device, i)); | |
7513 | btrfs_mark_buffer_dirty(eb); | |
7514 | ||
7515 | out: | |
7516 | btrfs_free_path(path); | |
7517 | return ret; | |
7518 | } | |
7519 | ||
7520 | /* | |
7521 | * called from commit_transaction. Writes all changed device stats to disk. | |
7522 | */ | |
7523 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7524 | struct btrfs_fs_info *fs_info) | |
7525 | { | |
733f4fbb SB |
7526 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7527 | struct btrfs_device *device; | |
addc3fa7 | 7528 | int stats_cnt; |
733f4fbb SB |
7529 | int ret = 0; |
7530 | ||
7531 | mutex_lock(&fs_devices->device_list_mutex); | |
7532 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7533 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7534 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7535 | continue; |
7536 | ||
9deae968 NB |
7537 | |
7538 | /* | |
7539 | * There is a LOAD-LOAD control dependency between the value of | |
7540 | * dev_stats_ccnt and updating the on-disk values which requires | |
7541 | * reading the in-memory counters. Such control dependencies | |
7542 | * require explicit read memory barriers. | |
7543 | * | |
7544 | * This memory barriers pairs with smp_mb__before_atomic in | |
7545 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7546 | * barrier implied by atomic_xchg in | |
7547 | * btrfs_dev_stats_read_and_reset | |
7548 | */ | |
7549 | smp_rmb(); | |
7550 | ||
5495f195 | 7551 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7552 | if (!ret) |
addc3fa7 | 7553 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7554 | } |
7555 | mutex_unlock(&fs_devices->device_list_mutex); | |
7556 | ||
7557 | return ret; | |
7558 | } | |
7559 | ||
442a4f63 SB |
7560 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7561 | { | |
7562 | btrfs_dev_stat_inc(dev, index); | |
7563 | btrfs_dev_stat_print_on_error(dev); | |
7564 | } | |
7565 | ||
48a3b636 | 7566 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7567 | { |
733f4fbb SB |
7568 | if (!dev->dev_stats_valid) |
7569 | return; | |
fb456252 | 7570 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7571 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7572 | rcu_str_deref(dev->name), |
442a4f63 SB |
7573 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7574 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7575 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7576 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7577 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7578 | } |
c11d2c23 | 7579 | |
733f4fbb SB |
7580 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7581 | { | |
a98cdb85 SB |
7582 | int i; |
7583 | ||
7584 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7585 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7586 | break; | |
7587 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7588 | return; /* all values == 0, suppress message */ | |
7589 | ||
fb456252 | 7590 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7591 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7592 | rcu_str_deref(dev->name), |
733f4fbb SB |
7593 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7594 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7595 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7596 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7597 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7598 | } | |
7599 | ||
2ff7e61e | 7600 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7601 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7602 | { |
7603 | struct btrfs_device *dev; | |
0b246afa | 7604 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7605 | int i; |
7606 | ||
7607 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7608 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7609 | true); | |
c11d2c23 SB |
7610 | mutex_unlock(&fs_devices->device_list_mutex); |
7611 | ||
7612 | if (!dev) { | |
0b246afa | 7613 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7614 | return -ENODEV; |
733f4fbb | 7615 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7616 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7617 | return -ENODEV; |
b27f7c0c | 7618 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7619 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7620 | if (stats->nr_items > i) | |
7621 | stats->values[i] = | |
7622 | btrfs_dev_stat_read_and_reset(dev, i); | |
7623 | else | |
7624 | btrfs_dev_stat_reset(dev, i); | |
7625 | } | |
7626 | } else { | |
7627 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7628 | if (stats->nr_items > i) | |
7629 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7630 | } | |
7631 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7632 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7633 | return 0; | |
7634 | } | |
a8a6dab7 | 7635 | |
da353f6b | 7636 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7637 | { |
7638 | struct buffer_head *bh; | |
7639 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7640 | int copy_num; |
a8a6dab7 | 7641 | |
12b1c263 AJ |
7642 | if (!bdev) |
7643 | return; | |
a8a6dab7 | 7644 | |
12b1c263 AJ |
7645 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7646 | copy_num++) { | |
a8a6dab7 | 7647 | |
12b1c263 AJ |
7648 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7649 | continue; | |
7650 | ||
7651 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7652 | ||
7653 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7654 | set_buffer_dirty(bh); | |
7655 | sync_dirty_buffer(bh); | |
7656 | brelse(bh); | |
7657 | } | |
7658 | ||
7659 | /* Notify udev that device has changed */ | |
7660 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7661 | ||
7662 | /* Update ctime/mtime for device path for libblkid */ | |
7663 | update_dev_time(device_path); | |
a8a6dab7 | 7664 | } |
935e5cc9 MX |
7665 | |
7666 | /* | |
7667 | * Update the size of all devices, which is used for writing out the | |
7668 | * super blocks. | |
7669 | */ | |
7670 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7671 | { | |
7672 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7673 | struct btrfs_device *curr, *next; | |
7674 | ||
7675 | if (list_empty(&fs_devices->resized_devices)) | |
7676 | return; | |
7677 | ||
7678 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7679 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7680 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7681 | resized_list) { | |
7682 | list_del_init(&curr->resized_list); | |
7683 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7684 | } | |
34441361 | 7685 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7686 | mutex_unlock(&fs_devices->device_list_mutex); |
7687 | } | |
ce7213c7 MX |
7688 | |
7689 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7690 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7691 | { |
e9b919b1 | 7692 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7693 | struct extent_map *em; |
7694 | struct map_lookup *map; | |
7695 | struct btrfs_device *dev; | |
7696 | int i; | |
7697 | ||
e9b919b1 | 7698 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7699 | return; |
7700 | ||
7701 | /* In order to kick the device replace finish process */ | |
34441361 | 7702 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7703 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7704 | map = em->map_lookup; |
ce7213c7 MX |
7705 | |
7706 | for (i = 0; i < map->num_stripes; i++) { | |
7707 | dev = map->stripes[i].dev; | |
7708 | dev->commit_bytes_used = dev->bytes_used; | |
7709 | } | |
7710 | } | |
34441361 | 7711 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7712 | } |
5a13f430 AJ |
7713 | |
7714 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7715 | { | |
7716 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7717 | while (fs_devices) { | |
7718 | fs_devices->fs_info = fs_info; | |
7719 | fs_devices = fs_devices->seed; | |
7720 | } | |
7721 | } | |
7722 | ||
7723 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7724 | { | |
7725 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7726 | while (fs_devices) { | |
7727 | fs_devices->fs_info = NULL; | |
7728 | fs_devices = fs_devices->seed; | |
7729 | } | |
7730 | } | |
46df06b8 DS |
7731 | |
7732 | /* | |
7733 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7734 | */ | |
7735 | int btrfs_bg_type_to_factor(u64 flags) | |
7736 | { | |
7737 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7738 | BTRFS_BLOCK_GROUP_RAID10)) | |
7739 | return 2; | |
7740 | return 1; | |
7741 | } | |
cf90d884 QW |
7742 | |
7743 | ||
7744 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7745 | { | |
7746 | int index = btrfs_bg_flags_to_raid_index(type); | |
7747 | int ncopies = btrfs_raid_array[index].ncopies; | |
7748 | int data_stripes; | |
7749 | ||
7750 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7751 | case BTRFS_BLOCK_GROUP_RAID5: | |
7752 | data_stripes = num_stripes - 1; | |
7753 | break; | |
7754 | case BTRFS_BLOCK_GROUP_RAID6: | |
7755 | data_stripes = num_stripes - 2; | |
7756 | break; | |
7757 | default: | |
7758 | data_stripes = num_stripes / ncopies; | |
7759 | break; | |
7760 | } | |
7761 | return div_u64(chunk_len, data_stripes); | |
7762 | } | |
7763 | ||
7764 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7765 | u64 chunk_offset, u64 devid, | |
7766 | u64 physical_offset, u64 physical_len) | |
7767 | { | |
7768 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7769 | struct extent_map *em; | |
7770 | struct map_lookup *map; | |
05a37c48 | 7771 | struct btrfs_device *dev; |
cf90d884 QW |
7772 | u64 stripe_len; |
7773 | bool found = false; | |
7774 | int ret = 0; | |
7775 | int i; | |
7776 | ||
7777 | read_lock(&em_tree->lock); | |
7778 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7779 | read_unlock(&em_tree->lock); | |
7780 | ||
7781 | if (!em) { | |
7782 | btrfs_err(fs_info, | |
7783 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7784 | physical_offset, devid); | |
7785 | ret = -EUCLEAN; | |
7786 | goto out; | |
7787 | } | |
7788 | ||
7789 | map = em->map_lookup; | |
7790 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7791 | if (physical_len != stripe_len) { | |
7792 | btrfs_err(fs_info, | |
7793 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7794 | physical_offset, devid, em->start, physical_len, | |
7795 | stripe_len); | |
7796 | ret = -EUCLEAN; | |
7797 | goto out; | |
7798 | } | |
7799 | ||
7800 | for (i = 0; i < map->num_stripes; i++) { | |
7801 | if (map->stripes[i].dev->devid == devid && | |
7802 | map->stripes[i].physical == physical_offset) { | |
7803 | found = true; | |
7804 | if (map->verified_stripes >= map->num_stripes) { | |
7805 | btrfs_err(fs_info, | |
7806 | "too many dev extents for chunk %llu found", | |
7807 | em->start); | |
7808 | ret = -EUCLEAN; | |
7809 | goto out; | |
7810 | } | |
7811 | map->verified_stripes++; | |
7812 | break; | |
7813 | } | |
7814 | } | |
7815 | if (!found) { | |
7816 | btrfs_err(fs_info, | |
7817 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7818 | physical_offset, devid); | |
7819 | ret = -EUCLEAN; | |
7820 | } | |
05a37c48 QW |
7821 | |
7822 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7823 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7824 | if (!dev) { |
7825 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7826 | ret = -EUCLEAN; | |
7827 | goto out; | |
7828 | } | |
1b3922a8 QW |
7829 | |
7830 | /* It's possible this device is a dummy for seed device */ | |
7831 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7832 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7833 | NULL, false); | |
1b3922a8 QW |
7834 | if (!dev) { |
7835 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7836 | devid); | |
7837 | ret = -EUCLEAN; | |
7838 | goto out; | |
7839 | } | |
7840 | } | |
7841 | ||
05a37c48 QW |
7842 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7843 | btrfs_err(fs_info, | |
7844 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7845 | devid, physical_offset, physical_len, | |
7846 | dev->disk_total_bytes); | |
7847 | ret = -EUCLEAN; | |
7848 | goto out; | |
7849 | } | |
cf90d884 QW |
7850 | out: |
7851 | free_extent_map(em); | |
7852 | return ret; | |
7853 | } | |
7854 | ||
7855 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7856 | { | |
7857 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7858 | struct extent_map *em; | |
7859 | struct rb_node *node; | |
7860 | int ret = 0; | |
7861 | ||
7862 | read_lock(&em_tree->lock); | |
07e1ce09 | 7863 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7864 | em = rb_entry(node, struct extent_map, rb_node); |
7865 | if (em->map_lookup->num_stripes != | |
7866 | em->map_lookup->verified_stripes) { | |
7867 | btrfs_err(fs_info, | |
7868 | "chunk %llu has missing dev extent, have %d expect %d", | |
7869 | em->start, em->map_lookup->verified_stripes, | |
7870 | em->map_lookup->num_stripes); | |
7871 | ret = -EUCLEAN; | |
7872 | goto out; | |
7873 | } | |
7874 | } | |
7875 | out: | |
7876 | read_unlock(&em_tree->lock); | |
7877 | return ret; | |
7878 | } | |
7879 | ||
7880 | /* | |
7881 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7882 | * later chunk allocation/free would cause unexpected behavior. | |
7883 | * | |
7884 | * NOTE: This will iterate through the whole device tree, which should be of | |
7885 | * the same size level as the chunk tree. This slightly increases mount time. | |
7886 | */ | |
7887 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7888 | { | |
7889 | struct btrfs_path *path; | |
7890 | struct btrfs_root *root = fs_info->dev_root; | |
7891 | struct btrfs_key key; | |
5eb19381 QW |
7892 | u64 prev_devid = 0; |
7893 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7894 | int ret = 0; |
7895 | ||
7896 | key.objectid = 1; | |
7897 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7898 | key.offset = 0; | |
7899 | ||
7900 | path = btrfs_alloc_path(); | |
7901 | if (!path) | |
7902 | return -ENOMEM; | |
7903 | ||
7904 | path->reada = READA_FORWARD; | |
7905 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7906 | if (ret < 0) | |
7907 | goto out; | |
7908 | ||
7909 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7910 | ret = btrfs_next_item(root, path); | |
7911 | if (ret < 0) | |
7912 | goto out; | |
7913 | /* No dev extents at all? Not good */ | |
7914 | if (ret > 0) { | |
7915 | ret = -EUCLEAN; | |
7916 | goto out; | |
7917 | } | |
7918 | } | |
7919 | while (1) { | |
7920 | struct extent_buffer *leaf = path->nodes[0]; | |
7921 | struct btrfs_dev_extent *dext; | |
7922 | int slot = path->slots[0]; | |
7923 | u64 chunk_offset; | |
7924 | u64 physical_offset; | |
7925 | u64 physical_len; | |
7926 | u64 devid; | |
7927 | ||
7928 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7929 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7930 | break; | |
7931 | devid = key.objectid; | |
7932 | physical_offset = key.offset; | |
7933 | ||
7934 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7935 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7936 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7937 | ||
5eb19381 QW |
7938 | /* Check if this dev extent overlaps with the previous one */ |
7939 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7940 | btrfs_err(fs_info, | |
7941 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7942 | devid, physical_offset, prev_dev_ext_end); | |
7943 | ret = -EUCLEAN; | |
7944 | goto out; | |
7945 | } | |
7946 | ||
cf90d884 QW |
7947 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7948 | physical_offset, physical_len); | |
7949 | if (ret < 0) | |
7950 | goto out; | |
5eb19381 QW |
7951 | prev_devid = devid; |
7952 | prev_dev_ext_end = physical_offset + physical_len; | |
7953 | ||
cf90d884 QW |
7954 | ret = btrfs_next_item(root, path); |
7955 | if (ret < 0) | |
7956 | goto out; | |
7957 | if (ret > 0) { | |
7958 | ret = 0; | |
7959 | break; | |
7960 | } | |
7961 | } | |
7962 | ||
7963 | /* Ensure all chunks have corresponding dev extents */ | |
7964 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7965 | out: | |
7966 | btrfs_free_path(path); | |
7967 | return ret; | |
7968 | } | |
eede2bf3 OS |
7969 | |
7970 | /* | |
7971 | * Check whether the given block group or device is pinned by any inode being | |
7972 | * used as a swapfile. | |
7973 | */ | |
7974 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7975 | { | |
7976 | struct btrfs_swapfile_pin *sp; | |
7977 | struct rb_node *node; | |
7978 | ||
7979 | spin_lock(&fs_info->swapfile_pins_lock); | |
7980 | node = fs_info->swapfile_pins.rb_node; | |
7981 | while (node) { | |
7982 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7983 | if (ptr < sp->ptr) | |
7984 | node = node->rb_left; | |
7985 | else if (ptr > sp->ptr) | |
7986 | node = node->rb_right; | |
7987 | else | |
7988 | break; | |
7989 | } | |
7990 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7991 | return node != NULL; | |
7992 | } |