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" |
82fc28fb | 30 | #include "tree-checker.h" |
0b86a832 | 31 | |
af902047 ZL |
32 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
33 | [BTRFS_RAID_RAID10] = { | |
34 | .sub_stripes = 2, | |
35 | .dev_stripes = 1, | |
36 | .devs_max = 0, /* 0 == as many as possible */ | |
37 | .devs_min = 4, | |
8789f4fe | 38 | .tolerated_failures = 1, |
af902047 ZL |
39 | .devs_increment = 2, |
40 | .ncopies = 2, | |
b50836ed | 41 | .nparity = 0, |
ed23467b | 42 | .raid_name = "raid10", |
41a6e891 | 43 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 44 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
45 | }, |
46 | [BTRFS_RAID_RAID1] = { | |
47 | .sub_stripes = 1, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 2, | |
50 | .devs_min = 2, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
b50836ed | 54 | .nparity = 0, |
ed23467b | 55 | .raid_name = "raid1", |
41a6e891 | 56 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 57 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 ZL |
58 | }, |
59 | [BTRFS_RAID_DUP] = { | |
60 | .sub_stripes = 1, | |
61 | .dev_stripes = 2, | |
62 | .devs_max = 1, | |
63 | .devs_min = 1, | |
8789f4fe | 64 | .tolerated_failures = 0, |
af902047 ZL |
65 | .devs_increment = 1, |
66 | .ncopies = 2, | |
b50836ed | 67 | .nparity = 0, |
ed23467b | 68 | .raid_name = "dup", |
41a6e891 | 69 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 70 | .mindev_error = 0, |
af902047 ZL |
71 | }, |
72 | [BTRFS_RAID_RAID0] = { | |
73 | .sub_stripes = 1, | |
74 | .dev_stripes = 1, | |
75 | .devs_max = 0, | |
76 | .devs_min = 2, | |
8789f4fe | 77 | .tolerated_failures = 0, |
af902047 ZL |
78 | .devs_increment = 1, |
79 | .ncopies = 1, | |
b50836ed | 80 | .nparity = 0, |
ed23467b | 81 | .raid_name = "raid0", |
41a6e891 | 82 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 83 | .mindev_error = 0, |
af902047 ZL |
84 | }, |
85 | [BTRFS_RAID_SINGLE] = { | |
86 | .sub_stripes = 1, | |
87 | .dev_stripes = 1, | |
88 | .devs_max = 1, | |
89 | .devs_min = 1, | |
8789f4fe | 90 | .tolerated_failures = 0, |
af902047 ZL |
91 | .devs_increment = 1, |
92 | .ncopies = 1, | |
b50836ed | 93 | .nparity = 0, |
ed23467b | 94 | .raid_name = "single", |
41a6e891 | 95 | .bg_flag = 0, |
f9fbcaa2 | 96 | .mindev_error = 0, |
af902047 ZL |
97 | }, |
98 | [BTRFS_RAID_RAID5] = { | |
99 | .sub_stripes = 1, | |
100 | .dev_stripes = 1, | |
101 | .devs_max = 0, | |
102 | .devs_min = 2, | |
8789f4fe | 103 | .tolerated_failures = 1, |
af902047 | 104 | .devs_increment = 1, |
da612e31 | 105 | .ncopies = 1, |
b50836ed | 106 | .nparity = 1, |
ed23467b | 107 | .raid_name = "raid5", |
41a6e891 | 108 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 109 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
110 | }, |
111 | [BTRFS_RAID_RAID6] = { | |
112 | .sub_stripes = 1, | |
113 | .dev_stripes = 1, | |
114 | .devs_max = 0, | |
115 | .devs_min = 3, | |
8789f4fe | 116 | .tolerated_failures = 2, |
af902047 | 117 | .devs_increment = 1, |
da612e31 | 118 | .ncopies = 1, |
b50836ed | 119 | .nparity = 2, |
ed23467b | 120 | .raid_name = "raid6", |
41a6e891 | 121 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 122 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
123 | }, |
124 | }; | |
125 | ||
ed23467b AJ |
126 | const char *get_raid_name(enum btrfs_raid_types type) |
127 | { | |
128 | if (type >= BTRFS_NR_RAID_TYPES) | |
129 | return NULL; | |
130 | ||
131 | return btrfs_raid_array[type].raid_name; | |
132 | } | |
133 | ||
f89e09cf AJ |
134 | /* |
135 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
136 | * bytes including terminating null byte. | |
137 | */ | |
138 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
139 | { | |
140 | int i; | |
141 | int ret; | |
142 | char *bp = buf; | |
143 | u64 flags = bg_flags; | |
144 | u32 size_bp = size_buf; | |
145 | ||
146 | if (!flags) { | |
147 | strcpy(bp, "NONE"); | |
148 | return; | |
149 | } | |
150 | ||
151 | #define DESCRIBE_FLAG(flag, desc) \ | |
152 | do { \ | |
153 | if (flags & (flag)) { \ | |
154 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
155 | if (ret < 0 || ret >= size_bp) \ | |
156 | goto out_overflow; \ | |
157 | size_bp -= ret; \ | |
158 | bp += ret; \ | |
159 | flags &= ~(flag); \ | |
160 | } \ | |
161 | } while (0) | |
162 | ||
163 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
164 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
165 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
166 | ||
167 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
168 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
169 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
170 | btrfs_raid_array[i].raid_name); | |
171 | #undef DESCRIBE_FLAG | |
172 | ||
173 | if (flags) { | |
174 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
175 | size_bp -= ret; | |
176 | } | |
177 | ||
178 | if (size_bp < size_buf) | |
179 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
180 | ||
181 | /* | |
182 | * The text is trimmed, it's up to the caller to provide sufficiently | |
183 | * large buffer | |
184 | */ | |
185 | out_overflow:; | |
186 | } | |
187 | ||
2b82032c | 188 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 189 | struct btrfs_fs_info *fs_info); |
2ff7e61e | 190 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 191 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 192 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 193 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
194 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
195 | enum btrfs_map_op op, | |
196 | u64 logical, u64 *length, | |
197 | struct btrfs_bio **bbio_ret, | |
198 | int mirror_num, int need_raid_map); | |
2b82032c | 199 | |
9c6b1c4d DS |
200 | /* |
201 | * Device locking | |
202 | * ============== | |
203 | * | |
204 | * There are several mutexes that protect manipulation of devices and low-level | |
205 | * structures like chunks but not block groups, extents or files | |
206 | * | |
207 | * uuid_mutex (global lock) | |
208 | * ------------------------ | |
209 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
210 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
211 | * device) or requested by the device= mount option | |
212 | * | |
213 | * the mutex can be very coarse and can cover long-running operations | |
214 | * | |
215 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 216 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
217 | * |
218 | * global::fs_devs - add, remove, updates to the global list | |
219 | * | |
220 | * does not protect: manipulation of the fs_devices::devices list! | |
221 | * | |
222 | * btrfs_device::name - renames (write side), read is RCU | |
223 | * | |
224 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
225 | * ------------------------------------------------ | |
226 | * protects updates to fs_devices::devices, ie. adding and deleting | |
227 | * | |
228 | * simple list traversal with read-only actions can be done with RCU protection | |
229 | * | |
230 | * may be used to exclude some operations from running concurrently without any | |
231 | * modifications to the list (see write_all_supers) | |
232 | * | |
9c6b1c4d DS |
233 | * balance_mutex |
234 | * ------------- | |
235 | * protects balance structures (status, state) and context accessed from | |
236 | * several places (internally, ioctl) | |
237 | * | |
238 | * chunk_mutex | |
239 | * ----------- | |
240 | * protects chunks, adding or removing during allocation, trim or when a new | |
241 | * device is added/removed | |
242 | * | |
243 | * cleaner_mutex | |
244 | * ------------- | |
245 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
246 | * cleaning together with relocation or delayed iputs | |
247 | * | |
248 | * | |
249 | * Lock nesting | |
250 | * ============ | |
251 | * | |
252 | * uuid_mutex | |
253 | * volume_mutex | |
254 | * device_list_mutex | |
255 | * chunk_mutex | |
256 | * balance_mutex | |
89595e80 AJ |
257 | * |
258 | * | |
259 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
260 | * ====================================== | |
261 | * | |
262 | * Maintains the exclusivity of the following operations that apply to the | |
263 | * whole filesystem and cannot run in parallel. | |
264 | * | |
265 | * - Balance (*) | |
266 | * - Device add | |
267 | * - Device remove | |
268 | * - Device replace (*) | |
269 | * - Resize | |
270 | * | |
271 | * The device operations (as above) can be in one of the following states: | |
272 | * | |
273 | * - Running state | |
274 | * - Paused state | |
275 | * - Completed state | |
276 | * | |
277 | * Only device operations marked with (*) can go into the Paused state for the | |
278 | * following reasons: | |
279 | * | |
280 | * - ioctl (only Balance can be Paused through ioctl) | |
281 | * - filesystem remounted as read-only | |
282 | * - filesystem unmounted and mounted as read-only | |
283 | * - system power-cycle and filesystem mounted as read-only | |
284 | * - filesystem or device errors leading to forced read-only | |
285 | * | |
286 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
287 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
288 | * A device operation in Paused or Running state can be canceled or resumed | |
289 | * either by ioctl (Balance only) or when remounted as read-write. | |
290 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
291 | * completed. | |
9c6b1c4d DS |
292 | */ |
293 | ||
67a2c45e | 294 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 295 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
296 | struct list_head *btrfs_get_fs_uuids(void) |
297 | { | |
298 | return &fs_uuids; | |
299 | } | |
8a4b83cc | 300 | |
2dfeca9b DS |
301 | /* |
302 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
303 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
304 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
305 | * |
306 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
307 | * The returned struct is not linked onto any lists and can be destroyed with | |
308 | * kfree() right away. | |
309 | */ | |
7239ff4b NB |
310 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
311 | const u8 *metadata_fsid) | |
2208a378 ID |
312 | { |
313 | struct btrfs_fs_devices *fs_devs; | |
314 | ||
78f2c9e6 | 315 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
316 | if (!fs_devs) |
317 | return ERR_PTR(-ENOMEM); | |
318 | ||
319 | mutex_init(&fs_devs->device_list_mutex); | |
320 | ||
321 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 322 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 | 323 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
c4babc5e | 324 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
325 | if (fsid) |
326 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 327 | |
7239ff4b NB |
328 | if (metadata_fsid) |
329 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
330 | else if (fsid) | |
331 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
332 | ||
2208a378 ID |
333 | return fs_devs; |
334 | } | |
335 | ||
a425f9d4 | 336 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf DS |
337 | { |
338 | rcu_string_free(device->name); | |
339 | bio_put(device->flush_bio); | |
340 | kfree(device); | |
341 | } | |
342 | ||
e4404d6e YZ |
343 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
344 | { | |
345 | struct btrfs_device *device; | |
346 | WARN_ON(fs_devices->opened); | |
347 | while (!list_empty(&fs_devices->devices)) { | |
348 | device = list_entry(fs_devices->devices.next, | |
349 | struct btrfs_device, dev_list); | |
350 | list_del(&device->dev_list); | |
a425f9d4 | 351 | btrfs_free_device(device); |
e4404d6e YZ |
352 | } |
353 | kfree(fs_devices); | |
354 | } | |
355 | ||
b8b8ff59 LC |
356 | static void btrfs_kobject_uevent(struct block_device *bdev, |
357 | enum kobject_action action) | |
358 | { | |
359 | int ret; | |
360 | ||
361 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
362 | if (ret) | |
efe120a0 | 363 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
364 | action, |
365 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
366 | &disk_to_dev(bdev->bd_disk)->kobj); | |
367 | } | |
368 | ||
ffc5a379 | 369 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
370 | { |
371 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 372 | |
2b82032c YZ |
373 | while (!list_empty(&fs_uuids)) { |
374 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
375 | struct btrfs_fs_devices, fs_list); |
376 | list_del(&fs_devices->fs_list); | |
e4404d6e | 377 | free_fs_devices(fs_devices); |
8a4b83cc | 378 | } |
8a4b83cc CM |
379 | } |
380 | ||
48dae9cf DS |
381 | /* |
382 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
383 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 384 | * btrfs_free_device. |
48dae9cf | 385 | */ |
12bd2fc0 ID |
386 | static struct btrfs_device *__alloc_device(void) |
387 | { | |
388 | struct btrfs_device *dev; | |
389 | ||
78f2c9e6 | 390 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
391 | if (!dev) |
392 | return ERR_PTR(-ENOMEM); | |
393 | ||
e0ae9994 DS |
394 | /* |
395 | * Preallocate a bio that's always going to be used for flushing device | |
396 | * barriers and matches the device lifespan | |
397 | */ | |
398 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
399 | if (!dev->flush_bio) { | |
400 | kfree(dev); | |
401 | return ERR_PTR(-ENOMEM); | |
402 | } | |
e0ae9994 | 403 | |
12bd2fc0 ID |
404 | INIT_LIST_HEAD(&dev->dev_list); |
405 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 406 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
407 | |
408 | spin_lock_init(&dev->io_lock); | |
409 | ||
12bd2fc0 | 410 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 411 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 412 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 413 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 414 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
12bd2fc0 ID |
415 | |
416 | return dev; | |
417 | } | |
418 | ||
7239ff4b NB |
419 | static noinline struct btrfs_fs_devices *find_fsid( |
420 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 421 | { |
8a4b83cc CM |
422 | struct btrfs_fs_devices *fs_devices; |
423 | ||
7239ff4b NB |
424 | ASSERT(fsid); |
425 | ||
7a62d0f0 NB |
426 | if (metadata_fsid) { |
427 | /* | |
428 | * Handle scanned device having completed its fsid change but | |
429 | * belonging to a fs_devices that was created by first scanning | |
430 | * a device which didn't have its fsid/metadata_uuid changed | |
431 | * at all and the CHANGING_FSID_V2 flag set. | |
432 | */ | |
433 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
434 | if (fs_devices->fsid_change && | |
435 | memcmp(metadata_fsid, fs_devices->fsid, | |
436 | BTRFS_FSID_SIZE) == 0 && | |
437 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
438 | BTRFS_FSID_SIZE) == 0) { | |
439 | return fs_devices; | |
440 | } | |
441 | } | |
cc5de4e7 NB |
442 | /* |
443 | * Handle scanned device having completed its fsid change but | |
444 | * belonging to a fs_devices that was created by a device that | |
445 | * has an outdated pair of fsid/metadata_uuid and | |
446 | * CHANGING_FSID_V2 flag set. | |
447 | */ | |
448 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
449 | if (fs_devices->fsid_change && | |
450 | memcmp(fs_devices->metadata_uuid, | |
451 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
452 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
453 | BTRFS_FSID_SIZE) == 0) { | |
454 | return fs_devices; | |
455 | } | |
456 | } | |
7a62d0f0 NB |
457 | } |
458 | ||
459 | /* Handle non-split brain cases */ | |
c4babc5e | 460 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
461 | if (metadata_fsid) { |
462 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
463 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
464 | BTRFS_FSID_SIZE) == 0) | |
465 | return fs_devices; | |
466 | } else { | |
467 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
468 | return fs_devices; | |
469 | } | |
8a4b83cc CM |
470 | } |
471 | return NULL; | |
472 | } | |
473 | ||
beaf8ab3 SB |
474 | static int |
475 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
476 | int flush, struct block_device **bdev, | |
477 | struct buffer_head **bh) | |
478 | { | |
479 | int ret; | |
480 | ||
481 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
482 | ||
483 | if (IS_ERR(*bdev)) { | |
484 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
485 | goto error; |
486 | } | |
487 | ||
488 | if (flush) | |
489 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 490 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
491 | if (ret) { |
492 | blkdev_put(*bdev, flags); | |
493 | goto error; | |
494 | } | |
495 | invalidate_bdev(*bdev); | |
496 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
497 | if (IS_ERR(*bh)) { |
498 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
499 | blkdev_put(*bdev, flags); |
500 | goto error; | |
501 | } | |
502 | ||
503 | return 0; | |
504 | ||
505 | error: | |
506 | *bdev = NULL; | |
507 | *bh = NULL; | |
508 | return ret; | |
509 | } | |
510 | ||
ffbd517d CM |
511 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
512 | struct bio *head, struct bio *tail) | |
513 | { | |
514 | ||
515 | struct bio *old_head; | |
516 | ||
517 | old_head = pending_bios->head; | |
518 | pending_bios->head = head; | |
519 | if (pending_bios->tail) | |
520 | tail->bi_next = old_head; | |
521 | else | |
522 | pending_bios->tail = tail; | |
523 | } | |
524 | ||
8b712842 CM |
525 | /* |
526 | * we try to collect pending bios for a device so we don't get a large | |
527 | * number of procs sending bios down to the same device. This greatly | |
528 | * improves the schedulers ability to collect and merge the bios. | |
529 | * | |
530 | * But, it also turns into a long list of bios to process and that is sure | |
531 | * to eventually make the worker thread block. The solution here is to | |
532 | * make some progress and then put this work struct back at the end of | |
533 | * the list if the block device is congested. This way, multiple devices | |
534 | * can make progress from a single worker thread. | |
535 | */ | |
143bede5 | 536 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 537 | { |
0b246afa | 538 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
539 | struct bio *pending; |
540 | struct backing_dev_info *bdi; | |
ffbd517d | 541 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
542 | struct bio *tail; |
543 | struct bio *cur; | |
544 | int again = 0; | |
ffbd517d | 545 | unsigned long num_run; |
d644d8a1 | 546 | unsigned long batch_run = 0; |
b765ead5 | 547 | unsigned long last_waited = 0; |
d84275c9 | 548 | int force_reg = 0; |
0e588859 | 549 | int sync_pending = 0; |
211588ad CM |
550 | struct blk_plug plug; |
551 | ||
552 | /* | |
553 | * this function runs all the bios we've collected for | |
554 | * a particular device. We don't want to wander off to | |
555 | * another device without first sending all of these down. | |
556 | * So, setup a plug here and finish it off before we return | |
557 | */ | |
558 | blk_start_plug(&plug); | |
8b712842 | 559 | |
efa7c9f9 | 560 | bdi = device->bdev->bd_bdi; |
b64a2851 | 561 | |
8b712842 CM |
562 | loop: |
563 | spin_lock(&device->io_lock); | |
564 | ||
a6837051 | 565 | loop_lock: |
d84275c9 | 566 | num_run = 0; |
ffbd517d | 567 | |
8b712842 CM |
568 | /* take all the bios off the list at once and process them |
569 | * later on (without the lock held). But, remember the | |
570 | * tail and other pointers so the bios can be properly reinserted | |
571 | * into the list if we hit congestion | |
572 | */ | |
d84275c9 | 573 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 574 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
575 | force_reg = 1; |
576 | } else { | |
ffbd517d | 577 | pending_bios = &device->pending_bios; |
d84275c9 CM |
578 | force_reg = 0; |
579 | } | |
ffbd517d CM |
580 | |
581 | pending = pending_bios->head; | |
582 | tail = pending_bios->tail; | |
8b712842 | 583 | WARN_ON(pending && !tail); |
8b712842 CM |
584 | |
585 | /* | |
586 | * if pending was null this time around, no bios need processing | |
587 | * at all and we can stop. Otherwise it'll loop back up again | |
588 | * and do an additional check so no bios are missed. | |
589 | * | |
590 | * device->running_pending is used to synchronize with the | |
591 | * schedule_bio code. | |
592 | */ | |
ffbd517d CM |
593 | if (device->pending_sync_bios.head == NULL && |
594 | device->pending_bios.head == NULL) { | |
8b712842 CM |
595 | again = 0; |
596 | device->running_pending = 0; | |
ffbd517d CM |
597 | } else { |
598 | again = 1; | |
599 | device->running_pending = 1; | |
8b712842 | 600 | } |
ffbd517d CM |
601 | |
602 | pending_bios->head = NULL; | |
603 | pending_bios->tail = NULL; | |
604 | ||
8b712842 CM |
605 | spin_unlock(&device->io_lock); |
606 | ||
d397712b | 607 | while (pending) { |
ffbd517d CM |
608 | |
609 | rmb(); | |
d84275c9 CM |
610 | /* we want to work on both lists, but do more bios on the |
611 | * sync list than the regular list | |
612 | */ | |
613 | if ((num_run > 32 && | |
614 | pending_bios != &device->pending_sync_bios && | |
615 | device->pending_sync_bios.head) || | |
616 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
617 | device->pending_bios.head)) { | |
ffbd517d CM |
618 | spin_lock(&device->io_lock); |
619 | requeue_list(pending_bios, pending, tail); | |
620 | goto loop_lock; | |
621 | } | |
622 | ||
8b712842 CM |
623 | cur = pending; |
624 | pending = pending->bi_next; | |
625 | cur->bi_next = NULL; | |
b64a2851 | 626 | |
dac56212 | 627 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 628 | |
2ab1ba68 CM |
629 | /* |
630 | * if we're doing the sync list, record that our | |
631 | * plug has some sync requests on it | |
632 | * | |
633 | * If we're doing the regular list and there are | |
634 | * sync requests sitting around, unplug before | |
635 | * we add more | |
636 | */ | |
637 | if (pending_bios == &device->pending_sync_bios) { | |
638 | sync_pending = 1; | |
639 | } else if (sync_pending) { | |
640 | blk_finish_plug(&plug); | |
641 | blk_start_plug(&plug); | |
642 | sync_pending = 0; | |
643 | } | |
644 | ||
4e49ea4a | 645 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
646 | num_run++; |
647 | batch_run++; | |
853d8ec4 DS |
648 | |
649 | cond_resched(); | |
8b712842 CM |
650 | |
651 | /* | |
652 | * we made progress, there is more work to do and the bdi | |
653 | * is now congested. Back off and let other work structs | |
654 | * run instead | |
655 | */ | |
57fd5a5f | 656 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 657 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 658 | struct io_context *ioc; |
8b712842 | 659 | |
b765ead5 CM |
660 | ioc = current->io_context; |
661 | ||
662 | /* | |
663 | * the main goal here is that we don't want to | |
664 | * block if we're going to be able to submit | |
665 | * more requests without blocking. | |
666 | * | |
667 | * This code does two great things, it pokes into | |
668 | * the elevator code from a filesystem _and_ | |
669 | * it makes assumptions about how batching works. | |
670 | */ | |
671 | if (ioc && ioc->nr_batch_requests > 0 && | |
672 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
673 | (last_waited == 0 || | |
674 | ioc->last_waited == last_waited)) { | |
675 | /* | |
676 | * we want to go through our batch of | |
677 | * requests and stop. So, we copy out | |
678 | * the ioc->last_waited time and test | |
679 | * against it before looping | |
680 | */ | |
681 | last_waited = ioc->last_waited; | |
853d8ec4 | 682 | cond_resched(); |
b765ead5 CM |
683 | continue; |
684 | } | |
8b712842 | 685 | spin_lock(&device->io_lock); |
ffbd517d | 686 | requeue_list(pending_bios, pending, tail); |
a6837051 | 687 | device->running_pending = 1; |
8b712842 CM |
688 | |
689 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
690 | btrfs_queue_work(fs_info->submit_workers, |
691 | &device->work); | |
8b712842 CM |
692 | goto done; |
693 | } | |
694 | } | |
ffbd517d | 695 | |
51684082 CM |
696 | cond_resched(); |
697 | if (again) | |
698 | goto loop; | |
699 | ||
700 | spin_lock(&device->io_lock); | |
701 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
702 | goto loop_lock; | |
703 | spin_unlock(&device->io_lock); | |
704 | ||
8b712842 | 705 | done: |
211588ad | 706 | blk_finish_plug(&plug); |
8b712842 CM |
707 | } |
708 | ||
b2950863 | 709 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
710 | { |
711 | struct btrfs_device *device; | |
712 | ||
713 | device = container_of(work, struct btrfs_device, work); | |
714 | run_scheduled_bios(device); | |
715 | } | |
716 | ||
70bc7088 AJ |
717 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
718 | { | |
719 | int found; | |
720 | ||
721 | rcu_read_lock(); | |
722 | found = strcmp(rcu_str_deref(device->name), path); | |
723 | rcu_read_unlock(); | |
724 | ||
725 | return found == 0; | |
726 | } | |
727 | ||
d8367db3 AJ |
728 | /* |
729 | * Search and remove all stale (devices which are not mounted) devices. | |
730 | * When both inputs are NULL, it will search and release all stale devices. | |
731 | * path: Optional. When provided will it release all unmounted devices | |
732 | * matching this path only. | |
733 | * skip_dev: Optional. Will skip this device when searching for the stale | |
734 | * devices. | |
70bc7088 AJ |
735 | * Return: 0 for success or if @path is NULL. |
736 | * -EBUSY if @path is a mounted device. | |
737 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 738 | */ |
70bc7088 | 739 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 740 | struct btrfs_device *skip_device) |
4fde46f0 | 741 | { |
fa6d2ae5 AJ |
742 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
743 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
744 | int ret = 0; |
745 | ||
746 | if (path) | |
747 | ret = -ENOENT; | |
4fde46f0 | 748 | |
fa6d2ae5 | 749 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 750 | |
70bc7088 | 751 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
752 | list_for_each_entry_safe(device, tmp_device, |
753 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 754 | if (skip_device && skip_device == device) |
d8367db3 | 755 | continue; |
fa6d2ae5 | 756 | if (path && !device->name) |
4fde46f0 | 757 | continue; |
70bc7088 | 758 | if (path && !device_path_matched(path, device)) |
38cf665d | 759 | continue; |
70bc7088 AJ |
760 | if (fs_devices->opened) { |
761 | /* for an already deleted device return 0 */ | |
762 | if (path && ret != 0) | |
763 | ret = -EBUSY; | |
764 | break; | |
765 | } | |
4fde46f0 | 766 | |
4fde46f0 | 767 | /* delete the stale device */ |
7bcb8164 AJ |
768 | fs_devices->num_devices--; |
769 | list_del(&device->dev_list); | |
770 | btrfs_free_device(device); | |
771 | ||
70bc7088 | 772 | ret = 0; |
7bcb8164 | 773 | if (fs_devices->num_devices == 0) |
fd649f10 | 774 | break; |
7bcb8164 AJ |
775 | } |
776 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 777 | |
7bcb8164 AJ |
778 | if (fs_devices->num_devices == 0) { |
779 | btrfs_sysfs_remove_fsid(fs_devices); | |
780 | list_del(&fs_devices->fs_list); | |
781 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
782 | } |
783 | } | |
70bc7088 AJ |
784 | |
785 | return ret; | |
4fde46f0 AJ |
786 | } |
787 | ||
0fb08bcc AJ |
788 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
789 | struct btrfs_device *device, fmode_t flags, | |
790 | void *holder) | |
791 | { | |
792 | struct request_queue *q; | |
793 | struct block_device *bdev; | |
794 | struct buffer_head *bh; | |
795 | struct btrfs_super_block *disk_super; | |
796 | u64 devid; | |
797 | int ret; | |
798 | ||
799 | if (device->bdev) | |
800 | return -EINVAL; | |
801 | if (!device->name) | |
802 | return -EINVAL; | |
803 | ||
804 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
805 | &bdev, &bh); | |
806 | if (ret) | |
807 | return ret; | |
808 | ||
809 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
810 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
811 | if (devid != device->devid) | |
812 | goto error_brelse; | |
813 | ||
814 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
815 | goto error_brelse; | |
816 | ||
817 | device->generation = btrfs_super_generation(disk_super); | |
818 | ||
819 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
820 | if (btrfs_super_incompat_flags(disk_super) & |
821 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
822 | pr_err( | |
823 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
824 | goto error_brelse; | |
825 | } | |
826 | ||
ebbede42 | 827 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
828 | fs_devices->seeding = 1; |
829 | } else { | |
ebbede42 AJ |
830 | if (bdev_read_only(bdev)) |
831 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
832 | else | |
833 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
834 | } |
835 | ||
836 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
837 | if (!blk_queue_nonrot(q)) |
838 | fs_devices->rotating = 1; | |
839 | ||
840 | device->bdev = bdev; | |
e12c9621 | 841 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
842 | device->mode = flags; |
843 | ||
844 | fs_devices->open_devices++; | |
ebbede42 AJ |
845 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
846 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 847 | fs_devices->rw_devices++; |
b1b8e386 | 848 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
849 | } |
850 | brelse(bh); | |
851 | ||
852 | return 0; | |
853 | ||
854 | error_brelse: | |
855 | brelse(bh); | |
856 | blkdev_put(bdev, flags); | |
857 | ||
858 | return -EINVAL; | |
859 | } | |
860 | ||
7a62d0f0 NB |
861 | /* |
862 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
863 | * being created with a disk that has already completed its fsid change. | |
864 | */ | |
865 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
866 | struct btrfs_super_block *disk_super) | |
867 | { | |
868 | struct btrfs_fs_devices *fs_devices; | |
869 | ||
870 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
871 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
872 | BTRFS_FSID_SIZE) != 0 && | |
873 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
874 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
875 | return fs_devices; | |
876 | } | |
877 | } | |
878 | ||
879 | return NULL; | |
880 | } | |
881 | ||
cc5de4e7 NB |
882 | |
883 | static struct btrfs_fs_devices *find_fsid_changed( | |
884 | struct btrfs_super_block *disk_super) | |
885 | { | |
886 | struct btrfs_fs_devices *fs_devices; | |
887 | ||
888 | /* | |
889 | * Handles the case where scanned device is part of an fs that had | |
890 | * multiple successful changes of FSID but curently device didn't | |
891 | * observe it. Meaning our fsid will be different than theirs. | |
892 | */ | |
893 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
894 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
895 | BTRFS_FSID_SIZE) != 0 && | |
896 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
897 | BTRFS_FSID_SIZE) == 0 && | |
898 | memcmp(fs_devices->fsid, disk_super->fsid, | |
899 | BTRFS_FSID_SIZE) != 0) { | |
900 | return fs_devices; | |
901 | } | |
902 | } | |
903 | ||
904 | return NULL; | |
905 | } | |
60999ca4 DS |
906 | /* |
907 | * Add new device to list of registered devices | |
908 | * | |
909 | * Returns: | |
e124ece5 AJ |
910 | * device pointer which was just added or updated when successful |
911 | * error pointer when failed | |
60999ca4 | 912 | */ |
e124ece5 | 913 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
914 | struct btrfs_super_block *disk_super, |
915 | bool *new_device_added) | |
8a4b83cc CM |
916 | { |
917 | struct btrfs_device *device; | |
7a62d0f0 | 918 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 919 | struct rcu_string *name; |
8a4b83cc | 920 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 921 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
922 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
923 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
924 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
925 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 926 | |
cc5de4e7 NB |
927 | if (fsid_change_in_progress) { |
928 | if (!has_metadata_uuid) { | |
929 | /* | |
930 | * When we have an image which has CHANGING_FSID_V2 set | |
931 | * it might belong to either a filesystem which has | |
932 | * disks with completed fsid change or it might belong | |
933 | * to fs with no UUID changes in effect, handle both. | |
934 | */ | |
935 | fs_devices = find_fsid_inprogress(disk_super); | |
936 | if (!fs_devices) | |
937 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
938 | } else { | |
939 | fs_devices = find_fsid_changed(disk_super); | |
940 | } | |
7a62d0f0 NB |
941 | } else if (has_metadata_uuid) { |
942 | fs_devices = find_fsid(disk_super->fsid, | |
943 | disk_super->metadata_uuid); | |
944 | } else { | |
7239ff4b | 945 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
946 | } |
947 | ||
8a4b83cc | 948 | |
8a4b83cc | 949 | if (!fs_devices) { |
7239ff4b NB |
950 | if (has_metadata_uuid) |
951 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
952 | disk_super->metadata_uuid); | |
953 | else | |
954 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
955 | ||
2208a378 | 956 | if (IS_ERR(fs_devices)) |
e124ece5 | 957 | return ERR_CAST(fs_devices); |
2208a378 | 958 | |
92900e51 AV |
959 | fs_devices->fsid_change = fsid_change_in_progress; |
960 | ||
9c6d173e | 961 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 962 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 963 | |
8a4b83cc CM |
964 | device = NULL; |
965 | } else { | |
9c6d173e | 966 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
967 | device = btrfs_find_device(fs_devices, devid, |
968 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
969 | |
970 | /* | |
971 | * If this disk has been pulled into an fs devices created by | |
972 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
973 | * metadata_uuid/fsid values of the fs_devices. | |
974 | */ | |
975 | if (has_metadata_uuid && fs_devices->fsid_change && | |
976 | found_transid > fs_devices->latest_generation) { | |
977 | memcpy(fs_devices->fsid, disk_super->fsid, | |
978 | BTRFS_FSID_SIZE); | |
979 | memcpy(fs_devices->metadata_uuid, | |
980 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
981 | ||
982 | fs_devices->fsid_change = false; | |
983 | } | |
8a4b83cc | 984 | } |
443f24fe | 985 | |
8a4b83cc | 986 | if (!device) { |
9c6d173e AJ |
987 | if (fs_devices->opened) { |
988 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 989 | return ERR_PTR(-EBUSY); |
9c6d173e | 990 | } |
2b82032c | 991 | |
12bd2fc0 ID |
992 | device = btrfs_alloc_device(NULL, &devid, |
993 | disk_super->dev_item.uuid); | |
994 | if (IS_ERR(device)) { | |
9c6d173e | 995 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 996 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 997 | return device; |
8a4b83cc | 998 | } |
606686ee JB |
999 | |
1000 | name = rcu_string_strdup(path, GFP_NOFS); | |
1001 | if (!name) { | |
a425f9d4 | 1002 | btrfs_free_device(device); |
9c6d173e | 1003 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1004 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 1005 | } |
606686ee | 1006 | rcu_assign_pointer(device->name, name); |
90519d66 | 1007 | |
1f78160c | 1008 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 1009 | fs_devices->num_devices++; |
e5e9a520 | 1010 | |
2b82032c | 1011 | device->fs_devices = fs_devices; |
4306a974 | 1012 | *new_device_added = true; |
327f18cc AJ |
1013 | |
1014 | if (disk_super->label[0]) | |
1015 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
1016 | disk_super->label, devid, found_transid, path); | |
1017 | else | |
1018 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
1019 | disk_super->fsid, devid, found_transid, path); | |
1020 | ||
606686ee | 1021 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
1022 | /* |
1023 | * When FS is already mounted. | |
1024 | * 1. If you are here and if the device->name is NULL that | |
1025 | * means this device was missing at time of FS mount. | |
1026 | * 2. If you are here and if the device->name is different | |
1027 | * from 'path' that means either | |
1028 | * a. The same device disappeared and reappeared with | |
1029 | * different name. or | |
1030 | * b. The missing-disk-which-was-replaced, has | |
1031 | * reappeared now. | |
1032 | * | |
1033 | * We must allow 1 and 2a above. But 2b would be a spurious | |
1034 | * and unintentional. | |
1035 | * | |
1036 | * Further in case of 1 and 2a above, the disk at 'path' | |
1037 | * would have missed some transaction when it was away and | |
1038 | * in case of 2a the stale bdev has to be updated as well. | |
1039 | * 2b must not be allowed at all time. | |
1040 | */ | |
1041 | ||
1042 | /* | |
0f23ae74 CM |
1043 | * For now, we do allow update to btrfs_fs_device through the |
1044 | * btrfs dev scan cli after FS has been mounted. We're still | |
1045 | * tracking a problem where systems fail mount by subvolume id | |
1046 | * when we reject replacement on a mounted FS. | |
b96de000 | 1047 | */ |
0f23ae74 | 1048 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
1049 | /* |
1050 | * That is if the FS is _not_ mounted and if you | |
1051 | * are here, that means there is more than one | |
1052 | * disk with same uuid and devid.We keep the one | |
1053 | * with larger generation number or the last-in if | |
1054 | * generation are equal. | |
1055 | */ | |
9c6d173e | 1056 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1057 | return ERR_PTR(-EEXIST); |
77bdae4d | 1058 | } |
b96de000 | 1059 | |
a9261d41 AJ |
1060 | /* |
1061 | * We are going to replace the device path for a given devid, | |
1062 | * make sure it's the same device if the device is mounted | |
1063 | */ | |
1064 | if (device->bdev) { | |
1065 | struct block_device *path_bdev; | |
1066 | ||
1067 | path_bdev = lookup_bdev(path); | |
1068 | if (IS_ERR(path_bdev)) { | |
1069 | mutex_unlock(&fs_devices->device_list_mutex); | |
1070 | return ERR_CAST(path_bdev); | |
1071 | } | |
1072 | ||
1073 | if (device->bdev != path_bdev) { | |
1074 | bdput(path_bdev); | |
1075 | mutex_unlock(&fs_devices->device_list_mutex); | |
1076 | btrfs_warn_in_rcu(device->fs_info, | |
1077 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1078 | disk_super->fsid, devid, | |
1079 | rcu_str_deref(device->name), path); | |
1080 | return ERR_PTR(-EEXIST); | |
1081 | } | |
1082 | bdput(path_bdev); | |
1083 | btrfs_info_in_rcu(device->fs_info, | |
1084 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1085 | disk_super->fsid, devid, | |
1086 | rcu_str_deref(device->name), path); | |
1087 | } | |
1088 | ||
606686ee | 1089 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1090 | if (!name) { |
1091 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1092 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1093 | } |
606686ee JB |
1094 | rcu_string_free(device->name); |
1095 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1096 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1097 | fs_devices->missing_devices--; |
e6e674bd | 1098 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1099 | } |
8a4b83cc CM |
1100 | } |
1101 | ||
77bdae4d AJ |
1102 | /* |
1103 | * Unmount does not free the btrfs_device struct but would zero | |
1104 | * generation along with most of the other members. So just update | |
1105 | * it back. We need it to pick the disk with largest generation | |
1106 | * (as above). | |
1107 | */ | |
d1a63002 | 1108 | if (!fs_devices->opened) { |
77bdae4d | 1109 | device->generation = found_transid; |
d1a63002 NB |
1110 | fs_devices->latest_generation = max_t(u64, found_transid, |
1111 | fs_devices->latest_generation); | |
1112 | } | |
77bdae4d | 1113 | |
f2788d2f AJ |
1114 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1115 | ||
9c6d173e | 1116 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1117 | return device; |
8a4b83cc CM |
1118 | } |
1119 | ||
e4404d6e YZ |
1120 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1121 | { | |
1122 | struct btrfs_fs_devices *fs_devices; | |
1123 | struct btrfs_device *device; | |
1124 | struct btrfs_device *orig_dev; | |
1125 | ||
7239ff4b | 1126 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1127 | if (IS_ERR(fs_devices)) |
1128 | return fs_devices; | |
e4404d6e | 1129 | |
adbbb863 | 1130 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1131 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1132 | |
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 | ||
228a73ab AJ |
1449 | int btrfs_forget_devices(const char *path) |
1450 | { | |
1451 | int ret; | |
1452 | ||
1453 | mutex_lock(&uuid_mutex); | |
1454 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1455 | mutex_unlock(&uuid_mutex); | |
1456 | ||
1457 | return ret; | |
1458 | } | |
1459 | ||
6f60cbd3 DS |
1460 | /* |
1461 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1462 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1463 | * is read via pagecache | |
1464 | */ | |
36350e95 GJ |
1465 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1466 | void *holder) | |
8a4b83cc CM |
1467 | { |
1468 | struct btrfs_super_block *disk_super; | |
4306a974 | 1469 | bool new_device_added = false; |
36350e95 | 1470 | struct btrfs_device *device = NULL; |
8a4b83cc | 1471 | struct block_device *bdev; |
6f60cbd3 | 1472 | struct page *page; |
6f60cbd3 | 1473 | u64 bytenr; |
8a4b83cc | 1474 | |
899f9307 DS |
1475 | lockdep_assert_held(&uuid_mutex); |
1476 | ||
6f60cbd3 DS |
1477 | /* |
1478 | * we would like to check all the supers, but that would make | |
1479 | * a btrfs mount succeed after a mkfs from a different FS. | |
1480 | * So, we need to add a special mount option to scan for | |
1481 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1482 | */ | |
1483 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1484 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1485 | |
1486 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1487 | if (IS_ERR(bdev)) |
36350e95 | 1488 | return ERR_CAST(bdev); |
6f60cbd3 | 1489 | |
05a5c55d | 1490 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1491 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1492 | goto error_bdev_put; |
05a5c55d | 1493 | } |
6f60cbd3 | 1494 | |
4306a974 | 1495 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1496 | if (!IS_ERR(device)) { |
4306a974 AJ |
1497 | if (new_device_added) |
1498 | btrfs_free_stale_devices(path, device); | |
1499 | } | |
6f60cbd3 | 1500 | |
6cf86a00 | 1501 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1502 | |
1503 | error_bdev_put: | |
d4d77629 | 1504 | blkdev_put(bdev, flags); |
b6ed73bc | 1505 | |
36350e95 | 1506 | return device; |
8a4b83cc | 1507 | } |
0b86a832 | 1508 | |
499f377f | 1509 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1510 | struct btrfs_device *device, |
1511 | u64 *start, u64 len) | |
1512 | { | |
fb456252 | 1513 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1514 | struct extent_map *em; |
499f377f | 1515 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1516 | int ret = 0; |
1b984508 | 1517 | u64 physical_start = *start; |
6df9a95e | 1518 | |
499f377f JM |
1519 | if (transaction) |
1520 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1521 | again: |
1522 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1523 | struct map_lookup *map; |
1524 | int i; | |
1525 | ||
95617d69 | 1526 | map = em->map_lookup; |
6df9a95e | 1527 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1528 | u64 end; |
1529 | ||
6df9a95e JB |
1530 | if (map->stripes[i].dev != device) |
1531 | continue; | |
1b984508 | 1532 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1533 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1534 | physical_start) |
6df9a95e | 1535 | continue; |
c152b63e FM |
1536 | /* |
1537 | * Make sure that while processing the pinned list we do | |
1538 | * not override our *start with a lower value, because | |
1539 | * we can have pinned chunks that fall within this | |
1540 | * device hole and that have lower physical addresses | |
1541 | * than the pending chunks we processed before. If we | |
1542 | * do not take this special care we can end up getting | |
1543 | * 2 pending chunks that start at the same physical | |
1544 | * device offsets because the end offset of a pinned | |
1545 | * chunk can be equal to the start offset of some | |
1546 | * pending chunk. | |
1547 | */ | |
1548 | end = map->stripes[i].physical + em->orig_block_len; | |
1549 | if (end > *start) { | |
1550 | *start = end; | |
1551 | ret = 1; | |
1552 | } | |
6df9a95e JB |
1553 | } |
1554 | } | |
499f377f JM |
1555 | if (search_list != &fs_info->pinned_chunks) { |
1556 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1557 | goto again; |
1558 | } | |
6df9a95e JB |
1559 | |
1560 | return ret; | |
1561 | } | |
1562 | ||
1563 | ||
0b86a832 | 1564 | /* |
499f377f JM |
1565 | * find_free_dev_extent_start - find free space in the specified device |
1566 | * @device: the device which we search the free space in | |
1567 | * @num_bytes: the size of the free space that we need | |
1568 | * @search_start: the position from which to begin the search | |
1569 | * @start: store the start of the free space. | |
1570 | * @len: the size of the free space. that we find, or the size | |
1571 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1572 | * |
0b86a832 CM |
1573 | * this uses a pretty simple search, the expectation is that it is |
1574 | * called very infrequently and that a given device has a small number | |
1575 | * of extents | |
7bfc837d MX |
1576 | * |
1577 | * @start is used to store the start of the free space if we find. But if we | |
1578 | * don't find suitable free space, it will be used to store the start position | |
1579 | * of the max free space. | |
1580 | * | |
1581 | * @len is used to store the size of the free space that we find. | |
1582 | * But if we don't find suitable free space, it is used to store the size of | |
1583 | * the max free space. | |
0b86a832 | 1584 | */ |
499f377f JM |
1585 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1586 | struct btrfs_device *device, u64 num_bytes, | |
1587 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1588 | { |
0b246afa JM |
1589 | struct btrfs_fs_info *fs_info = device->fs_info; |
1590 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1591 | struct btrfs_key key; |
7bfc837d | 1592 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1593 | struct btrfs_path *path; |
7bfc837d MX |
1594 | u64 hole_size; |
1595 | u64 max_hole_start; | |
1596 | u64 max_hole_size; | |
1597 | u64 extent_end; | |
0b86a832 CM |
1598 | u64 search_end = device->total_bytes; |
1599 | int ret; | |
7bfc837d | 1600 | int slot; |
0b86a832 | 1601 | struct extent_buffer *l; |
8cdc7c5b FM |
1602 | |
1603 | /* | |
1604 | * We don't want to overwrite the superblock on the drive nor any area | |
1605 | * used by the boot loader (grub for example), so we make sure to start | |
1606 | * at an offset of at least 1MB. | |
1607 | */ | |
0d0c71b3 | 1608 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1609 | |
6df9a95e JB |
1610 | path = btrfs_alloc_path(); |
1611 | if (!path) | |
1612 | return -ENOMEM; | |
f2ab7618 | 1613 | |
7bfc837d MX |
1614 | max_hole_start = search_start; |
1615 | max_hole_size = 0; | |
1616 | ||
f2ab7618 | 1617 | again: |
401e29c1 AJ |
1618 | if (search_start >= search_end || |
1619 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1620 | ret = -ENOSPC; |
6df9a95e | 1621 | goto out; |
7bfc837d MX |
1622 | } |
1623 | ||
e4058b54 | 1624 | path->reada = READA_FORWARD; |
6df9a95e JB |
1625 | path->search_commit_root = 1; |
1626 | path->skip_locking = 1; | |
7bfc837d | 1627 | |
0b86a832 CM |
1628 | key.objectid = device->devid; |
1629 | key.offset = search_start; | |
1630 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1631 | |
125ccb0a | 1632 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1633 | if (ret < 0) |
7bfc837d | 1634 | goto out; |
1fcbac58 YZ |
1635 | if (ret > 0) { |
1636 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1637 | if (ret < 0) | |
7bfc837d | 1638 | goto out; |
1fcbac58 | 1639 | } |
7bfc837d | 1640 | |
0b86a832 CM |
1641 | while (1) { |
1642 | l = path->nodes[0]; | |
1643 | slot = path->slots[0]; | |
1644 | if (slot >= btrfs_header_nritems(l)) { | |
1645 | ret = btrfs_next_leaf(root, path); | |
1646 | if (ret == 0) | |
1647 | continue; | |
1648 | if (ret < 0) | |
7bfc837d MX |
1649 | goto out; |
1650 | ||
1651 | break; | |
0b86a832 CM |
1652 | } |
1653 | btrfs_item_key_to_cpu(l, &key, slot); | |
1654 | ||
1655 | if (key.objectid < device->devid) | |
1656 | goto next; | |
1657 | ||
1658 | if (key.objectid > device->devid) | |
7bfc837d | 1659 | break; |
0b86a832 | 1660 | |
962a298f | 1661 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1662 | goto next; |
9779b72f | 1663 | |
7bfc837d MX |
1664 | if (key.offset > search_start) { |
1665 | hole_size = key.offset - search_start; | |
9779b72f | 1666 | |
6df9a95e JB |
1667 | /* |
1668 | * Have to check before we set max_hole_start, otherwise | |
1669 | * we could end up sending back this offset anyway. | |
1670 | */ | |
499f377f | 1671 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1672 | &search_start, |
1b984508 FL |
1673 | hole_size)) { |
1674 | if (key.offset >= search_start) { | |
1675 | hole_size = key.offset - search_start; | |
1676 | } else { | |
1677 | WARN_ON_ONCE(1); | |
1678 | hole_size = 0; | |
1679 | } | |
1680 | } | |
6df9a95e | 1681 | |
7bfc837d MX |
1682 | if (hole_size > max_hole_size) { |
1683 | max_hole_start = search_start; | |
1684 | max_hole_size = hole_size; | |
1685 | } | |
9779b72f | 1686 | |
7bfc837d MX |
1687 | /* |
1688 | * If this free space is greater than which we need, | |
1689 | * it must be the max free space that we have found | |
1690 | * until now, so max_hole_start must point to the start | |
1691 | * of this free space and the length of this free space | |
1692 | * is stored in max_hole_size. Thus, we return | |
1693 | * max_hole_start and max_hole_size and go back to the | |
1694 | * caller. | |
1695 | */ | |
1696 | if (hole_size >= num_bytes) { | |
1697 | ret = 0; | |
1698 | goto out; | |
0b86a832 CM |
1699 | } |
1700 | } | |
0b86a832 | 1701 | |
0b86a832 | 1702 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1703 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1704 | dev_extent); | |
1705 | if (extent_end > search_start) | |
1706 | search_start = extent_end; | |
0b86a832 CM |
1707 | next: |
1708 | path->slots[0]++; | |
1709 | cond_resched(); | |
1710 | } | |
0b86a832 | 1711 | |
38c01b96 | 1712 | /* |
1713 | * At this point, search_start should be the end of | |
1714 | * allocated dev extents, and when shrinking the device, | |
1715 | * search_end may be smaller than search_start. | |
1716 | */ | |
f2ab7618 | 1717 | if (search_end > search_start) { |
38c01b96 | 1718 | hole_size = search_end - search_start; |
1719 | ||
499f377f | 1720 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1721 | hole_size)) { |
1722 | btrfs_release_path(path); | |
1723 | goto again; | |
1724 | } | |
0b86a832 | 1725 | |
f2ab7618 ZL |
1726 | if (hole_size > max_hole_size) { |
1727 | max_hole_start = search_start; | |
1728 | max_hole_size = hole_size; | |
1729 | } | |
6df9a95e JB |
1730 | } |
1731 | ||
7bfc837d | 1732 | /* See above. */ |
f2ab7618 | 1733 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1734 | ret = -ENOSPC; |
1735 | else | |
1736 | ret = 0; | |
1737 | ||
1738 | out: | |
2b82032c | 1739 | btrfs_free_path(path); |
7bfc837d | 1740 | *start = max_hole_start; |
b2117a39 | 1741 | if (len) |
7bfc837d | 1742 | *len = max_hole_size; |
0b86a832 CM |
1743 | return ret; |
1744 | } | |
1745 | ||
499f377f JM |
1746 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1747 | struct btrfs_device *device, u64 num_bytes, | |
1748 | u64 *start, u64 *len) | |
1749 | { | |
499f377f | 1750 | /* FIXME use last free of some kind */ |
499f377f | 1751 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1752 | num_bytes, 0, start, len); |
499f377f JM |
1753 | } |
1754 | ||
b2950863 | 1755 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1756 | struct btrfs_device *device, |
2196d6e8 | 1757 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1758 | { |
0b246afa JM |
1759 | struct btrfs_fs_info *fs_info = device->fs_info; |
1760 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1761 | int ret; |
1762 | struct btrfs_path *path; | |
8f18cf13 | 1763 | struct btrfs_key key; |
a061fc8d CM |
1764 | struct btrfs_key found_key; |
1765 | struct extent_buffer *leaf = NULL; | |
1766 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1767 | |
1768 | path = btrfs_alloc_path(); | |
1769 | if (!path) | |
1770 | return -ENOMEM; | |
1771 | ||
1772 | key.objectid = device->devid; | |
1773 | key.offset = start; | |
1774 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1775 | again: |
8f18cf13 | 1776 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1777 | if (ret > 0) { |
1778 | ret = btrfs_previous_item(root, path, key.objectid, | |
1779 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1780 | if (ret) |
1781 | goto out; | |
a061fc8d CM |
1782 | leaf = path->nodes[0]; |
1783 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1784 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1785 | struct btrfs_dev_extent); | |
1786 | BUG_ON(found_key.offset > start || found_key.offset + | |
1787 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1788 | key = found_key; |
1789 | btrfs_release_path(path); | |
1790 | goto again; | |
a061fc8d CM |
1791 | } else if (ret == 0) { |
1792 | leaf = path->nodes[0]; | |
1793 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1794 | struct btrfs_dev_extent); | |
79787eaa | 1795 | } else { |
0b246afa | 1796 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1797 | goto out; |
a061fc8d | 1798 | } |
8f18cf13 | 1799 | |
2196d6e8 MX |
1800 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1801 | ||
8f18cf13 | 1802 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1803 | if (ret) { |
0b246afa JM |
1804 | btrfs_handle_fs_error(fs_info, ret, |
1805 | "Failed to remove dev extent item"); | |
13212b54 | 1806 | } else { |
3204d33c | 1807 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1808 | } |
b0b802d7 | 1809 | out: |
8f18cf13 CM |
1810 | btrfs_free_path(path); |
1811 | return ret; | |
1812 | } | |
1813 | ||
48a3b636 ES |
1814 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1815 | struct btrfs_device *device, | |
48a3b636 | 1816 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1817 | { |
1818 | int ret; | |
1819 | struct btrfs_path *path; | |
0b246afa JM |
1820 | struct btrfs_fs_info *fs_info = device->fs_info; |
1821 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1822 | struct btrfs_dev_extent *extent; |
1823 | struct extent_buffer *leaf; | |
1824 | struct btrfs_key key; | |
1825 | ||
e12c9621 | 1826 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1827 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1828 | path = btrfs_alloc_path(); |
1829 | if (!path) | |
1830 | return -ENOMEM; | |
1831 | ||
0b86a832 | 1832 | key.objectid = device->devid; |
2b82032c | 1833 | key.offset = start; |
0b86a832 CM |
1834 | key.type = BTRFS_DEV_EXTENT_KEY; |
1835 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1836 | sizeof(*extent)); | |
2cdcecbc MF |
1837 | if (ret) |
1838 | goto out; | |
0b86a832 CM |
1839 | |
1840 | leaf = path->nodes[0]; | |
1841 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1842 | struct btrfs_dev_extent); | |
b5d9071c NB |
1843 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1844 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1845 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1846 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1847 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1848 | ||
0b86a832 CM |
1849 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1850 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1851 | out: |
0b86a832 CM |
1852 | btrfs_free_path(path); |
1853 | return ret; | |
1854 | } | |
1855 | ||
6df9a95e | 1856 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1857 | { |
6df9a95e JB |
1858 | struct extent_map_tree *em_tree; |
1859 | struct extent_map *em; | |
1860 | struct rb_node *n; | |
1861 | u64 ret = 0; | |
0b86a832 | 1862 | |
6df9a95e JB |
1863 | em_tree = &fs_info->mapping_tree.map_tree; |
1864 | read_lock(&em_tree->lock); | |
07e1ce09 | 1865 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1866 | if (n) { |
1867 | em = rb_entry(n, struct extent_map, rb_node); | |
1868 | ret = em->start + em->len; | |
0b86a832 | 1869 | } |
6df9a95e JB |
1870 | read_unlock(&em_tree->lock); |
1871 | ||
0b86a832 CM |
1872 | return ret; |
1873 | } | |
1874 | ||
53f10659 ID |
1875 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1876 | u64 *devid_ret) | |
0b86a832 CM |
1877 | { |
1878 | int ret; | |
1879 | struct btrfs_key key; | |
1880 | struct btrfs_key found_key; | |
2b82032c YZ |
1881 | struct btrfs_path *path; |
1882 | ||
2b82032c YZ |
1883 | path = btrfs_alloc_path(); |
1884 | if (!path) | |
1885 | return -ENOMEM; | |
0b86a832 CM |
1886 | |
1887 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1888 | key.type = BTRFS_DEV_ITEM_KEY; | |
1889 | key.offset = (u64)-1; | |
1890 | ||
53f10659 | 1891 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1892 | if (ret < 0) |
1893 | goto error; | |
1894 | ||
79787eaa | 1895 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1896 | |
53f10659 ID |
1897 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1898 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1899 | BTRFS_DEV_ITEM_KEY); |
1900 | if (ret) { | |
53f10659 | 1901 | *devid_ret = 1; |
0b86a832 CM |
1902 | } else { |
1903 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1904 | path->slots[0]); | |
53f10659 | 1905 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1906 | } |
1907 | ret = 0; | |
1908 | error: | |
2b82032c | 1909 | btrfs_free_path(path); |
0b86a832 CM |
1910 | return ret; |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * the device information is stored in the chunk root | |
1915 | * the btrfs_device struct should be fully filled in | |
1916 | */ | |
c74a0b02 | 1917 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1918 | struct btrfs_device *device) |
0b86a832 CM |
1919 | { |
1920 | int ret; | |
1921 | struct btrfs_path *path; | |
1922 | struct btrfs_dev_item *dev_item; | |
1923 | struct extent_buffer *leaf; | |
1924 | struct btrfs_key key; | |
1925 | unsigned long ptr; | |
0b86a832 | 1926 | |
0b86a832 CM |
1927 | path = btrfs_alloc_path(); |
1928 | if (!path) | |
1929 | return -ENOMEM; | |
1930 | ||
0b86a832 CM |
1931 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1932 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1933 | key.offset = device->devid; |
0b86a832 | 1934 | |
8e87e856 NB |
1935 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1936 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1937 | if (ret) |
1938 | goto out; | |
1939 | ||
1940 | leaf = path->nodes[0]; | |
1941 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1942 | ||
1943 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1944 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1945 | btrfs_set_device_type(leaf, dev_item, device->type); |
1946 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1947 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1948 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1949 | btrfs_set_device_total_bytes(leaf, dev_item, |
1950 | btrfs_device_get_disk_total_bytes(device)); | |
1951 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1952 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1953 | btrfs_set_device_group(leaf, dev_item, 0); |
1954 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1955 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1956 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1957 | |
410ba3a2 | 1958 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1959 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1960 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1961 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1962 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1963 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1964 | |
2b82032c | 1965 | ret = 0; |
0b86a832 CM |
1966 | out: |
1967 | btrfs_free_path(path); | |
1968 | return ret; | |
1969 | } | |
8f18cf13 | 1970 | |
5a1972bd QW |
1971 | /* |
1972 | * Function to update ctime/mtime for a given device path. | |
1973 | * Mainly used for ctime/mtime based probe like libblkid. | |
1974 | */ | |
da353f6b | 1975 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1976 | { |
1977 | struct file *filp; | |
1978 | ||
1979 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1980 | if (IS_ERR(filp)) |
5a1972bd QW |
1981 | return; |
1982 | file_update_time(filp); | |
1983 | filp_close(filp, NULL); | |
5a1972bd QW |
1984 | } |
1985 | ||
5b4aacef | 1986 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1987 | struct btrfs_device *device) |
1988 | { | |
5b4aacef | 1989 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1990 | int ret; |
1991 | struct btrfs_path *path; | |
a061fc8d | 1992 | struct btrfs_key key; |
a061fc8d CM |
1993 | struct btrfs_trans_handle *trans; |
1994 | ||
a061fc8d CM |
1995 | path = btrfs_alloc_path(); |
1996 | if (!path) | |
1997 | return -ENOMEM; | |
1998 | ||
a22285a6 | 1999 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
2000 | if (IS_ERR(trans)) { |
2001 | btrfs_free_path(path); | |
2002 | return PTR_ERR(trans); | |
2003 | } | |
a061fc8d CM |
2004 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2005 | key.type = BTRFS_DEV_ITEM_KEY; | |
2006 | key.offset = device->devid; | |
2007 | ||
2008 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
2009 | if (ret) { |
2010 | if (ret > 0) | |
2011 | ret = -ENOENT; | |
2012 | btrfs_abort_transaction(trans, ret); | |
2013 | btrfs_end_transaction(trans); | |
a061fc8d CM |
2014 | goto out; |
2015 | } | |
2016 | ||
2017 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
2018 | if (ret) { |
2019 | btrfs_abort_transaction(trans, ret); | |
2020 | btrfs_end_transaction(trans); | |
2021 | } | |
2022 | ||
a061fc8d CM |
2023 | out: |
2024 | btrfs_free_path(path); | |
5e9f2ad5 NB |
2025 | if (!ret) |
2026 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
2027 | return ret; |
2028 | } | |
2029 | ||
3cc31a0d DS |
2030 | /* |
2031 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
2032 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
2033 | * replace. | |
2034 | */ | |
2035 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
2036 | u64 num_devices) | |
a061fc8d | 2037 | { |
a061fc8d | 2038 | u64 all_avail; |
de98ced9 | 2039 | unsigned seq; |
418775a2 | 2040 | int i; |
a061fc8d | 2041 | |
de98ced9 | 2042 | do { |
bd45ffbc | 2043 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2044 | |
bd45ffbc AJ |
2045 | all_avail = fs_info->avail_data_alloc_bits | |
2046 | fs_info->avail_system_alloc_bits | | |
2047 | fs_info->avail_metadata_alloc_bits; | |
2048 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2049 | |
418775a2 | 2050 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2051 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2052 | continue; |
a061fc8d | 2053 | |
418775a2 | 2054 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2055 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2056 | |
418775a2 DS |
2057 | if (ret) |
2058 | return ret; | |
2059 | } | |
53b381b3 DW |
2060 | } |
2061 | ||
bd45ffbc | 2062 | return 0; |
f1fa7f26 AJ |
2063 | } |
2064 | ||
c9162bdf OS |
2065 | static struct btrfs_device * btrfs_find_next_active_device( |
2066 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2067 | { |
2b82032c | 2068 | struct btrfs_device *next_device; |
88acff64 AJ |
2069 | |
2070 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2071 | if (next_device != device && | |
e6e674bd AJ |
2072 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2073 | && next_device->bdev) | |
88acff64 AJ |
2074 | return next_device; |
2075 | } | |
2076 | ||
2077 | return NULL; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2082 | * and replace it with the provided or the next active device, in the context | |
2083 | * where this function called, there should be always be another device (or | |
2084 | * this_dev) which is active. | |
2085 | */ | |
d6507cf1 NB |
2086 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2087 | struct btrfs_device *this_dev) | |
88acff64 | 2088 | { |
d6507cf1 | 2089 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2090 | struct btrfs_device *next_device; |
2091 | ||
2092 | if (this_dev) | |
2093 | next_device = this_dev; | |
2094 | else | |
2095 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2096 | device); | |
2097 | ASSERT(next_device); | |
2098 | ||
2099 | if (fs_info->sb->s_bdev && | |
2100 | (fs_info->sb->s_bdev == device->bdev)) | |
2101 | fs_info->sb->s_bdev = next_device->bdev; | |
2102 | ||
2103 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2104 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2105 | } | |
2106 | ||
1da73967 AJ |
2107 | /* |
2108 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2109 | * currently replaced. | |
2110 | */ | |
2111 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2112 | { | |
2113 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2114 | ||
cb5583dd | 2115 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2116 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2117 | ASSERT(num_devices > 1); | |
2118 | num_devices--; | |
2119 | } | |
cb5583dd | 2120 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2121 | |
2122 | return num_devices; | |
2123 | } | |
2124 | ||
da353f6b DS |
2125 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2126 | u64 devid) | |
f1fa7f26 AJ |
2127 | { |
2128 | struct btrfs_device *device; | |
1f78160c | 2129 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2130 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2131 | u64 num_devices; |
a061fc8d CM |
2132 | int ret = 0; |
2133 | ||
a061fc8d CM |
2134 | mutex_lock(&uuid_mutex); |
2135 | ||
1da73967 | 2136 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2137 | |
0b246afa | 2138 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2139 | if (ret) |
a061fc8d | 2140 | goto out; |
a061fc8d | 2141 | |
a27a94c2 NB |
2142 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2143 | ||
2144 | if (IS_ERR(device)) { | |
2145 | if (PTR_ERR(device) == -ENOENT && | |
2146 | strcmp(device_path, "missing") == 0) | |
2147 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2148 | else | |
2149 | ret = PTR_ERR(device); | |
53b381b3 | 2150 | goto out; |
a27a94c2 | 2151 | } |
dfe25020 | 2152 | |
eede2bf3 OS |
2153 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2154 | btrfs_warn_in_rcu(fs_info, | |
2155 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2156 | rcu_str_deref(device->name), device->devid); | |
2157 | ret = -ETXTBSY; | |
2158 | goto out; | |
2159 | } | |
2160 | ||
401e29c1 | 2161 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2162 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2163 | goto out; |
63a212ab SB |
2164 | } |
2165 | ||
ebbede42 AJ |
2166 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2167 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2168 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2169 | goto out; |
2b82032c YZ |
2170 | } |
2171 | ||
ebbede42 | 2172 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2173 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2174 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2175 | device->fs_devices->rw_devices--; |
34441361 | 2176 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2177 | } |
a061fc8d | 2178 | |
d7901554 | 2179 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2180 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2181 | mutex_lock(&uuid_mutex); |
a061fc8d | 2182 | if (ret) |
9b3517e9 | 2183 | goto error_undo; |
a061fc8d | 2184 | |
63a212ab SB |
2185 | /* |
2186 | * TODO: the superblock still includes this device in its num_devices | |
2187 | * counter although write_all_supers() is not locked out. This | |
2188 | * could give a filesystem state which requires a degraded mount. | |
2189 | */ | |
0b246afa | 2190 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 2191 | if (ret) |
9b3517e9 | 2192 | goto error_undo; |
a061fc8d | 2193 | |
e12c9621 | 2194 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 2195 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
2196 | |
2197 | /* | |
2198 | * the device list mutex makes sure that we don't change | |
2199 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2200 | * the device supers. Whoever is writing all supers, should |
2201 | * lock the device list mutex before getting the number of | |
2202 | * devices in the super block (super_copy). Conversely, | |
2203 | * whoever updates the number of devices in the super block | |
2204 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2205 | */ |
1f78160c | 2206 | |
41a52a0f AJ |
2207 | /* |
2208 | * In normal cases the cur_devices == fs_devices. But in case | |
2209 | * of deleting a seed device, the cur_devices should point to | |
2210 | * its own fs_devices listed under the fs_devices->seed. | |
2211 | */ | |
1f78160c | 2212 | cur_devices = device->fs_devices; |
b5185197 | 2213 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2214 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2215 | |
41a52a0f AJ |
2216 | cur_devices->num_devices--; |
2217 | cur_devices->total_devices--; | |
b4993e64 AJ |
2218 | /* Update total_devices of the parent fs_devices if it's seed */ |
2219 | if (cur_devices != fs_devices) | |
2220 | fs_devices->total_devices--; | |
2b82032c | 2221 | |
e6e674bd | 2222 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2223 | cur_devices->missing_devices--; |
cd02dca5 | 2224 | |
d6507cf1 | 2225 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2226 | |
0bfaa9c5 | 2227 | if (device->bdev) { |
41a52a0f | 2228 | cur_devices->open_devices--; |
0bfaa9c5 | 2229 | /* remove sysfs entry */ |
b5185197 | 2230 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2231 | } |
99994cde | 2232 | |
0b246afa JM |
2233 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2234 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2235 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2236 | |
cea67ab9 JM |
2237 | /* |
2238 | * at this point, the device is zero sized and detached from | |
2239 | * the devices list. All that's left is to zero out the old | |
2240 | * supers and free the device. | |
2241 | */ | |
ebbede42 | 2242 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2243 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2244 | ||
2245 | btrfs_close_bdev(device); | |
f06c5965 | 2246 | call_rcu(&device->rcu, free_device_rcu); |
cea67ab9 | 2247 | |
1f78160c | 2248 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2249 | while (fs_devices) { |
8321cf25 RS |
2250 | if (fs_devices->seed == cur_devices) { |
2251 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2252 | break; |
8321cf25 | 2253 | } |
e4404d6e | 2254 | fs_devices = fs_devices->seed; |
2b82032c | 2255 | } |
1f78160c | 2256 | cur_devices->seed = NULL; |
0226e0eb | 2257 | close_fs_devices(cur_devices); |
1f78160c | 2258 | free_fs_devices(cur_devices); |
2b82032c YZ |
2259 | } |
2260 | ||
a061fc8d CM |
2261 | out: |
2262 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2263 | return ret; |
24fc572f | 2264 | |
9b3517e9 | 2265 | error_undo: |
ebbede42 | 2266 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2267 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2268 | list_add(&device->dev_alloc_list, |
b5185197 | 2269 | &fs_devices->alloc_list); |
c3929c36 | 2270 | device->fs_devices->rw_devices++; |
34441361 | 2271 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2272 | } |
24fc572f | 2273 | goto out; |
a061fc8d CM |
2274 | } |
2275 | ||
68a9db5f | 2276 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2277 | { |
d51908ce AJ |
2278 | struct btrfs_fs_devices *fs_devices; |
2279 | ||
68a9db5f | 2280 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2281 | |
25e8e911 AJ |
2282 | /* |
2283 | * in case of fs with no seed, srcdev->fs_devices will point | |
2284 | * to fs_devices of fs_info. However when the dev being replaced is | |
2285 | * a seed dev it will point to the seed's local fs_devices. In short | |
2286 | * srcdev will have its correct fs_devices in both the cases. | |
2287 | */ | |
2288 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2289 | |
e93c89c1 | 2290 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2291 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2292 | fs_devices->num_devices--; |
e6e674bd | 2293 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2294 | fs_devices->missing_devices--; |
e93c89c1 | 2295 | |
ebbede42 | 2296 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2297 | fs_devices->rw_devices--; |
1357272f | 2298 | |
82372bc8 | 2299 | if (srcdev->bdev) |
d51908ce | 2300 | fs_devices->open_devices--; |
084b6e7c QW |
2301 | } |
2302 | ||
2303 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2304 | struct btrfs_device *srcdev) | |
2305 | { | |
2306 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2307 | |
ebbede42 | 2308 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2309 | /* zero out the old super if it is writable */ |
2310 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2311 | } | |
14238819 AJ |
2312 | |
2313 | btrfs_close_bdev(srcdev); | |
f06c5965 | 2314 | call_rcu(&srcdev->rcu, free_device_rcu); |
94d5f0c2 | 2315 | |
94d5f0c2 AJ |
2316 | /* if this is no devs we rather delete the fs_devices */ |
2317 | if (!fs_devices->num_devices) { | |
2318 | struct btrfs_fs_devices *tmp_fs_devices; | |
2319 | ||
6dd38f81 AJ |
2320 | /* |
2321 | * On a mounted FS, num_devices can't be zero unless it's a | |
2322 | * seed. In case of a seed device being replaced, the replace | |
2323 | * target added to the sprout FS, so there will be no more | |
2324 | * device left under the seed FS. | |
2325 | */ | |
2326 | ASSERT(fs_devices->seeding); | |
2327 | ||
94d5f0c2 AJ |
2328 | tmp_fs_devices = fs_info->fs_devices; |
2329 | while (tmp_fs_devices) { | |
2330 | if (tmp_fs_devices->seed == fs_devices) { | |
2331 | tmp_fs_devices->seed = fs_devices->seed; | |
2332 | break; | |
2333 | } | |
2334 | tmp_fs_devices = tmp_fs_devices->seed; | |
2335 | } | |
2336 | fs_devices->seed = NULL; | |
0226e0eb | 2337 | close_fs_devices(fs_devices); |
8bef8401 | 2338 | free_fs_devices(fs_devices); |
94d5f0c2 | 2339 | } |
e93c89c1 SB |
2340 | } |
2341 | ||
4f5ad7bd | 2342 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2343 | { |
4f5ad7bd | 2344 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2345 | |
e93c89c1 | 2346 | WARN_ON(!tgtdev); |
d9a071f0 | 2347 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2348 | |
d9a071f0 | 2349 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2350 | |
779bf3fe | 2351 | if (tgtdev->bdev) |
d9a071f0 | 2352 | fs_devices->open_devices--; |
779bf3fe | 2353 | |
d9a071f0 | 2354 | fs_devices->num_devices--; |
e93c89c1 | 2355 | |
d6507cf1 | 2356 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2357 | |
e93c89c1 | 2358 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2359 | |
d9a071f0 | 2360 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2361 | |
2362 | /* | |
2363 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2364 | * may lead to a call to btrfs_show_devname() which will try | |
2365 | * to hold device_list_mutex. And here this device | |
2366 | * is already out of device list, so we don't have to hold | |
2367 | * the device_list_mutex lock. | |
2368 | */ | |
2369 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2370 | |
2371 | btrfs_close_bdev(tgtdev); | |
f06c5965 | 2372 | call_rcu(&tgtdev->rcu, free_device_rcu); |
e93c89c1 SB |
2373 | } |
2374 | ||
b444ad46 NB |
2375 | static struct btrfs_device *btrfs_find_device_by_path( |
2376 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2377 | { |
2378 | int ret = 0; | |
2379 | struct btrfs_super_block *disk_super; | |
2380 | u64 devid; | |
2381 | u8 *dev_uuid; | |
2382 | struct block_device *bdev; | |
2383 | struct buffer_head *bh; | |
b444ad46 | 2384 | struct btrfs_device *device; |
7ba15b7d | 2385 | |
7ba15b7d | 2386 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2387 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2388 | if (ret) |
b444ad46 | 2389 | return ERR_PTR(ret); |
7ba15b7d SB |
2390 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2391 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2392 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2393 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2394 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2395 | disk_super->metadata_uuid, true); |
7239ff4b | 2396 | else |
e4319cd9 | 2397 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2398 | disk_super->fsid, true); |
7239ff4b | 2399 | |
7ba15b7d | 2400 | brelse(bh); |
b444ad46 NB |
2401 | if (!device) |
2402 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2403 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2404 | return device; |
7ba15b7d SB |
2405 | } |
2406 | ||
5c5c0df0 DS |
2407 | /* |
2408 | * Lookup a device given by device id, or the path if the id is 0. | |
2409 | */ | |
a27a94c2 | 2410 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2411 | struct btrfs_fs_info *fs_info, u64 devid, |
2412 | const char *device_path) | |
24e0474b | 2413 | { |
a27a94c2 | 2414 | struct btrfs_device *device; |
24e0474b | 2415 | |
5c5c0df0 | 2416 | if (devid) { |
e4319cd9 | 2417 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2418 | NULL, true); |
a27a94c2 NB |
2419 | if (!device) |
2420 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2421 | return device; |
2422 | } | |
2423 | ||
2424 | if (!device_path || !device_path[0]) | |
2425 | return ERR_PTR(-EINVAL); | |
2426 | ||
2427 | if (strcmp(device_path, "missing") == 0) { | |
2428 | /* Find first missing device */ | |
2429 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2430 | dev_list) { | |
2431 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2432 | &device->dev_state) && !device->bdev) | |
2433 | return device; | |
d95a830c | 2434 | } |
6e927ceb | 2435 | return ERR_PTR(-ENOENT); |
24e0474b | 2436 | } |
6e927ceb AJ |
2437 | |
2438 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2439 | } |
2440 | ||
2b82032c YZ |
2441 | /* |
2442 | * does all the dirty work required for changing file system's UUID. | |
2443 | */ | |
2ff7e61e | 2444 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2445 | { |
0b246afa | 2446 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2447 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2448 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2449 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2450 | struct btrfs_device *device; |
2451 | u64 super_flags; | |
2452 | ||
a32bf9a3 | 2453 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2454 | if (!fs_devices->seeding) |
2b82032c YZ |
2455 | return -EINVAL; |
2456 | ||
7239ff4b | 2457 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2458 | if (IS_ERR(seed_devices)) |
2459 | return PTR_ERR(seed_devices); | |
2b82032c | 2460 | |
e4404d6e YZ |
2461 | old_devices = clone_fs_devices(fs_devices); |
2462 | if (IS_ERR(old_devices)) { | |
2463 | kfree(seed_devices); | |
2464 | return PTR_ERR(old_devices); | |
2b82032c | 2465 | } |
e4404d6e | 2466 | |
c4babc5e | 2467 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2468 | |
e4404d6e YZ |
2469 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2470 | seed_devices->opened = 1; | |
2471 | INIT_LIST_HEAD(&seed_devices->devices); | |
2472 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2473 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2474 | |
321a4bf7 | 2475 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2476 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2477 | synchronize_rcu); | |
2196d6e8 MX |
2478 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2479 | device->fs_devices = seed_devices; | |
c9513edb | 2480 | |
34441361 | 2481 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2482 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2483 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2484 | |
2b82032c YZ |
2485 | fs_devices->seeding = 0; |
2486 | fs_devices->num_devices = 0; | |
2487 | fs_devices->open_devices = 0; | |
69611ac8 | 2488 | fs_devices->missing_devices = 0; |
69611ac8 | 2489 | fs_devices->rotating = 0; |
e4404d6e | 2490 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2491 | |
2492 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2493 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2494 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2495 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2496 | |
2b82032c YZ |
2497 | super_flags = btrfs_super_flags(disk_super) & |
2498 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2499 | btrfs_set_super_flags(disk_super, super_flags); | |
2500 | ||
2501 | return 0; | |
2502 | } | |
2503 | ||
2504 | /* | |
01327610 | 2505 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2506 | */ |
2507 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2508 | struct btrfs_fs_info *fs_info) |
2b82032c | 2509 | { |
5b4aacef | 2510 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2511 | struct btrfs_path *path; |
2512 | struct extent_buffer *leaf; | |
2513 | struct btrfs_dev_item *dev_item; | |
2514 | struct btrfs_device *device; | |
2515 | struct btrfs_key key; | |
44880fdc | 2516 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2517 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2518 | u64 devid; | |
2519 | int ret; | |
2520 | ||
2521 | path = btrfs_alloc_path(); | |
2522 | if (!path) | |
2523 | return -ENOMEM; | |
2524 | ||
2b82032c YZ |
2525 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2526 | key.offset = 0; | |
2527 | key.type = BTRFS_DEV_ITEM_KEY; | |
2528 | ||
2529 | while (1) { | |
2530 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2531 | if (ret < 0) | |
2532 | goto error; | |
2533 | ||
2534 | leaf = path->nodes[0]; | |
2535 | next_slot: | |
2536 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2537 | ret = btrfs_next_leaf(root, path); | |
2538 | if (ret > 0) | |
2539 | break; | |
2540 | if (ret < 0) | |
2541 | goto error; | |
2542 | leaf = path->nodes[0]; | |
2543 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2544 | btrfs_release_path(path); |
2b82032c YZ |
2545 | continue; |
2546 | } | |
2547 | ||
2548 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2549 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2550 | key.type != BTRFS_DEV_ITEM_KEY) | |
2551 | break; | |
2552 | ||
2553 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2554 | struct btrfs_dev_item); | |
2555 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2556 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2557 | BTRFS_UUID_SIZE); |
1473b24e | 2558 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2559 | BTRFS_FSID_SIZE); |
e4319cd9 | 2560 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2561 | fs_uuid, true); |
79787eaa | 2562 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2563 | |
2564 | if (device->fs_devices->seeding) { | |
2565 | btrfs_set_device_generation(leaf, dev_item, | |
2566 | device->generation); | |
2567 | btrfs_mark_buffer_dirty(leaf); | |
2568 | } | |
2569 | ||
2570 | path->slots[0]++; | |
2571 | goto next_slot; | |
2572 | } | |
2573 | ret = 0; | |
2574 | error: | |
2575 | btrfs_free_path(path); | |
2576 | return ret; | |
2577 | } | |
2578 | ||
da353f6b | 2579 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2580 | { |
5112febb | 2581 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2582 | struct request_queue *q; |
788f20eb CM |
2583 | struct btrfs_trans_handle *trans; |
2584 | struct btrfs_device *device; | |
2585 | struct block_device *bdev; | |
0b246afa | 2586 | struct super_block *sb = fs_info->sb; |
606686ee | 2587 | struct rcu_string *name; |
5da54bc1 | 2588 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2589 | u64 orig_super_total_bytes; |
2590 | u64 orig_super_num_devices; | |
2b82032c | 2591 | int seeding_dev = 0; |
788f20eb | 2592 | int ret = 0; |
7132a262 | 2593 | bool unlocked = false; |
788f20eb | 2594 | |
5da54bc1 | 2595 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2596 | return -EROFS; |
788f20eb | 2597 | |
a5d16333 | 2598 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2599 | fs_info->bdev_holder); |
7f59203a JB |
2600 | if (IS_ERR(bdev)) |
2601 | return PTR_ERR(bdev); | |
a2135011 | 2602 | |
5da54bc1 | 2603 | if (fs_devices->seeding) { |
2b82032c YZ |
2604 | seeding_dev = 1; |
2605 | down_write(&sb->s_umount); | |
2606 | mutex_lock(&uuid_mutex); | |
2607 | } | |
2608 | ||
8c8bee1d | 2609 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2610 | |
5da54bc1 | 2611 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2612 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2613 | if (device->bdev == bdev) { |
2614 | ret = -EEXIST; | |
d25628bd | 2615 | mutex_unlock( |
5da54bc1 | 2616 | &fs_devices->device_list_mutex); |
2b82032c | 2617 | goto error; |
788f20eb CM |
2618 | } |
2619 | } | |
5da54bc1 | 2620 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2621 | |
0b246afa | 2622 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2623 | if (IS_ERR(device)) { |
788f20eb | 2624 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2625 | ret = PTR_ERR(device); |
2b82032c | 2626 | goto error; |
788f20eb CM |
2627 | } |
2628 | ||
78f2c9e6 | 2629 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2630 | if (!name) { |
2b82032c | 2631 | ret = -ENOMEM; |
5c4cf6c9 | 2632 | goto error_free_device; |
788f20eb | 2633 | } |
606686ee | 2634 | rcu_assign_pointer(device->name, name); |
2b82032c | 2635 | |
a22285a6 | 2636 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2637 | if (IS_ERR(trans)) { |
98d5dc13 | 2638 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2639 | goto error_free_device; |
98d5dc13 TI |
2640 | } |
2641 | ||
d5e2003c | 2642 | q = bdev_get_queue(bdev); |
ebbede42 | 2643 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2644 | device->generation = trans->transid; |
0b246afa JM |
2645 | device->io_width = fs_info->sectorsize; |
2646 | device->io_align = fs_info->sectorsize; | |
2647 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2648 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2649 | fs_info->sectorsize); | |
2cc3c559 | 2650 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2651 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2652 | device->fs_info = fs_info; |
788f20eb | 2653 | device->bdev = bdev; |
e12c9621 | 2654 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2655 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2656 | device->mode = FMODE_EXCL; |
27087f37 | 2657 | device->dev_stats_valid = 1; |
9f6d2510 | 2658 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2659 | |
2b82032c | 2660 | if (seeding_dev) { |
1751e8a6 | 2661 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2662 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2663 | if (ret) { |
2664 | btrfs_abort_transaction(trans, ret); | |
2665 | goto error_trans; | |
2666 | } | |
2b82032c | 2667 | } |
788f20eb | 2668 | |
5da54bc1 | 2669 | device->fs_devices = fs_devices; |
e5e9a520 | 2670 | |
5da54bc1 | 2671 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2672 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2673 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2674 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2675 | fs_devices->num_devices++; | |
2676 | fs_devices->open_devices++; | |
2677 | fs_devices->rw_devices++; | |
2678 | fs_devices->total_devices++; | |
2679 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2680 | |
a5ed45f8 | 2681 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2682 | |
e884f4f0 | 2683 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2684 | fs_devices->rotating = 1; |
c289811c | 2685 | |
39379faa | 2686 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2687 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2688 | round_down(orig_super_total_bytes + device->total_bytes, |
2689 | fs_info->sectorsize)); | |
788f20eb | 2690 | |
39379faa NA |
2691 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2692 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2693 | orig_super_num_devices + 1); | |
0d39376a AJ |
2694 | |
2695 | /* add sysfs device entry */ | |
5da54bc1 | 2696 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2697 | |
2196d6e8 MX |
2698 | /* |
2699 | * we've got more storage, clear any full flags on the space | |
2700 | * infos | |
2701 | */ | |
0b246afa | 2702 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2703 | |
34441361 | 2704 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2705 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2706 | |
2b82032c | 2707 | if (seeding_dev) { |
34441361 | 2708 | mutex_lock(&fs_info->chunk_mutex); |
e4a4dce7 | 2709 | ret = init_first_rw_device(trans, fs_info); |
34441361 | 2710 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2711 | if (ret) { |
66642832 | 2712 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2713 | goto error_sysfs; |
005d6427 | 2714 | } |
2196d6e8 MX |
2715 | } |
2716 | ||
8e87e856 | 2717 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2718 | if (ret) { |
66642832 | 2719 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2720 | goto error_sysfs; |
2196d6e8 MX |
2721 | } |
2722 | ||
2723 | if (seeding_dev) { | |
2724 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2725 | ||
0b246afa | 2726 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2727 | if (ret) { |
66642832 | 2728 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2729 | goto error_sysfs; |
005d6427 | 2730 | } |
b2373f25 AJ |
2731 | |
2732 | /* Sprouting would change fsid of the mounted root, | |
2733 | * so rename the fsid on the sysfs | |
2734 | */ | |
2735 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
de37aa51 | 2736 | fs_info->fs_devices->fsid); |
5da54bc1 | 2737 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2738 | btrfs_warn(fs_info, |
2739 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2740 | } |
2741 | ||
3a45bb20 | 2742 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2743 | |
2b82032c YZ |
2744 | if (seeding_dev) { |
2745 | mutex_unlock(&uuid_mutex); | |
2746 | up_write(&sb->s_umount); | |
7132a262 | 2747 | unlocked = true; |
788f20eb | 2748 | |
79787eaa JM |
2749 | if (ret) /* transaction commit */ |
2750 | return ret; | |
2751 | ||
2ff7e61e | 2752 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2753 | if (ret < 0) |
0b246afa | 2754 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2755 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2756 | trans = btrfs_attach_transaction(root); |
2757 | if (IS_ERR(trans)) { | |
2758 | if (PTR_ERR(trans) == -ENOENT) | |
2759 | return 0; | |
7132a262 AJ |
2760 | ret = PTR_ERR(trans); |
2761 | trans = NULL; | |
2762 | goto error_sysfs; | |
671415b7 | 2763 | } |
3a45bb20 | 2764 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2765 | } |
c9e9f97b | 2766 | |
5a1972bd QW |
2767 | /* Update ctime/mtime for libblkid */ |
2768 | update_dev_time(device_path); | |
2b82032c | 2769 | return ret; |
79787eaa | 2770 | |
d31c32f6 | 2771 | error_sysfs: |
5da54bc1 | 2772 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2773 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2774 | mutex_lock(&fs_info->chunk_mutex); | |
2775 | list_del_rcu(&device->dev_list); | |
2776 | list_del(&device->dev_alloc_list); | |
2777 | fs_info->fs_devices->num_devices--; | |
2778 | fs_info->fs_devices->open_devices--; | |
2779 | fs_info->fs_devices->rw_devices--; | |
2780 | fs_info->fs_devices->total_devices--; | |
2781 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2782 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2783 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2784 | orig_super_total_bytes); | |
2785 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2786 | orig_super_num_devices); | |
2787 | mutex_unlock(&fs_info->chunk_mutex); | |
2788 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2789 | error_trans: |
0af2c4bf | 2790 | if (seeding_dev) |
1751e8a6 | 2791 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2792 | if (trans) |
2793 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2794 | error_free_device: |
a425f9d4 | 2795 | btrfs_free_device(device); |
2b82032c | 2796 | error: |
e525fd89 | 2797 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2798 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2799 | mutex_unlock(&uuid_mutex); |
2800 | up_write(&sb->s_umount); | |
2801 | } | |
c9e9f97b | 2802 | return ret; |
788f20eb CM |
2803 | } |
2804 | ||
d397712b CM |
2805 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2806 | struct btrfs_device *device) | |
0b86a832 CM |
2807 | { |
2808 | int ret; | |
2809 | struct btrfs_path *path; | |
0b246afa | 2810 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2811 | struct btrfs_dev_item *dev_item; |
2812 | struct extent_buffer *leaf; | |
2813 | struct btrfs_key key; | |
2814 | ||
0b86a832 CM |
2815 | path = btrfs_alloc_path(); |
2816 | if (!path) | |
2817 | return -ENOMEM; | |
2818 | ||
2819 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2820 | key.type = BTRFS_DEV_ITEM_KEY; | |
2821 | key.offset = device->devid; | |
2822 | ||
2823 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2824 | if (ret < 0) | |
2825 | goto out; | |
2826 | ||
2827 | if (ret > 0) { | |
2828 | ret = -ENOENT; | |
2829 | goto out; | |
2830 | } | |
2831 | ||
2832 | leaf = path->nodes[0]; | |
2833 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2834 | ||
2835 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2836 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2837 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2838 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2839 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2840 | btrfs_set_device_total_bytes(leaf, dev_item, |
2841 | btrfs_device_get_disk_total_bytes(device)); | |
2842 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2843 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2844 | btrfs_mark_buffer_dirty(leaf); |
2845 | ||
2846 | out: | |
2847 | btrfs_free_path(path); | |
2848 | return ret; | |
2849 | } | |
2850 | ||
2196d6e8 | 2851 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2852 | struct btrfs_device *device, u64 new_size) |
2853 | { | |
0b246afa JM |
2854 | struct btrfs_fs_info *fs_info = device->fs_info; |
2855 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2856 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2857 | u64 old_total; |
2858 | u64 diff; | |
8f18cf13 | 2859 | |
ebbede42 | 2860 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2861 | return -EACCES; |
2196d6e8 | 2862 | |
7dfb8be1 NB |
2863 | new_size = round_down(new_size, fs_info->sectorsize); |
2864 | ||
34441361 | 2865 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2866 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2867 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2868 | |
63a212ab | 2869 | if (new_size <= device->total_bytes || |
401e29c1 | 2870 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2871 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2872 | return -EINVAL; |
2196d6e8 | 2873 | } |
2b82032c | 2874 | |
0b246afa | 2875 | fs_devices = fs_info->fs_devices; |
2b82032c | 2876 | |
7dfb8be1 NB |
2877 | btrfs_set_super_total_bytes(super_copy, |
2878 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2879 | device->fs_devices->total_rw_bytes += diff; |
2880 | ||
7cc8e58d MX |
2881 | btrfs_device_set_total_bytes(device, new_size); |
2882 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2883 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2884 | if (list_empty(&device->resized_list)) |
2885 | list_add_tail(&device->resized_list, | |
2886 | &fs_devices->resized_devices); | |
34441361 | 2887 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2888 | |
8f18cf13 CM |
2889 | return btrfs_update_device(trans, device); |
2890 | } | |
2891 | ||
f4208794 | 2892 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2893 | { |
f4208794 | 2894 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2895 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2896 | int ret; |
2897 | struct btrfs_path *path; | |
2898 | struct btrfs_key key; | |
2899 | ||
8f18cf13 CM |
2900 | path = btrfs_alloc_path(); |
2901 | if (!path) | |
2902 | return -ENOMEM; | |
2903 | ||
408fbf19 | 2904 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2905 | key.offset = chunk_offset; |
2906 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2907 | ||
2908 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2909 | if (ret < 0) |
2910 | goto out; | |
2911 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2912 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2913 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2914 | ret = -ENOENT; |
2915 | goto out; | |
2916 | } | |
8f18cf13 CM |
2917 | |
2918 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2919 | if (ret < 0) |
0b246afa JM |
2920 | btrfs_handle_fs_error(fs_info, ret, |
2921 | "Failed to delete chunk item."); | |
79787eaa | 2922 | out: |
8f18cf13 | 2923 | btrfs_free_path(path); |
65a246c5 | 2924 | return ret; |
8f18cf13 CM |
2925 | } |
2926 | ||
408fbf19 | 2927 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2928 | { |
0b246afa | 2929 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2930 | struct btrfs_disk_key *disk_key; |
2931 | struct btrfs_chunk *chunk; | |
2932 | u8 *ptr; | |
2933 | int ret = 0; | |
2934 | u32 num_stripes; | |
2935 | u32 array_size; | |
2936 | u32 len = 0; | |
2937 | u32 cur; | |
2938 | struct btrfs_key key; | |
2939 | ||
34441361 | 2940 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2941 | array_size = btrfs_super_sys_array_size(super_copy); |
2942 | ||
2943 | ptr = super_copy->sys_chunk_array; | |
2944 | cur = 0; | |
2945 | ||
2946 | while (cur < array_size) { | |
2947 | disk_key = (struct btrfs_disk_key *)ptr; | |
2948 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2949 | ||
2950 | len = sizeof(*disk_key); | |
2951 | ||
2952 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2953 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2954 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2955 | len += btrfs_chunk_item_size(num_stripes); | |
2956 | } else { | |
2957 | ret = -EIO; | |
2958 | break; | |
2959 | } | |
408fbf19 | 2960 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2961 | key.offset == chunk_offset) { |
2962 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2963 | array_size -= len; | |
2964 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2965 | } else { | |
2966 | ptr += len; | |
2967 | cur += len; | |
2968 | } | |
2969 | } | |
34441361 | 2970 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2971 | return ret; |
2972 | } | |
2973 | ||
60ca842e OS |
2974 | /* |
2975 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2976 | * @logical: Logical block offset in bytes. | |
2977 | * @length: Length of extent in bytes. | |
2978 | * | |
2979 | * Return: Chunk mapping or ERR_PTR. | |
2980 | */ | |
2981 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2982 | u64 logical, u64 length) | |
592d92ee LB |
2983 | { |
2984 | struct extent_map_tree *em_tree; | |
2985 | struct extent_map *em; | |
2986 | ||
2987 | em_tree = &fs_info->mapping_tree.map_tree; | |
2988 | read_lock(&em_tree->lock); | |
2989 | em = lookup_extent_mapping(em_tree, logical, length); | |
2990 | read_unlock(&em_tree->lock); | |
2991 | ||
2992 | if (!em) { | |
2993 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2994 | logical, length); | |
2995 | return ERR_PTR(-EINVAL); | |
2996 | } | |
2997 | ||
2998 | if (em->start > logical || em->start + em->len < logical) { | |
2999 | btrfs_crit(fs_info, | |
3000 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
3001 | logical, length, em->start, em->start + em->len); | |
3002 | free_extent_map(em); | |
3003 | return ERR_PTR(-EINVAL); | |
3004 | } | |
3005 | ||
3006 | /* callers are responsible for dropping em's ref. */ | |
3007 | return em; | |
3008 | } | |
3009 | ||
97aff912 | 3010 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3011 | { |
97aff912 | 3012 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3013 | struct extent_map *em; |
3014 | struct map_lookup *map; | |
2196d6e8 | 3015 | u64 dev_extent_len = 0; |
47ab2a6c | 3016 | int i, ret = 0; |
0b246afa | 3017 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3018 | |
60ca842e | 3019 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3020 | if (IS_ERR(em)) { |
47ab2a6c JB |
3021 | /* |
3022 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3023 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3024 | * do anything we still error out. |
3025 | */ | |
3026 | ASSERT(0); | |
592d92ee | 3027 | return PTR_ERR(em); |
47ab2a6c | 3028 | } |
95617d69 | 3029 | map = em->map_lookup; |
34441361 | 3030 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 3031 | check_system_chunk(trans, map->type); |
34441361 | 3032 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 3033 | |
57ba4cb8 FM |
3034 | /* |
3035 | * Take the device list mutex to prevent races with the final phase of | |
3036 | * a device replace operation that replaces the device object associated | |
3037 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
3038 | */ | |
3039 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3040 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3041 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3042 | ret = btrfs_free_dev_extent(trans, device, |
3043 | map->stripes[i].physical, | |
3044 | &dev_extent_len); | |
47ab2a6c | 3045 | if (ret) { |
57ba4cb8 | 3046 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3047 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3048 | goto out; |
3049 | } | |
a061fc8d | 3050 | |
2196d6e8 | 3051 | if (device->bytes_used > 0) { |
34441361 | 3052 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3053 | btrfs_device_set_bytes_used(device, |
3054 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3055 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3056 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3057 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3058 | } |
a061fc8d | 3059 | |
64bc6c2a NB |
3060 | ret = btrfs_update_device(trans, device); |
3061 | if (ret) { | |
3062 | mutex_unlock(&fs_devices->device_list_mutex); | |
3063 | btrfs_abort_transaction(trans, ret); | |
3064 | goto out; | |
dfe25020 | 3065 | } |
8f18cf13 | 3066 | } |
57ba4cb8 FM |
3067 | mutex_unlock(&fs_devices->device_list_mutex); |
3068 | ||
f4208794 | 3069 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3070 | if (ret) { |
66642832 | 3071 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3072 | goto out; |
3073 | } | |
8f18cf13 | 3074 | |
6bccf3ab | 3075 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3076 | |
8f18cf13 | 3077 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3078 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3079 | if (ret) { |
66642832 | 3080 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3081 | goto out; |
3082 | } | |
8f18cf13 CM |
3083 | } |
3084 | ||
5a98ec01 | 3085 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3086 | if (ret) { |
66642832 | 3087 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3088 | goto out; |
3089 | } | |
2b82032c | 3090 | |
47ab2a6c | 3091 | out: |
2b82032c YZ |
3092 | /* once for us */ |
3093 | free_extent_map(em); | |
47ab2a6c JB |
3094 | return ret; |
3095 | } | |
2b82032c | 3096 | |
5b4aacef | 3097 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3098 | { |
5b4aacef | 3099 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3100 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3101 | int ret; |
2b82032c | 3102 | |
67c5e7d4 FM |
3103 | /* |
3104 | * Prevent races with automatic removal of unused block groups. | |
3105 | * After we relocate and before we remove the chunk with offset | |
3106 | * chunk_offset, automatic removal of the block group can kick in, | |
3107 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3108 | * | |
3109 | * Make sure to acquire this mutex before doing a tree search (dev | |
3110 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3111 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3112 | * we release the path used to search the chunk/dev tree and before | |
3113 | * the current task acquires this mutex and calls us. | |
3114 | */ | |
a32bf9a3 | 3115 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3116 | |
0b246afa | 3117 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
3118 | if (ret) |
3119 | return -ENOSPC; | |
3120 | ||
3121 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3122 | btrfs_scrub_pause(fs_info); |
0b246afa | 3123 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3124 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3125 | if (ret) |
3126 | return ret; | |
3127 | ||
75cb379d JM |
3128 | /* |
3129 | * We add the kobjects here (and after forcing data chunk creation) | |
3130 | * since relocation is the only place we'll create chunks of a new | |
3131 | * type at runtime. The only place where we'll remove the last | |
3132 | * chunk of a type is the call immediately below this one. Even | |
3133 | * so, we're protected against races with the cleaner thread since | |
3134 | * we're covered by the delete_unused_bgs_mutex. | |
3135 | */ | |
3136 | btrfs_add_raid_kobjects(fs_info); | |
3137 | ||
19c4d2f9 CM |
3138 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3139 | chunk_offset); | |
3140 | if (IS_ERR(trans)) { | |
3141 | ret = PTR_ERR(trans); | |
3142 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3143 | return ret; | |
3144 | } | |
3145 | ||
47ab2a6c | 3146 | /* |
19c4d2f9 CM |
3147 | * step two, delete the device extents and the |
3148 | * chunk tree entries | |
47ab2a6c | 3149 | */ |
97aff912 | 3150 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3151 | btrfs_end_transaction(trans); |
19c4d2f9 | 3152 | return ret; |
2b82032c YZ |
3153 | } |
3154 | ||
2ff7e61e | 3155 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3156 | { |
0b246afa | 3157 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3158 | struct btrfs_path *path; |
3159 | struct extent_buffer *leaf; | |
3160 | struct btrfs_chunk *chunk; | |
3161 | struct btrfs_key key; | |
3162 | struct btrfs_key found_key; | |
2b82032c | 3163 | u64 chunk_type; |
ba1bf481 JB |
3164 | bool retried = false; |
3165 | int failed = 0; | |
2b82032c YZ |
3166 | int ret; |
3167 | ||
3168 | path = btrfs_alloc_path(); | |
3169 | if (!path) | |
3170 | return -ENOMEM; | |
3171 | ||
ba1bf481 | 3172 | again: |
2b82032c YZ |
3173 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3174 | key.offset = (u64)-1; | |
3175 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3176 | ||
3177 | while (1) { | |
0b246afa | 3178 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3179 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3180 | if (ret < 0) { |
0b246afa | 3181 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3182 | goto error; |
67c5e7d4 | 3183 | } |
79787eaa | 3184 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3185 | |
3186 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3187 | key.type); | |
67c5e7d4 | 3188 | if (ret) |
0b246afa | 3189 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3190 | if (ret < 0) |
3191 | goto error; | |
3192 | if (ret > 0) | |
3193 | break; | |
1a40e23b | 3194 | |
2b82032c YZ |
3195 | leaf = path->nodes[0]; |
3196 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3197 | |
2b82032c YZ |
3198 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3199 | struct btrfs_chunk); | |
3200 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3201 | btrfs_release_path(path); |
8f18cf13 | 3202 | |
2b82032c | 3203 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3204 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3205 | if (ret == -ENOSPC) |
3206 | failed++; | |
14586651 HS |
3207 | else |
3208 | BUG_ON(ret); | |
2b82032c | 3209 | } |
0b246afa | 3210 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3211 | |
2b82032c YZ |
3212 | if (found_key.offset == 0) |
3213 | break; | |
3214 | key.offset = found_key.offset - 1; | |
3215 | } | |
3216 | ret = 0; | |
ba1bf481 JB |
3217 | if (failed && !retried) { |
3218 | failed = 0; | |
3219 | retried = true; | |
3220 | goto again; | |
fae7f21c | 3221 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3222 | ret = -ENOSPC; |
3223 | } | |
2b82032c YZ |
3224 | error: |
3225 | btrfs_free_path(path); | |
3226 | return ret; | |
8f18cf13 CM |
3227 | } |
3228 | ||
a6f93c71 LB |
3229 | /* |
3230 | * return 1 : allocate a data chunk successfully, | |
3231 | * return <0: errors during allocating a data chunk, | |
3232 | * return 0 : no need to allocate a data chunk. | |
3233 | */ | |
3234 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3235 | u64 chunk_offset) | |
3236 | { | |
3237 | struct btrfs_block_group_cache *cache; | |
3238 | u64 bytes_used; | |
3239 | u64 chunk_type; | |
3240 | ||
3241 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3242 | ASSERT(cache); | |
3243 | chunk_type = cache->flags; | |
3244 | btrfs_put_block_group(cache); | |
3245 | ||
3246 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3247 | spin_lock(&fs_info->data_sinfo->lock); | |
3248 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3249 | spin_unlock(&fs_info->data_sinfo->lock); | |
3250 | ||
3251 | if (!bytes_used) { | |
3252 | struct btrfs_trans_handle *trans; | |
3253 | int ret; | |
3254 | ||
3255 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3256 | if (IS_ERR(trans)) | |
3257 | return PTR_ERR(trans); | |
3258 | ||
43a7e99d | 3259 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3260 | BTRFS_BLOCK_GROUP_DATA); |
3261 | btrfs_end_transaction(trans); | |
3262 | if (ret < 0) | |
3263 | return ret; | |
3264 | ||
75cb379d JM |
3265 | btrfs_add_raid_kobjects(fs_info); |
3266 | ||
a6f93c71 LB |
3267 | return 1; |
3268 | } | |
3269 | } | |
3270 | return 0; | |
3271 | } | |
3272 | ||
6bccf3ab | 3273 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3274 | struct btrfs_balance_control *bctl) |
3275 | { | |
6bccf3ab | 3276 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3277 | struct btrfs_trans_handle *trans; |
3278 | struct btrfs_balance_item *item; | |
3279 | struct btrfs_disk_balance_args disk_bargs; | |
3280 | struct btrfs_path *path; | |
3281 | struct extent_buffer *leaf; | |
3282 | struct btrfs_key key; | |
3283 | int ret, err; | |
3284 | ||
3285 | path = btrfs_alloc_path(); | |
3286 | if (!path) | |
3287 | return -ENOMEM; | |
3288 | ||
3289 | trans = btrfs_start_transaction(root, 0); | |
3290 | if (IS_ERR(trans)) { | |
3291 | btrfs_free_path(path); | |
3292 | return PTR_ERR(trans); | |
3293 | } | |
3294 | ||
3295 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3296 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3297 | key.offset = 0; |
3298 | ||
3299 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3300 | sizeof(*item)); | |
3301 | if (ret) | |
3302 | goto out; | |
3303 | ||
3304 | leaf = path->nodes[0]; | |
3305 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3306 | ||
b159fa28 | 3307 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3308 | |
3309 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3310 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3311 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3312 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3313 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3314 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3315 | ||
3316 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3317 | ||
3318 | btrfs_mark_buffer_dirty(leaf); | |
3319 | out: | |
3320 | btrfs_free_path(path); | |
3a45bb20 | 3321 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3322 | if (err && !ret) |
3323 | ret = err; | |
3324 | return ret; | |
3325 | } | |
3326 | ||
6bccf3ab | 3327 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3328 | { |
6bccf3ab | 3329 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3330 | struct btrfs_trans_handle *trans; |
3331 | struct btrfs_path *path; | |
3332 | struct btrfs_key key; | |
3333 | int ret, err; | |
3334 | ||
3335 | path = btrfs_alloc_path(); | |
3336 | if (!path) | |
3337 | return -ENOMEM; | |
3338 | ||
3339 | trans = btrfs_start_transaction(root, 0); | |
3340 | if (IS_ERR(trans)) { | |
3341 | btrfs_free_path(path); | |
3342 | return PTR_ERR(trans); | |
3343 | } | |
3344 | ||
3345 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3346 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3347 | key.offset = 0; |
3348 | ||
3349 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3350 | if (ret < 0) | |
3351 | goto out; | |
3352 | if (ret > 0) { | |
3353 | ret = -ENOENT; | |
3354 | goto out; | |
3355 | } | |
3356 | ||
3357 | ret = btrfs_del_item(trans, root, path); | |
3358 | out: | |
3359 | btrfs_free_path(path); | |
3a45bb20 | 3360 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3361 | if (err && !ret) |
3362 | ret = err; | |
3363 | return ret; | |
3364 | } | |
3365 | ||
59641015 ID |
3366 | /* |
3367 | * This is a heuristic used to reduce the number of chunks balanced on | |
3368 | * resume after balance was interrupted. | |
3369 | */ | |
3370 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3371 | { | |
3372 | /* | |
3373 | * Turn on soft mode for chunk types that were being converted. | |
3374 | */ | |
3375 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3376 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3377 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3378 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3379 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3380 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3381 | ||
3382 | /* | |
3383 | * Turn on usage filter if is not already used. The idea is | |
3384 | * that chunks that we have already balanced should be | |
3385 | * reasonably full. Don't do it for chunks that are being | |
3386 | * converted - that will keep us from relocating unconverted | |
3387 | * (albeit full) chunks. | |
3388 | */ | |
3389 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3390 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3391 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3392 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3393 | bctl->data.usage = 90; | |
3394 | } | |
3395 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3396 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3397 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3398 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3399 | bctl->sys.usage = 90; | |
3400 | } | |
3401 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3402 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3403 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3404 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3405 | bctl->meta.usage = 90; | |
3406 | } | |
3407 | } | |
3408 | ||
149196a2 DS |
3409 | /* |
3410 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3411 | */ | |
3412 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3413 | { |
3414 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3415 | int ret; |
c9e9f97b ID |
3416 | |
3417 | BUG_ON(!fs_info->balance_ctl); | |
3418 | ||
3419 | spin_lock(&fs_info->balance_lock); | |
3420 | fs_info->balance_ctl = NULL; | |
3421 | spin_unlock(&fs_info->balance_lock); | |
3422 | ||
3423 | kfree(bctl); | |
149196a2 DS |
3424 | ret = del_balance_item(fs_info); |
3425 | if (ret) | |
3426 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3427 | } |
3428 | ||
ed25e9b2 ID |
3429 | /* |
3430 | * Balance filters. Return 1 if chunk should be filtered out | |
3431 | * (should not be balanced). | |
3432 | */ | |
899c81ea | 3433 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3434 | struct btrfs_balance_args *bargs) |
3435 | { | |
899c81ea ID |
3436 | chunk_type = chunk_to_extended(chunk_type) & |
3437 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3438 | |
899c81ea | 3439 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3440 | return 0; |
3441 | ||
3442 | return 1; | |
3443 | } | |
3444 | ||
dba72cb3 | 3445 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3446 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3447 | { |
3448 | struct btrfs_block_group_cache *cache; | |
3449 | u64 chunk_used; | |
3450 | u64 user_thresh_min; | |
3451 | u64 user_thresh_max; | |
3452 | int ret = 1; | |
3453 | ||
3454 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3455 | chunk_used = btrfs_block_group_used(&cache->item); | |
3456 | ||
3457 | if (bargs->usage_min == 0) | |
3458 | user_thresh_min = 0; | |
3459 | else | |
3460 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3461 | bargs->usage_min); | |
3462 | ||
3463 | if (bargs->usage_max == 0) | |
3464 | user_thresh_max = 1; | |
3465 | else if (bargs->usage_max > 100) | |
3466 | user_thresh_max = cache->key.offset; | |
3467 | else | |
3468 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3469 | bargs->usage_max); | |
3470 | ||
3471 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3472 | ret = 0; | |
3473 | ||
3474 | btrfs_put_block_group(cache); | |
3475 | return ret; | |
3476 | } | |
3477 | ||
dba72cb3 | 3478 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3479 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3480 | { |
3481 | struct btrfs_block_group_cache *cache; | |
3482 | u64 chunk_used, user_thresh; | |
3483 | int ret = 1; | |
3484 | ||
3485 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3486 | chunk_used = btrfs_block_group_used(&cache->item); | |
3487 | ||
bc309467 | 3488 | if (bargs->usage_min == 0) |
3e39cea6 | 3489 | user_thresh = 1; |
a105bb88 ID |
3490 | else if (bargs->usage > 100) |
3491 | user_thresh = cache->key.offset; | |
3492 | else | |
3493 | user_thresh = div_factor_fine(cache->key.offset, | |
3494 | bargs->usage); | |
3495 | ||
5ce5b3c0 ID |
3496 | if (chunk_used < user_thresh) |
3497 | ret = 0; | |
3498 | ||
3499 | btrfs_put_block_group(cache); | |
3500 | return ret; | |
3501 | } | |
3502 | ||
409d404b ID |
3503 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3504 | struct btrfs_chunk *chunk, | |
3505 | struct btrfs_balance_args *bargs) | |
3506 | { | |
3507 | struct btrfs_stripe *stripe; | |
3508 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3509 | int i; | |
3510 | ||
3511 | for (i = 0; i < num_stripes; i++) { | |
3512 | stripe = btrfs_stripe_nr(chunk, i); | |
3513 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3514 | return 0; | |
3515 | } | |
3516 | ||
3517 | return 1; | |
3518 | } | |
3519 | ||
94e60d5a ID |
3520 | /* [pstart, pend) */ |
3521 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3522 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3523 | struct btrfs_balance_args *bargs) |
3524 | { | |
3525 | struct btrfs_stripe *stripe; | |
3526 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3527 | u64 stripe_offset; | |
3528 | u64 stripe_length; | |
3529 | int factor; | |
3530 | int i; | |
3531 | ||
3532 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3533 | return 0; | |
3534 | ||
3535 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3536 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3537 | factor = num_stripes / 2; | |
3538 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3539 | factor = num_stripes - 1; | |
3540 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3541 | factor = num_stripes - 2; | |
3542 | } else { | |
3543 | factor = num_stripes; | |
3544 | } | |
94e60d5a ID |
3545 | |
3546 | for (i = 0; i < num_stripes; i++) { | |
3547 | stripe = btrfs_stripe_nr(chunk, i); | |
3548 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3549 | continue; | |
3550 | ||
3551 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3552 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3553 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3554 | |
3555 | if (stripe_offset < bargs->pend && | |
3556 | stripe_offset + stripe_length > bargs->pstart) | |
3557 | return 0; | |
3558 | } | |
3559 | ||
3560 | return 1; | |
3561 | } | |
3562 | ||
ea67176a ID |
3563 | /* [vstart, vend) */ |
3564 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3565 | struct btrfs_chunk *chunk, | |
3566 | u64 chunk_offset, | |
3567 | struct btrfs_balance_args *bargs) | |
3568 | { | |
3569 | if (chunk_offset < bargs->vend && | |
3570 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3571 | /* at least part of the chunk is inside this vrange */ | |
3572 | return 0; | |
3573 | ||
3574 | return 1; | |
3575 | } | |
3576 | ||
dee32d0a GAP |
3577 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3578 | struct btrfs_chunk *chunk, | |
3579 | struct btrfs_balance_args *bargs) | |
3580 | { | |
3581 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3582 | ||
3583 | if (bargs->stripes_min <= num_stripes | |
3584 | && num_stripes <= bargs->stripes_max) | |
3585 | return 0; | |
3586 | ||
3587 | return 1; | |
3588 | } | |
3589 | ||
899c81ea | 3590 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3591 | struct btrfs_balance_args *bargs) |
3592 | { | |
3593 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3594 | return 0; | |
3595 | ||
899c81ea ID |
3596 | chunk_type = chunk_to_extended(chunk_type) & |
3597 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3598 | |
899c81ea | 3599 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3600 | return 1; |
3601 | ||
3602 | return 0; | |
3603 | } | |
3604 | ||
2ff7e61e | 3605 | static int should_balance_chunk(struct btrfs_fs_info *fs_info, |
f43ffb60 ID |
3606 | struct extent_buffer *leaf, |
3607 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3608 | { | |
0b246afa | 3609 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3610 | struct btrfs_balance_args *bargs = NULL; |
3611 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3612 | ||
3613 | /* type filter */ | |
3614 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3615 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3616 | return 0; | |
3617 | } | |
3618 | ||
3619 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3620 | bargs = &bctl->data; | |
3621 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3622 | bargs = &bctl->sys; | |
3623 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3624 | bargs = &bctl->meta; | |
3625 | ||
ed25e9b2 ID |
3626 | /* profiles filter */ |
3627 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3628 | chunk_profiles_filter(chunk_type, bargs)) { | |
3629 | return 0; | |
5ce5b3c0 ID |
3630 | } |
3631 | ||
3632 | /* usage filter */ | |
3633 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3634 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3635 | return 0; |
bc309467 | 3636 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3637 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3638 | return 0; |
409d404b ID |
3639 | } |
3640 | ||
3641 | /* devid filter */ | |
3642 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3643 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3644 | return 0; | |
94e60d5a ID |
3645 | } |
3646 | ||
3647 | /* drange filter, makes sense only with devid filter */ | |
3648 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3649 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3650 | return 0; |
ea67176a ID |
3651 | } |
3652 | ||
3653 | /* vrange filter */ | |
3654 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3655 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3656 | return 0; | |
ed25e9b2 ID |
3657 | } |
3658 | ||
dee32d0a GAP |
3659 | /* stripes filter */ |
3660 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3661 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3662 | return 0; | |
3663 | } | |
3664 | ||
cfa4c961 ID |
3665 | /* soft profile changing mode */ |
3666 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3667 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3668 | return 0; | |
3669 | } | |
3670 | ||
7d824b6f DS |
3671 | /* |
3672 | * limited by count, must be the last filter | |
3673 | */ | |
3674 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3675 | if (bargs->limit == 0) | |
3676 | return 0; | |
3677 | else | |
3678 | bargs->limit--; | |
12907fc7 DS |
3679 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3680 | /* | |
3681 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3682 | * determined here because we do not have the global information |
12907fc7 DS |
3683 | * about the count of all chunks that satisfy the filters. |
3684 | */ | |
3685 | if (bargs->limit_max == 0) | |
3686 | return 0; | |
3687 | else | |
3688 | bargs->limit_max--; | |
7d824b6f DS |
3689 | } |
3690 | ||
f43ffb60 ID |
3691 | return 1; |
3692 | } | |
3693 | ||
c9e9f97b | 3694 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3695 | { |
19a39dce | 3696 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3697 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3698 | u64 chunk_type; |
f43ffb60 | 3699 | struct btrfs_chunk *chunk; |
5a488b9d | 3700 | struct btrfs_path *path = NULL; |
ec44a35c | 3701 | struct btrfs_key key; |
ec44a35c | 3702 | struct btrfs_key found_key; |
f43ffb60 ID |
3703 | struct extent_buffer *leaf; |
3704 | int slot; | |
c9e9f97b ID |
3705 | int ret; |
3706 | int enospc_errors = 0; | |
19a39dce | 3707 | bool counting = true; |
12907fc7 | 3708 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3709 | u64 limit_data = bctl->data.limit; |
3710 | u64 limit_meta = bctl->meta.limit; | |
3711 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3712 | u32 count_data = 0; |
3713 | u32 count_meta = 0; | |
3714 | u32 count_sys = 0; | |
2c9fe835 | 3715 | int chunk_reserved = 0; |
ec44a35c | 3716 | |
ec44a35c | 3717 | path = btrfs_alloc_path(); |
17e9f796 MF |
3718 | if (!path) { |
3719 | ret = -ENOMEM; | |
3720 | goto error; | |
3721 | } | |
19a39dce ID |
3722 | |
3723 | /* zero out stat counters */ | |
3724 | spin_lock(&fs_info->balance_lock); | |
3725 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3726 | spin_unlock(&fs_info->balance_lock); | |
3727 | again: | |
7d824b6f | 3728 | if (!counting) { |
12907fc7 DS |
3729 | /* |
3730 | * The single value limit and min/max limits use the same bytes | |
3731 | * in the | |
3732 | */ | |
7d824b6f DS |
3733 | bctl->data.limit = limit_data; |
3734 | bctl->meta.limit = limit_meta; | |
3735 | bctl->sys.limit = limit_sys; | |
3736 | } | |
ec44a35c CM |
3737 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3738 | key.offset = (u64)-1; | |
3739 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3740 | ||
d397712b | 3741 | while (1) { |
19a39dce | 3742 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3743 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3744 | ret = -ECANCELED; |
3745 | goto error; | |
3746 | } | |
3747 | ||
67c5e7d4 | 3748 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3749 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3750 | if (ret < 0) { |
3751 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3752 | goto error; |
67c5e7d4 | 3753 | } |
ec44a35c CM |
3754 | |
3755 | /* | |
3756 | * this shouldn't happen, it means the last relocate | |
3757 | * failed | |
3758 | */ | |
3759 | if (ret == 0) | |
c9e9f97b | 3760 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3761 | |
3762 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3763 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3764 | if (ret) { |
67c5e7d4 | 3765 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3766 | ret = 0; |
ec44a35c | 3767 | break; |
c9e9f97b | 3768 | } |
7d9eb12c | 3769 | |
f43ffb60 ID |
3770 | leaf = path->nodes[0]; |
3771 | slot = path->slots[0]; | |
3772 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3773 | |
67c5e7d4 FM |
3774 | if (found_key.objectid != key.objectid) { |
3775 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3776 | break; |
67c5e7d4 | 3777 | } |
7d9eb12c | 3778 | |
f43ffb60 | 3779 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3780 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3781 | |
19a39dce ID |
3782 | if (!counting) { |
3783 | spin_lock(&fs_info->balance_lock); | |
3784 | bctl->stat.considered++; | |
3785 | spin_unlock(&fs_info->balance_lock); | |
3786 | } | |
3787 | ||
2ff7e61e | 3788 | ret = should_balance_chunk(fs_info, leaf, chunk, |
f43ffb60 | 3789 | found_key.offset); |
2c9fe835 | 3790 | |
b3b4aa74 | 3791 | btrfs_release_path(path); |
67c5e7d4 FM |
3792 | if (!ret) { |
3793 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3794 | goto loop; |
67c5e7d4 | 3795 | } |
f43ffb60 | 3796 | |
19a39dce | 3797 | if (counting) { |
67c5e7d4 | 3798 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3799 | spin_lock(&fs_info->balance_lock); |
3800 | bctl->stat.expected++; | |
3801 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3802 | |
3803 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3804 | count_data++; | |
3805 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3806 | count_sys++; | |
3807 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3808 | count_meta++; | |
3809 | ||
3810 | goto loop; | |
3811 | } | |
3812 | ||
3813 | /* | |
3814 | * Apply limit_min filter, no need to check if the LIMITS | |
3815 | * filter is used, limit_min is 0 by default | |
3816 | */ | |
3817 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3818 | count_data < bctl->data.limit_min) | |
3819 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3820 | count_meta < bctl->meta.limit_min) | |
3821 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3822 | count_sys < bctl->sys.limit_min)) { | |
3823 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3824 | goto loop; |
3825 | } | |
3826 | ||
a6f93c71 LB |
3827 | if (!chunk_reserved) { |
3828 | /* | |
3829 | * We may be relocating the only data chunk we have, | |
3830 | * which could potentially end up with losing data's | |
3831 | * raid profile, so lets allocate an empty one in | |
3832 | * advance. | |
3833 | */ | |
3834 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3835 | found_key.offset); | |
2c9fe835 ZL |
3836 | if (ret < 0) { |
3837 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3838 | goto error; | |
a6f93c71 LB |
3839 | } else if (ret == 1) { |
3840 | chunk_reserved = 1; | |
2c9fe835 | 3841 | } |
2c9fe835 ZL |
3842 | } |
3843 | ||
5b4aacef | 3844 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3845 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3846 | if (ret == -ENOSPC) { |
c9e9f97b | 3847 | enospc_errors++; |
eede2bf3 OS |
3848 | } else if (ret == -ETXTBSY) { |
3849 | btrfs_info(fs_info, | |
3850 | "skipping relocation of block group %llu due to active swapfile", | |
3851 | found_key.offset); | |
3852 | ret = 0; | |
3853 | } else if (ret) { | |
3854 | goto error; | |
19a39dce ID |
3855 | } else { |
3856 | spin_lock(&fs_info->balance_lock); | |
3857 | bctl->stat.completed++; | |
3858 | spin_unlock(&fs_info->balance_lock); | |
3859 | } | |
f43ffb60 | 3860 | loop: |
795a3321 ID |
3861 | if (found_key.offset == 0) |
3862 | break; | |
ba1bf481 | 3863 | key.offset = found_key.offset - 1; |
ec44a35c | 3864 | } |
c9e9f97b | 3865 | |
19a39dce ID |
3866 | if (counting) { |
3867 | btrfs_release_path(path); | |
3868 | counting = false; | |
3869 | goto again; | |
3870 | } | |
ec44a35c CM |
3871 | error: |
3872 | btrfs_free_path(path); | |
c9e9f97b | 3873 | if (enospc_errors) { |
efe120a0 | 3874 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3875 | enospc_errors); |
c9e9f97b ID |
3876 | if (!ret) |
3877 | ret = -ENOSPC; | |
3878 | } | |
3879 | ||
ec44a35c CM |
3880 | return ret; |
3881 | } | |
3882 | ||
0c460c0d ID |
3883 | /** |
3884 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3885 | * @flags: profile to validate | |
3886 | * @extended: if true @flags is treated as an extended profile | |
3887 | */ | |
3888 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3889 | { | |
3890 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3891 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3892 | ||
3893 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3894 | ||
3895 | /* 1) check that all other bits are zeroed */ | |
3896 | if (flags & ~mask) | |
3897 | return 0; | |
3898 | ||
3899 | /* 2) see if profile is reduced */ | |
3900 | if (flags == 0) | |
3901 | return !extended; /* "0" is valid for usual profiles */ | |
3902 | ||
3903 | /* true if exactly one bit set */ | |
818255fe | 3904 | return is_power_of_2(flags); |
0c460c0d ID |
3905 | } |
3906 | ||
837d5b6e ID |
3907 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3908 | { | |
a7e99c69 ID |
3909 | /* cancel requested || normal exit path */ |
3910 | return atomic_read(&fs_info->balance_cancel_req) || | |
3911 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3912 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3913 | } |
3914 | ||
bdcd3c97 AM |
3915 | /* Non-zero return value signifies invalidity */ |
3916 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3917 | u64 allowed) | |
3918 | { | |
3919 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3920 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3921 | (bctl_arg->target & ~allowed))); | |
3922 | } | |
3923 | ||
56fc37d9 AJ |
3924 | /* |
3925 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3926 | * @size_buf including the terminating null. The output may be trimmed if it | |
3927 | * does not fit into the provided buffer. | |
3928 | */ | |
3929 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3930 | u32 size_buf) | |
3931 | { | |
3932 | int ret; | |
3933 | u32 size_bp = size_buf; | |
3934 | char *bp = buf; | |
3935 | u64 flags = bargs->flags; | |
3936 | char tmp_buf[128] = {'\0'}; | |
3937 | ||
3938 | if (!flags) | |
3939 | return; | |
3940 | ||
3941 | #define CHECK_APPEND_NOARG(a) \ | |
3942 | do { \ | |
3943 | ret = snprintf(bp, size_bp, (a)); \ | |
3944 | if (ret < 0 || ret >= size_bp) \ | |
3945 | goto out_overflow; \ | |
3946 | size_bp -= ret; \ | |
3947 | bp += ret; \ | |
3948 | } while (0) | |
3949 | ||
3950 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3951 | do { \ | |
3952 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3953 | if (ret < 0 || ret >= size_bp) \ | |
3954 | goto out_overflow; \ | |
3955 | size_bp -= ret; \ | |
3956 | bp += ret; \ | |
3957 | } while (0) | |
3958 | ||
3959 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3960 | do { \ | |
3961 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3962 | if (ret < 0 || ret >= size_bp) \ | |
3963 | goto out_overflow; \ | |
3964 | size_bp -= ret; \ | |
3965 | bp += ret; \ | |
3966 | } while (0) | |
3967 | ||
3968 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3969 | int index = btrfs_bg_flags_to_raid_index(bargs->target); | |
3970 | ||
3971 | CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index)); | |
3972 | } | |
3973 | ||
3974 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3975 | CHECK_APPEND_NOARG("soft,"); | |
3976 | ||
3977 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3978 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3979 | sizeof(tmp_buf)); | |
3980 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3981 | } | |
3982 | ||
3983 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3984 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3985 | ||
3986 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3987 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3988 | bargs->usage_min, bargs->usage_max); | |
3989 | ||
3990 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3991 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3992 | ||
3993 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3994 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3995 | bargs->pstart, bargs->pend); | |
3996 | ||
3997 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3998 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3999 | bargs->vstart, bargs->vend); | |
4000 | ||
4001 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4002 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4003 | ||
4004 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4005 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4006 | bargs->limit_min, bargs->limit_max); | |
4007 | ||
4008 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4009 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4010 | bargs->stripes_min, bargs->stripes_max); | |
4011 | ||
4012 | #undef CHECK_APPEND_2ARG | |
4013 | #undef CHECK_APPEND_1ARG | |
4014 | #undef CHECK_APPEND_NOARG | |
4015 | ||
4016 | out_overflow: | |
4017 | ||
4018 | if (size_bp < size_buf) | |
4019 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4020 | else | |
4021 | buf[0] = '\0'; | |
4022 | } | |
4023 | ||
4024 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4025 | { | |
4026 | u32 size_buf = 1024; | |
4027 | char tmp_buf[192] = {'\0'}; | |
4028 | char *buf; | |
4029 | char *bp; | |
4030 | u32 size_bp = size_buf; | |
4031 | int ret; | |
4032 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4033 | ||
4034 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4035 | if (!buf) | |
4036 | return; | |
4037 | ||
4038 | bp = buf; | |
4039 | ||
4040 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4041 | do { \ | |
4042 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4043 | if (ret < 0 || ret >= size_bp) \ | |
4044 | goto out_overflow; \ | |
4045 | size_bp -= ret; \ | |
4046 | bp += ret; \ | |
4047 | } while (0) | |
4048 | ||
4049 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4050 | CHECK_APPEND_1ARG("%s", "-f "); | |
4051 | ||
4052 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4053 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4054 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4055 | } | |
4056 | ||
4057 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4058 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4059 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4060 | } | |
4061 | ||
4062 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4063 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4064 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4065 | } | |
4066 | ||
4067 | #undef CHECK_APPEND_1ARG | |
4068 | ||
4069 | out_overflow: | |
4070 | ||
4071 | if (size_bp < size_buf) | |
4072 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4073 | btrfs_info(fs_info, "balance: %s %s", | |
4074 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4075 | "resume" : "start", buf); | |
4076 | ||
4077 | kfree(buf); | |
4078 | } | |
4079 | ||
c9e9f97b | 4080 | /* |
dccdb07b | 4081 | * Should be called with balance mutexe held |
c9e9f97b | 4082 | */ |
6fcf6e2b DS |
4083 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4084 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4085 | struct btrfs_ioctl_balance_args *bargs) |
4086 | { | |
14506127 | 4087 | u64 meta_target, data_target; |
f43ffb60 | 4088 | u64 allowed; |
e4837f8f | 4089 | int mixed = 0; |
c9e9f97b | 4090 | int ret; |
8dabb742 | 4091 | u64 num_devices; |
de98ced9 | 4092 | unsigned seq; |
5a8067c0 | 4093 | bool reducing_integrity; |
c9e9f97b | 4094 | |
837d5b6e | 4095 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4096 | atomic_read(&fs_info->balance_pause_req) || |
4097 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4098 | ret = -EINVAL; |
4099 | goto out; | |
4100 | } | |
4101 | ||
e4837f8f ID |
4102 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4103 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4104 | mixed = 1; | |
4105 | ||
f43ffb60 ID |
4106 | /* |
4107 | * In case of mixed groups both data and meta should be picked, | |
4108 | * and identical options should be given for both of them. | |
4109 | */ | |
e4837f8f ID |
4110 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4111 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4112 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4113 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4114 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4115 | btrfs_err(fs_info, |
6dac13f8 | 4116 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4117 | ret = -EINVAL; |
4118 | goto out; | |
4119 | } | |
4120 | } | |
4121 | ||
1da73967 AJ |
4122 | num_devices = btrfs_num_devices(fs_info); |
4123 | ||
88be159c AH |
4124 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
4125 | if (num_devices > 1) | |
e4d8ec0f | 4126 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
4127 | if (num_devices > 2) |
4128 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
4129 | if (num_devices > 3) | |
4130 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
4131 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 4132 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
4133 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
4134 | ||
5d163e0e | 4135 | btrfs_err(fs_info, |
6dac13f8 AJ |
4136 | "balance: invalid convert data profile %s", |
4137 | get_raid_name(index)); | |
e4d8ec0f ID |
4138 | ret = -EINVAL; |
4139 | goto out; | |
4140 | } | |
bdcd3c97 | 4141 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
4142 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
4143 | ||
efe120a0 | 4144 | btrfs_err(fs_info, |
6dac13f8 AJ |
4145 | "balance: invalid convert metadata profile %s", |
4146 | get_raid_name(index)); | |
e4d8ec0f ID |
4147 | ret = -EINVAL; |
4148 | goto out; | |
4149 | } | |
bdcd3c97 | 4150 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
4151 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
4152 | ||
efe120a0 | 4153 | btrfs_err(fs_info, |
6dac13f8 AJ |
4154 | "balance: invalid convert system profile %s", |
4155 | get_raid_name(index)); | |
e4d8ec0f ID |
4156 | ret = -EINVAL; |
4157 | goto out; | |
4158 | } | |
4159 | ||
e4d8ec0f ID |
4160 | /* allow to reduce meta or sys integrity only if force set */ |
4161 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
4162 | BTRFS_BLOCK_GROUP_RAID10 | |
4163 | BTRFS_BLOCK_GROUP_RAID5 | | |
4164 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
4165 | do { |
4166 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4167 | ||
4168 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4169 | (fs_info->avail_system_alloc_bits & allowed) && | |
4170 | !(bctl->sys.target & allowed)) || | |
4171 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4172 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4173 | !(bctl->meta.target & allowed))) |
4174 | reducing_integrity = true; | |
4175 | else | |
4176 | reducing_integrity = false; | |
4177 | ||
4178 | /* if we're not converting, the target field is uninitialized */ | |
4179 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4180 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4181 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4182 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4183 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4184 | |
5a8067c0 FM |
4185 | if (reducing_integrity) { |
4186 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4187 | btrfs_info(fs_info, | |
4188 | "balance: force reducing metadata integrity"); | |
4189 | } else { | |
4190 | btrfs_err(fs_info, | |
4191 | "balance: reduces metadata integrity, use --force if you want this"); | |
4192 | ret = -EINVAL; | |
4193 | goto out; | |
4194 | } | |
4195 | } | |
4196 | ||
14506127 AB |
4197 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4198 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
4199 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
4200 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
4201 | ||
ee592d07 | 4202 | btrfs_warn(fs_info, |
6dac13f8 AJ |
4203 | "balance: metadata profile %s has lower redundancy than data profile %s", |
4204 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
4205 | } |
4206 | ||
6bccf3ab | 4207 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4208 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4209 | goto out; |
4210 | ||
59641015 ID |
4211 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4212 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4213 | BUG_ON(fs_info->balance_ctl); |
4214 | spin_lock(&fs_info->balance_lock); | |
4215 | fs_info->balance_ctl = bctl; | |
4216 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4217 | } else { |
4218 | BUG_ON(ret != -EEXIST); | |
4219 | spin_lock(&fs_info->balance_lock); | |
4220 | update_balance_args(bctl); | |
4221 | spin_unlock(&fs_info->balance_lock); | |
4222 | } | |
c9e9f97b | 4223 | |
3009a62f DS |
4224 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4225 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4226 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4227 | mutex_unlock(&fs_info->balance_mutex); |
4228 | ||
4229 | ret = __btrfs_balance(fs_info); | |
4230 | ||
4231 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4232 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4233 | btrfs_info(fs_info, "balance: paused"); | |
4234 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4235 | btrfs_info(fs_info, "balance: canceled"); | |
4236 | else | |
4237 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4238 | ||
3009a62f | 4239 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4240 | |
4241 | if (bargs) { | |
4242 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4243 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4244 | } |
4245 | ||
3a01aa7a ID |
4246 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4247 | balance_need_close(fs_info)) { | |
149196a2 | 4248 | reset_balance_state(fs_info); |
a17c95df | 4249 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4250 | } |
4251 | ||
837d5b6e | 4252 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4253 | |
4254 | return ret; | |
4255 | out: | |
59641015 | 4256 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4257 | reset_balance_state(fs_info); |
a17c95df | 4258 | else |
59641015 | 4259 | kfree(bctl); |
a17c95df DS |
4260 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4261 | ||
59641015 ID |
4262 | return ret; |
4263 | } | |
4264 | ||
4265 | static int balance_kthread(void *data) | |
4266 | { | |
2b6ba629 | 4267 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4268 | int ret = 0; |
59641015 | 4269 | |
59641015 | 4270 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4271 | if (fs_info->balance_ctl) |
6fcf6e2b | 4272 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4273 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4274 | |
59641015 ID |
4275 | return ret; |
4276 | } | |
4277 | ||
2b6ba629 ID |
4278 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4279 | { | |
4280 | struct task_struct *tsk; | |
4281 | ||
1354e1a1 | 4282 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4283 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4284 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4285 | return 0; |
4286 | } | |
1354e1a1 | 4287 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4288 | |
3cdde224 | 4289 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4290 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4291 | return 0; |
4292 | } | |
4293 | ||
02ee654d AJ |
4294 | /* |
4295 | * A ro->rw remount sequence should continue with the paused balance | |
4296 | * regardless of who pauses it, system or the user as of now, so set | |
4297 | * the resume flag. | |
4298 | */ | |
4299 | spin_lock(&fs_info->balance_lock); | |
4300 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4301 | spin_unlock(&fs_info->balance_lock); | |
4302 | ||
2b6ba629 | 4303 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4304 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4305 | } |
4306 | ||
68310a5e | 4307 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4308 | { |
59641015 ID |
4309 | struct btrfs_balance_control *bctl; |
4310 | struct btrfs_balance_item *item; | |
4311 | struct btrfs_disk_balance_args disk_bargs; | |
4312 | struct btrfs_path *path; | |
4313 | struct extent_buffer *leaf; | |
4314 | struct btrfs_key key; | |
4315 | int ret; | |
4316 | ||
4317 | path = btrfs_alloc_path(); | |
4318 | if (!path) | |
4319 | return -ENOMEM; | |
4320 | ||
59641015 | 4321 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4322 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4323 | key.offset = 0; |
4324 | ||
68310a5e | 4325 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4326 | if (ret < 0) |
68310a5e | 4327 | goto out; |
59641015 ID |
4328 | if (ret > 0) { /* ret = -ENOENT; */ |
4329 | ret = 0; | |
68310a5e ID |
4330 | goto out; |
4331 | } | |
4332 | ||
4333 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4334 | if (!bctl) { | |
4335 | ret = -ENOMEM; | |
4336 | goto out; | |
59641015 ID |
4337 | } |
4338 | ||
4339 | leaf = path->nodes[0]; | |
4340 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4341 | ||
68310a5e ID |
4342 | bctl->flags = btrfs_balance_flags(leaf, item); |
4343 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4344 | |
4345 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4346 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4347 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4348 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4349 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4350 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4351 | ||
eee95e3f DS |
4352 | /* |
4353 | * This should never happen, as the paused balance state is recovered | |
4354 | * during mount without any chance of other exclusive ops to collide. | |
4355 | * | |
4356 | * This gives the exclusive op status to balance and keeps in paused | |
4357 | * state until user intervention (cancel or umount). If the ownership | |
4358 | * cannot be assigned, show a message but do not fail. The balance | |
4359 | * is in a paused state and must have fs_info::balance_ctl properly | |
4360 | * set up. | |
4361 | */ | |
4362 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4363 | btrfs_warn(fs_info, | |
6dac13f8 | 4364 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4365 | |
68310a5e | 4366 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4367 | BUG_ON(fs_info->balance_ctl); |
4368 | spin_lock(&fs_info->balance_lock); | |
4369 | fs_info->balance_ctl = bctl; | |
4370 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4371 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4372 | out: |
4373 | btrfs_free_path(path); | |
ec44a35c CM |
4374 | return ret; |
4375 | } | |
4376 | ||
837d5b6e ID |
4377 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4378 | { | |
4379 | int ret = 0; | |
4380 | ||
4381 | mutex_lock(&fs_info->balance_mutex); | |
4382 | if (!fs_info->balance_ctl) { | |
4383 | mutex_unlock(&fs_info->balance_mutex); | |
4384 | return -ENOTCONN; | |
4385 | } | |
4386 | ||
3009a62f | 4387 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4388 | atomic_inc(&fs_info->balance_pause_req); |
4389 | mutex_unlock(&fs_info->balance_mutex); | |
4390 | ||
4391 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4392 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4393 | |
4394 | mutex_lock(&fs_info->balance_mutex); | |
4395 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4396 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4397 | atomic_dec(&fs_info->balance_pause_req); |
4398 | } else { | |
4399 | ret = -ENOTCONN; | |
4400 | } | |
4401 | ||
4402 | mutex_unlock(&fs_info->balance_mutex); | |
4403 | return ret; | |
4404 | } | |
4405 | ||
a7e99c69 ID |
4406 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4407 | { | |
4408 | mutex_lock(&fs_info->balance_mutex); | |
4409 | if (!fs_info->balance_ctl) { | |
4410 | mutex_unlock(&fs_info->balance_mutex); | |
4411 | return -ENOTCONN; | |
4412 | } | |
4413 | ||
cf7d20f4 DS |
4414 | /* |
4415 | * A paused balance with the item stored on disk can be resumed at | |
4416 | * mount time if the mount is read-write. Otherwise it's still paused | |
4417 | * and we must not allow cancelling as it deletes the item. | |
4418 | */ | |
4419 | if (sb_rdonly(fs_info->sb)) { | |
4420 | mutex_unlock(&fs_info->balance_mutex); | |
4421 | return -EROFS; | |
4422 | } | |
4423 | ||
a7e99c69 ID |
4424 | atomic_inc(&fs_info->balance_cancel_req); |
4425 | /* | |
4426 | * if we are running just wait and return, balance item is | |
4427 | * deleted in btrfs_balance in this case | |
4428 | */ | |
3009a62f | 4429 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4430 | mutex_unlock(&fs_info->balance_mutex); |
4431 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4432 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4433 | mutex_lock(&fs_info->balance_mutex); |
4434 | } else { | |
a7e99c69 | 4435 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4436 | /* |
4437 | * Lock released to allow other waiters to continue, we'll | |
4438 | * reexamine the status again. | |
4439 | */ | |
a7e99c69 ID |
4440 | mutex_lock(&fs_info->balance_mutex); |
4441 | ||
a17c95df | 4442 | if (fs_info->balance_ctl) { |
149196a2 | 4443 | reset_balance_state(fs_info); |
a17c95df | 4444 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4445 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4446 | } |
a7e99c69 ID |
4447 | } |
4448 | ||
3009a62f DS |
4449 | BUG_ON(fs_info->balance_ctl || |
4450 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4451 | atomic_dec(&fs_info->balance_cancel_req); |
4452 | mutex_unlock(&fs_info->balance_mutex); | |
4453 | return 0; | |
4454 | } | |
4455 | ||
803b2f54 SB |
4456 | static int btrfs_uuid_scan_kthread(void *data) |
4457 | { | |
4458 | struct btrfs_fs_info *fs_info = data; | |
4459 | struct btrfs_root *root = fs_info->tree_root; | |
4460 | struct btrfs_key key; | |
803b2f54 SB |
4461 | struct btrfs_path *path = NULL; |
4462 | int ret = 0; | |
4463 | struct extent_buffer *eb; | |
4464 | int slot; | |
4465 | struct btrfs_root_item root_item; | |
4466 | u32 item_size; | |
f45388f3 | 4467 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4468 | |
4469 | path = btrfs_alloc_path(); | |
4470 | if (!path) { | |
4471 | ret = -ENOMEM; | |
4472 | goto out; | |
4473 | } | |
4474 | ||
4475 | key.objectid = 0; | |
4476 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4477 | key.offset = 0; | |
4478 | ||
803b2f54 | 4479 | while (1) { |
7c829b72 AJ |
4480 | ret = btrfs_search_forward(root, &key, path, |
4481 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4482 | if (ret) { |
4483 | if (ret > 0) | |
4484 | ret = 0; | |
4485 | break; | |
4486 | } | |
4487 | ||
4488 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4489 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4490 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4491 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4492 | goto skip; | |
4493 | ||
4494 | eb = path->nodes[0]; | |
4495 | slot = path->slots[0]; | |
4496 | item_size = btrfs_item_size_nr(eb, slot); | |
4497 | if (item_size < sizeof(root_item)) | |
4498 | goto skip; | |
4499 | ||
803b2f54 SB |
4500 | read_extent_buffer(eb, &root_item, |
4501 | btrfs_item_ptr_offset(eb, slot), | |
4502 | (int)sizeof(root_item)); | |
4503 | if (btrfs_root_refs(&root_item) == 0) | |
4504 | goto skip; | |
f45388f3 FDBM |
4505 | |
4506 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4507 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4508 | if (trans) | |
4509 | goto update_tree; | |
4510 | ||
4511 | btrfs_release_path(path); | |
803b2f54 SB |
4512 | /* |
4513 | * 1 - subvol uuid item | |
4514 | * 1 - received_subvol uuid item | |
4515 | */ | |
4516 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4517 | if (IS_ERR(trans)) { | |
4518 | ret = PTR_ERR(trans); | |
4519 | break; | |
4520 | } | |
f45388f3 FDBM |
4521 | continue; |
4522 | } else { | |
4523 | goto skip; | |
4524 | } | |
4525 | update_tree: | |
4526 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4527 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4528 | BTRFS_UUID_KEY_SUBVOL, |
4529 | key.objectid); | |
4530 | if (ret < 0) { | |
efe120a0 | 4531 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4532 | ret); |
803b2f54 SB |
4533 | break; |
4534 | } | |
4535 | } | |
4536 | ||
4537 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4538 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4539 | root_item.received_uuid, |
4540 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4541 | key.objectid); | |
4542 | if (ret < 0) { | |
efe120a0 | 4543 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4544 | ret); |
803b2f54 SB |
4545 | break; |
4546 | } | |
4547 | } | |
4548 | ||
f45388f3 | 4549 | skip: |
803b2f54 | 4550 | if (trans) { |
3a45bb20 | 4551 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4552 | trans = NULL; |
803b2f54 SB |
4553 | if (ret) |
4554 | break; | |
4555 | } | |
4556 | ||
803b2f54 SB |
4557 | btrfs_release_path(path); |
4558 | if (key.offset < (u64)-1) { | |
4559 | key.offset++; | |
4560 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4561 | key.offset = 0; | |
4562 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4563 | } else if (key.objectid < (u64)-1) { | |
4564 | key.offset = 0; | |
4565 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4566 | key.objectid++; | |
4567 | } else { | |
4568 | break; | |
4569 | } | |
4570 | cond_resched(); | |
4571 | } | |
4572 | ||
4573 | out: | |
4574 | btrfs_free_path(path); | |
f45388f3 | 4575 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4576 | btrfs_end_transaction(trans); |
803b2f54 | 4577 | if (ret) |
efe120a0 | 4578 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4579 | else |
afcdd129 | 4580 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4581 | up(&fs_info->uuid_tree_rescan_sem); |
4582 | return 0; | |
4583 | } | |
4584 | ||
70f80175 SB |
4585 | /* |
4586 | * Callback for btrfs_uuid_tree_iterate(). | |
4587 | * returns: | |
4588 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4589 | * < 0 if an error occurred. |
70f80175 SB |
4590 | * > 0 if the check failed, which means the caller shall remove the entry. |
4591 | */ | |
4592 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4593 | u8 *uuid, u8 type, u64 subid) | |
4594 | { | |
4595 | struct btrfs_key key; | |
4596 | int ret = 0; | |
4597 | struct btrfs_root *subvol_root; | |
4598 | ||
4599 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4600 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4601 | goto out; | |
4602 | ||
4603 | key.objectid = subid; | |
4604 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4605 | key.offset = (u64)-1; | |
4606 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4607 | if (IS_ERR(subvol_root)) { | |
4608 | ret = PTR_ERR(subvol_root); | |
4609 | if (ret == -ENOENT) | |
4610 | ret = 1; | |
4611 | goto out; | |
4612 | } | |
4613 | ||
4614 | switch (type) { | |
4615 | case BTRFS_UUID_KEY_SUBVOL: | |
4616 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4617 | ret = 1; | |
4618 | break; | |
4619 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4620 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4621 | BTRFS_UUID_SIZE)) | |
4622 | ret = 1; | |
4623 | break; | |
4624 | } | |
4625 | ||
4626 | out: | |
4627 | return ret; | |
4628 | } | |
4629 | ||
4630 | static int btrfs_uuid_rescan_kthread(void *data) | |
4631 | { | |
4632 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4633 | int ret; | |
4634 | ||
4635 | /* | |
4636 | * 1st step is to iterate through the existing UUID tree and | |
4637 | * to delete all entries that contain outdated data. | |
4638 | * 2nd step is to add all missing entries to the UUID tree. | |
4639 | */ | |
4640 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4641 | if (ret < 0) { | |
efe120a0 | 4642 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4643 | up(&fs_info->uuid_tree_rescan_sem); |
4644 | return ret; | |
4645 | } | |
4646 | return btrfs_uuid_scan_kthread(data); | |
4647 | } | |
4648 | ||
f7a81ea4 SB |
4649 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4650 | { | |
4651 | struct btrfs_trans_handle *trans; | |
4652 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4653 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4654 | struct task_struct *task; |
4655 | int ret; | |
f7a81ea4 SB |
4656 | |
4657 | /* | |
4658 | * 1 - root node | |
4659 | * 1 - root item | |
4660 | */ | |
4661 | trans = btrfs_start_transaction(tree_root, 2); | |
4662 | if (IS_ERR(trans)) | |
4663 | return PTR_ERR(trans); | |
4664 | ||
4665 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4666 | BTRFS_UUID_TREE_OBJECTID); | |
4667 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4668 | ret = PTR_ERR(uuid_root); |
66642832 | 4669 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4670 | btrfs_end_transaction(trans); |
6d13f549 | 4671 | return ret; |
f7a81ea4 SB |
4672 | } |
4673 | ||
4674 | fs_info->uuid_root = uuid_root; | |
4675 | ||
3a45bb20 | 4676 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4677 | if (ret) |
4678 | return ret; | |
4679 | ||
4680 | down(&fs_info->uuid_tree_rescan_sem); | |
4681 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4682 | if (IS_ERR(task)) { | |
70f80175 | 4683 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4684 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4685 | up(&fs_info->uuid_tree_rescan_sem); |
4686 | return PTR_ERR(task); | |
4687 | } | |
4688 | ||
4689 | return 0; | |
f7a81ea4 | 4690 | } |
803b2f54 | 4691 | |
70f80175 SB |
4692 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4693 | { | |
4694 | struct task_struct *task; | |
4695 | ||
4696 | down(&fs_info->uuid_tree_rescan_sem); | |
4697 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4698 | if (IS_ERR(task)) { | |
4699 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4700 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4701 | up(&fs_info->uuid_tree_rescan_sem); |
4702 | return PTR_ERR(task); | |
4703 | } | |
4704 | ||
4705 | return 0; | |
4706 | } | |
4707 | ||
8f18cf13 CM |
4708 | /* |
4709 | * shrinking a device means finding all of the device extents past | |
4710 | * the new size, and then following the back refs to the chunks. | |
4711 | * The chunk relocation code actually frees the device extent | |
4712 | */ | |
4713 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4714 | { | |
0b246afa JM |
4715 | struct btrfs_fs_info *fs_info = device->fs_info; |
4716 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4717 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4718 | struct btrfs_dev_extent *dev_extent = NULL; |
4719 | struct btrfs_path *path; | |
4720 | u64 length; | |
8f18cf13 CM |
4721 | u64 chunk_offset; |
4722 | int ret; | |
4723 | int slot; | |
ba1bf481 JB |
4724 | int failed = 0; |
4725 | bool retried = false; | |
53e489bc | 4726 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4727 | struct extent_buffer *l; |
4728 | struct btrfs_key key; | |
0b246afa | 4729 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4730 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4731 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 NB |
4732 | u64 diff; |
4733 | ||
4734 | new_size = round_down(new_size, fs_info->sectorsize); | |
0e4324a4 | 4735 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4736 | |
401e29c1 | 4737 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4738 | return -EINVAL; |
4739 | ||
8f18cf13 CM |
4740 | path = btrfs_alloc_path(); |
4741 | if (!path) | |
4742 | return -ENOMEM; | |
4743 | ||
0338dff6 | 4744 | path->reada = READA_BACK; |
8f18cf13 | 4745 | |
34441361 | 4746 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4747 | |
7cc8e58d | 4748 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4749 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4750 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4751 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4752 | } |
34441361 | 4753 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 4754 | |
ba1bf481 | 4755 | again: |
8f18cf13 CM |
4756 | key.objectid = device->devid; |
4757 | key.offset = (u64)-1; | |
4758 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4759 | ||
213e64da | 4760 | do { |
0b246afa | 4761 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4762 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4763 | if (ret < 0) { |
0b246afa | 4764 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4765 | goto done; |
67c5e7d4 | 4766 | } |
8f18cf13 CM |
4767 | |
4768 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4769 | if (ret) |
0b246afa | 4770 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4771 | if (ret < 0) |
4772 | goto done; | |
4773 | if (ret) { | |
4774 | ret = 0; | |
b3b4aa74 | 4775 | btrfs_release_path(path); |
bf1fb512 | 4776 | break; |
8f18cf13 CM |
4777 | } |
4778 | ||
4779 | l = path->nodes[0]; | |
4780 | slot = path->slots[0]; | |
4781 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4782 | ||
ba1bf481 | 4783 | if (key.objectid != device->devid) { |
0b246afa | 4784 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4785 | btrfs_release_path(path); |
bf1fb512 | 4786 | break; |
ba1bf481 | 4787 | } |
8f18cf13 CM |
4788 | |
4789 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4790 | length = btrfs_dev_extent_length(l, dev_extent); | |
4791 | ||
ba1bf481 | 4792 | if (key.offset + length <= new_size) { |
0b246afa | 4793 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4794 | btrfs_release_path(path); |
d6397bae | 4795 | break; |
ba1bf481 | 4796 | } |
8f18cf13 | 4797 | |
8f18cf13 | 4798 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4799 | btrfs_release_path(path); |
8f18cf13 | 4800 | |
a6f93c71 LB |
4801 | /* |
4802 | * We may be relocating the only data chunk we have, | |
4803 | * which could potentially end up with losing data's | |
4804 | * raid profile, so lets allocate an empty one in | |
4805 | * advance. | |
4806 | */ | |
4807 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4808 | if (ret < 0) { | |
4809 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4810 | goto done; | |
4811 | } | |
4812 | ||
0b246afa JM |
4813 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4814 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4815 | if (ret == -ENOSPC) { |
ba1bf481 | 4816 | failed++; |
eede2bf3 OS |
4817 | } else if (ret) { |
4818 | if (ret == -ETXTBSY) { | |
4819 | btrfs_warn(fs_info, | |
4820 | "could not shrink block group %llu due to active swapfile", | |
4821 | chunk_offset); | |
4822 | } | |
4823 | goto done; | |
4824 | } | |
213e64da | 4825 | } while (key.offset-- > 0); |
ba1bf481 JB |
4826 | |
4827 | if (failed && !retried) { | |
4828 | failed = 0; | |
4829 | retried = true; | |
4830 | goto again; | |
4831 | } else if (failed && retried) { | |
4832 | ret = -ENOSPC; | |
ba1bf481 | 4833 | goto done; |
8f18cf13 CM |
4834 | } |
4835 | ||
d6397bae | 4836 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4837 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4838 | if (IS_ERR(trans)) { |
4839 | ret = PTR_ERR(trans); | |
4840 | goto done; | |
4841 | } | |
4842 | ||
34441361 | 4843 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc FM |
4844 | |
4845 | /* | |
4846 | * We checked in the above loop all device extents that were already in | |
4847 | * the device tree. However before we have updated the device's | |
4848 | * total_bytes to the new size, we might have had chunk allocations that | |
4849 | * have not complete yet (new block groups attached to transaction | |
4850 | * handles), and therefore their device extents were not yet in the | |
4851 | * device tree and we missed them in the loop above. So if we have any | |
4852 | * pending chunk using a device extent that overlaps the device range | |
4853 | * that we can not use anymore, commit the current transaction and | |
4854 | * repeat the search on the device tree - this way we guarantee we will | |
4855 | * not have chunks using device extents that end beyond 'new_size'. | |
4856 | */ | |
4857 | if (!checked_pending_chunks) { | |
4858 | u64 start = new_size; | |
4859 | u64 len = old_size - new_size; | |
4860 | ||
499f377f JM |
4861 | if (contains_pending_extent(trans->transaction, device, |
4862 | &start, len)) { | |
34441361 | 4863 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc FM |
4864 | checked_pending_chunks = true; |
4865 | failed = 0; | |
4866 | retried = false; | |
3a45bb20 | 4867 | ret = btrfs_commit_transaction(trans); |
53e489bc FM |
4868 | if (ret) |
4869 | goto done; | |
4870 | goto again; | |
4871 | } | |
4872 | } | |
4873 | ||
7cc8e58d | 4874 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4875 | if (list_empty(&device->resized_list)) |
4876 | list_add_tail(&device->resized_list, | |
0b246afa | 4877 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4878 | |
d6397bae | 4879 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4880 | btrfs_set_super_total_bytes(super_copy, |
4881 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4882 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4883 | |
4884 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4885 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4886 | if (ret < 0) { |
4887 | btrfs_abort_transaction(trans, ret); | |
4888 | btrfs_end_transaction(trans); | |
4889 | } else { | |
4890 | ret = btrfs_commit_transaction(trans); | |
4891 | } | |
8f18cf13 CM |
4892 | done: |
4893 | btrfs_free_path(path); | |
53e489bc | 4894 | if (ret) { |
34441361 | 4895 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4896 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4897 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4898 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4899 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4900 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4901 | } |
8f18cf13 CM |
4902 | return ret; |
4903 | } | |
4904 | ||
2ff7e61e | 4905 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4906 | struct btrfs_key *key, |
4907 | struct btrfs_chunk *chunk, int item_size) | |
4908 | { | |
0b246afa | 4909 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4910 | struct btrfs_disk_key disk_key; |
4911 | u32 array_size; | |
4912 | u8 *ptr; | |
4913 | ||
34441361 | 4914 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4915 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4916 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4917 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4918 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4919 | return -EFBIG; |
fe48a5c0 | 4920 | } |
0b86a832 CM |
4921 | |
4922 | ptr = super_copy->sys_chunk_array + array_size; | |
4923 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4924 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4925 | ptr += sizeof(disk_key); | |
4926 | memcpy(ptr, chunk, item_size); | |
4927 | item_size += sizeof(disk_key); | |
4928 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4929 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4930 | |
0b86a832 CM |
4931 | return 0; |
4932 | } | |
4933 | ||
73c5de00 AJ |
4934 | /* |
4935 | * sort the devices in descending order by max_avail, total_avail | |
4936 | */ | |
4937 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4938 | { |
73c5de00 AJ |
4939 | const struct btrfs_device_info *di_a = a; |
4940 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4941 | |
73c5de00 | 4942 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4943 | return -1; |
73c5de00 | 4944 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4945 | return 1; |
73c5de00 AJ |
4946 | if (di_a->total_avail > di_b->total_avail) |
4947 | return -1; | |
4948 | if (di_a->total_avail < di_b->total_avail) | |
4949 | return 1; | |
4950 | return 0; | |
b2117a39 | 4951 | } |
0b86a832 | 4952 | |
53b381b3 DW |
4953 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4954 | { | |
ffe2d203 | 4955 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4956 | return; |
4957 | ||
ceda0864 | 4958 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4959 | } |
4960 | ||
73c5de00 | 4961 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4962 | u64 start, u64 type) |
b2117a39 | 4963 | { |
2ff7e61e | 4964 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4965 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4966 | struct btrfs_device *device; |
73c5de00 AJ |
4967 | struct map_lookup *map = NULL; |
4968 | struct extent_map_tree *em_tree; | |
4969 | struct extent_map *em; | |
4970 | struct btrfs_device_info *devices_info = NULL; | |
4971 | u64 total_avail; | |
4972 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4973 | int data_stripes; /* number of stripes that count for |
4974 | block group size */ | |
73c5de00 AJ |
4975 | int sub_stripes; /* sub_stripes info for map */ |
4976 | int dev_stripes; /* stripes per dev */ | |
4977 | int devs_max; /* max devs to use */ | |
4978 | int devs_min; /* min devs needed */ | |
4979 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4980 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4981 | int nparity; /* number of stripes worth of bytes to |
4982 | store parity information */ | |
73c5de00 AJ |
4983 | int ret; |
4984 | u64 max_stripe_size; | |
4985 | u64 max_chunk_size; | |
4986 | u64 stripe_size; | |
23f0ff1e | 4987 | u64 chunk_size; |
73c5de00 AJ |
4988 | int ndevs; |
4989 | int i; | |
4990 | int j; | |
31e50229 | 4991 | int index; |
593060d7 | 4992 | |
0c460c0d | 4993 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4994 | |
4117f207 QW |
4995 | if (list_empty(&fs_devices->alloc_list)) { |
4996 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4997 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4998 | return -ENOSPC; |
4117f207 | 4999 | } |
b2117a39 | 5000 | |
3e72ee88 | 5001 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 5002 | |
31e50229 LB |
5003 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
5004 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5005 | devs_max = btrfs_raid_array[index].devs_max; | |
5006 | devs_min = btrfs_raid_array[index].devs_min; | |
5007 | devs_increment = btrfs_raid_array[index].devs_increment; | |
5008 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 5009 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 5010 | |
9b3f68b9 | 5011 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 5012 | max_stripe_size = SZ_1G; |
fce466ea | 5013 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 5014 | if (!devs_max) |
062d4d1f | 5015 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 5016 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 5017 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
5018 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
5019 | max_stripe_size = SZ_1G; | |
1100373f | 5020 | else |
ee22184b | 5021 | max_stripe_size = SZ_256M; |
73c5de00 | 5022 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 5023 | if (!devs_max) |
062d4d1f | 5024 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 5025 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 5026 | max_stripe_size = SZ_32M; |
73c5de00 | 5027 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
5028 | if (!devs_max) |
5029 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 5030 | } else { |
351fd353 | 5031 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 5032 | type); |
290342f6 | 5033 | BUG(); |
9b3f68b9 CM |
5034 | } |
5035 | ||
52042d8e | 5036 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
5037 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
5038 | max_chunk_size); | |
9b3f68b9 | 5039 | |
31e818fe | 5040 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
5041 | GFP_NOFS); |
5042 | if (!devices_info) | |
5043 | return -ENOMEM; | |
0cad8a11 | 5044 | |
9f680ce0 | 5045 | /* |
73c5de00 AJ |
5046 | * in the first pass through the devices list, we gather information |
5047 | * about the available holes on each device. | |
9f680ce0 | 5048 | */ |
73c5de00 | 5049 | ndevs = 0; |
ebcc9301 | 5050 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
5051 | u64 max_avail; |
5052 | u64 dev_offset; | |
b2117a39 | 5053 | |
ebbede42 | 5054 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5055 | WARN(1, KERN_ERR |
efe120a0 | 5056 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5057 | continue; |
5058 | } | |
b2117a39 | 5059 | |
e12c9621 AJ |
5060 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5061 | &device->dev_state) || | |
401e29c1 | 5062 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5063 | continue; |
b2117a39 | 5064 | |
73c5de00 AJ |
5065 | if (device->total_bytes > device->bytes_used) |
5066 | total_avail = device->total_bytes - device->bytes_used; | |
5067 | else | |
5068 | total_avail = 0; | |
38c01b96 | 5069 | |
5070 | /* If there is no space on this device, skip it. */ | |
5071 | if (total_avail == 0) | |
5072 | continue; | |
b2117a39 | 5073 | |
6df9a95e | 5074 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
5075 | max_stripe_size * dev_stripes, |
5076 | &dev_offset, &max_avail); | |
5077 | if (ret && ret != -ENOSPC) | |
5078 | goto error; | |
b2117a39 | 5079 | |
73c5de00 AJ |
5080 | if (ret == 0) |
5081 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5082 | |
4117f207 QW |
5083 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5084 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5085 | btrfs_debug(info, | |
5086 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5087 | __func__, device->devid, max_avail, | |
5088 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5089 | continue; |
4117f207 | 5090 | } |
b2117a39 | 5091 | |
063d006f ES |
5092 | if (ndevs == fs_devices->rw_devices) { |
5093 | WARN(1, "%s: found more than %llu devices\n", | |
5094 | __func__, fs_devices->rw_devices); | |
5095 | break; | |
5096 | } | |
73c5de00 AJ |
5097 | devices_info[ndevs].dev_offset = dev_offset; |
5098 | devices_info[ndevs].max_avail = max_avail; | |
5099 | devices_info[ndevs].total_avail = total_avail; | |
5100 | devices_info[ndevs].dev = device; | |
5101 | ++ndevs; | |
5102 | } | |
b2117a39 | 5103 | |
73c5de00 AJ |
5104 | /* |
5105 | * now sort the devices by hole size / available space | |
5106 | */ | |
5107 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5108 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5109 | |
73c5de00 | 5110 | /* round down to number of usable stripes */ |
e5600fd6 | 5111 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5112 | |
ba89b802 | 5113 | if (ndevs < devs_min) { |
73c5de00 | 5114 | ret = -ENOSPC; |
4117f207 QW |
5115 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5116 | btrfs_debug(info, | |
5117 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5118 | __func__, ndevs, devs_min); |
4117f207 | 5119 | } |
73c5de00 | 5120 | goto error; |
b2117a39 | 5121 | } |
9f680ce0 | 5122 | |
f148ef4d NB |
5123 | ndevs = min(ndevs, devs_max); |
5124 | ||
73c5de00 | 5125 | /* |
92e222df HK |
5126 | * The primary goal is to maximize the number of stripes, so use as |
5127 | * many devices as possible, even if the stripes are not maximum sized. | |
5128 | * | |
5129 | * The DUP profile stores more than one stripe per device, the | |
5130 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5131 | */ |
92e222df | 5132 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5133 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5134 | |
53b381b3 DW |
5135 | /* |
5136 | * this will have to be fixed for RAID1 and RAID10 over | |
5137 | * more drives | |
5138 | */ | |
b50836ed | 5139 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5140 | |
5141 | /* | |
5142 | * Use the number of data stripes to figure out how big this chunk | |
5143 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5144 | * and compare that answer with the max chunk size. If it's higher, |
5145 | * we try to reduce stripe_size. | |
86db2578 CM |
5146 | */ |
5147 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5148 | /* |
baf92114 HK |
5149 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5150 | * then use it, unless it ends up being even bigger than the | |
5151 | * previous value we had already. | |
86db2578 | 5152 | */ |
baf92114 HK |
5153 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5154 | data_stripes), SZ_16M), | |
793ff2c8 | 5155 | stripe_size); |
86db2578 CM |
5156 | } |
5157 | ||
37db63a4 | 5158 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5159 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5160 | |
5161 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5162 | if (!map) { | |
5163 | ret = -ENOMEM; | |
5164 | goto error; | |
5165 | } | |
5166 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5167 | |
73c5de00 AJ |
5168 | for (i = 0; i < ndevs; ++i) { |
5169 | for (j = 0; j < dev_stripes; ++j) { | |
5170 | int s = i * dev_stripes + j; | |
5171 | map->stripes[s].dev = devices_info[i].dev; | |
5172 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5173 | j * stripe_size; | |
6324fbf3 | 5174 | } |
6324fbf3 | 5175 | } |
500ceed8 NB |
5176 | map->stripe_len = BTRFS_STRIPE_LEN; |
5177 | map->io_align = BTRFS_STRIPE_LEN; | |
5178 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5179 | map->type = type; |
2b82032c | 5180 | map->sub_stripes = sub_stripes; |
0b86a832 | 5181 | |
23f0ff1e | 5182 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5183 | |
23f0ff1e | 5184 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5185 | |
172ddd60 | 5186 | em = alloc_extent_map(); |
2b82032c | 5187 | if (!em) { |
298a8f9c | 5188 | kfree(map); |
b2117a39 MX |
5189 | ret = -ENOMEM; |
5190 | goto error; | |
593060d7 | 5191 | } |
298a8f9c | 5192 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5193 | em->map_lookup = map; |
2b82032c | 5194 | em->start = start; |
23f0ff1e | 5195 | em->len = chunk_size; |
2b82032c YZ |
5196 | em->block_start = 0; |
5197 | em->block_len = em->len; | |
6df9a95e | 5198 | em->orig_block_len = stripe_size; |
593060d7 | 5199 | |
0b246afa | 5200 | em_tree = &info->mapping_tree.map_tree; |
890871be | 5201 | write_lock(&em_tree->lock); |
09a2a8f9 | 5202 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5203 | if (ret) { |
1efb72a3 | 5204 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5205 | free_extent_map(em); |
1dd4602f | 5206 | goto error; |
0f5d42b2 | 5207 | } |
0b86a832 | 5208 | |
1efb72a3 NB |
5209 | list_add_tail(&em->list, &trans->transaction->pending_chunks); |
5210 | refcount_inc(&em->refs); | |
5211 | write_unlock(&em_tree->lock); | |
5212 | ||
23f0ff1e | 5213 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5214 | if (ret) |
5215 | goto error_del_extent; | |
2b82032c | 5216 | |
2f29df4f HK |
5217 | for (i = 0; i < map->num_stripes; i++) |
5218 | btrfs_device_set_bytes_used(map->stripes[i].dev, | |
5219 | map->stripes[i].dev->bytes_used + stripe_size); | |
43530c46 | 5220 | |
a5ed45f8 | 5221 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5222 | |
0f5d42b2 | 5223 | free_extent_map(em); |
0b246afa | 5224 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5225 | |
b2117a39 | 5226 | kfree(devices_info); |
2b82032c | 5227 | return 0; |
b2117a39 | 5228 | |
6df9a95e | 5229 | error_del_extent: |
0f5d42b2 JB |
5230 | write_lock(&em_tree->lock); |
5231 | remove_extent_mapping(em_tree, em); | |
5232 | write_unlock(&em_tree->lock); | |
5233 | ||
5234 | /* One for our allocation */ | |
5235 | free_extent_map(em); | |
5236 | /* One for the tree reference */ | |
5237 | free_extent_map(em); | |
495e64f4 FM |
5238 | /* One for the pending_chunks list reference */ |
5239 | free_extent_map(em); | |
b2117a39 | 5240 | error: |
b2117a39 MX |
5241 | kfree(devices_info); |
5242 | return ret; | |
2b82032c YZ |
5243 | } |
5244 | ||
6df9a95e | 5245 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5246 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5247 | { |
97aff912 | 5248 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5249 | struct btrfs_root *extent_root = fs_info->extent_root; |
5250 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5251 | struct btrfs_key key; |
2b82032c YZ |
5252 | struct btrfs_device *device; |
5253 | struct btrfs_chunk *chunk; | |
5254 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5255 | struct extent_map *em; |
5256 | struct map_lookup *map; | |
5257 | size_t item_size; | |
5258 | u64 dev_offset; | |
5259 | u64 stripe_size; | |
5260 | int i = 0; | |
140e639f | 5261 | int ret = 0; |
2b82032c | 5262 | |
60ca842e | 5263 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5264 | if (IS_ERR(em)) |
5265 | return PTR_ERR(em); | |
6df9a95e | 5266 | |
95617d69 | 5267 | map = em->map_lookup; |
6df9a95e JB |
5268 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5269 | stripe_size = em->orig_block_len; | |
5270 | ||
2b82032c | 5271 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5272 | if (!chunk) { |
5273 | ret = -ENOMEM; | |
5274 | goto out; | |
5275 | } | |
5276 | ||
50460e37 FM |
5277 | /* |
5278 | * Take the device list mutex to prevent races with the final phase of | |
5279 | * a device replace operation that replaces the device object associated | |
5280 | * with the map's stripes, because the device object's id can change | |
5281 | * at any time during that final phase of the device replace operation | |
5282 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5283 | */ | |
0b246afa | 5284 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5285 | for (i = 0; i < map->num_stripes; i++) { |
5286 | device = map->stripes[i].dev; | |
5287 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5288 | |
0b86a832 | 5289 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5290 | if (ret) |
50460e37 | 5291 | break; |
b5d9071c NB |
5292 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5293 | dev_offset, stripe_size); | |
6df9a95e | 5294 | if (ret) |
50460e37 FM |
5295 | break; |
5296 | } | |
5297 | if (ret) { | |
0b246afa | 5298 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5299 | goto out; |
2b82032c YZ |
5300 | } |
5301 | ||
2b82032c | 5302 | stripe = &chunk->stripe; |
6df9a95e JB |
5303 | for (i = 0; i < map->num_stripes; i++) { |
5304 | device = map->stripes[i].dev; | |
5305 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5306 | |
e17cade2 CM |
5307 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5308 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5309 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5310 | stripe++; |
0b86a832 | 5311 | } |
0b246afa | 5312 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5313 | |
2b82032c | 5314 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5315 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5316 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5317 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5318 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5319 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5320 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5321 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5322 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5323 | |
2b82032c YZ |
5324 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5325 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5326 | key.offset = chunk_offset; | |
0b86a832 | 5327 | |
2b82032c | 5328 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5329 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5330 | /* | |
5331 | * TODO: Cleanup of inserted chunk root in case of | |
5332 | * failure. | |
5333 | */ | |
2ff7e61e | 5334 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5335 | } |
1abe9b8a | 5336 | |
6df9a95e | 5337 | out: |
0b86a832 | 5338 | kfree(chunk); |
6df9a95e | 5339 | free_extent_map(em); |
4ed1d16e | 5340 | return ret; |
2b82032c | 5341 | } |
0b86a832 | 5342 | |
2b82032c | 5343 | /* |
52042d8e AG |
5344 | * Chunk allocation falls into two parts. The first part does work |
5345 | * that makes the new allocated chunk usable, but does not do any operation | |
5346 | * that modifies the chunk tree. The second part does the work that | |
5347 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5348 | * bootstrap process of adding storage to a seed btrfs. |
5349 | */ | |
c216b203 | 5350 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5351 | { |
5352 | u64 chunk_offset; | |
2b82032c | 5353 | |
c216b203 NB |
5354 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5355 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5356 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5357 | } |
5358 | ||
d397712b | 5359 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
e4a4dce7 | 5360 | struct btrfs_fs_info *fs_info) |
2b82032c YZ |
5361 | { |
5362 | u64 chunk_offset; | |
5363 | u64 sys_chunk_offset; | |
2b82032c | 5364 | u64 alloc_profile; |
2b82032c YZ |
5365 | int ret; |
5366 | ||
6df9a95e | 5367 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5368 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5369 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5370 | if (ret) |
5371 | return ret; | |
2b82032c | 5372 | |
0b246afa | 5373 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5374 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5375 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5376 | return ret; |
2b82032c YZ |
5377 | } |
5378 | ||
d20983b4 MX |
5379 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5380 | { | |
5381 | int max_errors; | |
5382 | ||
5383 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5384 | BTRFS_BLOCK_GROUP_RAID10 | | |
5385 | BTRFS_BLOCK_GROUP_RAID5 | | |
5386 | BTRFS_BLOCK_GROUP_DUP)) { | |
5387 | max_errors = 1; | |
5388 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5389 | max_errors = 2; | |
5390 | } else { | |
5391 | max_errors = 0; | |
005d6427 | 5392 | } |
2b82032c | 5393 | |
d20983b4 | 5394 | return max_errors; |
2b82032c YZ |
5395 | } |
5396 | ||
2ff7e61e | 5397 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5398 | { |
5399 | struct extent_map *em; | |
5400 | struct map_lookup *map; | |
2b82032c | 5401 | int readonly = 0; |
d20983b4 | 5402 | int miss_ndevs = 0; |
2b82032c YZ |
5403 | int i; |
5404 | ||
60ca842e | 5405 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5406 | if (IS_ERR(em)) |
2b82032c YZ |
5407 | return 1; |
5408 | ||
95617d69 | 5409 | map = em->map_lookup; |
2b82032c | 5410 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5411 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5412 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5413 | miss_ndevs++; |
5414 | continue; | |
5415 | } | |
ebbede42 AJ |
5416 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5417 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5418 | readonly = 1; |
d20983b4 | 5419 | goto end; |
2b82032c YZ |
5420 | } |
5421 | } | |
d20983b4 MX |
5422 | |
5423 | /* | |
5424 | * If the number of missing devices is larger than max errors, | |
5425 | * we can not write the data into that chunk successfully, so | |
5426 | * set it readonly. | |
5427 | */ | |
5428 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5429 | readonly = 1; | |
5430 | end: | |
0b86a832 | 5431 | free_extent_map(em); |
2b82032c | 5432 | return readonly; |
0b86a832 CM |
5433 | } |
5434 | ||
5435 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5436 | { | |
a8067e02 | 5437 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5438 | } |
5439 | ||
5440 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5441 | { | |
5442 | struct extent_map *em; | |
5443 | ||
d397712b | 5444 | while (1) { |
890871be | 5445 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5446 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5447 | if (em) | |
5448 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5449 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5450 | if (!em) |
5451 | break; | |
0b86a832 CM |
5452 | /* once for us */ |
5453 | free_extent_map(em); | |
5454 | /* once for the tree */ | |
5455 | free_extent_map(em); | |
5456 | } | |
5457 | } | |
5458 | ||
5d964051 | 5459 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5460 | { |
5461 | struct extent_map *em; | |
5462 | struct map_lookup *map; | |
f188591e CM |
5463 | int ret; |
5464 | ||
60ca842e | 5465 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5466 | if (IS_ERR(em)) |
5467 | /* | |
5468 | * We could return errors for these cases, but that could get | |
5469 | * ugly and we'd probably do the same thing which is just not do | |
5470 | * anything else and exit, so return 1 so the callers don't try | |
5471 | * to use other copies. | |
5472 | */ | |
fb7669b5 | 5473 | return 1; |
fb7669b5 | 5474 | |
95617d69 | 5475 | map = em->map_lookup; |
f188591e CM |
5476 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5477 | ret = map->num_stripes; | |
321aecc6 CM |
5478 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5479 | ret = map->sub_stripes; | |
53b381b3 DW |
5480 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5481 | ret = 2; | |
5482 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5483 | /* |
5484 | * There could be two corrupted data stripes, we need | |
5485 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5486 | * |
8810f751 LB |
5487 | * Fail a stripe at a time on every retry except the |
5488 | * stripe under reconstruction. | |
5489 | */ | |
5490 | ret = map->num_stripes; | |
f188591e CM |
5491 | else |
5492 | ret = 1; | |
5493 | free_extent_map(em); | |
ad6d620e | 5494 | |
cb5583dd | 5495 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5496 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5497 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5498 | ret++; |
cb5583dd | 5499 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5500 | |
f188591e CM |
5501 | return ret; |
5502 | } | |
5503 | ||
2ff7e61e | 5504 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5505 | u64 logical) |
5506 | { | |
5507 | struct extent_map *em; | |
5508 | struct map_lookup *map; | |
0b246afa | 5509 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5510 | |
60ca842e | 5511 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5512 | |
69f03f13 NB |
5513 | if (!WARN_ON(IS_ERR(em))) { |
5514 | map = em->map_lookup; | |
5515 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5516 | len = map->stripe_len * nr_data_stripes(map); | |
5517 | free_extent_map(em); | |
5518 | } | |
53b381b3 DW |
5519 | return len; |
5520 | } | |
5521 | ||
e4ff5fb5 | 5522 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5523 | { |
5524 | struct extent_map *em; | |
5525 | struct map_lookup *map; | |
53b381b3 DW |
5526 | int ret = 0; |
5527 | ||
60ca842e | 5528 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5529 | |
69f03f13 NB |
5530 | if(!WARN_ON(IS_ERR(em))) { |
5531 | map = em->map_lookup; | |
5532 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5533 | ret = 1; | |
5534 | free_extent_map(em); | |
5535 | } | |
53b381b3 DW |
5536 | return ret; |
5537 | } | |
5538 | ||
30d9861f | 5539 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5540 | struct map_lookup *map, int first, |
8ba0ae78 | 5541 | int dev_replace_is_ongoing) |
dfe25020 CM |
5542 | { |
5543 | int i; | |
99f92a7c | 5544 | int num_stripes; |
8ba0ae78 | 5545 | int preferred_mirror; |
30d9861f SB |
5546 | int tolerance; |
5547 | struct btrfs_device *srcdev; | |
5548 | ||
99f92a7c AJ |
5549 | ASSERT((map->type & |
5550 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5551 | ||
5552 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5553 | num_stripes = map->sub_stripes; | |
5554 | else | |
5555 | num_stripes = map->num_stripes; | |
5556 | ||
8ba0ae78 AJ |
5557 | preferred_mirror = first + current->pid % num_stripes; |
5558 | ||
30d9861f SB |
5559 | if (dev_replace_is_ongoing && |
5560 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5561 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5562 | srcdev = fs_info->dev_replace.srcdev; | |
5563 | else | |
5564 | srcdev = NULL; | |
5565 | ||
5566 | /* | |
5567 | * try to avoid the drive that is the source drive for a | |
5568 | * dev-replace procedure, only choose it if no other non-missing | |
5569 | * mirror is available | |
5570 | */ | |
5571 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5572 | if (map->stripes[preferred_mirror].dev->bdev && |
5573 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5574 | return preferred_mirror; | |
99f92a7c | 5575 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5576 | if (map->stripes[i].dev->bdev && |
5577 | (tolerance || map->stripes[i].dev != srcdev)) | |
5578 | return i; | |
5579 | } | |
dfe25020 | 5580 | } |
30d9861f | 5581 | |
dfe25020 CM |
5582 | /* we couldn't find one that doesn't fail. Just return something |
5583 | * and the io error handling code will clean up eventually | |
5584 | */ | |
8ba0ae78 | 5585 | return preferred_mirror; |
dfe25020 CM |
5586 | } |
5587 | ||
53b381b3 DW |
5588 | static inline int parity_smaller(u64 a, u64 b) |
5589 | { | |
5590 | return a > b; | |
5591 | } | |
5592 | ||
5593 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5594 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5595 | { |
5596 | struct btrfs_bio_stripe s; | |
5597 | int i; | |
5598 | u64 l; | |
5599 | int again = 1; | |
5600 | ||
5601 | while (again) { | |
5602 | again = 0; | |
cc7539ed | 5603 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5604 | if (parity_smaller(bbio->raid_map[i], |
5605 | bbio->raid_map[i+1])) { | |
53b381b3 | 5606 | s = bbio->stripes[i]; |
8e5cfb55 | 5607 | l = bbio->raid_map[i]; |
53b381b3 | 5608 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5609 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5610 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5611 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5612 | |
53b381b3 DW |
5613 | again = 1; |
5614 | } | |
5615 | } | |
5616 | } | |
5617 | } | |
5618 | ||
6e9606d2 ZL |
5619 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5620 | { | |
5621 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5622 | /* the size of the btrfs_bio */ |
6e9606d2 | 5623 | sizeof(struct btrfs_bio) + |
e57cf21e | 5624 | /* plus the variable array for the stripes */ |
6e9606d2 | 5625 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5626 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5627 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5628 | /* |
5629 | * plus the raid_map, which includes both the tgt dev | |
5630 | * and the stripes | |
5631 | */ | |
5632 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5633 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5634 | |
5635 | atomic_set(&bbio->error, 0); | |
140475ae | 5636 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5637 | |
5638 | return bbio; | |
5639 | } | |
5640 | ||
5641 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5642 | { | |
140475ae ER |
5643 | WARN_ON(!refcount_read(&bbio->refs)); |
5644 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5645 | } |
5646 | ||
5647 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5648 | { | |
5649 | if (!bbio) | |
5650 | return; | |
140475ae | 5651 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5652 | kfree(bbio); |
5653 | } | |
5654 | ||
0b3d4cd3 LB |
5655 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5656 | /* | |
5657 | * Please note that, discard won't be sent to target device of device | |
5658 | * replace. | |
5659 | */ | |
5660 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5661 | u64 logical, u64 length, | |
5662 | struct btrfs_bio **bbio_ret) | |
5663 | { | |
5664 | struct extent_map *em; | |
5665 | struct map_lookup *map; | |
5666 | struct btrfs_bio *bbio; | |
5667 | u64 offset; | |
5668 | u64 stripe_nr; | |
5669 | u64 stripe_nr_end; | |
5670 | u64 stripe_end_offset; | |
5671 | u64 stripe_cnt; | |
5672 | u64 stripe_len; | |
5673 | u64 stripe_offset; | |
5674 | u64 num_stripes; | |
5675 | u32 stripe_index; | |
5676 | u32 factor = 0; | |
5677 | u32 sub_stripes = 0; | |
5678 | u64 stripes_per_dev = 0; | |
5679 | u32 remaining_stripes = 0; | |
5680 | u32 last_stripe = 0; | |
5681 | int ret = 0; | |
5682 | int i; | |
5683 | ||
5684 | /* discard always return a bbio */ | |
5685 | ASSERT(bbio_ret); | |
5686 | ||
60ca842e | 5687 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5688 | if (IS_ERR(em)) |
5689 | return PTR_ERR(em); | |
5690 | ||
5691 | map = em->map_lookup; | |
5692 | /* we don't discard raid56 yet */ | |
5693 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5694 | ret = -EOPNOTSUPP; | |
5695 | goto out; | |
5696 | } | |
5697 | ||
5698 | offset = logical - em->start; | |
5699 | length = min_t(u64, em->len - offset, length); | |
5700 | ||
5701 | stripe_len = map->stripe_len; | |
5702 | /* | |
5703 | * stripe_nr counts the total number of stripes we have to stride | |
5704 | * to get to this block | |
5705 | */ | |
5706 | stripe_nr = div64_u64(offset, stripe_len); | |
5707 | ||
5708 | /* stripe_offset is the offset of this block in its stripe */ | |
5709 | stripe_offset = offset - stripe_nr * stripe_len; | |
5710 | ||
5711 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5712 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5713 | stripe_cnt = stripe_nr_end - stripe_nr; |
5714 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5715 | (offset + length); | |
5716 | /* | |
5717 | * after this, stripe_nr is the number of stripes on this | |
5718 | * device we have to walk to find the data, and stripe_index is | |
5719 | * the number of our device in the stripe array | |
5720 | */ | |
5721 | num_stripes = 1; | |
5722 | stripe_index = 0; | |
5723 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5724 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5725 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5726 | sub_stripes = 1; | |
5727 | else | |
5728 | sub_stripes = map->sub_stripes; | |
5729 | ||
5730 | factor = map->num_stripes / sub_stripes; | |
5731 | num_stripes = min_t(u64, map->num_stripes, | |
5732 | sub_stripes * stripe_cnt); | |
5733 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5734 | stripe_index *= sub_stripes; | |
5735 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5736 | &remaining_stripes); | |
5737 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5738 | last_stripe *= sub_stripes; | |
5739 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5740 | BTRFS_BLOCK_GROUP_DUP)) { | |
5741 | num_stripes = map->num_stripes; | |
5742 | } else { | |
5743 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5744 | &stripe_index); | |
5745 | } | |
5746 | ||
5747 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5748 | if (!bbio) { | |
5749 | ret = -ENOMEM; | |
5750 | goto out; | |
5751 | } | |
5752 | ||
5753 | for (i = 0; i < num_stripes; i++) { | |
5754 | bbio->stripes[i].physical = | |
5755 | map->stripes[stripe_index].physical + | |
5756 | stripe_offset + stripe_nr * map->stripe_len; | |
5757 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5758 | ||
5759 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5760 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5761 | bbio->stripes[i].length = stripes_per_dev * | |
5762 | map->stripe_len; | |
5763 | ||
5764 | if (i / sub_stripes < remaining_stripes) | |
5765 | bbio->stripes[i].length += | |
5766 | map->stripe_len; | |
5767 | ||
5768 | /* | |
5769 | * Special for the first stripe and | |
5770 | * the last stripe: | |
5771 | * | |
5772 | * |-------|...|-------| | |
5773 | * |----------| | |
5774 | * off end_off | |
5775 | */ | |
5776 | if (i < sub_stripes) | |
5777 | bbio->stripes[i].length -= | |
5778 | stripe_offset; | |
5779 | ||
5780 | if (stripe_index >= last_stripe && | |
5781 | stripe_index <= (last_stripe + | |
5782 | sub_stripes - 1)) | |
5783 | bbio->stripes[i].length -= | |
5784 | stripe_end_offset; | |
5785 | ||
5786 | if (i == sub_stripes - 1) | |
5787 | stripe_offset = 0; | |
5788 | } else { | |
5789 | bbio->stripes[i].length = length; | |
5790 | } | |
5791 | ||
5792 | stripe_index++; | |
5793 | if (stripe_index == map->num_stripes) { | |
5794 | stripe_index = 0; | |
5795 | stripe_nr++; | |
5796 | } | |
5797 | } | |
5798 | ||
5799 | *bbio_ret = bbio; | |
5800 | bbio->map_type = map->type; | |
5801 | bbio->num_stripes = num_stripes; | |
5802 | out: | |
5803 | free_extent_map(em); | |
5804 | return ret; | |
5805 | } | |
5806 | ||
5ab56090 LB |
5807 | /* |
5808 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5809 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5810 | * left cursor can also be read from the target drive. | |
5811 | * | |
5812 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5813 | * array of stripes. | |
5814 | * For READ, it also needs to be supported using the same mirror number. | |
5815 | * | |
5816 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5817 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5818 | * case. | |
5819 | */ | |
5820 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5821 | u64 logical, u64 length, | |
5822 | u64 srcdev_devid, int *mirror_num, | |
5823 | u64 *physical) | |
5824 | { | |
5825 | struct btrfs_bio *bbio = NULL; | |
5826 | int num_stripes; | |
5827 | int index_srcdev = 0; | |
5828 | int found = 0; | |
5829 | u64 physical_of_found = 0; | |
5830 | int i; | |
5831 | int ret = 0; | |
5832 | ||
5833 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5834 | logical, &length, &bbio, 0, 0); | |
5835 | if (ret) { | |
5836 | ASSERT(bbio == NULL); | |
5837 | return ret; | |
5838 | } | |
5839 | ||
5840 | num_stripes = bbio->num_stripes; | |
5841 | if (*mirror_num > num_stripes) { | |
5842 | /* | |
5843 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5844 | * that means that the requested area is not left of the left | |
5845 | * cursor | |
5846 | */ | |
5847 | btrfs_put_bbio(bbio); | |
5848 | return -EIO; | |
5849 | } | |
5850 | ||
5851 | /* | |
5852 | * process the rest of the function using the mirror_num of the source | |
5853 | * drive. Therefore look it up first. At the end, patch the device | |
5854 | * pointer to the one of the target drive. | |
5855 | */ | |
5856 | for (i = 0; i < num_stripes; i++) { | |
5857 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5858 | continue; | |
5859 | ||
5860 | /* | |
5861 | * In case of DUP, in order to keep it simple, only add the | |
5862 | * mirror with the lowest physical address | |
5863 | */ | |
5864 | if (found && | |
5865 | physical_of_found <= bbio->stripes[i].physical) | |
5866 | continue; | |
5867 | ||
5868 | index_srcdev = i; | |
5869 | found = 1; | |
5870 | physical_of_found = bbio->stripes[i].physical; | |
5871 | } | |
5872 | ||
5873 | btrfs_put_bbio(bbio); | |
5874 | ||
5875 | ASSERT(found); | |
5876 | if (!found) | |
5877 | return -EIO; | |
5878 | ||
5879 | *mirror_num = index_srcdev + 1; | |
5880 | *physical = physical_of_found; | |
5881 | return ret; | |
5882 | } | |
5883 | ||
73c0f228 LB |
5884 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5885 | struct btrfs_bio **bbio_ret, | |
5886 | struct btrfs_dev_replace *dev_replace, | |
5887 | int *num_stripes_ret, int *max_errors_ret) | |
5888 | { | |
5889 | struct btrfs_bio *bbio = *bbio_ret; | |
5890 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5891 | int tgtdev_indexes = 0; | |
5892 | int num_stripes = *num_stripes_ret; | |
5893 | int max_errors = *max_errors_ret; | |
5894 | int i; | |
5895 | ||
5896 | if (op == BTRFS_MAP_WRITE) { | |
5897 | int index_where_to_add; | |
5898 | ||
5899 | /* | |
5900 | * duplicate the write operations while the dev replace | |
5901 | * procedure is running. Since the copying of the old disk to | |
5902 | * the new disk takes place at run time while the filesystem is | |
5903 | * mounted writable, the regular write operations to the old | |
5904 | * disk have to be duplicated to go to the new disk as well. | |
5905 | * | |
5906 | * Note that device->missing is handled by the caller, and that | |
5907 | * the write to the old disk is already set up in the stripes | |
5908 | * array. | |
5909 | */ | |
5910 | index_where_to_add = num_stripes; | |
5911 | for (i = 0; i < num_stripes; i++) { | |
5912 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5913 | /* write to new disk, too */ | |
5914 | struct btrfs_bio_stripe *new = | |
5915 | bbio->stripes + index_where_to_add; | |
5916 | struct btrfs_bio_stripe *old = | |
5917 | bbio->stripes + i; | |
5918 | ||
5919 | new->physical = old->physical; | |
5920 | new->length = old->length; | |
5921 | new->dev = dev_replace->tgtdev; | |
5922 | bbio->tgtdev_map[i] = index_where_to_add; | |
5923 | index_where_to_add++; | |
5924 | max_errors++; | |
5925 | tgtdev_indexes++; | |
5926 | } | |
5927 | } | |
5928 | num_stripes = index_where_to_add; | |
5929 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5930 | int index_srcdev = 0; | |
5931 | int found = 0; | |
5932 | u64 physical_of_found = 0; | |
5933 | ||
5934 | /* | |
5935 | * During the dev-replace procedure, the target drive can also | |
5936 | * be used to read data in case it is needed to repair a corrupt | |
5937 | * block elsewhere. This is possible if the requested area is | |
5938 | * left of the left cursor. In this area, the target drive is a | |
5939 | * full copy of the source drive. | |
5940 | */ | |
5941 | for (i = 0; i < num_stripes; i++) { | |
5942 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5943 | /* | |
5944 | * In case of DUP, in order to keep it simple, | |
5945 | * only add the mirror with the lowest physical | |
5946 | * address | |
5947 | */ | |
5948 | if (found && | |
5949 | physical_of_found <= | |
5950 | bbio->stripes[i].physical) | |
5951 | continue; | |
5952 | index_srcdev = i; | |
5953 | found = 1; | |
5954 | physical_of_found = bbio->stripes[i].physical; | |
5955 | } | |
5956 | } | |
5957 | if (found) { | |
5958 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5959 | bbio->stripes + num_stripes; | |
5960 | ||
5961 | tgtdev_stripe->physical = physical_of_found; | |
5962 | tgtdev_stripe->length = | |
5963 | bbio->stripes[index_srcdev].length; | |
5964 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5965 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5966 | ||
5967 | tgtdev_indexes++; | |
5968 | num_stripes++; | |
5969 | } | |
5970 | } | |
5971 | ||
5972 | *num_stripes_ret = num_stripes; | |
5973 | *max_errors_ret = max_errors; | |
5974 | bbio->num_tgtdevs = tgtdev_indexes; | |
5975 | *bbio_ret = bbio; | |
5976 | } | |
5977 | ||
2b19a1fe LB |
5978 | static bool need_full_stripe(enum btrfs_map_op op) |
5979 | { | |
5980 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5981 | } | |
5982 | ||
cf8cddd3 CH |
5983 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5984 | enum btrfs_map_op op, | |
f2d8d74d | 5985 | u64 logical, u64 *length, |
a1d3c478 | 5986 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5987 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5988 | { |
5989 | struct extent_map *em; | |
5990 | struct map_lookup *map; | |
0b86a832 | 5991 | u64 offset; |
593060d7 CM |
5992 | u64 stripe_offset; |
5993 | u64 stripe_nr; | |
53b381b3 | 5994 | u64 stripe_len; |
9d644a62 | 5995 | u32 stripe_index; |
cea9e445 | 5996 | int i; |
de11cc12 | 5997 | int ret = 0; |
f2d8d74d | 5998 | int num_stripes; |
a236aed1 | 5999 | int max_errors = 0; |
2c8cdd6e | 6000 | int tgtdev_indexes = 0; |
a1d3c478 | 6001 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6002 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6003 | int dev_replace_is_ongoing = 0; | |
6004 | int num_alloc_stripes; | |
ad6d620e SB |
6005 | int patch_the_first_stripe_for_dev_replace = 0; |
6006 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6007 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 6008 | |
0b3d4cd3 LB |
6009 | if (op == BTRFS_MAP_DISCARD) |
6010 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6011 | *length, bbio_ret); | |
6012 | ||
60ca842e | 6013 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
6014 | if (IS_ERR(em)) |
6015 | return PTR_ERR(em); | |
0b86a832 | 6016 | |
95617d69 | 6017 | map = em->map_lookup; |
0b86a832 | 6018 | offset = logical - em->start; |
593060d7 | 6019 | |
53b381b3 | 6020 | stripe_len = map->stripe_len; |
593060d7 CM |
6021 | stripe_nr = offset; |
6022 | /* | |
6023 | * stripe_nr counts the total number of stripes we have to stride | |
6024 | * to get to this block | |
6025 | */ | |
47c5713f | 6026 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 6027 | |
53b381b3 | 6028 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 6029 | if (offset < stripe_offset) { |
5d163e0e JM |
6030 | btrfs_crit(fs_info, |
6031 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
6032 | stripe_offset, offset, em->start, logical, |
6033 | stripe_len); | |
6034 | free_extent_map(em); | |
6035 | return -EINVAL; | |
6036 | } | |
593060d7 CM |
6037 | |
6038 | /* stripe_offset is the offset of this block in its stripe*/ | |
6039 | stripe_offset = offset - stripe_offset; | |
6040 | ||
53b381b3 | 6041 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 6042 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
6043 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
6044 | raid56_full_stripe_start = offset; | |
6045 | ||
6046 | /* allow a write of a full stripe, but make sure we don't | |
6047 | * allow straddling of stripes | |
6048 | */ | |
47c5713f DS |
6049 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
6050 | full_stripe_len); | |
53b381b3 DW |
6051 | raid56_full_stripe_start *= full_stripe_len; |
6052 | } | |
6053 | ||
0b3d4cd3 | 6054 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
6055 | u64 max_len; |
6056 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
6057 | For other RAID types and for RAID[56] reads, just allow a single | |
6058 | stripe (on a single disk). */ | |
ffe2d203 | 6059 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 6060 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
6061 | max_len = stripe_len * nr_data_stripes(map) - |
6062 | (offset - raid56_full_stripe_start); | |
6063 | } else { | |
6064 | /* we limit the length of each bio to what fits in a stripe */ | |
6065 | max_len = stripe_len - stripe_offset; | |
6066 | } | |
6067 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
6068 | } else { |
6069 | *length = em->len - offset; | |
6070 | } | |
f2d8d74d | 6071 | |
da12fe54 NB |
6072 | /* |
6073 | * This is for when we're called from btrfs_bio_fits_in_stripe and all | |
6074 | * it cares about is the length | |
6075 | */ | |
a1d3c478 | 6076 | if (!bbio_ret) |
cea9e445 CM |
6077 | goto out; |
6078 | ||
cb5583dd | 6079 | down_read(&dev_replace->rwsem); |
472262f3 | 6080 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6081 | /* |
6082 | * Hold the semaphore for read during the whole operation, write is | |
6083 | * requested at commit time but must wait. | |
6084 | */ | |
472262f3 | 6085 | if (!dev_replace_is_ongoing) |
cb5583dd | 6086 | up_read(&dev_replace->rwsem); |
472262f3 | 6087 | |
ad6d620e | 6088 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6089 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6090 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6091 | dev_replace->srcdev->devid, | |
6092 | &mirror_num, | |
6093 | &physical_to_patch_in_first_stripe); | |
6094 | if (ret) | |
ad6d620e | 6095 | goto out; |
5ab56090 LB |
6096 | else |
6097 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6098 | } else if (mirror_num > map->num_stripes) { |
6099 | mirror_num = 0; | |
6100 | } | |
6101 | ||
f2d8d74d | 6102 | num_stripes = 1; |
cea9e445 | 6103 | stripe_index = 0; |
fce3bb9a | 6104 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6105 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6106 | &stripe_index); | |
de483734 | 6107 | if (!need_full_stripe(op)) |
28e1cc7d | 6108 | mirror_num = 1; |
fce3bb9a | 6109 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 6110 | if (need_full_stripe(op)) |
f2d8d74d | 6111 | num_stripes = map->num_stripes; |
2fff734f | 6112 | else if (mirror_num) |
f188591e | 6113 | stripe_index = mirror_num - 1; |
dfe25020 | 6114 | else { |
30d9861f | 6115 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6116 | dev_replace_is_ongoing); |
a1d3c478 | 6117 | mirror_num = stripe_index + 1; |
dfe25020 | 6118 | } |
2fff734f | 6119 | |
611f0e00 | 6120 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6121 | if (need_full_stripe(op)) { |
f2d8d74d | 6122 | num_stripes = map->num_stripes; |
a1d3c478 | 6123 | } else if (mirror_num) { |
f188591e | 6124 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6125 | } else { |
6126 | mirror_num = 1; | |
6127 | } | |
2fff734f | 6128 | |
321aecc6 | 6129 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6130 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6131 | |
47c5713f | 6132 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6133 | stripe_index *= map->sub_stripes; |
6134 | ||
de483734 | 6135 | if (need_full_stripe(op)) |
f2d8d74d | 6136 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6137 | else if (mirror_num) |
6138 | stripe_index += mirror_num - 1; | |
dfe25020 | 6139 | else { |
3e74317a | 6140 | int old_stripe_index = stripe_index; |
30d9861f SB |
6141 | stripe_index = find_live_mirror(fs_info, map, |
6142 | stripe_index, | |
30d9861f | 6143 | dev_replace_is_ongoing); |
3e74317a | 6144 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6145 | } |
53b381b3 | 6146 | |
ffe2d203 | 6147 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6148 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6149 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6150 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 6151 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
6152 | |
6153 | /* RAID[56] write or recovery. Return all stripes */ | |
6154 | num_stripes = map->num_stripes; | |
6155 | max_errors = nr_parity_stripes(map); | |
6156 | ||
53b381b3 DW |
6157 | *length = map->stripe_len; |
6158 | stripe_index = 0; | |
6159 | stripe_offset = 0; | |
6160 | } else { | |
6161 | /* | |
6162 | * Mirror #0 or #1 means the original data block. | |
6163 | * Mirror #2 is RAID5 parity block. | |
6164 | * Mirror #3 is RAID6 Q block. | |
6165 | */ | |
47c5713f DS |
6166 | stripe_nr = div_u64_rem(stripe_nr, |
6167 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
6168 | if (mirror_num > 1) |
6169 | stripe_index = nr_data_stripes(map) + | |
6170 | mirror_num - 2; | |
6171 | ||
6172 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6173 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6174 | &stripe_index); | |
de483734 | 6175 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6176 | mirror_num = 1; |
53b381b3 | 6177 | } |
8790d502 CM |
6178 | } else { |
6179 | /* | |
47c5713f DS |
6180 | * after this, stripe_nr is the number of stripes on this |
6181 | * device we have to walk to find the data, and stripe_index is | |
6182 | * the number of our device in the stripe array | |
8790d502 | 6183 | */ |
47c5713f DS |
6184 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6185 | &stripe_index); | |
a1d3c478 | 6186 | mirror_num = stripe_index + 1; |
8790d502 | 6187 | } |
e042d1ec | 6188 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6189 | btrfs_crit(fs_info, |
6190 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6191 | stripe_index, map->num_stripes); |
6192 | ret = -EINVAL; | |
6193 | goto out; | |
6194 | } | |
cea9e445 | 6195 | |
472262f3 | 6196 | num_alloc_stripes = num_stripes; |
6fad823f | 6197 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6198 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6199 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6200 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6201 | num_alloc_stripes++; |
2c8cdd6e | 6202 | tgtdev_indexes = num_stripes; |
ad6d620e | 6203 | } |
2c8cdd6e | 6204 | |
6e9606d2 | 6205 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6206 | if (!bbio) { |
6207 | ret = -ENOMEM; | |
6208 | goto out; | |
6209 | } | |
6fad823f | 6210 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6211 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6212 | |
8e5cfb55 | 6213 | /* build raid_map */ |
2b19a1fe LB |
6214 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6215 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6216 | u64 tmp; |
9d644a62 | 6217 | unsigned rot; |
8e5cfb55 ZL |
6218 | |
6219 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6220 | sizeof(struct btrfs_bio_stripe) * | |
6221 | num_alloc_stripes + | |
6222 | sizeof(int) * tgtdev_indexes); | |
6223 | ||
6224 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6225 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6226 | |
6227 | /* Fill in the logical address of each stripe */ | |
6228 | tmp = stripe_nr * nr_data_stripes(map); | |
6229 | for (i = 0; i < nr_data_stripes(map); i++) | |
6230 | bbio->raid_map[(i+rot) % num_stripes] = | |
6231 | em->start + (tmp + i) * map->stripe_len; | |
6232 | ||
6233 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6234 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6235 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6236 | RAID6_Q_STRIPE; | |
6237 | } | |
6238 | ||
b89203f7 | 6239 | |
0b3d4cd3 LB |
6240 | for (i = 0; i < num_stripes; i++) { |
6241 | bbio->stripes[i].physical = | |
6242 | map->stripes[stripe_index].physical + | |
6243 | stripe_offset + | |
6244 | stripe_nr * map->stripe_len; | |
6245 | bbio->stripes[i].dev = | |
6246 | map->stripes[stripe_index].dev; | |
6247 | stripe_index++; | |
593060d7 | 6248 | } |
de11cc12 | 6249 | |
2b19a1fe | 6250 | if (need_full_stripe(op)) |
d20983b4 | 6251 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6252 | |
8e5cfb55 ZL |
6253 | if (bbio->raid_map) |
6254 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6255 | |
73c0f228 | 6256 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6257 | need_full_stripe(op)) { |
73c0f228 LB |
6258 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6259 | &max_errors); | |
472262f3 SB |
6260 | } |
6261 | ||
de11cc12 | 6262 | *bbio_ret = bbio; |
10f11900 | 6263 | bbio->map_type = map->type; |
de11cc12 LZ |
6264 | bbio->num_stripes = num_stripes; |
6265 | bbio->max_errors = max_errors; | |
6266 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6267 | |
6268 | /* | |
6269 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6270 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6271 | * available as a mirror | |
6272 | */ | |
6273 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6274 | WARN_ON(num_stripes > 1); | |
6275 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6276 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6277 | bbio->mirror_num = map->num_stripes + 1; | |
6278 | } | |
cea9e445 | 6279 | out: |
73beece9 | 6280 | if (dev_replace_is_ongoing) { |
53176dde DS |
6281 | lockdep_assert_held(&dev_replace->rwsem); |
6282 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6283 | up_read(&dev_replace->rwsem); |
73beece9 | 6284 | } |
0b86a832 | 6285 | free_extent_map(em); |
de11cc12 | 6286 | return ret; |
0b86a832 CM |
6287 | } |
6288 | ||
cf8cddd3 | 6289 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6290 | u64 logical, u64 *length, |
a1d3c478 | 6291 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6292 | { |
b3d3fa51 | 6293 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6294 | mirror_num, 0); |
f2d8d74d CM |
6295 | } |
6296 | ||
af8e2d1d | 6297 | /* For Scrub/replace */ |
cf8cddd3 | 6298 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6299 | u64 logical, u64 *length, |
825ad4c9 | 6300 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6301 | { |
825ad4c9 | 6302 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6303 | } |
6304 | ||
63a9c7b9 NB |
6305 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6306 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6307 | { |
a512bbf8 YZ |
6308 | struct extent_map *em; |
6309 | struct map_lookup *map; | |
6310 | u64 *buf; | |
6311 | u64 bytenr; | |
6312 | u64 length; | |
6313 | u64 stripe_nr; | |
53b381b3 | 6314 | u64 rmap_len; |
a512bbf8 YZ |
6315 | int i, j, nr = 0; |
6316 | ||
60ca842e | 6317 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6318 | if (IS_ERR(em)) |
835d974f | 6319 | return -EIO; |
835d974f | 6320 | |
95617d69 | 6321 | map = em->map_lookup; |
a512bbf8 | 6322 | length = em->len; |
53b381b3 DW |
6323 | rmap_len = map->stripe_len; |
6324 | ||
a512bbf8 | 6325 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6326 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6327 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6328 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6329 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6330 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6331 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6332 | } | |
a512bbf8 | 6333 | |
31e818fe | 6334 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6335 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6336 | |
6337 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6338 | if (map->stripes[i].physical > physical || |
6339 | map->stripes[i].physical + length <= physical) | |
6340 | continue; | |
6341 | ||
6342 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6343 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6344 | |
6345 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6346 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6347 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6348 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6349 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6350 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6351 | * Alternatively, just use rmap_len below instead of | |
6352 | * map->stripe_len */ | |
6353 | ||
6354 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6355 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6356 | for (j = 0; j < nr; j++) { |
6357 | if (buf[j] == bytenr) | |
6358 | break; | |
6359 | } | |
934d375b CM |
6360 | if (j == nr) { |
6361 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6362 | buf[nr++] = bytenr; |
934d375b | 6363 | } |
a512bbf8 YZ |
6364 | } |
6365 | ||
a512bbf8 YZ |
6366 | *logical = buf; |
6367 | *naddrs = nr; | |
53b381b3 | 6368 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6369 | |
6370 | free_extent_map(em); | |
6371 | return 0; | |
f2d8d74d CM |
6372 | } |
6373 | ||
4246a0b6 | 6374 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6375 | { |
326e1dbb MS |
6376 | bio->bi_private = bbio->private; |
6377 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6378 | bio_endio(bio); |
326e1dbb | 6379 | |
6e9606d2 | 6380 | btrfs_put_bbio(bbio); |
8408c716 MX |
6381 | } |
6382 | ||
4246a0b6 | 6383 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6384 | { |
9be3395b | 6385 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6386 | int is_orig_bio = 0; |
8790d502 | 6387 | |
4e4cbee9 | 6388 | if (bio->bi_status) { |
a1d3c478 | 6389 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6390 | if (bio->bi_status == BLK_STS_IOERR || |
6391 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6392 | unsigned int stripe_index = |
9be3395b | 6393 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6394 | struct btrfs_device *dev; |
442a4f63 SB |
6395 | |
6396 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6397 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6398 | if (dev->bdev) { |
37226b21 | 6399 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6400 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6401 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6402 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6403 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6404 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6405 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6406 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6407 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6408 | } |
442a4f63 SB |
6409 | } |
6410 | } | |
8790d502 | 6411 | |
a1d3c478 | 6412 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6413 | is_orig_bio = 1; |
6414 | ||
c404e0dc MX |
6415 | btrfs_bio_counter_dec(bbio->fs_info); |
6416 | ||
a1d3c478 | 6417 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6418 | if (!is_orig_bio) { |
6419 | bio_put(bio); | |
a1d3c478 | 6420 | bio = bbio->orig_bio; |
7d2b4daa | 6421 | } |
c7b22bb1 | 6422 | |
9be3395b | 6423 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6424 | /* only send an error to the higher layers if it is |
53b381b3 | 6425 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6426 | */ |
a1d3c478 | 6427 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6428 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6429 | } else { |
1259ab75 CM |
6430 | /* |
6431 | * this bio is actually up to date, we didn't | |
6432 | * go over the max number of errors | |
6433 | */ | |
2dbe0c77 | 6434 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6435 | } |
c55f1396 | 6436 | |
4246a0b6 | 6437 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6438 | } else if (!is_orig_bio) { |
8790d502 CM |
6439 | bio_put(bio); |
6440 | } | |
8790d502 CM |
6441 | } |
6442 | ||
8b712842 CM |
6443 | /* |
6444 | * see run_scheduled_bios for a description of why bios are collected for | |
6445 | * async submit. | |
6446 | * | |
6447 | * This will add one bio to the pending list for a device and make sure | |
6448 | * the work struct is scheduled. | |
6449 | */ | |
2ff7e61e | 6450 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6451 | struct bio *bio) |
8b712842 | 6452 | { |
0b246afa | 6453 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6454 | int should_queue = 1; |
ffbd517d | 6455 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6456 | |
6457 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6458 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6459 | btrfsic_submit_bio(bio); |
143bede5 | 6460 | return; |
8b712842 CM |
6461 | } |
6462 | ||
492bb6de | 6463 | WARN_ON(bio->bi_next); |
8b712842 | 6464 | bio->bi_next = NULL; |
8b712842 CM |
6465 | |
6466 | spin_lock(&device->io_lock); | |
67f055c7 | 6467 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6468 | pending_bios = &device->pending_sync_bios; |
6469 | else | |
6470 | pending_bios = &device->pending_bios; | |
8b712842 | 6471 | |
ffbd517d CM |
6472 | if (pending_bios->tail) |
6473 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6474 | |
ffbd517d CM |
6475 | pending_bios->tail = bio; |
6476 | if (!pending_bios->head) | |
6477 | pending_bios->head = bio; | |
8b712842 CM |
6478 | if (device->running_pending) |
6479 | should_queue = 0; | |
6480 | ||
6481 | spin_unlock(&device->io_lock); | |
6482 | ||
6483 | if (should_queue) | |
0b246afa | 6484 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6485 | } |
6486 | ||
2ff7e61e JM |
6487 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6488 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6489 | { |
6490 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6491 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6492 | |
6493 | bio->bi_private = bbio; | |
9be3395b | 6494 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6495 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6496 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6497 | btrfs_debug_in_rcu(fs_info, |
6498 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6499 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6500 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6501 | bio->bi_iter.bi_size); | |
74d46992 | 6502 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6503 | |
2ff7e61e | 6504 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6505 | |
de1ee92a | 6506 | if (async) |
2ff7e61e | 6507 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6508 | else |
4e49ea4a | 6509 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6510 | } |
6511 | ||
de1ee92a JB |
6512 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6513 | { | |
6514 | atomic_inc(&bbio->error); | |
6515 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6516 | /* Should be the original bio. */ |
8408c716 MX |
6517 | WARN_ON(bio != bbio->orig_bio); |
6518 | ||
9be3395b | 6519 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6520 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6521 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6522 | bio->bi_status = BLK_STS_IOERR; | |
6523 | else | |
6524 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6525 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6526 | } |
6527 | } | |
6528 | ||
58efbc9f OS |
6529 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6530 | int mirror_num, int async_submit) | |
0b86a832 | 6531 | { |
0b86a832 | 6532 | struct btrfs_device *dev; |
8790d502 | 6533 | struct bio *first_bio = bio; |
4f024f37 | 6534 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6535 | u64 length = 0; |
6536 | u64 map_length; | |
0b86a832 | 6537 | int ret; |
08da757d ZL |
6538 | int dev_nr; |
6539 | int total_devs; | |
a1d3c478 | 6540 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6541 | |
4f024f37 | 6542 | length = bio->bi_iter.bi_size; |
0b86a832 | 6543 | map_length = length; |
cea9e445 | 6544 | |
0b246afa | 6545 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6546 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6547 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6548 | if (ret) { |
0b246afa | 6549 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6550 | return errno_to_blk_status(ret); |
c404e0dc | 6551 | } |
cea9e445 | 6552 | |
a1d3c478 | 6553 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6554 | bbio->orig_bio = first_bio; |
6555 | bbio->private = first_bio->bi_private; | |
6556 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6557 | bbio->fs_info = fs_info; |
53b381b3 DW |
6558 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6559 | ||
ad1ba2a0 | 6560 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6561 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6562 | /* In this case, map_length has been set to the length of |
6563 | a single stripe; not the whole write */ | |
37226b21 | 6564 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6565 | ret = raid56_parity_write(fs_info, bio, bbio, |
6566 | map_length); | |
53b381b3 | 6567 | } else { |
2ff7e61e JM |
6568 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6569 | map_length, mirror_num, 1); | |
53b381b3 | 6570 | } |
4245215d | 6571 | |
0b246afa | 6572 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6573 | return errno_to_blk_status(ret); |
53b381b3 DW |
6574 | } |
6575 | ||
cea9e445 | 6576 | if (map_length < length) { |
0b246afa | 6577 | btrfs_crit(fs_info, |
5d163e0e JM |
6578 | "mapping failed logical %llu bio len %llu len %llu", |
6579 | logical, length, map_length); | |
cea9e445 CM |
6580 | BUG(); |
6581 | } | |
a1d3c478 | 6582 | |
08da757d | 6583 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6584 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6585 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6586 | &dev->dev_state) || | |
ebbede42 AJ |
6587 | (bio_op(first_bio) == REQ_OP_WRITE && |
6588 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6589 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6590 | continue; |
6591 | } | |
6592 | ||
3aa8e074 | 6593 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6594 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6595 | else |
a1d3c478 | 6596 | bio = first_bio; |
de1ee92a | 6597 | |
2ff7e61e JM |
6598 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6599 | dev_nr, async_submit); | |
8790d502 | 6600 | } |
0b246afa | 6601 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6602 | return BLK_STS_OK; |
0b86a832 CM |
6603 | } |
6604 | ||
09ba3bc9 AJ |
6605 | /* |
6606 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6607 | * return NULL. | |
6608 | * | |
6609 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6610 | * only devid is used. | |
6611 | * | |
6612 | * If @seed is true, traverse through the seed devices. | |
6613 | */ | |
e4319cd9 | 6614 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6615 | u64 devid, u8 *uuid, u8 *fsid, |
6616 | bool seed) | |
0b86a832 | 6617 | { |
2b82032c | 6618 | struct btrfs_device *device; |
2b82032c | 6619 | |
e4319cd9 | 6620 | while (fs_devices) { |
2b82032c | 6621 | if (!fsid || |
e4319cd9 | 6622 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6623 | list_for_each_entry(device, &fs_devices->devices, |
6624 | dev_list) { | |
6625 | if (device->devid == devid && | |
6626 | (!uuid || memcmp(device->uuid, uuid, | |
6627 | BTRFS_UUID_SIZE) == 0)) | |
6628 | return device; | |
6629 | } | |
2b82032c | 6630 | } |
09ba3bc9 AJ |
6631 | if (seed) |
6632 | fs_devices = fs_devices->seed; | |
6633 | else | |
6634 | return NULL; | |
2b82032c YZ |
6635 | } |
6636 | return NULL; | |
0b86a832 CM |
6637 | } |
6638 | ||
2ff7e61e | 6639 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6640 | u64 devid, u8 *dev_uuid) |
6641 | { | |
6642 | struct btrfs_device *device; | |
dfe25020 | 6643 | |
12bd2fc0 ID |
6644 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6645 | if (IS_ERR(device)) | |
adfb69af | 6646 | return device; |
12bd2fc0 ID |
6647 | |
6648 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6649 | device->fs_devices = fs_devices; |
dfe25020 | 6650 | fs_devices->num_devices++; |
12bd2fc0 | 6651 | |
e6e674bd | 6652 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6653 | fs_devices->missing_devices++; |
12bd2fc0 | 6654 | |
dfe25020 CM |
6655 | return device; |
6656 | } | |
6657 | ||
12bd2fc0 ID |
6658 | /** |
6659 | * btrfs_alloc_device - allocate struct btrfs_device | |
6660 | * @fs_info: used only for generating a new devid, can be NULL if | |
6661 | * devid is provided (i.e. @devid != NULL). | |
6662 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6663 | * is generated. | |
6664 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6665 | * is generated. | |
6666 | * | |
6667 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6668 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6669 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6670 | */ |
6671 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6672 | const u64 *devid, | |
6673 | const u8 *uuid) | |
6674 | { | |
6675 | struct btrfs_device *dev; | |
6676 | u64 tmp; | |
6677 | ||
fae7f21c | 6678 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6679 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6680 | |
6681 | dev = __alloc_device(); | |
6682 | if (IS_ERR(dev)) | |
6683 | return dev; | |
6684 | ||
6685 | if (devid) | |
6686 | tmp = *devid; | |
6687 | else { | |
6688 | int ret; | |
6689 | ||
6690 | ret = find_next_devid(fs_info, &tmp); | |
6691 | if (ret) { | |
a425f9d4 | 6692 | btrfs_free_device(dev); |
12bd2fc0 ID |
6693 | return ERR_PTR(ret); |
6694 | } | |
6695 | } | |
6696 | dev->devid = tmp; | |
6697 | ||
6698 | if (uuid) | |
6699 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6700 | else | |
6701 | generate_random_uuid(dev->uuid); | |
6702 | ||
9e0af237 LB |
6703 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6704 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6705 | |
6706 | return dev; | |
6707 | } | |
6708 | ||
5a2b8e60 | 6709 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6710 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6711 | { |
2b902dfc AJ |
6712 | if (error) |
6713 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6714 | devid, uuid); | |
6715 | else | |
6716 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6717 | devid, uuid); | |
5a2b8e60 AJ |
6718 | } |
6719 | ||
2ff7e61e | 6720 | static int read_one_chunk(struct btrfs_fs_info *fs_info, struct btrfs_key *key, |
e06cd3dd LB |
6721 | struct extent_buffer *leaf, |
6722 | struct btrfs_chunk *chunk) | |
6723 | { | |
0b246afa | 6724 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6725 | struct map_lookup *map; |
6726 | struct extent_map *em; | |
6727 | u64 logical; | |
6728 | u64 length; | |
e06cd3dd LB |
6729 | u64 devid; |
6730 | u8 uuid[BTRFS_UUID_SIZE]; | |
6731 | int num_stripes; | |
6732 | int ret; | |
6733 | int i; | |
6734 | ||
6735 | logical = key->offset; | |
6736 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6737 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6738 | ||
075cb3c7 QW |
6739 | /* |
6740 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6741 | * as chunk item in tree block is already verified by tree-checker. | |
6742 | */ | |
6743 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
6744 | ret = btrfs_check_chunk_valid(fs_info, leaf, chunk, logical); | |
6745 | if (ret) | |
6746 | return ret; | |
6747 | } | |
a061fc8d | 6748 | |
890871be | 6749 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6750 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6751 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6752 | |
6753 | /* already mapped? */ | |
6754 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6755 | free_extent_map(em); | |
0b86a832 CM |
6756 | return 0; |
6757 | } else if (em) { | |
6758 | free_extent_map(em); | |
6759 | } | |
0b86a832 | 6760 | |
172ddd60 | 6761 | em = alloc_extent_map(); |
0b86a832 CM |
6762 | if (!em) |
6763 | return -ENOMEM; | |
593060d7 | 6764 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6765 | if (!map) { |
6766 | free_extent_map(em); | |
6767 | return -ENOMEM; | |
6768 | } | |
6769 | ||
298a8f9c | 6770 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6771 | em->map_lookup = map; |
0b86a832 CM |
6772 | em->start = logical; |
6773 | em->len = length; | |
70c8a91c | 6774 | em->orig_start = 0; |
0b86a832 | 6775 | em->block_start = 0; |
c8b97818 | 6776 | em->block_len = em->len; |
0b86a832 | 6777 | |
593060d7 CM |
6778 | map->num_stripes = num_stripes; |
6779 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6780 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6781 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6782 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6783 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6784 | map->verified_stripes = 0; |
593060d7 CM |
6785 | for (i = 0; i < num_stripes; i++) { |
6786 | map->stripes[i].physical = | |
6787 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6788 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6789 | read_extent_buffer(leaf, uuid, (unsigned long) |
6790 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6791 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6792 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6793 | devid, uuid, NULL, true); |
3cdde224 | 6794 | if (!map->stripes[i].dev && |
0b246afa | 6795 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6796 | free_extent_map(em); |
2b902dfc | 6797 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6798 | return -ENOENT; |
593060d7 | 6799 | } |
dfe25020 CM |
6800 | if (!map->stripes[i].dev) { |
6801 | map->stripes[i].dev = | |
2ff7e61e JM |
6802 | add_missing_dev(fs_info->fs_devices, devid, |
6803 | uuid); | |
adfb69af | 6804 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6805 | free_extent_map(em); |
adfb69af AJ |
6806 | btrfs_err(fs_info, |
6807 | "failed to init missing dev %llu: %ld", | |
6808 | devid, PTR_ERR(map->stripes[i].dev)); | |
6809 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6810 | } |
2b902dfc | 6811 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6812 | } |
e12c9621 AJ |
6813 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6814 | &(map->stripes[i].dev->dev_state)); | |
6815 | ||
0b86a832 CM |
6816 | } |
6817 | ||
890871be | 6818 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6819 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6820 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6821 | if (ret < 0) { |
6822 | btrfs_err(fs_info, | |
6823 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6824 | em->start, em->len, ret); | |
6825 | } | |
0b86a832 CM |
6826 | free_extent_map(em); |
6827 | ||
64f64f43 | 6828 | return ret; |
0b86a832 CM |
6829 | } |
6830 | ||
143bede5 | 6831 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6832 | struct btrfs_dev_item *dev_item, |
6833 | struct btrfs_device *device) | |
6834 | { | |
6835 | unsigned long ptr; | |
0b86a832 CM |
6836 | |
6837 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6838 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6839 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6840 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6841 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6842 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6843 | device->type = btrfs_device_type(leaf, dev_item); |
6844 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6845 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6846 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6847 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6848 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6849 | |
410ba3a2 | 6850 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6851 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6852 | } |
6853 | ||
2ff7e61e | 6854 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6855 | u8 *fsid) |
2b82032c YZ |
6856 | { |
6857 | struct btrfs_fs_devices *fs_devices; | |
6858 | int ret; | |
6859 | ||
a32bf9a3 | 6860 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6861 | ASSERT(fsid); |
2b82032c | 6862 | |
0b246afa | 6863 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6864 | while (fs_devices) { |
44880fdc | 6865 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6866 | return fs_devices; |
6867 | ||
2b82032c YZ |
6868 | fs_devices = fs_devices->seed; |
6869 | } | |
6870 | ||
7239ff4b | 6871 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6872 | if (!fs_devices) { |
0b246afa | 6873 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6874 | return ERR_PTR(-ENOENT); |
6875 | ||
7239ff4b | 6876 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6877 | if (IS_ERR(fs_devices)) |
6878 | return fs_devices; | |
6879 | ||
6880 | fs_devices->seeding = 1; | |
6881 | fs_devices->opened = 1; | |
6882 | return fs_devices; | |
2b82032c | 6883 | } |
e4404d6e YZ |
6884 | |
6885 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6886 | if (IS_ERR(fs_devices)) |
6887 | return fs_devices; | |
2b82032c | 6888 | |
897fb573 | 6889 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6890 | if (ret) { |
6891 | free_fs_devices(fs_devices); | |
5f375835 | 6892 | fs_devices = ERR_PTR(ret); |
2b82032c | 6893 | goto out; |
48d28232 | 6894 | } |
2b82032c YZ |
6895 | |
6896 | if (!fs_devices->seeding) { | |
0226e0eb | 6897 | close_fs_devices(fs_devices); |
e4404d6e | 6898 | free_fs_devices(fs_devices); |
5f375835 | 6899 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6900 | goto out; |
6901 | } | |
6902 | ||
0b246afa JM |
6903 | fs_devices->seed = fs_info->fs_devices->seed; |
6904 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6905 | out: |
5f375835 | 6906 | return fs_devices; |
2b82032c YZ |
6907 | } |
6908 | ||
2ff7e61e | 6909 | static int read_one_dev(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
6910 | struct extent_buffer *leaf, |
6911 | struct btrfs_dev_item *dev_item) | |
6912 | { | |
0b246afa | 6913 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6914 | struct btrfs_device *device; |
6915 | u64 devid; | |
6916 | int ret; | |
44880fdc | 6917 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6918 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6919 | ||
0b86a832 | 6920 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6921 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6922 | BTRFS_UUID_SIZE); |
1473b24e | 6923 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6924 | BTRFS_FSID_SIZE); |
2b82032c | 6925 | |
de37aa51 | 6926 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6927 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6928 | if (IS_ERR(fs_devices)) |
6929 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6930 | } |
6931 | ||
e4319cd9 | 6932 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6933 | fs_uuid, true); |
5f375835 | 6934 | if (!device) { |
c5502451 | 6935 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6936 | btrfs_report_missing_device(fs_info, devid, |
6937 | dev_uuid, true); | |
45dbdbc9 | 6938 | return -ENOENT; |
c5502451 | 6939 | } |
2b82032c | 6940 | |
2ff7e61e | 6941 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6942 | if (IS_ERR(device)) { |
6943 | btrfs_err(fs_info, | |
6944 | "failed to add missing dev %llu: %ld", | |
6945 | devid, PTR_ERR(device)); | |
6946 | return PTR_ERR(device); | |
6947 | } | |
2b902dfc | 6948 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6949 | } else { |
c5502451 | 6950 | if (!device->bdev) { |
2b902dfc AJ |
6951 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6952 | btrfs_report_missing_device(fs_info, | |
6953 | devid, dev_uuid, true); | |
45dbdbc9 | 6954 | return -ENOENT; |
2b902dfc AJ |
6955 | } |
6956 | btrfs_report_missing_device(fs_info, devid, | |
6957 | dev_uuid, false); | |
c5502451 | 6958 | } |
5f375835 | 6959 | |
e6e674bd AJ |
6960 | if (!device->bdev && |
6961 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6962 | /* |
6963 | * this happens when a device that was properly setup | |
6964 | * in the device info lists suddenly goes bad. | |
6965 | * device->bdev is NULL, and so we have to set | |
6966 | * device->missing to one here | |
6967 | */ | |
5f375835 | 6968 | device->fs_devices->missing_devices++; |
e6e674bd | 6969 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6970 | } |
5f375835 MX |
6971 | |
6972 | /* Move the device to its own fs_devices */ | |
6973 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6974 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6975 | &device->dev_state)); | |
5f375835 MX |
6976 | |
6977 | list_move(&device->dev_list, &fs_devices->devices); | |
6978 | device->fs_devices->num_devices--; | |
6979 | fs_devices->num_devices++; | |
6980 | ||
6981 | device->fs_devices->missing_devices--; | |
6982 | fs_devices->missing_devices++; | |
6983 | ||
6984 | device->fs_devices = fs_devices; | |
6985 | } | |
2b82032c YZ |
6986 | } |
6987 | ||
0b246afa | 6988 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6989 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6990 | if (device->generation != |
6991 | btrfs_device_generation(leaf, dev_item)) | |
6992 | return -EINVAL; | |
6324fbf3 | 6993 | } |
0b86a832 CM |
6994 | |
6995 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6996 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6997 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6998 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6999 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7000 | atomic64_add(device->total_bytes - device->bytes_used, |
7001 | &fs_info->free_chunk_space); | |
2bf64758 | 7002 | } |
0b86a832 | 7003 | ret = 0; |
0b86a832 CM |
7004 | return ret; |
7005 | } | |
7006 | ||
6bccf3ab | 7007 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7008 | { |
6bccf3ab | 7009 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 7010 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7011 | struct extent_buffer *sb; |
0b86a832 | 7012 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7013 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7014 | u8 *array_ptr; |
7015 | unsigned long sb_array_offset; | |
84eed90f | 7016 | int ret = 0; |
0b86a832 CM |
7017 | u32 num_stripes; |
7018 | u32 array_size; | |
7019 | u32 len = 0; | |
1ffb22cf | 7020 | u32 cur_offset; |
e06cd3dd | 7021 | u64 type; |
84eed90f | 7022 | struct btrfs_key key; |
0b86a832 | 7023 | |
0b246afa | 7024 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7025 | /* |
7026 | * This will create extent buffer of nodesize, superblock size is | |
7027 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7028 | * overallocate but we can keep it as-is, only the first page is used. | |
7029 | */ | |
2ff7e61e | 7030 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7031 | if (IS_ERR(sb)) |
7032 | return PTR_ERR(sb); | |
4db8c528 | 7033 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7034 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7035 | /* |
01327610 | 7036 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7037 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7038 | * pages up-to-date when the page is larger: extent does not cover the |
7039 | * whole page and consequently check_page_uptodate does not find all | |
7040 | * the page's extents up-to-date (the hole beyond sb), | |
7041 | * write_extent_buffer then triggers a WARN_ON. | |
7042 | * | |
7043 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7044 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7045 | * to silence the warning eg. on PowerPC 64. | |
7046 | */ | |
09cbfeaf | 7047 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7048 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7049 | |
a061fc8d | 7050 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7051 | array_size = btrfs_super_sys_array_size(super_copy); |
7052 | ||
1ffb22cf DS |
7053 | array_ptr = super_copy->sys_chunk_array; |
7054 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7055 | cur_offset = 0; | |
0b86a832 | 7056 | |
1ffb22cf DS |
7057 | while (cur_offset < array_size) { |
7058 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7059 | len = sizeof(*disk_key); |
7060 | if (cur_offset + len > array_size) | |
7061 | goto out_short_read; | |
7062 | ||
0b86a832 CM |
7063 | btrfs_disk_key_to_cpu(&key, disk_key); |
7064 | ||
1ffb22cf DS |
7065 | array_ptr += len; |
7066 | sb_array_offset += len; | |
7067 | cur_offset += len; | |
0b86a832 | 7068 | |
0d81ba5d | 7069 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7070 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7071 | /* |
7072 | * At least one btrfs_chunk with one stripe must be | |
7073 | * present, exact stripe count check comes afterwards | |
7074 | */ | |
7075 | len = btrfs_chunk_item_size(1); | |
7076 | if (cur_offset + len > array_size) | |
7077 | goto out_short_read; | |
7078 | ||
7079 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7080 | if (!num_stripes) { |
ab8d0fc4 JM |
7081 | btrfs_err(fs_info, |
7082 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7083 | num_stripes, cur_offset); |
7084 | ret = -EIO; | |
7085 | break; | |
7086 | } | |
7087 | ||
e06cd3dd LB |
7088 | type = btrfs_chunk_type(sb, chunk); |
7089 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7090 | btrfs_err(fs_info, |
e06cd3dd LB |
7091 | "invalid chunk type %llu in sys_array at offset %u", |
7092 | type, cur_offset); | |
7093 | ret = -EIO; | |
7094 | break; | |
7095 | } | |
7096 | ||
e3540eab DS |
7097 | len = btrfs_chunk_item_size(num_stripes); |
7098 | if (cur_offset + len > array_size) | |
7099 | goto out_short_read; | |
7100 | ||
2ff7e61e | 7101 | ret = read_one_chunk(fs_info, &key, sb, chunk); |
84eed90f CM |
7102 | if (ret) |
7103 | break; | |
0b86a832 | 7104 | } else { |
ab8d0fc4 JM |
7105 | btrfs_err(fs_info, |
7106 | "unexpected item type %u in sys_array at offset %u", | |
7107 | (u32)key.type, cur_offset); | |
84eed90f CM |
7108 | ret = -EIO; |
7109 | break; | |
0b86a832 | 7110 | } |
1ffb22cf DS |
7111 | array_ptr += len; |
7112 | sb_array_offset += len; | |
7113 | cur_offset += len; | |
0b86a832 | 7114 | } |
d865177a | 7115 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7116 | free_extent_buffer_stale(sb); |
84eed90f | 7117 | return ret; |
e3540eab DS |
7118 | |
7119 | out_short_read: | |
ab8d0fc4 | 7120 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7121 | len, cur_offset); |
d865177a | 7122 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7123 | free_extent_buffer_stale(sb); |
e3540eab | 7124 | return -EIO; |
0b86a832 CM |
7125 | } |
7126 | ||
21634a19 QW |
7127 | /* |
7128 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7129 | * | |
6528b99d AJ |
7130 | * If the @failing_dev is specified, it's accounted as missing. |
7131 | * | |
21634a19 QW |
7132 | * Return true if all chunks meet the minimal RW mount requirements. |
7133 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7134 | */ | |
6528b99d AJ |
7135 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7136 | struct btrfs_device *failing_dev) | |
21634a19 QW |
7137 | { |
7138 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
7139 | struct extent_map *em; | |
7140 | u64 next_start = 0; | |
7141 | bool ret = true; | |
7142 | ||
7143 | read_lock(&map_tree->map_tree.lock); | |
7144 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
7145 | read_unlock(&map_tree->map_tree.lock); | |
7146 | /* No chunk at all? Return false anyway */ | |
7147 | if (!em) { | |
7148 | ret = false; | |
7149 | goto out; | |
7150 | } | |
7151 | while (em) { | |
7152 | struct map_lookup *map; | |
7153 | int missing = 0; | |
7154 | int max_tolerated; | |
7155 | int i; | |
7156 | ||
7157 | map = em->map_lookup; | |
7158 | max_tolerated = | |
7159 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7160 | map->type); | |
7161 | for (i = 0; i < map->num_stripes; i++) { | |
7162 | struct btrfs_device *dev = map->stripes[i].dev; | |
7163 | ||
e6e674bd AJ |
7164 | if (!dev || !dev->bdev || |
7165 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7166 | dev->last_flush_error) |
7167 | missing++; | |
6528b99d AJ |
7168 | else if (failing_dev && failing_dev == dev) |
7169 | missing++; | |
21634a19 QW |
7170 | } |
7171 | if (missing > max_tolerated) { | |
6528b99d AJ |
7172 | if (!failing_dev) |
7173 | btrfs_warn(fs_info, | |
52042d8e | 7174 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7175 | em->start, missing, max_tolerated); |
7176 | free_extent_map(em); | |
7177 | ret = false; | |
7178 | goto out; | |
7179 | } | |
7180 | next_start = extent_map_end(em); | |
7181 | free_extent_map(em); | |
7182 | ||
7183 | read_lock(&map_tree->map_tree.lock); | |
7184 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
7185 | (u64)(-1) - next_start); | |
7186 | read_unlock(&map_tree->map_tree.lock); | |
7187 | } | |
7188 | out: | |
7189 | return ret; | |
7190 | } | |
7191 | ||
5b4aacef | 7192 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7193 | { |
5b4aacef | 7194 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7195 | struct btrfs_path *path; |
7196 | struct extent_buffer *leaf; | |
7197 | struct btrfs_key key; | |
7198 | struct btrfs_key found_key; | |
7199 | int ret; | |
7200 | int slot; | |
99e3ecfc | 7201 | u64 total_dev = 0; |
0b86a832 | 7202 | |
0b86a832 CM |
7203 | path = btrfs_alloc_path(); |
7204 | if (!path) | |
7205 | return -ENOMEM; | |
7206 | ||
3dd0f7a3 AJ |
7207 | /* |
7208 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7209 | * otherwise we don't need it. | |
7210 | */ | |
b367e47f | 7211 | mutex_lock(&uuid_mutex); |
34441361 | 7212 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7213 | |
395927a9 FDBM |
7214 | /* |
7215 | * Read all device items, and then all the chunk items. All | |
7216 | * device items are found before any chunk item (their object id | |
7217 | * is smaller than the lowest possible object id for a chunk | |
7218 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7219 | */ |
7220 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7221 | key.offset = 0; | |
7222 | key.type = 0; | |
0b86a832 | 7223 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7224 | if (ret < 0) |
7225 | goto error; | |
d397712b | 7226 | while (1) { |
0b86a832 CM |
7227 | leaf = path->nodes[0]; |
7228 | slot = path->slots[0]; | |
7229 | if (slot >= btrfs_header_nritems(leaf)) { | |
7230 | ret = btrfs_next_leaf(root, path); | |
7231 | if (ret == 0) | |
7232 | continue; | |
7233 | if (ret < 0) | |
7234 | goto error; | |
7235 | break; | |
7236 | } | |
7237 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7238 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7239 | struct btrfs_dev_item *dev_item; | |
7240 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7241 | struct btrfs_dev_item); |
2ff7e61e | 7242 | ret = read_one_dev(fs_info, leaf, dev_item); |
395927a9 FDBM |
7243 | if (ret) |
7244 | goto error; | |
99e3ecfc | 7245 | total_dev++; |
0b86a832 CM |
7246 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7247 | struct btrfs_chunk *chunk; | |
7248 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2ff7e61e | 7249 | ret = read_one_chunk(fs_info, &found_key, leaf, chunk); |
2b82032c YZ |
7250 | if (ret) |
7251 | goto error; | |
0b86a832 CM |
7252 | } |
7253 | path->slots[0]++; | |
7254 | } | |
99e3ecfc LB |
7255 | |
7256 | /* | |
7257 | * After loading chunk tree, we've got all device information, | |
7258 | * do another round of validation checks. | |
7259 | */ | |
0b246afa JM |
7260 | if (total_dev != fs_info->fs_devices->total_devices) { |
7261 | btrfs_err(fs_info, | |
99e3ecfc | 7262 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7263 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7264 | total_dev); |
7265 | ret = -EINVAL; | |
7266 | goto error; | |
7267 | } | |
0b246afa JM |
7268 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7269 | fs_info->fs_devices->total_rw_bytes) { | |
7270 | btrfs_err(fs_info, | |
99e3ecfc | 7271 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7272 | btrfs_super_total_bytes(fs_info->super_copy), |
7273 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7274 | ret = -EINVAL; |
7275 | goto error; | |
7276 | } | |
0b86a832 CM |
7277 | ret = 0; |
7278 | error: | |
34441361 | 7279 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7280 | mutex_unlock(&uuid_mutex); |
7281 | ||
2b82032c | 7282 | btrfs_free_path(path); |
0b86a832 CM |
7283 | return ret; |
7284 | } | |
442a4f63 | 7285 | |
cb517eab MX |
7286 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7287 | { | |
7288 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7289 | struct btrfs_device *device; | |
7290 | ||
29cc83f6 LB |
7291 | while (fs_devices) { |
7292 | mutex_lock(&fs_devices->device_list_mutex); | |
7293 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7294 | device->fs_info = fs_info; |
29cc83f6 LB |
7295 | mutex_unlock(&fs_devices->device_list_mutex); |
7296 | ||
7297 | fs_devices = fs_devices->seed; | |
7298 | } | |
cb517eab MX |
7299 | } |
7300 | ||
733f4fbb SB |
7301 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7302 | { | |
7303 | int i; | |
7304 | ||
7305 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7306 | btrfs_dev_stat_reset(dev, i); | |
7307 | } | |
7308 | ||
7309 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7310 | { | |
7311 | struct btrfs_key key; | |
7312 | struct btrfs_key found_key; | |
7313 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7314 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7315 | struct extent_buffer *eb; | |
7316 | int slot; | |
7317 | int ret = 0; | |
7318 | struct btrfs_device *device; | |
7319 | struct btrfs_path *path = NULL; | |
7320 | int i; | |
7321 | ||
7322 | path = btrfs_alloc_path(); | |
7323 | if (!path) { | |
7324 | ret = -ENOMEM; | |
7325 | goto out; | |
7326 | } | |
7327 | ||
7328 | mutex_lock(&fs_devices->device_list_mutex); | |
7329 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7330 | int item_size; | |
7331 | struct btrfs_dev_stats_item *ptr; | |
7332 | ||
242e2956 DS |
7333 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7334 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7335 | key.offset = device->devid; |
7336 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7337 | if (ret) { | |
733f4fbb SB |
7338 | __btrfs_reset_dev_stats(device); |
7339 | device->dev_stats_valid = 1; | |
7340 | btrfs_release_path(path); | |
7341 | continue; | |
7342 | } | |
7343 | slot = path->slots[0]; | |
7344 | eb = path->nodes[0]; | |
7345 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7346 | item_size = btrfs_item_size_nr(eb, slot); | |
7347 | ||
7348 | ptr = btrfs_item_ptr(eb, slot, | |
7349 | struct btrfs_dev_stats_item); | |
7350 | ||
7351 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7352 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7353 | btrfs_dev_stat_set(device, i, | |
7354 | btrfs_dev_stats_value(eb, ptr, i)); | |
7355 | else | |
7356 | btrfs_dev_stat_reset(device, i); | |
7357 | } | |
7358 | ||
7359 | device->dev_stats_valid = 1; | |
7360 | btrfs_dev_stat_print_on_load(device); | |
7361 | btrfs_release_path(path); | |
7362 | } | |
7363 | mutex_unlock(&fs_devices->device_list_mutex); | |
7364 | ||
7365 | out: | |
7366 | btrfs_free_path(path); | |
7367 | return ret < 0 ? ret : 0; | |
7368 | } | |
7369 | ||
7370 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7371 | struct btrfs_device *device) |
7372 | { | |
5495f195 | 7373 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7374 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7375 | struct btrfs_path *path; |
7376 | struct btrfs_key key; | |
7377 | struct extent_buffer *eb; | |
7378 | struct btrfs_dev_stats_item *ptr; | |
7379 | int ret; | |
7380 | int i; | |
7381 | ||
242e2956 DS |
7382 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7383 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7384 | key.offset = device->devid; |
7385 | ||
7386 | path = btrfs_alloc_path(); | |
fa252992 DS |
7387 | if (!path) |
7388 | return -ENOMEM; | |
733f4fbb SB |
7389 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7390 | if (ret < 0) { | |
0b246afa | 7391 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7392 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7393 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7394 | goto out; |
7395 | } | |
7396 | ||
7397 | if (ret == 0 && | |
7398 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7399 | /* need to delete old one and insert a new one */ | |
7400 | ret = btrfs_del_item(trans, dev_root, path); | |
7401 | if (ret != 0) { | |
0b246afa | 7402 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7403 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7404 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7405 | goto out; |
7406 | } | |
7407 | ret = 1; | |
7408 | } | |
7409 | ||
7410 | if (ret == 1) { | |
7411 | /* need to insert a new item */ | |
7412 | btrfs_release_path(path); | |
7413 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7414 | &key, sizeof(*ptr)); | |
7415 | if (ret < 0) { | |
0b246afa | 7416 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7417 | "insert dev_stats item for device %s failed %d", |
7418 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7419 | goto out; |
7420 | } | |
7421 | } | |
7422 | ||
7423 | eb = path->nodes[0]; | |
7424 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7425 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7426 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7427 | btrfs_dev_stat_read(device, i)); | |
7428 | btrfs_mark_buffer_dirty(eb); | |
7429 | ||
7430 | out: | |
7431 | btrfs_free_path(path); | |
7432 | return ret; | |
7433 | } | |
7434 | ||
7435 | /* | |
7436 | * called from commit_transaction. Writes all changed device stats to disk. | |
7437 | */ | |
7438 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7439 | struct btrfs_fs_info *fs_info) | |
7440 | { | |
733f4fbb SB |
7441 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7442 | struct btrfs_device *device; | |
addc3fa7 | 7443 | int stats_cnt; |
733f4fbb SB |
7444 | int ret = 0; |
7445 | ||
7446 | mutex_lock(&fs_devices->device_list_mutex); | |
7447 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7448 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7449 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7450 | continue; |
7451 | ||
9deae968 NB |
7452 | |
7453 | /* | |
7454 | * There is a LOAD-LOAD control dependency between the value of | |
7455 | * dev_stats_ccnt and updating the on-disk values which requires | |
7456 | * reading the in-memory counters. Such control dependencies | |
7457 | * require explicit read memory barriers. | |
7458 | * | |
7459 | * This memory barriers pairs with smp_mb__before_atomic in | |
7460 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7461 | * barrier implied by atomic_xchg in | |
7462 | * btrfs_dev_stats_read_and_reset | |
7463 | */ | |
7464 | smp_rmb(); | |
7465 | ||
5495f195 | 7466 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7467 | if (!ret) |
addc3fa7 | 7468 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7469 | } |
7470 | mutex_unlock(&fs_devices->device_list_mutex); | |
7471 | ||
7472 | return ret; | |
7473 | } | |
7474 | ||
442a4f63 SB |
7475 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7476 | { | |
7477 | btrfs_dev_stat_inc(dev, index); | |
7478 | btrfs_dev_stat_print_on_error(dev); | |
7479 | } | |
7480 | ||
48a3b636 | 7481 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7482 | { |
733f4fbb SB |
7483 | if (!dev->dev_stats_valid) |
7484 | return; | |
fb456252 | 7485 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7486 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7487 | rcu_str_deref(dev->name), |
442a4f63 SB |
7488 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7489 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7490 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7491 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7492 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7493 | } |
c11d2c23 | 7494 | |
733f4fbb SB |
7495 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7496 | { | |
a98cdb85 SB |
7497 | int i; |
7498 | ||
7499 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7500 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7501 | break; | |
7502 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7503 | return; /* all values == 0, suppress message */ | |
7504 | ||
fb456252 | 7505 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7506 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7507 | rcu_str_deref(dev->name), |
733f4fbb SB |
7508 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7509 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7510 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7511 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7512 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7513 | } | |
7514 | ||
2ff7e61e | 7515 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7516 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7517 | { |
7518 | struct btrfs_device *dev; | |
0b246afa | 7519 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7520 | int i; |
7521 | ||
7522 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7523 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7524 | true); | |
c11d2c23 SB |
7525 | mutex_unlock(&fs_devices->device_list_mutex); |
7526 | ||
7527 | if (!dev) { | |
0b246afa | 7528 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7529 | return -ENODEV; |
733f4fbb | 7530 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7531 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7532 | return -ENODEV; |
b27f7c0c | 7533 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7534 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7535 | if (stats->nr_items > i) | |
7536 | stats->values[i] = | |
7537 | btrfs_dev_stat_read_and_reset(dev, i); | |
7538 | else | |
7539 | btrfs_dev_stat_reset(dev, i); | |
7540 | } | |
7541 | } else { | |
7542 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7543 | if (stats->nr_items > i) | |
7544 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7545 | } | |
7546 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7547 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7548 | return 0; | |
7549 | } | |
a8a6dab7 | 7550 | |
da353f6b | 7551 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7552 | { |
7553 | struct buffer_head *bh; | |
7554 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7555 | int copy_num; |
a8a6dab7 | 7556 | |
12b1c263 AJ |
7557 | if (!bdev) |
7558 | return; | |
a8a6dab7 | 7559 | |
12b1c263 AJ |
7560 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7561 | copy_num++) { | |
a8a6dab7 | 7562 | |
12b1c263 AJ |
7563 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7564 | continue; | |
7565 | ||
7566 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7567 | ||
7568 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7569 | set_buffer_dirty(bh); | |
7570 | sync_dirty_buffer(bh); | |
7571 | brelse(bh); | |
7572 | } | |
7573 | ||
7574 | /* Notify udev that device has changed */ | |
7575 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7576 | ||
7577 | /* Update ctime/mtime for device path for libblkid */ | |
7578 | update_dev_time(device_path); | |
a8a6dab7 | 7579 | } |
935e5cc9 MX |
7580 | |
7581 | /* | |
7582 | * Update the size of all devices, which is used for writing out the | |
7583 | * super blocks. | |
7584 | */ | |
7585 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7586 | { | |
7587 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7588 | struct btrfs_device *curr, *next; | |
7589 | ||
7590 | if (list_empty(&fs_devices->resized_devices)) | |
7591 | return; | |
7592 | ||
7593 | mutex_lock(&fs_devices->device_list_mutex); | |
34441361 | 7594 | mutex_lock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7595 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7596 | resized_list) { | |
7597 | list_del_init(&curr->resized_list); | |
7598 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7599 | } | |
34441361 | 7600 | mutex_unlock(&fs_info->chunk_mutex); |
935e5cc9 MX |
7601 | mutex_unlock(&fs_devices->device_list_mutex); |
7602 | } | |
ce7213c7 MX |
7603 | |
7604 | /* Must be invoked during the transaction commit */ | |
e9b919b1 | 7605 | void btrfs_update_commit_device_bytes_used(struct btrfs_transaction *trans) |
ce7213c7 | 7606 | { |
e9b919b1 | 7607 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ce7213c7 MX |
7608 | struct extent_map *em; |
7609 | struct map_lookup *map; | |
7610 | struct btrfs_device *dev; | |
7611 | int i; | |
7612 | ||
e9b919b1 | 7613 | if (list_empty(&trans->pending_chunks)) |
ce7213c7 MX |
7614 | return; |
7615 | ||
7616 | /* In order to kick the device replace finish process */ | |
34441361 | 7617 | mutex_lock(&fs_info->chunk_mutex); |
e9b919b1 | 7618 | list_for_each_entry(em, &trans->pending_chunks, list) { |
95617d69 | 7619 | map = em->map_lookup; |
ce7213c7 MX |
7620 | |
7621 | for (i = 0; i < map->num_stripes; i++) { | |
7622 | dev = map->stripes[i].dev; | |
7623 | dev->commit_bytes_used = dev->bytes_used; | |
7624 | } | |
7625 | } | |
34441361 | 7626 | mutex_unlock(&fs_info->chunk_mutex); |
ce7213c7 | 7627 | } |
5a13f430 AJ |
7628 | |
7629 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7630 | { | |
7631 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7632 | while (fs_devices) { | |
7633 | fs_devices->fs_info = fs_info; | |
7634 | fs_devices = fs_devices->seed; | |
7635 | } | |
7636 | } | |
7637 | ||
7638 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7639 | { | |
7640 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7641 | while (fs_devices) { | |
7642 | fs_devices->fs_info = NULL; | |
7643 | fs_devices = fs_devices->seed; | |
7644 | } | |
7645 | } | |
46df06b8 DS |
7646 | |
7647 | /* | |
7648 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7649 | */ | |
7650 | int btrfs_bg_type_to_factor(u64 flags) | |
7651 | { | |
7652 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7653 | BTRFS_BLOCK_GROUP_RAID10)) | |
7654 | return 2; | |
7655 | return 1; | |
7656 | } | |
cf90d884 QW |
7657 | |
7658 | ||
7659 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) | |
7660 | { | |
7661 | int index = btrfs_bg_flags_to_raid_index(type); | |
7662 | int ncopies = btrfs_raid_array[index].ncopies; | |
7663 | int data_stripes; | |
7664 | ||
7665 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
7666 | case BTRFS_BLOCK_GROUP_RAID5: | |
7667 | data_stripes = num_stripes - 1; | |
7668 | break; | |
7669 | case BTRFS_BLOCK_GROUP_RAID6: | |
7670 | data_stripes = num_stripes - 2; | |
7671 | break; | |
7672 | default: | |
7673 | data_stripes = num_stripes / ncopies; | |
7674 | break; | |
7675 | } | |
7676 | return div_u64(chunk_len, data_stripes); | |
7677 | } | |
7678 | ||
7679 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7680 | u64 chunk_offset, u64 devid, | |
7681 | u64 physical_offset, u64 physical_len) | |
7682 | { | |
7683 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7684 | struct extent_map *em; | |
7685 | struct map_lookup *map; | |
05a37c48 | 7686 | struct btrfs_device *dev; |
cf90d884 QW |
7687 | u64 stripe_len; |
7688 | bool found = false; | |
7689 | int ret = 0; | |
7690 | int i; | |
7691 | ||
7692 | read_lock(&em_tree->lock); | |
7693 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7694 | read_unlock(&em_tree->lock); | |
7695 | ||
7696 | if (!em) { | |
7697 | btrfs_err(fs_info, | |
7698 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7699 | physical_offset, devid); | |
7700 | ret = -EUCLEAN; | |
7701 | goto out; | |
7702 | } | |
7703 | ||
7704 | map = em->map_lookup; | |
7705 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7706 | if (physical_len != stripe_len) { | |
7707 | btrfs_err(fs_info, | |
7708 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7709 | physical_offset, devid, em->start, physical_len, | |
7710 | stripe_len); | |
7711 | ret = -EUCLEAN; | |
7712 | goto out; | |
7713 | } | |
7714 | ||
7715 | for (i = 0; i < map->num_stripes; i++) { | |
7716 | if (map->stripes[i].dev->devid == devid && | |
7717 | map->stripes[i].physical == physical_offset) { | |
7718 | found = true; | |
7719 | if (map->verified_stripes >= map->num_stripes) { | |
7720 | btrfs_err(fs_info, | |
7721 | "too many dev extents for chunk %llu found", | |
7722 | em->start); | |
7723 | ret = -EUCLEAN; | |
7724 | goto out; | |
7725 | } | |
7726 | map->verified_stripes++; | |
7727 | break; | |
7728 | } | |
7729 | } | |
7730 | if (!found) { | |
7731 | btrfs_err(fs_info, | |
7732 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7733 | physical_offset, devid); | |
7734 | ret = -EUCLEAN; | |
7735 | } | |
05a37c48 QW |
7736 | |
7737 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7738 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7739 | if (!dev) { |
7740 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7741 | ret = -EUCLEAN; | |
7742 | goto out; | |
7743 | } | |
1b3922a8 QW |
7744 | |
7745 | /* It's possible this device is a dummy for seed device */ | |
7746 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7747 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7748 | NULL, false); | |
1b3922a8 QW |
7749 | if (!dev) { |
7750 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7751 | devid); | |
7752 | ret = -EUCLEAN; | |
7753 | goto out; | |
7754 | } | |
7755 | } | |
7756 | ||
05a37c48 QW |
7757 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7758 | btrfs_err(fs_info, | |
7759 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7760 | devid, physical_offset, physical_len, | |
7761 | dev->disk_total_bytes); | |
7762 | ret = -EUCLEAN; | |
7763 | goto out; | |
7764 | } | |
cf90d884 QW |
7765 | out: |
7766 | free_extent_map(em); | |
7767 | return ret; | |
7768 | } | |
7769 | ||
7770 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7771 | { | |
7772 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7773 | struct extent_map *em; | |
7774 | struct rb_node *node; | |
7775 | int ret = 0; | |
7776 | ||
7777 | read_lock(&em_tree->lock); | |
07e1ce09 | 7778 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7779 | em = rb_entry(node, struct extent_map, rb_node); |
7780 | if (em->map_lookup->num_stripes != | |
7781 | em->map_lookup->verified_stripes) { | |
7782 | btrfs_err(fs_info, | |
7783 | "chunk %llu has missing dev extent, have %d expect %d", | |
7784 | em->start, em->map_lookup->verified_stripes, | |
7785 | em->map_lookup->num_stripes); | |
7786 | ret = -EUCLEAN; | |
7787 | goto out; | |
7788 | } | |
7789 | } | |
7790 | out: | |
7791 | read_unlock(&em_tree->lock); | |
7792 | return ret; | |
7793 | } | |
7794 | ||
7795 | /* | |
7796 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7797 | * later chunk allocation/free would cause unexpected behavior. | |
7798 | * | |
7799 | * NOTE: This will iterate through the whole device tree, which should be of | |
7800 | * the same size level as the chunk tree. This slightly increases mount time. | |
7801 | */ | |
7802 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7803 | { | |
7804 | struct btrfs_path *path; | |
7805 | struct btrfs_root *root = fs_info->dev_root; | |
7806 | struct btrfs_key key; | |
5eb19381 QW |
7807 | u64 prev_devid = 0; |
7808 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7809 | int ret = 0; |
7810 | ||
7811 | key.objectid = 1; | |
7812 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7813 | key.offset = 0; | |
7814 | ||
7815 | path = btrfs_alloc_path(); | |
7816 | if (!path) | |
7817 | return -ENOMEM; | |
7818 | ||
7819 | path->reada = READA_FORWARD; | |
7820 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7821 | if (ret < 0) | |
7822 | goto out; | |
7823 | ||
7824 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7825 | ret = btrfs_next_item(root, path); | |
7826 | if (ret < 0) | |
7827 | goto out; | |
7828 | /* No dev extents at all? Not good */ | |
7829 | if (ret > 0) { | |
7830 | ret = -EUCLEAN; | |
7831 | goto out; | |
7832 | } | |
7833 | } | |
7834 | while (1) { | |
7835 | struct extent_buffer *leaf = path->nodes[0]; | |
7836 | struct btrfs_dev_extent *dext; | |
7837 | int slot = path->slots[0]; | |
7838 | u64 chunk_offset; | |
7839 | u64 physical_offset; | |
7840 | u64 physical_len; | |
7841 | u64 devid; | |
7842 | ||
7843 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7844 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7845 | break; | |
7846 | devid = key.objectid; | |
7847 | physical_offset = key.offset; | |
7848 | ||
7849 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7850 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7851 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7852 | ||
5eb19381 QW |
7853 | /* Check if this dev extent overlaps with the previous one */ |
7854 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7855 | btrfs_err(fs_info, | |
7856 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7857 | devid, physical_offset, prev_dev_ext_end); | |
7858 | ret = -EUCLEAN; | |
7859 | goto out; | |
7860 | } | |
7861 | ||
cf90d884 QW |
7862 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7863 | physical_offset, physical_len); | |
7864 | if (ret < 0) | |
7865 | goto out; | |
5eb19381 QW |
7866 | prev_devid = devid; |
7867 | prev_dev_ext_end = physical_offset + physical_len; | |
7868 | ||
cf90d884 QW |
7869 | ret = btrfs_next_item(root, path); |
7870 | if (ret < 0) | |
7871 | goto out; | |
7872 | if (ret > 0) { | |
7873 | ret = 0; | |
7874 | break; | |
7875 | } | |
7876 | } | |
7877 | ||
7878 | /* Ensure all chunks have corresponding dev extents */ | |
7879 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7880 | out: | |
7881 | btrfs_free_path(path); | |
7882 | return ret; | |
7883 | } | |
eede2bf3 OS |
7884 | |
7885 | /* | |
7886 | * Check whether the given block group or device is pinned by any inode being | |
7887 | * used as a swapfile. | |
7888 | */ | |
7889 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7890 | { | |
7891 | struct btrfs_swapfile_pin *sp; | |
7892 | struct rb_node *node; | |
7893 | ||
7894 | spin_lock(&fs_info->swapfile_pins_lock); | |
7895 | node = fs_info->swapfile_pins.rb_node; | |
7896 | while (node) { | |
7897 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7898 | if (ptr < sp->ptr) | |
7899 | node = node->rb_left; | |
7900 | else if (ptr > sp->ptr) | |
7901 | node = node->rb_right; | |
7902 | else | |
7903 | break; | |
7904 | } | |
7905 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7906 | return node != NULL; | |
7907 | } |