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> |
784352fe | 17 | #include "misc.h" |
0b86a832 CM |
18 | #include "ctree.h" |
19 | #include "extent_map.h" | |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "print-tree.h" | |
23 | #include "volumes.h" | |
53b381b3 | 24 | #include "raid56.h" |
8b712842 | 25 | #include "async-thread.h" |
21adbd5c | 26 | #include "check-integrity.h" |
606686ee | 27 | #include "rcu-string.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
0b86a832 | 33 | |
af902047 ZL |
34 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
35 | [BTRFS_RAID_RAID10] = { | |
36 | .sub_stripes = 2, | |
37 | .dev_stripes = 1, | |
38 | .devs_max = 0, /* 0 == as many as possible */ | |
39 | .devs_min = 4, | |
8789f4fe | 40 | .tolerated_failures = 1, |
af902047 ZL |
41 | .devs_increment = 2, |
42 | .ncopies = 2, | |
b50836ed | 43 | .nparity = 0, |
ed23467b | 44 | .raid_name = "raid10", |
41a6e891 | 45 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 46 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
47 | }, |
48 | [BTRFS_RAID_RAID1] = { | |
49 | .sub_stripes = 1, | |
50 | .dev_stripes = 1, | |
51 | .devs_max = 2, | |
52 | .devs_min = 2, | |
8789f4fe | 53 | .tolerated_failures = 1, |
af902047 ZL |
54 | .devs_increment = 2, |
55 | .ncopies = 2, | |
b50836ed | 56 | .nparity = 0, |
ed23467b | 57 | .raid_name = "raid1", |
41a6e891 | 58 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 59 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 60 | }, |
47e6f742 DS |
61 | [BTRFS_RAID_RAID1C3] = { |
62 | .sub_stripes = 1, | |
63 | .dev_stripes = 1, | |
cf93e15e | 64 | .devs_max = 3, |
47e6f742 DS |
65 | .devs_min = 3, |
66 | .tolerated_failures = 2, | |
67 | .devs_increment = 3, | |
68 | .ncopies = 3, | |
db26a024 | 69 | .nparity = 0, |
47e6f742 DS |
70 | .raid_name = "raid1c3", |
71 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
72 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
73 | }, | |
8d6fac00 DS |
74 | [BTRFS_RAID_RAID1C4] = { |
75 | .sub_stripes = 1, | |
76 | .dev_stripes = 1, | |
cf93e15e | 77 | .devs_max = 4, |
8d6fac00 DS |
78 | .devs_min = 4, |
79 | .tolerated_failures = 3, | |
80 | .devs_increment = 4, | |
81 | .ncopies = 4, | |
db26a024 | 82 | .nparity = 0, |
8d6fac00 DS |
83 | .raid_name = "raid1c4", |
84 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
85 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
86 | }, | |
af902047 ZL |
87 | [BTRFS_RAID_DUP] = { |
88 | .sub_stripes = 1, | |
89 | .dev_stripes = 2, | |
90 | .devs_max = 1, | |
91 | .devs_min = 1, | |
8789f4fe | 92 | .tolerated_failures = 0, |
af902047 ZL |
93 | .devs_increment = 1, |
94 | .ncopies = 2, | |
b50836ed | 95 | .nparity = 0, |
ed23467b | 96 | .raid_name = "dup", |
41a6e891 | 97 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 98 | .mindev_error = 0, |
af902047 ZL |
99 | }, |
100 | [BTRFS_RAID_RAID0] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 2, | |
8789f4fe | 105 | .tolerated_failures = 0, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 1, | |
b50836ed | 108 | .nparity = 0, |
ed23467b | 109 | .raid_name = "raid0", |
41a6e891 | 110 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 111 | .mindev_error = 0, |
af902047 ZL |
112 | }, |
113 | [BTRFS_RAID_SINGLE] = { | |
114 | .sub_stripes = 1, | |
115 | .dev_stripes = 1, | |
116 | .devs_max = 1, | |
117 | .devs_min = 1, | |
8789f4fe | 118 | .tolerated_failures = 0, |
af902047 ZL |
119 | .devs_increment = 1, |
120 | .ncopies = 1, | |
b50836ed | 121 | .nparity = 0, |
ed23467b | 122 | .raid_name = "single", |
41a6e891 | 123 | .bg_flag = 0, |
f9fbcaa2 | 124 | .mindev_error = 0, |
af902047 ZL |
125 | }, |
126 | [BTRFS_RAID_RAID5] = { | |
127 | .sub_stripes = 1, | |
128 | .dev_stripes = 1, | |
129 | .devs_max = 0, | |
130 | .devs_min = 2, | |
8789f4fe | 131 | .tolerated_failures = 1, |
af902047 | 132 | .devs_increment = 1, |
da612e31 | 133 | .ncopies = 1, |
b50836ed | 134 | .nparity = 1, |
ed23467b | 135 | .raid_name = "raid5", |
41a6e891 | 136 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 137 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
138 | }, |
139 | [BTRFS_RAID_RAID6] = { | |
140 | .sub_stripes = 1, | |
141 | .dev_stripes = 1, | |
142 | .devs_max = 0, | |
143 | .devs_min = 3, | |
8789f4fe | 144 | .tolerated_failures = 2, |
af902047 | 145 | .devs_increment = 1, |
da612e31 | 146 | .ncopies = 1, |
b50836ed | 147 | .nparity = 2, |
ed23467b | 148 | .raid_name = "raid6", |
41a6e891 | 149 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 150 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
151 | }, |
152 | }; | |
153 | ||
158da513 | 154 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 155 | { |
158da513 DS |
156 | const int index = btrfs_bg_flags_to_raid_index(flags); |
157 | ||
158 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
159 | return NULL; |
160 | ||
158da513 | 161 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
162 | } |
163 | ||
f89e09cf AJ |
164 | /* |
165 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
166 | * bytes including terminating null byte. | |
167 | */ | |
168 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
169 | { | |
170 | int i; | |
171 | int ret; | |
172 | char *bp = buf; | |
173 | u64 flags = bg_flags; | |
174 | u32 size_bp = size_buf; | |
175 | ||
176 | if (!flags) { | |
177 | strcpy(bp, "NONE"); | |
178 | return; | |
179 | } | |
180 | ||
181 | #define DESCRIBE_FLAG(flag, desc) \ | |
182 | do { \ | |
183 | if (flags & (flag)) { \ | |
184 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
185 | if (ret < 0 || ret >= size_bp) \ | |
186 | goto out_overflow; \ | |
187 | size_bp -= ret; \ | |
188 | bp += ret; \ | |
189 | flags &= ~(flag); \ | |
190 | } \ | |
191 | } while (0) | |
192 | ||
193 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
194 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
195 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
196 | ||
197 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
198 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
199 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
200 | btrfs_raid_array[i].raid_name); | |
201 | #undef DESCRIBE_FLAG | |
202 | ||
203 | if (flags) { | |
204 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
205 | size_bp -= ret; | |
206 | } | |
207 | ||
208 | if (size_bp < size_buf) | |
209 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
210 | ||
211 | /* | |
212 | * The text is trimmed, it's up to the caller to provide sufficiently | |
213 | * large buffer | |
214 | */ | |
215 | out_overflow:; | |
216 | } | |
217 | ||
6f8e0fc7 | 218 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 219 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 220 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 221 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
222 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
223 | enum btrfs_map_op op, | |
224 | u64 logical, u64 *length, | |
225 | struct btrfs_bio **bbio_ret, | |
226 | int mirror_num, int need_raid_map); | |
2b82032c | 227 | |
9c6b1c4d DS |
228 | /* |
229 | * Device locking | |
230 | * ============== | |
231 | * | |
232 | * There are several mutexes that protect manipulation of devices and low-level | |
233 | * structures like chunks but not block groups, extents or files | |
234 | * | |
235 | * uuid_mutex (global lock) | |
236 | * ------------------------ | |
237 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
238 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
239 | * device) or requested by the device= mount option | |
240 | * | |
241 | * the mutex can be very coarse and can cover long-running operations | |
242 | * | |
243 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 244 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
245 | * |
246 | * global::fs_devs - add, remove, updates to the global list | |
247 | * | |
248 | * does not protect: manipulation of the fs_devices::devices list! | |
249 | * | |
250 | * btrfs_device::name - renames (write side), read is RCU | |
251 | * | |
252 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
253 | * ------------------------------------------------ | |
254 | * protects updates to fs_devices::devices, ie. adding and deleting | |
255 | * | |
256 | * simple list traversal with read-only actions can be done with RCU protection | |
257 | * | |
258 | * may be used to exclude some operations from running concurrently without any | |
259 | * modifications to the list (see write_all_supers) | |
260 | * | |
9c6b1c4d DS |
261 | * balance_mutex |
262 | * ------------- | |
263 | * protects balance structures (status, state) and context accessed from | |
264 | * several places (internally, ioctl) | |
265 | * | |
266 | * chunk_mutex | |
267 | * ----------- | |
268 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
269 | * device is added/removed. Additionally it also protects post_commit_list of |
270 | * individual devices, since they can be added to the transaction's | |
271 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
272 | * |
273 | * cleaner_mutex | |
274 | * ------------- | |
275 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
276 | * cleaning together with relocation or delayed iputs | |
277 | * | |
278 | * | |
279 | * Lock nesting | |
280 | * ============ | |
281 | * | |
282 | * uuid_mutex | |
283 | * volume_mutex | |
284 | * device_list_mutex | |
285 | * chunk_mutex | |
286 | * balance_mutex | |
89595e80 AJ |
287 | * |
288 | * | |
289 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
290 | * ====================================== | |
291 | * | |
292 | * Maintains the exclusivity of the following operations that apply to the | |
293 | * whole filesystem and cannot run in parallel. | |
294 | * | |
295 | * - Balance (*) | |
296 | * - Device add | |
297 | * - Device remove | |
298 | * - Device replace (*) | |
299 | * - Resize | |
300 | * | |
301 | * The device operations (as above) can be in one of the following states: | |
302 | * | |
303 | * - Running state | |
304 | * - Paused state | |
305 | * - Completed state | |
306 | * | |
307 | * Only device operations marked with (*) can go into the Paused state for the | |
308 | * following reasons: | |
309 | * | |
310 | * - ioctl (only Balance can be Paused through ioctl) | |
311 | * - filesystem remounted as read-only | |
312 | * - filesystem unmounted and mounted as read-only | |
313 | * - system power-cycle and filesystem mounted as read-only | |
314 | * - filesystem or device errors leading to forced read-only | |
315 | * | |
316 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
317 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
318 | * A device operation in Paused or Running state can be canceled or resumed | |
319 | * either by ioctl (Balance only) or when remounted as read-write. | |
320 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
321 | * completed. | |
9c6b1c4d DS |
322 | */ |
323 | ||
67a2c45e | 324 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 325 | static LIST_HEAD(fs_uuids); |
4143cb8b | 326 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
327 | { |
328 | return &fs_uuids; | |
329 | } | |
8a4b83cc | 330 | |
2dfeca9b DS |
331 | /* |
332 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
333 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
334 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
335 | * |
336 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
337 | * The returned struct is not linked onto any lists and can be destroyed with | |
338 | * kfree() right away. | |
339 | */ | |
7239ff4b NB |
340 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
341 | const u8 *metadata_fsid) | |
2208a378 ID |
342 | { |
343 | struct btrfs_fs_devices *fs_devs; | |
344 | ||
78f2c9e6 | 345 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
346 | if (!fs_devs) |
347 | return ERR_PTR(-ENOMEM); | |
348 | ||
349 | mutex_init(&fs_devs->device_list_mutex); | |
350 | ||
351 | INIT_LIST_HEAD(&fs_devs->devices); | |
352 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 353 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
354 | if (fsid) |
355 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 356 | |
7239ff4b NB |
357 | if (metadata_fsid) |
358 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
359 | else if (fsid) | |
360 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
361 | ||
2208a378 ID |
362 | return fs_devs; |
363 | } | |
364 | ||
a425f9d4 | 365 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 366 | { |
bbbf7243 | 367 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 368 | rcu_string_free(device->name); |
1c11b63e | 369 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
370 | bio_put(device->flush_bio); |
371 | kfree(device); | |
372 | } | |
373 | ||
e4404d6e YZ |
374 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
375 | { | |
376 | struct btrfs_device *device; | |
377 | WARN_ON(fs_devices->opened); | |
378 | while (!list_empty(&fs_devices->devices)) { | |
379 | device = list_entry(fs_devices->devices.next, | |
380 | struct btrfs_device, dev_list); | |
381 | list_del(&device->dev_list); | |
a425f9d4 | 382 | btrfs_free_device(device); |
e4404d6e YZ |
383 | } |
384 | kfree(fs_devices); | |
385 | } | |
386 | ||
ffc5a379 | 387 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
388 | { |
389 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 390 | |
2b82032c YZ |
391 | while (!list_empty(&fs_uuids)) { |
392 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
393 | struct btrfs_fs_devices, fs_list); |
394 | list_del(&fs_devices->fs_list); | |
e4404d6e | 395 | free_fs_devices(fs_devices); |
8a4b83cc | 396 | } |
8a4b83cc CM |
397 | } |
398 | ||
48dae9cf DS |
399 | /* |
400 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
401 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 402 | * btrfs_free_device. |
48dae9cf | 403 | */ |
12bd2fc0 ID |
404 | static struct btrfs_device *__alloc_device(void) |
405 | { | |
406 | struct btrfs_device *dev; | |
407 | ||
78f2c9e6 | 408 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
409 | if (!dev) |
410 | return ERR_PTR(-ENOMEM); | |
411 | ||
e0ae9994 DS |
412 | /* |
413 | * Preallocate a bio that's always going to be used for flushing device | |
414 | * barriers and matches the device lifespan | |
415 | */ | |
416 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
417 | if (!dev->flush_bio) { | |
418 | kfree(dev); | |
419 | return ERR_PTR(-ENOMEM); | |
420 | } | |
e0ae9994 | 421 | |
12bd2fc0 ID |
422 | INIT_LIST_HEAD(&dev->dev_list); |
423 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 424 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 | 425 | |
12bd2fc0 | 426 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 427 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 428 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 429 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 430 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
1c11b63e | 431 | extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL); |
12bd2fc0 ID |
432 | |
433 | return dev; | |
434 | } | |
435 | ||
7239ff4b NB |
436 | static noinline struct btrfs_fs_devices *find_fsid( |
437 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 438 | { |
8a4b83cc CM |
439 | struct btrfs_fs_devices *fs_devices; |
440 | ||
7239ff4b NB |
441 | ASSERT(fsid); |
442 | ||
7a62d0f0 NB |
443 | if (metadata_fsid) { |
444 | /* | |
445 | * Handle scanned device having completed its fsid change but | |
446 | * belonging to a fs_devices that was created by first scanning | |
447 | * a device which didn't have its fsid/metadata_uuid changed | |
448 | * at all and the CHANGING_FSID_V2 flag set. | |
449 | */ | |
450 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
451 | if (fs_devices->fsid_change && | |
452 | memcmp(metadata_fsid, fs_devices->fsid, | |
453 | BTRFS_FSID_SIZE) == 0 && | |
454 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
455 | BTRFS_FSID_SIZE) == 0) { | |
456 | return fs_devices; | |
457 | } | |
458 | } | |
cc5de4e7 NB |
459 | /* |
460 | * Handle scanned device having completed its fsid change but | |
461 | * belonging to a fs_devices that was created by a device that | |
462 | * has an outdated pair of fsid/metadata_uuid and | |
463 | * CHANGING_FSID_V2 flag set. | |
464 | */ | |
465 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
466 | if (fs_devices->fsid_change && | |
467 | memcmp(fs_devices->metadata_uuid, | |
468 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
469 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
470 | BTRFS_FSID_SIZE) == 0) { | |
471 | return fs_devices; | |
472 | } | |
473 | } | |
7a62d0f0 NB |
474 | } |
475 | ||
476 | /* Handle non-split brain cases */ | |
c4babc5e | 477 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
478 | if (metadata_fsid) { |
479 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
480 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
481 | BTRFS_FSID_SIZE) == 0) | |
482 | return fs_devices; | |
483 | } else { | |
484 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
485 | return fs_devices; | |
486 | } | |
8a4b83cc CM |
487 | } |
488 | return NULL; | |
489 | } | |
490 | ||
beaf8ab3 SB |
491 | static int |
492 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
493 | int flush, struct block_device **bdev, | |
494 | struct buffer_head **bh) | |
495 | { | |
496 | int ret; | |
497 | ||
498 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
499 | ||
500 | if (IS_ERR(*bdev)) { | |
501 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
502 | goto error; |
503 | } | |
504 | ||
505 | if (flush) | |
506 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 507 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
508 | if (ret) { |
509 | blkdev_put(*bdev, flags); | |
510 | goto error; | |
511 | } | |
512 | invalidate_bdev(*bdev); | |
513 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
514 | if (IS_ERR(*bh)) { |
515 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
516 | blkdev_put(*bdev, flags); |
517 | goto error; | |
518 | } | |
519 | ||
520 | return 0; | |
521 | ||
522 | error: | |
523 | *bdev = NULL; | |
524 | *bh = NULL; | |
525 | return ret; | |
526 | } | |
527 | ||
70bc7088 AJ |
528 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
529 | { | |
530 | int found; | |
531 | ||
532 | rcu_read_lock(); | |
533 | found = strcmp(rcu_str_deref(device->name), path); | |
534 | rcu_read_unlock(); | |
535 | ||
536 | return found == 0; | |
537 | } | |
538 | ||
d8367db3 AJ |
539 | /* |
540 | * Search and remove all stale (devices which are not mounted) devices. | |
541 | * When both inputs are NULL, it will search and release all stale devices. | |
542 | * path: Optional. When provided will it release all unmounted devices | |
543 | * matching this path only. | |
544 | * skip_dev: Optional. Will skip this device when searching for the stale | |
545 | * devices. | |
70bc7088 AJ |
546 | * Return: 0 for success or if @path is NULL. |
547 | * -EBUSY if @path is a mounted device. | |
548 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 549 | */ |
70bc7088 | 550 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 551 | struct btrfs_device *skip_device) |
4fde46f0 | 552 | { |
fa6d2ae5 AJ |
553 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
554 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
555 | int ret = 0; |
556 | ||
557 | if (path) | |
558 | ret = -ENOENT; | |
4fde46f0 | 559 | |
fa6d2ae5 | 560 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 561 | |
70bc7088 | 562 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
563 | list_for_each_entry_safe(device, tmp_device, |
564 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 565 | if (skip_device && skip_device == device) |
d8367db3 | 566 | continue; |
fa6d2ae5 | 567 | if (path && !device->name) |
4fde46f0 | 568 | continue; |
70bc7088 | 569 | if (path && !device_path_matched(path, device)) |
38cf665d | 570 | continue; |
70bc7088 AJ |
571 | if (fs_devices->opened) { |
572 | /* for an already deleted device return 0 */ | |
573 | if (path && ret != 0) | |
574 | ret = -EBUSY; | |
575 | break; | |
576 | } | |
4fde46f0 | 577 | |
4fde46f0 | 578 | /* delete the stale device */ |
7bcb8164 AJ |
579 | fs_devices->num_devices--; |
580 | list_del(&device->dev_list); | |
581 | btrfs_free_device(device); | |
582 | ||
70bc7088 | 583 | ret = 0; |
7bcb8164 | 584 | if (fs_devices->num_devices == 0) |
fd649f10 | 585 | break; |
7bcb8164 AJ |
586 | } |
587 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 588 | |
7bcb8164 AJ |
589 | if (fs_devices->num_devices == 0) { |
590 | btrfs_sysfs_remove_fsid(fs_devices); | |
591 | list_del(&fs_devices->fs_list); | |
592 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
593 | } |
594 | } | |
70bc7088 AJ |
595 | |
596 | return ret; | |
4fde46f0 AJ |
597 | } |
598 | ||
0fb08bcc AJ |
599 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
600 | struct btrfs_device *device, fmode_t flags, | |
601 | void *holder) | |
602 | { | |
603 | struct request_queue *q; | |
604 | struct block_device *bdev; | |
605 | struct buffer_head *bh; | |
606 | struct btrfs_super_block *disk_super; | |
607 | u64 devid; | |
608 | int ret; | |
609 | ||
610 | if (device->bdev) | |
611 | return -EINVAL; | |
612 | if (!device->name) | |
613 | return -EINVAL; | |
614 | ||
615 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
616 | &bdev, &bh); | |
617 | if (ret) | |
618 | return ret; | |
619 | ||
620 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
621 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
622 | if (devid != device->devid) | |
623 | goto error_brelse; | |
624 | ||
625 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
626 | goto error_brelse; | |
627 | ||
628 | device->generation = btrfs_super_generation(disk_super); | |
629 | ||
630 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
631 | if (btrfs_super_incompat_flags(disk_super) & |
632 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
633 | pr_err( | |
634 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
635 | goto error_brelse; | |
636 | } | |
637 | ||
ebbede42 | 638 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 639 | fs_devices->seeding = true; |
0fb08bcc | 640 | } else { |
ebbede42 AJ |
641 | if (bdev_read_only(bdev)) |
642 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
643 | else | |
644 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
645 | } |
646 | ||
647 | q = bdev_get_queue(bdev); | |
0fb08bcc | 648 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 649 | fs_devices->rotating = true; |
0fb08bcc AJ |
650 | |
651 | device->bdev = bdev; | |
e12c9621 | 652 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
653 | device->mode = flags; |
654 | ||
655 | fs_devices->open_devices++; | |
ebbede42 AJ |
656 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
657 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 658 | fs_devices->rw_devices++; |
b1b8e386 | 659 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
660 | } |
661 | brelse(bh); | |
662 | ||
663 | return 0; | |
664 | ||
665 | error_brelse: | |
666 | brelse(bh); | |
667 | blkdev_put(bdev, flags); | |
668 | ||
669 | return -EINVAL; | |
670 | } | |
671 | ||
7a62d0f0 NB |
672 | /* |
673 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
674 | * being created with a disk that has already completed its fsid change. | |
675 | */ | |
676 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
677 | struct btrfs_super_block *disk_super) | |
678 | { | |
679 | struct btrfs_fs_devices *fs_devices; | |
680 | ||
681 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
682 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
683 | BTRFS_FSID_SIZE) != 0 && | |
684 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
685 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
686 | return fs_devices; | |
687 | } | |
688 | } | |
689 | ||
690 | return NULL; | |
691 | } | |
692 | ||
cc5de4e7 NB |
693 | |
694 | static struct btrfs_fs_devices *find_fsid_changed( | |
695 | struct btrfs_super_block *disk_super) | |
696 | { | |
697 | struct btrfs_fs_devices *fs_devices; | |
698 | ||
699 | /* | |
700 | * Handles the case where scanned device is part of an fs that had | |
701 | * multiple successful changes of FSID but curently device didn't | |
702 | * observe it. Meaning our fsid will be different than theirs. | |
703 | */ | |
704 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
705 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
706 | BTRFS_FSID_SIZE) != 0 && | |
707 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
708 | BTRFS_FSID_SIZE) == 0 && | |
709 | memcmp(fs_devices->fsid, disk_super->fsid, | |
710 | BTRFS_FSID_SIZE) != 0) { | |
711 | return fs_devices; | |
712 | } | |
713 | } | |
714 | ||
715 | return NULL; | |
716 | } | |
60999ca4 DS |
717 | /* |
718 | * Add new device to list of registered devices | |
719 | * | |
720 | * Returns: | |
e124ece5 AJ |
721 | * device pointer which was just added or updated when successful |
722 | * error pointer when failed | |
60999ca4 | 723 | */ |
e124ece5 | 724 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
725 | struct btrfs_super_block *disk_super, |
726 | bool *new_device_added) | |
8a4b83cc CM |
727 | { |
728 | struct btrfs_device *device; | |
7a62d0f0 | 729 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 730 | struct rcu_string *name; |
8a4b83cc | 731 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 732 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
733 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
734 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
735 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
736 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 737 | |
cc5de4e7 NB |
738 | if (fsid_change_in_progress) { |
739 | if (!has_metadata_uuid) { | |
740 | /* | |
741 | * When we have an image which has CHANGING_FSID_V2 set | |
742 | * it might belong to either a filesystem which has | |
743 | * disks with completed fsid change or it might belong | |
744 | * to fs with no UUID changes in effect, handle both. | |
745 | */ | |
746 | fs_devices = find_fsid_inprogress(disk_super); | |
747 | if (!fs_devices) | |
748 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
749 | } else { | |
750 | fs_devices = find_fsid_changed(disk_super); | |
751 | } | |
7a62d0f0 NB |
752 | } else if (has_metadata_uuid) { |
753 | fs_devices = find_fsid(disk_super->fsid, | |
754 | disk_super->metadata_uuid); | |
755 | } else { | |
7239ff4b | 756 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
757 | } |
758 | ||
8a4b83cc | 759 | |
8a4b83cc | 760 | if (!fs_devices) { |
7239ff4b NB |
761 | if (has_metadata_uuid) |
762 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
763 | disk_super->metadata_uuid); | |
764 | else | |
765 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
766 | ||
2208a378 | 767 | if (IS_ERR(fs_devices)) |
e124ece5 | 768 | return ERR_CAST(fs_devices); |
2208a378 | 769 | |
92900e51 AV |
770 | fs_devices->fsid_change = fsid_change_in_progress; |
771 | ||
9c6d173e | 772 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 773 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 774 | |
8a4b83cc CM |
775 | device = NULL; |
776 | } else { | |
9c6d173e | 777 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
778 | device = btrfs_find_device(fs_devices, devid, |
779 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
780 | |
781 | /* | |
782 | * If this disk has been pulled into an fs devices created by | |
783 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
784 | * metadata_uuid/fsid values of the fs_devices. | |
785 | */ | |
786 | if (has_metadata_uuid && fs_devices->fsid_change && | |
787 | found_transid > fs_devices->latest_generation) { | |
788 | memcpy(fs_devices->fsid, disk_super->fsid, | |
789 | BTRFS_FSID_SIZE); | |
790 | memcpy(fs_devices->metadata_uuid, | |
791 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
792 | ||
793 | fs_devices->fsid_change = false; | |
794 | } | |
8a4b83cc | 795 | } |
443f24fe | 796 | |
8a4b83cc | 797 | if (!device) { |
9c6d173e AJ |
798 | if (fs_devices->opened) { |
799 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 800 | return ERR_PTR(-EBUSY); |
9c6d173e | 801 | } |
2b82032c | 802 | |
12bd2fc0 ID |
803 | device = btrfs_alloc_device(NULL, &devid, |
804 | disk_super->dev_item.uuid); | |
805 | if (IS_ERR(device)) { | |
9c6d173e | 806 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 807 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 808 | return device; |
8a4b83cc | 809 | } |
606686ee JB |
810 | |
811 | name = rcu_string_strdup(path, GFP_NOFS); | |
812 | if (!name) { | |
a425f9d4 | 813 | btrfs_free_device(device); |
9c6d173e | 814 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 815 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 816 | } |
606686ee | 817 | rcu_assign_pointer(device->name, name); |
90519d66 | 818 | |
1f78160c | 819 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 820 | fs_devices->num_devices++; |
e5e9a520 | 821 | |
2b82032c | 822 | device->fs_devices = fs_devices; |
4306a974 | 823 | *new_device_added = true; |
327f18cc AJ |
824 | |
825 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
826 | pr_info( |
827 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
828 | disk_super->label, devid, found_transid, path, | |
829 | current->comm, task_pid_nr(current)); | |
327f18cc | 830 | else |
aa6c0df7 AJ |
831 | pr_info( |
832 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
833 | disk_super->fsid, devid, found_transid, path, | |
834 | current->comm, task_pid_nr(current)); | |
327f18cc | 835 | |
606686ee | 836 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
837 | /* |
838 | * When FS is already mounted. | |
839 | * 1. If you are here and if the device->name is NULL that | |
840 | * means this device was missing at time of FS mount. | |
841 | * 2. If you are here and if the device->name is different | |
842 | * from 'path' that means either | |
843 | * a. The same device disappeared and reappeared with | |
844 | * different name. or | |
845 | * b. The missing-disk-which-was-replaced, has | |
846 | * reappeared now. | |
847 | * | |
848 | * We must allow 1 and 2a above. But 2b would be a spurious | |
849 | * and unintentional. | |
850 | * | |
851 | * Further in case of 1 and 2a above, the disk at 'path' | |
852 | * would have missed some transaction when it was away and | |
853 | * in case of 2a the stale bdev has to be updated as well. | |
854 | * 2b must not be allowed at all time. | |
855 | */ | |
856 | ||
857 | /* | |
0f23ae74 CM |
858 | * For now, we do allow update to btrfs_fs_device through the |
859 | * btrfs dev scan cli after FS has been mounted. We're still | |
860 | * tracking a problem where systems fail mount by subvolume id | |
861 | * when we reject replacement on a mounted FS. | |
b96de000 | 862 | */ |
0f23ae74 | 863 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
864 | /* |
865 | * That is if the FS is _not_ mounted and if you | |
866 | * are here, that means there is more than one | |
867 | * disk with same uuid and devid.We keep the one | |
868 | * with larger generation number or the last-in if | |
869 | * generation are equal. | |
870 | */ | |
9c6d173e | 871 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 872 | return ERR_PTR(-EEXIST); |
77bdae4d | 873 | } |
b96de000 | 874 | |
a9261d41 AJ |
875 | /* |
876 | * We are going to replace the device path for a given devid, | |
877 | * make sure it's the same device if the device is mounted | |
878 | */ | |
879 | if (device->bdev) { | |
880 | struct block_device *path_bdev; | |
881 | ||
882 | path_bdev = lookup_bdev(path); | |
883 | if (IS_ERR(path_bdev)) { | |
884 | mutex_unlock(&fs_devices->device_list_mutex); | |
885 | return ERR_CAST(path_bdev); | |
886 | } | |
887 | ||
888 | if (device->bdev != path_bdev) { | |
889 | bdput(path_bdev); | |
890 | mutex_unlock(&fs_devices->device_list_mutex); | |
891 | btrfs_warn_in_rcu(device->fs_info, | |
892 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
893 | disk_super->fsid, devid, | |
894 | rcu_str_deref(device->name), path); | |
895 | return ERR_PTR(-EEXIST); | |
896 | } | |
897 | bdput(path_bdev); | |
898 | btrfs_info_in_rcu(device->fs_info, | |
899 | "device fsid %pU devid %llu moved old:%s new:%s", | |
900 | disk_super->fsid, devid, | |
901 | rcu_str_deref(device->name), path); | |
902 | } | |
903 | ||
606686ee | 904 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
905 | if (!name) { |
906 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 907 | return ERR_PTR(-ENOMEM); |
9c6d173e | 908 | } |
606686ee JB |
909 | rcu_string_free(device->name); |
910 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 911 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 912 | fs_devices->missing_devices--; |
e6e674bd | 913 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 914 | } |
8a4b83cc CM |
915 | } |
916 | ||
77bdae4d AJ |
917 | /* |
918 | * Unmount does not free the btrfs_device struct but would zero | |
919 | * generation along with most of the other members. So just update | |
920 | * it back. We need it to pick the disk with largest generation | |
921 | * (as above). | |
922 | */ | |
d1a63002 | 923 | if (!fs_devices->opened) { |
77bdae4d | 924 | device->generation = found_transid; |
d1a63002 NB |
925 | fs_devices->latest_generation = max_t(u64, found_transid, |
926 | fs_devices->latest_generation); | |
927 | } | |
77bdae4d | 928 | |
f2788d2f AJ |
929 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
930 | ||
9c6d173e | 931 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 932 | return device; |
8a4b83cc CM |
933 | } |
934 | ||
e4404d6e YZ |
935 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
936 | { | |
937 | struct btrfs_fs_devices *fs_devices; | |
938 | struct btrfs_device *device; | |
939 | struct btrfs_device *orig_dev; | |
d2979aa2 | 940 | int ret = 0; |
e4404d6e | 941 | |
7239ff4b | 942 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
943 | if (IS_ERR(fs_devices)) |
944 | return fs_devices; | |
e4404d6e | 945 | |
adbbb863 | 946 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 947 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
948 | |
949 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
950 | struct rcu_string *name; |
951 | ||
12bd2fc0 ID |
952 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
953 | orig_dev->uuid); | |
d2979aa2 AJ |
954 | if (IS_ERR(device)) { |
955 | ret = PTR_ERR(device); | |
e4404d6e | 956 | goto error; |
d2979aa2 | 957 | } |
e4404d6e | 958 | |
606686ee JB |
959 | /* |
960 | * This is ok to do without rcu read locked because we hold the | |
961 | * uuid mutex so nothing we touch in here is going to disappear. | |
962 | */ | |
e755f780 | 963 | if (orig_dev->name) { |
78f2c9e6 DS |
964 | name = rcu_string_strdup(orig_dev->name->str, |
965 | GFP_KERNEL); | |
e755f780 | 966 | if (!name) { |
a425f9d4 | 967 | btrfs_free_device(device); |
d2979aa2 | 968 | ret = -ENOMEM; |
e755f780 AJ |
969 | goto error; |
970 | } | |
971 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 972 | } |
e4404d6e | 973 | |
e4404d6e YZ |
974 | list_add(&device->dev_list, &fs_devices->devices); |
975 | device->fs_devices = fs_devices; | |
976 | fs_devices->num_devices++; | |
977 | } | |
adbbb863 | 978 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
979 | return fs_devices; |
980 | error: | |
adbbb863 | 981 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 982 | free_fs_devices(fs_devices); |
d2979aa2 | 983 | return ERR_PTR(ret); |
e4404d6e YZ |
984 | } |
985 | ||
9b99b115 AJ |
986 | /* |
987 | * After we have read the system tree and know devids belonging to | |
988 | * this filesystem, remove the device which does not belong there. | |
989 | */ | |
990 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 991 | { |
c6e30871 | 992 | struct btrfs_device *device, *next; |
443f24fe | 993 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 994 | |
dfe25020 CM |
995 | mutex_lock(&uuid_mutex); |
996 | again: | |
46224705 | 997 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 998 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
999 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1000 | &device->dev_state)) { | |
401e29c1 AJ |
1001 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1002 | &device->dev_state) && | |
1003 | (!latest_dev || | |
1004 | device->generation > latest_dev->generation)) { | |
443f24fe | 1005 | latest_dev = device; |
a6b0d5c8 | 1006 | } |
2b82032c | 1007 | continue; |
a6b0d5c8 | 1008 | } |
2b82032c | 1009 | |
8dabb742 SB |
1010 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1011 | /* | |
1012 | * In the first step, keep the device which has | |
1013 | * the correct fsid and the devid that is used | |
1014 | * for the dev_replace procedure. | |
1015 | * In the second step, the dev_replace state is | |
1016 | * read from the device tree and it is known | |
1017 | * whether the procedure is really active or | |
1018 | * not, which means whether this device is | |
1019 | * used or whether it should be removed. | |
1020 | */ | |
401e29c1 AJ |
1021 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1022 | &device->dev_state)) { | |
8dabb742 SB |
1023 | continue; |
1024 | } | |
1025 | } | |
2b82032c | 1026 | if (device->bdev) { |
d4d77629 | 1027 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1028 | device->bdev = NULL; |
1029 | fs_devices->open_devices--; | |
1030 | } | |
ebbede42 | 1031 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1032 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1033 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1034 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1035 | &device->dev_state)) | |
8dabb742 | 1036 | fs_devices->rw_devices--; |
2b82032c | 1037 | } |
e4404d6e YZ |
1038 | list_del_init(&device->dev_list); |
1039 | fs_devices->num_devices--; | |
a425f9d4 | 1040 | btrfs_free_device(device); |
dfe25020 | 1041 | } |
2b82032c YZ |
1042 | |
1043 | if (fs_devices->seed) { | |
1044 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1045 | goto again; |
1046 | } | |
1047 | ||
443f24fe | 1048 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1049 | |
dfe25020 | 1050 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1051 | } |
a0af469b | 1052 | |
14238819 AJ |
1053 | static void btrfs_close_bdev(struct btrfs_device *device) |
1054 | { | |
08ffcae8 DS |
1055 | if (!device->bdev) |
1056 | return; | |
1057 | ||
ebbede42 | 1058 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1059 | sync_blockdev(device->bdev); |
1060 | invalidate_bdev(device->bdev); | |
1061 | } | |
1062 | ||
08ffcae8 | 1063 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1064 | } |
1065 | ||
959b1c04 | 1066 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1067 | { |
1068 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1069 | struct btrfs_device *new_device; | |
1070 | struct rcu_string *name; | |
1071 | ||
ebbede42 | 1072 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1073 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1074 | list_del_init(&device->dev_alloc_list); | |
1075 | fs_devices->rw_devices--; | |
1076 | } | |
1077 | ||
e6e674bd | 1078 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1079 | fs_devices->missing_devices--; |
1080 | ||
959b1c04 | 1081 | btrfs_close_bdev(device); |
3fff3975 JT |
1082 | if (device->bdev) |
1083 | fs_devices->open_devices--; | |
959b1c04 | 1084 | |
f448341a AJ |
1085 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1086 | device->uuid); | |
1087 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1088 | ||
1089 | /* Safe because we are under uuid_mutex */ | |
1090 | if (device->name) { | |
1091 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1092 | BUG_ON(!name); /* -ENOMEM */ | |
1093 | rcu_assign_pointer(new_device->name, name); | |
1094 | } | |
1095 | ||
1096 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1097 | new_device->fs_devices = device->fs_devices; | |
959b1c04 | 1098 | |
8e75fd89 NB |
1099 | synchronize_rcu(); |
1100 | btrfs_free_device(device); | |
f448341a AJ |
1101 | } |
1102 | ||
0226e0eb | 1103 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1104 | { |
2037a093 | 1105 | struct btrfs_device *device, *tmp; |
e4404d6e | 1106 | |
2b82032c YZ |
1107 | if (--fs_devices->opened > 0) |
1108 | return 0; | |
8a4b83cc | 1109 | |
c9513edb | 1110 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1111 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1112 | btrfs_close_one_device(device); |
8a4b83cc | 1113 | } |
c9513edb XG |
1114 | mutex_unlock(&fs_devices->device_list_mutex); |
1115 | ||
e4404d6e YZ |
1116 | WARN_ON(fs_devices->open_devices); |
1117 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1118 | fs_devices->opened = 0; |
0395d84f | 1119 | fs_devices->seeding = false; |
2b82032c | 1120 | |
8a4b83cc CM |
1121 | return 0; |
1122 | } | |
1123 | ||
2b82032c YZ |
1124 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1125 | { | |
e4404d6e | 1126 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1127 | int ret; |
1128 | ||
1129 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1130 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1131 | if (!fs_devices->opened) { |
1132 | seed_devices = fs_devices->seed; | |
1133 | fs_devices->seed = NULL; | |
1134 | } | |
2b82032c | 1135 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1136 | |
1137 | while (seed_devices) { | |
1138 | fs_devices = seed_devices; | |
1139 | seed_devices = fs_devices->seed; | |
0226e0eb | 1140 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1141 | free_fs_devices(fs_devices); |
1142 | } | |
2b82032c YZ |
1143 | return ret; |
1144 | } | |
1145 | ||
897fb573 | 1146 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1147 | fmode_t flags, void *holder) |
8a4b83cc | 1148 | { |
8a4b83cc | 1149 | struct btrfs_device *device; |
443f24fe | 1150 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1151 | int ret = 0; |
8a4b83cc | 1152 | |
d4d77629 TH |
1153 | flags |= FMODE_EXCL; |
1154 | ||
f117e290 | 1155 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1156 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1157 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1158 | continue; |
a0af469b | 1159 | |
9f050db4 AJ |
1160 | if (!latest_dev || |
1161 | device->generation > latest_dev->generation) | |
1162 | latest_dev = device; | |
8a4b83cc | 1163 | } |
a0af469b | 1164 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1165 | ret = -EINVAL; |
a0af469b CM |
1166 | goto out; |
1167 | } | |
2b82032c | 1168 | fs_devices->opened = 1; |
443f24fe | 1169 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1170 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1171 | out: |
2b82032c YZ |
1172 | return ret; |
1173 | } | |
1174 | ||
f8e10cd3 AJ |
1175 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1176 | { | |
1177 | struct btrfs_device *dev1, *dev2; | |
1178 | ||
1179 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1180 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1181 | ||
1182 | if (dev1->devid < dev2->devid) | |
1183 | return -1; | |
1184 | else if (dev1->devid > dev2->devid) | |
1185 | return 1; | |
1186 | return 0; | |
1187 | } | |
1188 | ||
2b82032c | 1189 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1190 | fmode_t flags, void *holder) |
2b82032c YZ |
1191 | { |
1192 | int ret; | |
1193 | ||
f5194e34 DS |
1194 | lockdep_assert_held(&uuid_mutex); |
1195 | ||
542c5908 | 1196 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1197 | if (fs_devices->opened) { |
e4404d6e YZ |
1198 | fs_devices->opened++; |
1199 | ret = 0; | |
2b82032c | 1200 | } else { |
f8e10cd3 | 1201 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1202 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1203 | } |
542c5908 AJ |
1204 | mutex_unlock(&fs_devices->device_list_mutex); |
1205 | ||
8a4b83cc CM |
1206 | return ret; |
1207 | } | |
1208 | ||
c9162bdf | 1209 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1210 | { |
1211 | kunmap(page); | |
1212 | put_page(page); | |
1213 | } | |
1214 | ||
c9162bdf OS |
1215 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1216 | struct page **page, | |
1217 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1218 | { |
1219 | void *p; | |
1220 | pgoff_t index; | |
1221 | ||
1222 | /* make sure our super fits in the device */ | |
1223 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1224 | return 1; | |
1225 | ||
1226 | /* make sure our super fits in the page */ | |
1227 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1228 | return 1; | |
1229 | ||
1230 | /* make sure our super doesn't straddle pages on disk */ | |
1231 | index = bytenr >> PAGE_SHIFT; | |
1232 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1233 | return 1; | |
1234 | ||
1235 | /* pull in the page with our super */ | |
1236 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1237 | index, GFP_KERNEL); | |
1238 | ||
1239 | if (IS_ERR_OR_NULL(*page)) | |
1240 | return 1; | |
1241 | ||
1242 | p = kmap(*page); | |
1243 | ||
1244 | /* align our pointer to the offset of the super block */ | |
7073017a | 1245 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1246 | |
1247 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1248 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1249 | btrfs_release_disk_super(*page); | |
1250 | return 1; | |
1251 | } | |
1252 | ||
1253 | if ((*disk_super)->label[0] && | |
1254 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1255 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1256 | ||
1257 | return 0; | |
1258 | } | |
1259 | ||
228a73ab AJ |
1260 | int btrfs_forget_devices(const char *path) |
1261 | { | |
1262 | int ret; | |
1263 | ||
1264 | mutex_lock(&uuid_mutex); | |
1265 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1266 | mutex_unlock(&uuid_mutex); | |
1267 | ||
1268 | return ret; | |
1269 | } | |
1270 | ||
6f60cbd3 DS |
1271 | /* |
1272 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1273 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1274 | * is read via pagecache | |
1275 | */ | |
36350e95 GJ |
1276 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1277 | void *holder) | |
8a4b83cc CM |
1278 | { |
1279 | struct btrfs_super_block *disk_super; | |
4306a974 | 1280 | bool new_device_added = false; |
36350e95 | 1281 | struct btrfs_device *device = NULL; |
8a4b83cc | 1282 | struct block_device *bdev; |
6f60cbd3 | 1283 | struct page *page; |
6f60cbd3 | 1284 | u64 bytenr; |
8a4b83cc | 1285 | |
899f9307 DS |
1286 | lockdep_assert_held(&uuid_mutex); |
1287 | ||
6f60cbd3 DS |
1288 | /* |
1289 | * we would like to check all the supers, but that would make | |
1290 | * a btrfs mount succeed after a mkfs from a different FS. | |
1291 | * So, we need to add a special mount option to scan for | |
1292 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1293 | */ | |
1294 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1295 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1296 | |
1297 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1298 | if (IS_ERR(bdev)) |
36350e95 | 1299 | return ERR_CAST(bdev); |
6f60cbd3 | 1300 | |
05a5c55d | 1301 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1302 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1303 | goto error_bdev_put; |
05a5c55d | 1304 | } |
6f60cbd3 | 1305 | |
4306a974 | 1306 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1307 | if (!IS_ERR(device)) { |
4306a974 AJ |
1308 | if (new_device_added) |
1309 | btrfs_free_stale_devices(path, device); | |
1310 | } | |
6f60cbd3 | 1311 | |
6cf86a00 | 1312 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1313 | |
1314 | error_bdev_put: | |
d4d77629 | 1315 | blkdev_put(bdev, flags); |
b6ed73bc | 1316 | |
36350e95 | 1317 | return device; |
8a4b83cc | 1318 | } |
0b86a832 | 1319 | |
1c11b63e JM |
1320 | /* |
1321 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1322 | * such is found, record the end of it in *start | |
1323 | */ | |
1c11b63e JM |
1324 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1325 | u64 len) | |
6df9a95e | 1326 | { |
1c11b63e | 1327 | u64 physical_start, physical_end; |
6df9a95e | 1328 | |
1c11b63e | 1329 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1330 | |
1c11b63e JM |
1331 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1332 | &physical_start, &physical_end, | |
1333 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1334 | |
1c11b63e JM |
1335 | if (in_range(physical_start, *start, len) || |
1336 | in_range(*start, physical_start, | |
1337 | physical_end - physical_start)) { | |
1338 | *start = physical_end + 1; | |
1339 | return true; | |
6df9a95e JB |
1340 | } |
1341 | } | |
1c11b63e | 1342 | return false; |
6df9a95e JB |
1343 | } |
1344 | ||
1345 | ||
0b86a832 | 1346 | /* |
499f377f JM |
1347 | * find_free_dev_extent_start - find free space in the specified device |
1348 | * @device: the device which we search the free space in | |
1349 | * @num_bytes: the size of the free space that we need | |
1350 | * @search_start: the position from which to begin the search | |
1351 | * @start: store the start of the free space. | |
1352 | * @len: the size of the free space. that we find, or the size | |
1353 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1354 | * |
0b86a832 CM |
1355 | * this uses a pretty simple search, the expectation is that it is |
1356 | * called very infrequently and that a given device has a small number | |
1357 | * of extents | |
7bfc837d MX |
1358 | * |
1359 | * @start is used to store the start of the free space if we find. But if we | |
1360 | * don't find suitable free space, it will be used to store the start position | |
1361 | * of the max free space. | |
1362 | * | |
1363 | * @len is used to store the size of the free space that we find. | |
1364 | * But if we don't find suitable free space, it is used to store the size of | |
1365 | * the max free space. | |
135da976 QW |
1366 | * |
1367 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1368 | * check to ensure dev extents are not double allocated. | |
1369 | * This makes the function safe to allocate dev extents but may not report | |
1370 | * correct usable device space, as device extent freed in current transaction | |
1371 | * is not reported as avaiable. | |
0b86a832 | 1372 | */ |
9e3246a5 QW |
1373 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1374 | u64 num_bytes, u64 search_start, u64 *start, | |
1375 | u64 *len) | |
0b86a832 | 1376 | { |
0b246afa JM |
1377 | struct btrfs_fs_info *fs_info = device->fs_info; |
1378 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1379 | struct btrfs_key key; |
7bfc837d | 1380 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1381 | struct btrfs_path *path; |
7bfc837d MX |
1382 | u64 hole_size; |
1383 | u64 max_hole_start; | |
1384 | u64 max_hole_size; | |
1385 | u64 extent_end; | |
0b86a832 CM |
1386 | u64 search_end = device->total_bytes; |
1387 | int ret; | |
7bfc837d | 1388 | int slot; |
0b86a832 | 1389 | struct extent_buffer *l; |
8cdc7c5b FM |
1390 | |
1391 | /* | |
1392 | * We don't want to overwrite the superblock on the drive nor any area | |
1393 | * used by the boot loader (grub for example), so we make sure to start | |
1394 | * at an offset of at least 1MB. | |
1395 | */ | |
0d0c71b3 | 1396 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1397 | |
6df9a95e JB |
1398 | path = btrfs_alloc_path(); |
1399 | if (!path) | |
1400 | return -ENOMEM; | |
f2ab7618 | 1401 | |
7bfc837d MX |
1402 | max_hole_start = search_start; |
1403 | max_hole_size = 0; | |
1404 | ||
f2ab7618 | 1405 | again: |
401e29c1 AJ |
1406 | if (search_start >= search_end || |
1407 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1408 | ret = -ENOSPC; |
6df9a95e | 1409 | goto out; |
7bfc837d MX |
1410 | } |
1411 | ||
e4058b54 | 1412 | path->reada = READA_FORWARD; |
6df9a95e JB |
1413 | path->search_commit_root = 1; |
1414 | path->skip_locking = 1; | |
7bfc837d | 1415 | |
0b86a832 CM |
1416 | key.objectid = device->devid; |
1417 | key.offset = search_start; | |
1418 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1419 | |
125ccb0a | 1420 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1421 | if (ret < 0) |
7bfc837d | 1422 | goto out; |
1fcbac58 YZ |
1423 | if (ret > 0) { |
1424 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1425 | if (ret < 0) | |
7bfc837d | 1426 | goto out; |
1fcbac58 | 1427 | } |
7bfc837d | 1428 | |
0b86a832 CM |
1429 | while (1) { |
1430 | l = path->nodes[0]; | |
1431 | slot = path->slots[0]; | |
1432 | if (slot >= btrfs_header_nritems(l)) { | |
1433 | ret = btrfs_next_leaf(root, path); | |
1434 | if (ret == 0) | |
1435 | continue; | |
1436 | if (ret < 0) | |
7bfc837d MX |
1437 | goto out; |
1438 | ||
1439 | break; | |
0b86a832 CM |
1440 | } |
1441 | btrfs_item_key_to_cpu(l, &key, slot); | |
1442 | ||
1443 | if (key.objectid < device->devid) | |
1444 | goto next; | |
1445 | ||
1446 | if (key.objectid > device->devid) | |
7bfc837d | 1447 | break; |
0b86a832 | 1448 | |
962a298f | 1449 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1450 | goto next; |
9779b72f | 1451 | |
7bfc837d MX |
1452 | if (key.offset > search_start) { |
1453 | hole_size = key.offset - search_start; | |
9779b72f | 1454 | |
6df9a95e JB |
1455 | /* |
1456 | * Have to check before we set max_hole_start, otherwise | |
1457 | * we could end up sending back this offset anyway. | |
1458 | */ | |
1c11b63e | 1459 | if (contains_pending_extent(device, &search_start, |
1b984508 | 1460 | hole_size)) { |
1c11b63e | 1461 | if (key.offset >= search_start) |
1b984508 | 1462 | hole_size = key.offset - search_start; |
1c11b63e | 1463 | else |
1b984508 | 1464 | hole_size = 0; |
1b984508 | 1465 | } |
6df9a95e | 1466 | |
7bfc837d MX |
1467 | if (hole_size > max_hole_size) { |
1468 | max_hole_start = search_start; | |
1469 | max_hole_size = hole_size; | |
1470 | } | |
9779b72f | 1471 | |
7bfc837d MX |
1472 | /* |
1473 | * If this free space is greater than which we need, | |
1474 | * it must be the max free space that we have found | |
1475 | * until now, so max_hole_start must point to the start | |
1476 | * of this free space and the length of this free space | |
1477 | * is stored in max_hole_size. Thus, we return | |
1478 | * max_hole_start and max_hole_size and go back to the | |
1479 | * caller. | |
1480 | */ | |
1481 | if (hole_size >= num_bytes) { | |
1482 | ret = 0; | |
1483 | goto out; | |
0b86a832 CM |
1484 | } |
1485 | } | |
0b86a832 | 1486 | |
0b86a832 | 1487 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1488 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1489 | dev_extent); | |
1490 | if (extent_end > search_start) | |
1491 | search_start = extent_end; | |
0b86a832 CM |
1492 | next: |
1493 | path->slots[0]++; | |
1494 | cond_resched(); | |
1495 | } | |
0b86a832 | 1496 | |
38c01b96 | 1497 | /* |
1498 | * At this point, search_start should be the end of | |
1499 | * allocated dev extents, and when shrinking the device, | |
1500 | * search_end may be smaller than search_start. | |
1501 | */ | |
f2ab7618 | 1502 | if (search_end > search_start) { |
38c01b96 | 1503 | hole_size = search_end - search_start; |
1504 | ||
1c11b63e | 1505 | if (contains_pending_extent(device, &search_start, hole_size)) { |
f2ab7618 ZL |
1506 | btrfs_release_path(path); |
1507 | goto again; | |
1508 | } | |
0b86a832 | 1509 | |
f2ab7618 ZL |
1510 | if (hole_size > max_hole_size) { |
1511 | max_hole_start = search_start; | |
1512 | max_hole_size = hole_size; | |
1513 | } | |
6df9a95e JB |
1514 | } |
1515 | ||
7bfc837d | 1516 | /* See above. */ |
f2ab7618 | 1517 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1518 | ret = -ENOSPC; |
1519 | else | |
1520 | ret = 0; | |
1521 | ||
1522 | out: | |
2b82032c | 1523 | btrfs_free_path(path); |
7bfc837d | 1524 | *start = max_hole_start; |
b2117a39 | 1525 | if (len) |
7bfc837d | 1526 | *len = max_hole_size; |
0b86a832 CM |
1527 | return ret; |
1528 | } | |
1529 | ||
60dfdf25 | 1530 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1531 | u64 *start, u64 *len) |
1532 | { | |
499f377f | 1533 | /* FIXME use last free of some kind */ |
60dfdf25 | 1534 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1535 | } |
1536 | ||
b2950863 | 1537 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1538 | struct btrfs_device *device, |
2196d6e8 | 1539 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1540 | { |
0b246afa JM |
1541 | struct btrfs_fs_info *fs_info = device->fs_info; |
1542 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1543 | int ret; |
1544 | struct btrfs_path *path; | |
8f18cf13 | 1545 | struct btrfs_key key; |
a061fc8d CM |
1546 | struct btrfs_key found_key; |
1547 | struct extent_buffer *leaf = NULL; | |
1548 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1549 | |
1550 | path = btrfs_alloc_path(); | |
1551 | if (!path) | |
1552 | return -ENOMEM; | |
1553 | ||
1554 | key.objectid = device->devid; | |
1555 | key.offset = start; | |
1556 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1557 | again: |
8f18cf13 | 1558 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1559 | if (ret > 0) { |
1560 | ret = btrfs_previous_item(root, path, key.objectid, | |
1561 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1562 | if (ret) |
1563 | goto out; | |
a061fc8d CM |
1564 | leaf = path->nodes[0]; |
1565 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1566 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1567 | struct btrfs_dev_extent); | |
1568 | BUG_ON(found_key.offset > start || found_key.offset + | |
1569 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1570 | key = found_key; |
1571 | btrfs_release_path(path); | |
1572 | goto again; | |
a061fc8d CM |
1573 | } else if (ret == 0) { |
1574 | leaf = path->nodes[0]; | |
1575 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1576 | struct btrfs_dev_extent); | |
79787eaa | 1577 | } else { |
0b246afa | 1578 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1579 | goto out; |
a061fc8d | 1580 | } |
8f18cf13 | 1581 | |
2196d6e8 MX |
1582 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1583 | ||
8f18cf13 | 1584 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1585 | if (ret) { |
0b246afa JM |
1586 | btrfs_handle_fs_error(fs_info, ret, |
1587 | "Failed to remove dev extent item"); | |
13212b54 | 1588 | } else { |
3204d33c | 1589 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1590 | } |
b0b802d7 | 1591 | out: |
8f18cf13 CM |
1592 | btrfs_free_path(path); |
1593 | return ret; | |
1594 | } | |
1595 | ||
48a3b636 ES |
1596 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1597 | struct btrfs_device *device, | |
48a3b636 | 1598 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1599 | { |
1600 | int ret; | |
1601 | struct btrfs_path *path; | |
0b246afa JM |
1602 | struct btrfs_fs_info *fs_info = device->fs_info; |
1603 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1604 | struct btrfs_dev_extent *extent; |
1605 | struct extent_buffer *leaf; | |
1606 | struct btrfs_key key; | |
1607 | ||
e12c9621 | 1608 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1609 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1610 | path = btrfs_alloc_path(); |
1611 | if (!path) | |
1612 | return -ENOMEM; | |
1613 | ||
0b86a832 | 1614 | key.objectid = device->devid; |
2b82032c | 1615 | key.offset = start; |
0b86a832 CM |
1616 | key.type = BTRFS_DEV_EXTENT_KEY; |
1617 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1618 | sizeof(*extent)); | |
2cdcecbc MF |
1619 | if (ret) |
1620 | goto out; | |
0b86a832 CM |
1621 | |
1622 | leaf = path->nodes[0]; | |
1623 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1624 | struct btrfs_dev_extent); | |
b5d9071c NB |
1625 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1626 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1627 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1628 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1629 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1630 | ||
0b86a832 CM |
1631 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1632 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1633 | out: |
0b86a832 CM |
1634 | btrfs_free_path(path); |
1635 | return ret; | |
1636 | } | |
1637 | ||
6df9a95e | 1638 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1639 | { |
6df9a95e JB |
1640 | struct extent_map_tree *em_tree; |
1641 | struct extent_map *em; | |
1642 | struct rb_node *n; | |
1643 | u64 ret = 0; | |
0b86a832 | 1644 | |
c8bf1b67 | 1645 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1646 | read_lock(&em_tree->lock); |
07e1ce09 | 1647 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1648 | if (n) { |
1649 | em = rb_entry(n, struct extent_map, rb_node); | |
1650 | ret = em->start + em->len; | |
0b86a832 | 1651 | } |
6df9a95e JB |
1652 | read_unlock(&em_tree->lock); |
1653 | ||
0b86a832 CM |
1654 | return ret; |
1655 | } | |
1656 | ||
53f10659 ID |
1657 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1658 | u64 *devid_ret) | |
0b86a832 CM |
1659 | { |
1660 | int ret; | |
1661 | struct btrfs_key key; | |
1662 | struct btrfs_key found_key; | |
2b82032c YZ |
1663 | struct btrfs_path *path; |
1664 | ||
2b82032c YZ |
1665 | path = btrfs_alloc_path(); |
1666 | if (!path) | |
1667 | return -ENOMEM; | |
0b86a832 CM |
1668 | |
1669 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1670 | key.type = BTRFS_DEV_ITEM_KEY; | |
1671 | key.offset = (u64)-1; | |
1672 | ||
53f10659 | 1673 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1674 | if (ret < 0) |
1675 | goto error; | |
1676 | ||
a06dee4d AJ |
1677 | if (ret == 0) { |
1678 | /* Corruption */ | |
1679 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1680 | ret = -EUCLEAN; | |
1681 | goto error; | |
1682 | } | |
0b86a832 | 1683 | |
53f10659 ID |
1684 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1685 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1686 | BTRFS_DEV_ITEM_KEY); |
1687 | if (ret) { | |
53f10659 | 1688 | *devid_ret = 1; |
0b86a832 CM |
1689 | } else { |
1690 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1691 | path->slots[0]); | |
53f10659 | 1692 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1693 | } |
1694 | ret = 0; | |
1695 | error: | |
2b82032c | 1696 | btrfs_free_path(path); |
0b86a832 CM |
1697 | return ret; |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * the device information is stored in the chunk root | |
1702 | * the btrfs_device struct should be fully filled in | |
1703 | */ | |
c74a0b02 | 1704 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1705 | struct btrfs_device *device) |
0b86a832 CM |
1706 | { |
1707 | int ret; | |
1708 | struct btrfs_path *path; | |
1709 | struct btrfs_dev_item *dev_item; | |
1710 | struct extent_buffer *leaf; | |
1711 | struct btrfs_key key; | |
1712 | unsigned long ptr; | |
0b86a832 | 1713 | |
0b86a832 CM |
1714 | path = btrfs_alloc_path(); |
1715 | if (!path) | |
1716 | return -ENOMEM; | |
1717 | ||
0b86a832 CM |
1718 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1719 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1720 | key.offset = device->devid; |
0b86a832 | 1721 | |
8e87e856 NB |
1722 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1723 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1724 | if (ret) |
1725 | goto out; | |
1726 | ||
1727 | leaf = path->nodes[0]; | |
1728 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1729 | ||
1730 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1731 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1732 | btrfs_set_device_type(leaf, dev_item, device->type); |
1733 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1734 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1735 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1736 | btrfs_set_device_total_bytes(leaf, dev_item, |
1737 | btrfs_device_get_disk_total_bytes(device)); | |
1738 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1739 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1740 | btrfs_set_device_group(leaf, dev_item, 0); |
1741 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1742 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1743 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1744 | |
410ba3a2 | 1745 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1746 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1747 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1748 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1749 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1750 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1751 | |
2b82032c | 1752 | ret = 0; |
0b86a832 CM |
1753 | out: |
1754 | btrfs_free_path(path); | |
1755 | return ret; | |
1756 | } | |
8f18cf13 | 1757 | |
5a1972bd QW |
1758 | /* |
1759 | * Function to update ctime/mtime for a given device path. | |
1760 | * Mainly used for ctime/mtime based probe like libblkid. | |
1761 | */ | |
da353f6b | 1762 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1763 | { |
1764 | struct file *filp; | |
1765 | ||
1766 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1767 | if (IS_ERR(filp)) |
5a1972bd QW |
1768 | return; |
1769 | file_update_time(filp); | |
1770 | filp_close(filp, NULL); | |
5a1972bd QW |
1771 | } |
1772 | ||
f331a952 | 1773 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1774 | { |
f331a952 | 1775 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1776 | int ret; |
1777 | struct btrfs_path *path; | |
a061fc8d | 1778 | struct btrfs_key key; |
a061fc8d CM |
1779 | struct btrfs_trans_handle *trans; |
1780 | ||
a061fc8d CM |
1781 | path = btrfs_alloc_path(); |
1782 | if (!path) | |
1783 | return -ENOMEM; | |
1784 | ||
a22285a6 | 1785 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1786 | if (IS_ERR(trans)) { |
1787 | btrfs_free_path(path); | |
1788 | return PTR_ERR(trans); | |
1789 | } | |
a061fc8d CM |
1790 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1791 | key.type = BTRFS_DEV_ITEM_KEY; | |
1792 | key.offset = device->devid; | |
1793 | ||
1794 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1795 | if (ret) { |
1796 | if (ret > 0) | |
1797 | ret = -ENOENT; | |
1798 | btrfs_abort_transaction(trans, ret); | |
1799 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1800 | goto out; |
1801 | } | |
1802 | ||
1803 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1804 | if (ret) { |
1805 | btrfs_abort_transaction(trans, ret); | |
1806 | btrfs_end_transaction(trans); | |
1807 | } | |
1808 | ||
a061fc8d CM |
1809 | out: |
1810 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1811 | if (!ret) |
1812 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1813 | return ret; |
1814 | } | |
1815 | ||
3cc31a0d DS |
1816 | /* |
1817 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1818 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1819 | * replace. | |
1820 | */ | |
1821 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1822 | u64 num_devices) | |
a061fc8d | 1823 | { |
a061fc8d | 1824 | u64 all_avail; |
de98ced9 | 1825 | unsigned seq; |
418775a2 | 1826 | int i; |
a061fc8d | 1827 | |
de98ced9 | 1828 | do { |
bd45ffbc | 1829 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1830 | |
bd45ffbc AJ |
1831 | all_avail = fs_info->avail_data_alloc_bits | |
1832 | fs_info->avail_system_alloc_bits | | |
1833 | fs_info->avail_metadata_alloc_bits; | |
1834 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1835 | |
418775a2 | 1836 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1837 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1838 | continue; |
a061fc8d | 1839 | |
418775a2 | 1840 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1841 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1842 | |
418775a2 DS |
1843 | if (ret) |
1844 | return ret; | |
1845 | } | |
53b381b3 DW |
1846 | } |
1847 | ||
bd45ffbc | 1848 | return 0; |
f1fa7f26 AJ |
1849 | } |
1850 | ||
c9162bdf OS |
1851 | static struct btrfs_device * btrfs_find_next_active_device( |
1852 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1853 | { |
2b82032c | 1854 | struct btrfs_device *next_device; |
88acff64 AJ |
1855 | |
1856 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1857 | if (next_device != device && | |
e6e674bd AJ |
1858 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1859 | && next_device->bdev) | |
88acff64 AJ |
1860 | return next_device; |
1861 | } | |
1862 | ||
1863 | return NULL; | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1868 | * and replace it with the provided or the next active device, in the context | |
1869 | * where this function called, there should be always be another device (or | |
1870 | * this_dev) which is active. | |
1871 | */ | |
b105e927 | 1872 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
d6507cf1 | 1873 | struct btrfs_device *this_dev) |
88acff64 | 1874 | { |
d6507cf1 | 1875 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1876 | struct btrfs_device *next_device; |
1877 | ||
1878 | if (this_dev) | |
1879 | next_device = this_dev; | |
1880 | else | |
1881 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1882 | device); | |
1883 | ASSERT(next_device); | |
1884 | ||
1885 | if (fs_info->sb->s_bdev && | |
1886 | (fs_info->sb->s_bdev == device->bdev)) | |
1887 | fs_info->sb->s_bdev = next_device->bdev; | |
1888 | ||
1889 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1890 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1891 | } | |
1892 | ||
1da73967 AJ |
1893 | /* |
1894 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1895 | * currently replaced. | |
1896 | */ | |
1897 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1898 | { | |
1899 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1900 | ||
cb5583dd | 1901 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1902 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
1903 | ASSERT(num_devices > 1); | |
1904 | num_devices--; | |
1905 | } | |
cb5583dd | 1906 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1907 | |
1908 | return num_devices; | |
1909 | } | |
1910 | ||
da353f6b DS |
1911 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1912 | u64 devid) | |
f1fa7f26 AJ |
1913 | { |
1914 | struct btrfs_device *device; | |
1f78160c | 1915 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 1916 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 1917 | u64 num_devices; |
a061fc8d CM |
1918 | int ret = 0; |
1919 | ||
a061fc8d CM |
1920 | mutex_lock(&uuid_mutex); |
1921 | ||
1da73967 | 1922 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 1923 | |
0b246afa | 1924 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1925 | if (ret) |
a061fc8d | 1926 | goto out; |
a061fc8d | 1927 | |
a27a94c2 NB |
1928 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
1929 | ||
1930 | if (IS_ERR(device)) { | |
1931 | if (PTR_ERR(device) == -ENOENT && | |
1932 | strcmp(device_path, "missing") == 0) | |
1933 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
1934 | else | |
1935 | ret = PTR_ERR(device); | |
53b381b3 | 1936 | goto out; |
a27a94c2 | 1937 | } |
dfe25020 | 1938 | |
eede2bf3 OS |
1939 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
1940 | btrfs_warn_in_rcu(fs_info, | |
1941 | "cannot remove device %s (devid %llu) due to active swapfile", | |
1942 | rcu_str_deref(device->name), device->devid); | |
1943 | ret = -ETXTBSY; | |
1944 | goto out; | |
1945 | } | |
1946 | ||
401e29c1 | 1947 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1948 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1949 | goto out; |
63a212ab SB |
1950 | } |
1951 | ||
ebbede42 AJ |
1952 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1953 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1954 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1955 | goto out; |
2b82032c YZ |
1956 | } |
1957 | ||
ebbede42 | 1958 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 1959 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 1960 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1961 | device->fs_devices->rw_devices--; |
34441361 | 1962 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 1963 | } |
a061fc8d | 1964 | |
d7901554 | 1965 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1966 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1967 | mutex_lock(&uuid_mutex); |
a061fc8d | 1968 | if (ret) |
9b3517e9 | 1969 | goto error_undo; |
a061fc8d | 1970 | |
63a212ab SB |
1971 | /* |
1972 | * TODO: the superblock still includes this device in its num_devices | |
1973 | * counter although write_all_supers() is not locked out. This | |
1974 | * could give a filesystem state which requires a degraded mount. | |
1975 | */ | |
f331a952 | 1976 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 1977 | if (ret) |
9b3517e9 | 1978 | goto error_undo; |
a061fc8d | 1979 | |
e12c9621 | 1980 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 1981 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
1982 | |
1983 | /* | |
1984 | * the device list mutex makes sure that we don't change | |
1985 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1986 | * the device supers. Whoever is writing all supers, should |
1987 | * lock the device list mutex before getting the number of | |
1988 | * devices in the super block (super_copy). Conversely, | |
1989 | * whoever updates the number of devices in the super block | |
1990 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1991 | */ |
1f78160c | 1992 | |
41a52a0f AJ |
1993 | /* |
1994 | * In normal cases the cur_devices == fs_devices. But in case | |
1995 | * of deleting a seed device, the cur_devices should point to | |
1996 | * its own fs_devices listed under the fs_devices->seed. | |
1997 | */ | |
1f78160c | 1998 | cur_devices = device->fs_devices; |
b5185197 | 1999 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2000 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2001 | |
41a52a0f AJ |
2002 | cur_devices->num_devices--; |
2003 | cur_devices->total_devices--; | |
b4993e64 AJ |
2004 | /* Update total_devices of the parent fs_devices if it's seed */ |
2005 | if (cur_devices != fs_devices) | |
2006 | fs_devices->total_devices--; | |
2b82032c | 2007 | |
e6e674bd | 2008 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2009 | cur_devices->missing_devices--; |
cd02dca5 | 2010 | |
d6507cf1 | 2011 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2012 | |
0bfaa9c5 | 2013 | if (device->bdev) { |
41a52a0f | 2014 | cur_devices->open_devices--; |
0bfaa9c5 | 2015 | /* remove sysfs entry */ |
b5185197 | 2016 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2017 | } |
99994cde | 2018 | |
0b246afa JM |
2019 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2020 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2021 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2022 | |
cea67ab9 JM |
2023 | /* |
2024 | * at this point, the device is zero sized and detached from | |
2025 | * the devices list. All that's left is to zero out the old | |
2026 | * supers and free the device. | |
2027 | */ | |
ebbede42 | 2028 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2029 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2030 | ||
2031 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2032 | synchronize_rcu(); |
2033 | btrfs_free_device(device); | |
cea67ab9 | 2034 | |
1f78160c | 2035 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2036 | while (fs_devices) { |
8321cf25 RS |
2037 | if (fs_devices->seed == cur_devices) { |
2038 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2039 | break; |
8321cf25 | 2040 | } |
e4404d6e | 2041 | fs_devices = fs_devices->seed; |
2b82032c | 2042 | } |
1f78160c | 2043 | cur_devices->seed = NULL; |
0226e0eb | 2044 | close_fs_devices(cur_devices); |
1f78160c | 2045 | free_fs_devices(cur_devices); |
2b82032c YZ |
2046 | } |
2047 | ||
a061fc8d CM |
2048 | out: |
2049 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2050 | return ret; |
24fc572f | 2051 | |
9b3517e9 | 2052 | error_undo: |
ebbede42 | 2053 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2054 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2055 | list_add(&device->dev_alloc_list, |
b5185197 | 2056 | &fs_devices->alloc_list); |
c3929c36 | 2057 | device->fs_devices->rw_devices++; |
34441361 | 2058 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2059 | } |
24fc572f | 2060 | goto out; |
a061fc8d CM |
2061 | } |
2062 | ||
68a9db5f | 2063 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2064 | { |
d51908ce AJ |
2065 | struct btrfs_fs_devices *fs_devices; |
2066 | ||
68a9db5f | 2067 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2068 | |
25e8e911 AJ |
2069 | /* |
2070 | * in case of fs with no seed, srcdev->fs_devices will point | |
2071 | * to fs_devices of fs_info. However when the dev being replaced is | |
2072 | * a seed dev it will point to the seed's local fs_devices. In short | |
2073 | * srcdev will have its correct fs_devices in both the cases. | |
2074 | */ | |
2075 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2076 | |
e93c89c1 | 2077 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2078 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2079 | fs_devices->num_devices--; |
e6e674bd | 2080 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2081 | fs_devices->missing_devices--; |
e93c89c1 | 2082 | |
ebbede42 | 2083 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2084 | fs_devices->rw_devices--; |
1357272f | 2085 | |
82372bc8 | 2086 | if (srcdev->bdev) |
d51908ce | 2087 | fs_devices->open_devices--; |
084b6e7c QW |
2088 | } |
2089 | ||
65237ee3 | 2090 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c | 2091 | { |
65237ee3 | 2092 | struct btrfs_fs_info *fs_info = srcdev->fs_info; |
084b6e7c | 2093 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
e93c89c1 | 2094 | |
ebbede42 | 2095 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2096 | /* zero out the old super if it is writable */ |
2097 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2098 | } | |
14238819 AJ |
2099 | |
2100 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2101 | synchronize_rcu(); |
2102 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2103 | |
94d5f0c2 AJ |
2104 | /* if this is no devs we rather delete the fs_devices */ |
2105 | if (!fs_devices->num_devices) { | |
2106 | struct btrfs_fs_devices *tmp_fs_devices; | |
2107 | ||
6dd38f81 AJ |
2108 | /* |
2109 | * On a mounted FS, num_devices can't be zero unless it's a | |
2110 | * seed. In case of a seed device being replaced, the replace | |
2111 | * target added to the sprout FS, so there will be no more | |
2112 | * device left under the seed FS. | |
2113 | */ | |
2114 | ASSERT(fs_devices->seeding); | |
2115 | ||
94d5f0c2 AJ |
2116 | tmp_fs_devices = fs_info->fs_devices; |
2117 | while (tmp_fs_devices) { | |
2118 | if (tmp_fs_devices->seed == fs_devices) { | |
2119 | tmp_fs_devices->seed = fs_devices->seed; | |
2120 | break; | |
2121 | } | |
2122 | tmp_fs_devices = tmp_fs_devices->seed; | |
2123 | } | |
2124 | fs_devices->seed = NULL; | |
0226e0eb | 2125 | close_fs_devices(fs_devices); |
8bef8401 | 2126 | free_fs_devices(fs_devices); |
94d5f0c2 | 2127 | } |
e93c89c1 SB |
2128 | } |
2129 | ||
4f5ad7bd | 2130 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2131 | { |
4f5ad7bd | 2132 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2133 | |
d9a071f0 | 2134 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2135 | |
d9a071f0 | 2136 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2137 | |
779bf3fe | 2138 | if (tgtdev->bdev) |
d9a071f0 | 2139 | fs_devices->open_devices--; |
779bf3fe | 2140 | |
d9a071f0 | 2141 | fs_devices->num_devices--; |
e93c89c1 | 2142 | |
d6507cf1 | 2143 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2144 | |
e93c89c1 | 2145 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2146 | |
d9a071f0 | 2147 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2148 | |
2149 | /* | |
2150 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2151 | * may lead to a call to btrfs_show_devname() which will try | |
2152 | * to hold device_list_mutex. And here this device | |
2153 | * is already out of device list, so we don't have to hold | |
2154 | * the device_list_mutex lock. | |
2155 | */ | |
2156 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2157 | |
2158 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2159 | synchronize_rcu(); |
2160 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2161 | } |
2162 | ||
b444ad46 NB |
2163 | static struct btrfs_device *btrfs_find_device_by_path( |
2164 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2165 | { |
2166 | int ret = 0; | |
2167 | struct btrfs_super_block *disk_super; | |
2168 | u64 devid; | |
2169 | u8 *dev_uuid; | |
2170 | struct block_device *bdev; | |
2171 | struct buffer_head *bh; | |
b444ad46 | 2172 | struct btrfs_device *device; |
7ba15b7d | 2173 | |
7ba15b7d | 2174 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2175 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2176 | if (ret) |
b444ad46 | 2177 | return ERR_PTR(ret); |
7ba15b7d SB |
2178 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2179 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2180 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2181 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2182 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2183 | disk_super->metadata_uuid, true); |
7239ff4b | 2184 | else |
e4319cd9 | 2185 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2186 | disk_super->fsid, true); |
7239ff4b | 2187 | |
7ba15b7d | 2188 | brelse(bh); |
b444ad46 NB |
2189 | if (!device) |
2190 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2191 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2192 | return device; |
7ba15b7d SB |
2193 | } |
2194 | ||
5c5c0df0 DS |
2195 | /* |
2196 | * Lookup a device given by device id, or the path if the id is 0. | |
2197 | */ | |
a27a94c2 | 2198 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2199 | struct btrfs_fs_info *fs_info, u64 devid, |
2200 | const char *device_path) | |
24e0474b | 2201 | { |
a27a94c2 | 2202 | struct btrfs_device *device; |
24e0474b | 2203 | |
5c5c0df0 | 2204 | if (devid) { |
e4319cd9 | 2205 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2206 | NULL, true); |
a27a94c2 NB |
2207 | if (!device) |
2208 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2209 | return device; |
2210 | } | |
2211 | ||
2212 | if (!device_path || !device_path[0]) | |
2213 | return ERR_PTR(-EINVAL); | |
2214 | ||
2215 | if (strcmp(device_path, "missing") == 0) { | |
2216 | /* Find first missing device */ | |
2217 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2218 | dev_list) { | |
2219 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2220 | &device->dev_state) && !device->bdev) | |
2221 | return device; | |
d95a830c | 2222 | } |
6e927ceb | 2223 | return ERR_PTR(-ENOENT); |
24e0474b | 2224 | } |
6e927ceb AJ |
2225 | |
2226 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2227 | } |
2228 | ||
2b82032c YZ |
2229 | /* |
2230 | * does all the dirty work required for changing file system's UUID. | |
2231 | */ | |
2ff7e61e | 2232 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2233 | { |
0b246afa | 2234 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2235 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2236 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2237 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2238 | struct btrfs_device *device; |
2239 | u64 super_flags; | |
2240 | ||
a32bf9a3 | 2241 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2242 | if (!fs_devices->seeding) |
2b82032c YZ |
2243 | return -EINVAL; |
2244 | ||
7239ff4b | 2245 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2246 | if (IS_ERR(seed_devices)) |
2247 | return PTR_ERR(seed_devices); | |
2b82032c | 2248 | |
e4404d6e YZ |
2249 | old_devices = clone_fs_devices(fs_devices); |
2250 | if (IS_ERR(old_devices)) { | |
2251 | kfree(seed_devices); | |
2252 | return PTR_ERR(old_devices); | |
2b82032c | 2253 | } |
e4404d6e | 2254 | |
c4babc5e | 2255 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2256 | |
e4404d6e YZ |
2257 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2258 | seed_devices->opened = 1; | |
2259 | INIT_LIST_HEAD(&seed_devices->devices); | |
2260 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2261 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2262 | |
321a4bf7 | 2263 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2264 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2265 | synchronize_rcu); | |
2196d6e8 MX |
2266 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2267 | device->fs_devices = seed_devices; | |
c9513edb | 2268 | |
34441361 | 2269 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2270 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2271 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2272 | |
0395d84f | 2273 | fs_devices->seeding = false; |
2b82032c YZ |
2274 | fs_devices->num_devices = 0; |
2275 | fs_devices->open_devices = 0; | |
69611ac8 | 2276 | fs_devices->missing_devices = 0; |
7f0432d0 | 2277 | fs_devices->rotating = false; |
e4404d6e | 2278 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2279 | |
2280 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2281 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2282 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2283 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2284 | |
2b82032c YZ |
2285 | super_flags = btrfs_super_flags(disk_super) & |
2286 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2287 | btrfs_set_super_flags(disk_super, super_flags); | |
2288 | ||
2289 | return 0; | |
2290 | } | |
2291 | ||
2292 | /* | |
01327610 | 2293 | * Store the expected generation for seed devices in device items. |
2b82032c | 2294 | */ |
5c466629 | 2295 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2296 | { |
5c466629 | 2297 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2298 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2299 | struct btrfs_path *path; |
2300 | struct extent_buffer *leaf; | |
2301 | struct btrfs_dev_item *dev_item; | |
2302 | struct btrfs_device *device; | |
2303 | struct btrfs_key key; | |
44880fdc | 2304 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2305 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2306 | u64 devid; | |
2307 | int ret; | |
2308 | ||
2309 | path = btrfs_alloc_path(); | |
2310 | if (!path) | |
2311 | return -ENOMEM; | |
2312 | ||
2b82032c YZ |
2313 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2314 | key.offset = 0; | |
2315 | key.type = BTRFS_DEV_ITEM_KEY; | |
2316 | ||
2317 | while (1) { | |
2318 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2319 | if (ret < 0) | |
2320 | goto error; | |
2321 | ||
2322 | leaf = path->nodes[0]; | |
2323 | next_slot: | |
2324 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2325 | ret = btrfs_next_leaf(root, path); | |
2326 | if (ret > 0) | |
2327 | break; | |
2328 | if (ret < 0) | |
2329 | goto error; | |
2330 | leaf = path->nodes[0]; | |
2331 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2332 | btrfs_release_path(path); |
2b82032c YZ |
2333 | continue; |
2334 | } | |
2335 | ||
2336 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2337 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2338 | key.type != BTRFS_DEV_ITEM_KEY) | |
2339 | break; | |
2340 | ||
2341 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2342 | struct btrfs_dev_item); | |
2343 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2344 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2345 | BTRFS_UUID_SIZE); |
1473b24e | 2346 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2347 | BTRFS_FSID_SIZE); |
e4319cd9 | 2348 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2349 | fs_uuid, true); |
79787eaa | 2350 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2351 | |
2352 | if (device->fs_devices->seeding) { | |
2353 | btrfs_set_device_generation(leaf, dev_item, | |
2354 | device->generation); | |
2355 | btrfs_mark_buffer_dirty(leaf); | |
2356 | } | |
2357 | ||
2358 | path->slots[0]++; | |
2359 | goto next_slot; | |
2360 | } | |
2361 | ret = 0; | |
2362 | error: | |
2363 | btrfs_free_path(path); | |
2364 | return ret; | |
2365 | } | |
2366 | ||
da353f6b | 2367 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2368 | { |
5112febb | 2369 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2370 | struct request_queue *q; |
788f20eb CM |
2371 | struct btrfs_trans_handle *trans; |
2372 | struct btrfs_device *device; | |
2373 | struct block_device *bdev; | |
0b246afa | 2374 | struct super_block *sb = fs_info->sb; |
606686ee | 2375 | struct rcu_string *name; |
5da54bc1 | 2376 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2377 | u64 orig_super_total_bytes; |
2378 | u64 orig_super_num_devices; | |
2b82032c | 2379 | int seeding_dev = 0; |
788f20eb | 2380 | int ret = 0; |
7132a262 | 2381 | bool unlocked = false; |
788f20eb | 2382 | |
5da54bc1 | 2383 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2384 | return -EROFS; |
788f20eb | 2385 | |
a5d16333 | 2386 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2387 | fs_info->bdev_holder); |
7f59203a JB |
2388 | if (IS_ERR(bdev)) |
2389 | return PTR_ERR(bdev); | |
a2135011 | 2390 | |
5da54bc1 | 2391 | if (fs_devices->seeding) { |
2b82032c YZ |
2392 | seeding_dev = 1; |
2393 | down_write(&sb->s_umount); | |
2394 | mutex_lock(&uuid_mutex); | |
2395 | } | |
2396 | ||
8c8bee1d | 2397 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2398 | |
5da54bc1 | 2399 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2400 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2401 | if (device->bdev == bdev) { |
2402 | ret = -EEXIST; | |
d25628bd | 2403 | mutex_unlock( |
5da54bc1 | 2404 | &fs_devices->device_list_mutex); |
2b82032c | 2405 | goto error; |
788f20eb CM |
2406 | } |
2407 | } | |
5da54bc1 | 2408 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2409 | |
0b246afa | 2410 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2411 | if (IS_ERR(device)) { |
788f20eb | 2412 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2413 | ret = PTR_ERR(device); |
2b82032c | 2414 | goto error; |
788f20eb CM |
2415 | } |
2416 | ||
78f2c9e6 | 2417 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2418 | if (!name) { |
2b82032c | 2419 | ret = -ENOMEM; |
5c4cf6c9 | 2420 | goto error_free_device; |
788f20eb | 2421 | } |
606686ee | 2422 | rcu_assign_pointer(device->name, name); |
2b82032c | 2423 | |
a22285a6 | 2424 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2425 | if (IS_ERR(trans)) { |
98d5dc13 | 2426 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2427 | goto error_free_device; |
98d5dc13 TI |
2428 | } |
2429 | ||
d5e2003c | 2430 | q = bdev_get_queue(bdev); |
ebbede42 | 2431 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2432 | device->generation = trans->transid; |
0b246afa JM |
2433 | device->io_width = fs_info->sectorsize; |
2434 | device->io_align = fs_info->sectorsize; | |
2435 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2436 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2437 | fs_info->sectorsize); | |
2cc3c559 | 2438 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2439 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2440 | device->fs_info = fs_info; |
788f20eb | 2441 | device->bdev = bdev; |
e12c9621 | 2442 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2443 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2444 | device->mode = FMODE_EXCL; |
27087f37 | 2445 | device->dev_stats_valid = 1; |
9f6d2510 | 2446 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2447 | |
2b82032c | 2448 | if (seeding_dev) { |
1751e8a6 | 2449 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2450 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2451 | if (ret) { |
2452 | btrfs_abort_transaction(trans, ret); | |
2453 | goto error_trans; | |
2454 | } | |
2b82032c | 2455 | } |
788f20eb | 2456 | |
5da54bc1 | 2457 | device->fs_devices = fs_devices; |
e5e9a520 | 2458 | |
5da54bc1 | 2459 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2460 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2461 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2462 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2463 | fs_devices->num_devices++; | |
2464 | fs_devices->open_devices++; | |
2465 | fs_devices->rw_devices++; | |
2466 | fs_devices->total_devices++; | |
2467 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2468 | |
a5ed45f8 | 2469 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2470 | |
e884f4f0 | 2471 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2472 | fs_devices->rotating = true; |
c289811c | 2473 | |
39379faa | 2474 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2475 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2476 | round_down(orig_super_total_bytes + device->total_bytes, |
2477 | fs_info->sectorsize)); | |
788f20eb | 2478 | |
39379faa NA |
2479 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2480 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2481 | orig_super_num_devices + 1); | |
0d39376a AJ |
2482 | |
2483 | /* add sysfs device entry */ | |
5da54bc1 | 2484 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2485 | |
2196d6e8 MX |
2486 | /* |
2487 | * we've got more storage, clear any full flags on the space | |
2488 | * infos | |
2489 | */ | |
0b246afa | 2490 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2491 | |
34441361 | 2492 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2493 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2494 | |
2b82032c | 2495 | if (seeding_dev) { |
34441361 | 2496 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2497 | ret = init_first_rw_device(trans); |
34441361 | 2498 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2499 | if (ret) { |
66642832 | 2500 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2501 | goto error_sysfs; |
005d6427 | 2502 | } |
2196d6e8 MX |
2503 | } |
2504 | ||
8e87e856 | 2505 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2506 | if (ret) { |
66642832 | 2507 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2508 | goto error_sysfs; |
2196d6e8 MX |
2509 | } |
2510 | ||
2511 | if (seeding_dev) { | |
5c466629 | 2512 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2513 | if (ret) { |
66642832 | 2514 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2515 | goto error_sysfs; |
005d6427 | 2516 | } |
b2373f25 | 2517 | |
f93c3997 DS |
2518 | btrfs_sysfs_update_sprout_fsid(fs_devices, |
2519 | fs_info->fs_devices->fsid); | |
2b82032c YZ |
2520 | } |
2521 | ||
3a45bb20 | 2522 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2523 | |
2b82032c YZ |
2524 | if (seeding_dev) { |
2525 | mutex_unlock(&uuid_mutex); | |
2526 | up_write(&sb->s_umount); | |
7132a262 | 2527 | unlocked = true; |
788f20eb | 2528 | |
79787eaa JM |
2529 | if (ret) /* transaction commit */ |
2530 | return ret; | |
2531 | ||
2ff7e61e | 2532 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2533 | if (ret < 0) |
0b246afa | 2534 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2535 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2536 | trans = btrfs_attach_transaction(root); |
2537 | if (IS_ERR(trans)) { | |
2538 | if (PTR_ERR(trans) == -ENOENT) | |
2539 | return 0; | |
7132a262 AJ |
2540 | ret = PTR_ERR(trans); |
2541 | trans = NULL; | |
2542 | goto error_sysfs; | |
671415b7 | 2543 | } |
3a45bb20 | 2544 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2545 | } |
c9e9f97b | 2546 | |
5a1972bd QW |
2547 | /* Update ctime/mtime for libblkid */ |
2548 | update_dev_time(device_path); | |
2b82032c | 2549 | return ret; |
79787eaa | 2550 | |
d31c32f6 | 2551 | error_sysfs: |
5da54bc1 | 2552 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2553 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2554 | mutex_lock(&fs_info->chunk_mutex); | |
2555 | list_del_rcu(&device->dev_list); | |
2556 | list_del(&device->dev_alloc_list); | |
2557 | fs_info->fs_devices->num_devices--; | |
2558 | fs_info->fs_devices->open_devices--; | |
2559 | fs_info->fs_devices->rw_devices--; | |
2560 | fs_info->fs_devices->total_devices--; | |
2561 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2562 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2563 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2564 | orig_super_total_bytes); | |
2565 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2566 | orig_super_num_devices); | |
2567 | mutex_unlock(&fs_info->chunk_mutex); | |
2568 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2569 | error_trans: |
0af2c4bf | 2570 | if (seeding_dev) |
1751e8a6 | 2571 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2572 | if (trans) |
2573 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2574 | error_free_device: |
a425f9d4 | 2575 | btrfs_free_device(device); |
2b82032c | 2576 | error: |
e525fd89 | 2577 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2578 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2579 | mutex_unlock(&uuid_mutex); |
2580 | up_write(&sb->s_umount); | |
2581 | } | |
c9e9f97b | 2582 | return ret; |
788f20eb CM |
2583 | } |
2584 | ||
d397712b CM |
2585 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2586 | struct btrfs_device *device) | |
0b86a832 CM |
2587 | { |
2588 | int ret; | |
2589 | struct btrfs_path *path; | |
0b246afa | 2590 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2591 | struct btrfs_dev_item *dev_item; |
2592 | struct extent_buffer *leaf; | |
2593 | struct btrfs_key key; | |
2594 | ||
0b86a832 CM |
2595 | path = btrfs_alloc_path(); |
2596 | if (!path) | |
2597 | return -ENOMEM; | |
2598 | ||
2599 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2600 | key.type = BTRFS_DEV_ITEM_KEY; | |
2601 | key.offset = device->devid; | |
2602 | ||
2603 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2604 | if (ret < 0) | |
2605 | goto out; | |
2606 | ||
2607 | if (ret > 0) { | |
2608 | ret = -ENOENT; | |
2609 | goto out; | |
2610 | } | |
2611 | ||
2612 | leaf = path->nodes[0]; | |
2613 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2614 | ||
2615 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2616 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2617 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2618 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2619 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2620 | btrfs_set_device_total_bytes(leaf, dev_item, |
2621 | btrfs_device_get_disk_total_bytes(device)); | |
2622 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2623 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2624 | btrfs_mark_buffer_dirty(leaf); |
2625 | ||
2626 | out: | |
2627 | btrfs_free_path(path); | |
2628 | return ret; | |
2629 | } | |
2630 | ||
2196d6e8 | 2631 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2632 | struct btrfs_device *device, u64 new_size) |
2633 | { | |
0b246afa JM |
2634 | struct btrfs_fs_info *fs_info = device->fs_info; |
2635 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2636 | u64 old_total; |
2637 | u64 diff; | |
8f18cf13 | 2638 | |
ebbede42 | 2639 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2640 | return -EACCES; |
2196d6e8 | 2641 | |
7dfb8be1 NB |
2642 | new_size = round_down(new_size, fs_info->sectorsize); |
2643 | ||
34441361 | 2644 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2645 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2646 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2647 | |
63a212ab | 2648 | if (new_size <= device->total_bytes || |
401e29c1 | 2649 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2650 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2651 | return -EINVAL; |
2196d6e8 | 2652 | } |
2b82032c | 2653 | |
7dfb8be1 NB |
2654 | btrfs_set_super_total_bytes(super_copy, |
2655 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2656 | device->fs_devices->total_rw_bytes += diff; |
2657 | ||
7cc8e58d MX |
2658 | btrfs_device_set_total_bytes(device, new_size); |
2659 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2660 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2661 | if (list_empty(&device->post_commit_list)) |
2662 | list_add_tail(&device->post_commit_list, | |
2663 | &trans->transaction->dev_update_list); | |
34441361 | 2664 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2665 | |
8f18cf13 CM |
2666 | return btrfs_update_device(trans, device); |
2667 | } | |
2668 | ||
f4208794 | 2669 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2670 | { |
f4208794 | 2671 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2672 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2673 | int ret; |
2674 | struct btrfs_path *path; | |
2675 | struct btrfs_key key; | |
2676 | ||
8f18cf13 CM |
2677 | path = btrfs_alloc_path(); |
2678 | if (!path) | |
2679 | return -ENOMEM; | |
2680 | ||
408fbf19 | 2681 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2682 | key.offset = chunk_offset; |
2683 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2684 | ||
2685 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2686 | if (ret < 0) |
2687 | goto out; | |
2688 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2689 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2690 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2691 | ret = -ENOENT; |
2692 | goto out; | |
2693 | } | |
8f18cf13 CM |
2694 | |
2695 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2696 | if (ret < 0) |
0b246afa JM |
2697 | btrfs_handle_fs_error(fs_info, ret, |
2698 | "Failed to delete chunk item."); | |
79787eaa | 2699 | out: |
8f18cf13 | 2700 | btrfs_free_path(path); |
65a246c5 | 2701 | return ret; |
8f18cf13 CM |
2702 | } |
2703 | ||
408fbf19 | 2704 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2705 | { |
0b246afa | 2706 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2707 | struct btrfs_disk_key *disk_key; |
2708 | struct btrfs_chunk *chunk; | |
2709 | u8 *ptr; | |
2710 | int ret = 0; | |
2711 | u32 num_stripes; | |
2712 | u32 array_size; | |
2713 | u32 len = 0; | |
2714 | u32 cur; | |
2715 | struct btrfs_key key; | |
2716 | ||
34441361 | 2717 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2718 | array_size = btrfs_super_sys_array_size(super_copy); |
2719 | ||
2720 | ptr = super_copy->sys_chunk_array; | |
2721 | cur = 0; | |
2722 | ||
2723 | while (cur < array_size) { | |
2724 | disk_key = (struct btrfs_disk_key *)ptr; | |
2725 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2726 | ||
2727 | len = sizeof(*disk_key); | |
2728 | ||
2729 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2730 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2731 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2732 | len += btrfs_chunk_item_size(num_stripes); | |
2733 | } else { | |
2734 | ret = -EIO; | |
2735 | break; | |
2736 | } | |
408fbf19 | 2737 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2738 | key.offset == chunk_offset) { |
2739 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2740 | array_size -= len; | |
2741 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2742 | } else { | |
2743 | ptr += len; | |
2744 | cur += len; | |
2745 | } | |
2746 | } | |
34441361 | 2747 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2748 | return ret; |
2749 | } | |
2750 | ||
60ca842e OS |
2751 | /* |
2752 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2753 | * @logical: Logical block offset in bytes. | |
2754 | * @length: Length of extent in bytes. | |
2755 | * | |
2756 | * Return: Chunk mapping or ERR_PTR. | |
2757 | */ | |
2758 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2759 | u64 logical, u64 length) | |
592d92ee LB |
2760 | { |
2761 | struct extent_map_tree *em_tree; | |
2762 | struct extent_map *em; | |
2763 | ||
c8bf1b67 | 2764 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2765 | read_lock(&em_tree->lock); |
2766 | em = lookup_extent_mapping(em_tree, logical, length); | |
2767 | read_unlock(&em_tree->lock); | |
2768 | ||
2769 | if (!em) { | |
2770 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2771 | logical, length); | |
2772 | return ERR_PTR(-EINVAL); | |
2773 | } | |
2774 | ||
2775 | if (em->start > logical || em->start + em->len < logical) { | |
2776 | btrfs_crit(fs_info, | |
2777 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2778 | logical, length, em->start, em->start + em->len); | |
2779 | free_extent_map(em); | |
2780 | return ERR_PTR(-EINVAL); | |
2781 | } | |
2782 | ||
2783 | /* callers are responsible for dropping em's ref. */ | |
2784 | return em; | |
2785 | } | |
2786 | ||
97aff912 | 2787 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2788 | { |
97aff912 | 2789 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2790 | struct extent_map *em; |
2791 | struct map_lookup *map; | |
2196d6e8 | 2792 | u64 dev_extent_len = 0; |
47ab2a6c | 2793 | int i, ret = 0; |
0b246afa | 2794 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2795 | |
60ca842e | 2796 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2797 | if (IS_ERR(em)) { |
47ab2a6c JB |
2798 | /* |
2799 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2800 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2801 | * do anything we still error out. |
2802 | */ | |
2803 | ASSERT(0); | |
592d92ee | 2804 | return PTR_ERR(em); |
47ab2a6c | 2805 | } |
95617d69 | 2806 | map = em->map_lookup; |
34441361 | 2807 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2808 | check_system_chunk(trans, map->type); |
34441361 | 2809 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2810 | |
57ba4cb8 FM |
2811 | /* |
2812 | * Take the device list mutex to prevent races with the final phase of | |
2813 | * a device replace operation that replaces the device object associated | |
2814 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2815 | */ | |
2816 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2817 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2818 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2819 | ret = btrfs_free_dev_extent(trans, device, |
2820 | map->stripes[i].physical, | |
2821 | &dev_extent_len); | |
47ab2a6c | 2822 | if (ret) { |
57ba4cb8 | 2823 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2824 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2825 | goto out; |
2826 | } | |
a061fc8d | 2827 | |
2196d6e8 | 2828 | if (device->bytes_used > 0) { |
34441361 | 2829 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2830 | btrfs_device_set_bytes_used(device, |
2831 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2832 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2833 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2834 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2835 | } |
a061fc8d | 2836 | |
64bc6c2a NB |
2837 | ret = btrfs_update_device(trans, device); |
2838 | if (ret) { | |
2839 | mutex_unlock(&fs_devices->device_list_mutex); | |
2840 | btrfs_abort_transaction(trans, ret); | |
2841 | goto out; | |
dfe25020 | 2842 | } |
8f18cf13 | 2843 | } |
57ba4cb8 FM |
2844 | mutex_unlock(&fs_devices->device_list_mutex); |
2845 | ||
f4208794 | 2846 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2847 | if (ret) { |
66642832 | 2848 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2849 | goto out; |
2850 | } | |
8f18cf13 | 2851 | |
6bccf3ab | 2852 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2853 | |
8f18cf13 | 2854 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2855 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2856 | if (ret) { |
66642832 | 2857 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2858 | goto out; |
2859 | } | |
8f18cf13 CM |
2860 | } |
2861 | ||
5a98ec01 | 2862 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2863 | if (ret) { |
66642832 | 2864 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2865 | goto out; |
2866 | } | |
2b82032c | 2867 | |
47ab2a6c | 2868 | out: |
2b82032c YZ |
2869 | /* once for us */ |
2870 | free_extent_map(em); | |
47ab2a6c JB |
2871 | return ret; |
2872 | } | |
2b82032c | 2873 | |
5b4aacef | 2874 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2875 | { |
5b4aacef | 2876 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2877 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2878 | int ret; |
2b82032c | 2879 | |
67c5e7d4 FM |
2880 | /* |
2881 | * Prevent races with automatic removal of unused block groups. | |
2882 | * After we relocate and before we remove the chunk with offset | |
2883 | * chunk_offset, automatic removal of the block group can kick in, | |
2884 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2885 | * | |
2886 | * Make sure to acquire this mutex before doing a tree search (dev | |
2887 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2888 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2889 | * we release the path used to search the chunk/dev tree and before | |
2890 | * the current task acquires this mutex and calls us. | |
2891 | */ | |
a32bf9a3 | 2892 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2893 | |
47ab2a6c | 2894 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 2895 | btrfs_scrub_pause(fs_info); |
0b246afa | 2896 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2897 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2898 | if (ret) |
2899 | return ret; | |
2900 | ||
19c4d2f9 CM |
2901 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2902 | chunk_offset); | |
2903 | if (IS_ERR(trans)) { | |
2904 | ret = PTR_ERR(trans); | |
2905 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2906 | return ret; | |
2907 | } | |
2908 | ||
47ab2a6c | 2909 | /* |
19c4d2f9 CM |
2910 | * step two, delete the device extents and the |
2911 | * chunk tree entries | |
47ab2a6c | 2912 | */ |
97aff912 | 2913 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 2914 | btrfs_end_transaction(trans); |
19c4d2f9 | 2915 | return ret; |
2b82032c YZ |
2916 | } |
2917 | ||
2ff7e61e | 2918 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2919 | { |
0b246afa | 2920 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2921 | struct btrfs_path *path; |
2922 | struct extent_buffer *leaf; | |
2923 | struct btrfs_chunk *chunk; | |
2924 | struct btrfs_key key; | |
2925 | struct btrfs_key found_key; | |
2b82032c | 2926 | u64 chunk_type; |
ba1bf481 JB |
2927 | bool retried = false; |
2928 | int failed = 0; | |
2b82032c YZ |
2929 | int ret; |
2930 | ||
2931 | path = btrfs_alloc_path(); | |
2932 | if (!path) | |
2933 | return -ENOMEM; | |
2934 | ||
ba1bf481 | 2935 | again: |
2b82032c YZ |
2936 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2937 | key.offset = (u64)-1; | |
2938 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2939 | ||
2940 | while (1) { | |
0b246afa | 2941 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2942 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2943 | if (ret < 0) { |
0b246afa | 2944 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2945 | goto error; |
67c5e7d4 | 2946 | } |
79787eaa | 2947 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2948 | |
2949 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2950 | key.type); | |
67c5e7d4 | 2951 | if (ret) |
0b246afa | 2952 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
2953 | if (ret < 0) |
2954 | goto error; | |
2955 | if (ret > 0) | |
2956 | break; | |
1a40e23b | 2957 | |
2b82032c YZ |
2958 | leaf = path->nodes[0]; |
2959 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2960 | |
2b82032c YZ |
2961 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2962 | struct btrfs_chunk); | |
2963 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2964 | btrfs_release_path(path); |
8f18cf13 | 2965 | |
2b82032c | 2966 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 2967 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
2968 | if (ret == -ENOSPC) |
2969 | failed++; | |
14586651 HS |
2970 | else |
2971 | BUG_ON(ret); | |
2b82032c | 2972 | } |
0b246afa | 2973 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 2974 | |
2b82032c YZ |
2975 | if (found_key.offset == 0) |
2976 | break; | |
2977 | key.offset = found_key.offset - 1; | |
2978 | } | |
2979 | ret = 0; | |
ba1bf481 JB |
2980 | if (failed && !retried) { |
2981 | failed = 0; | |
2982 | retried = true; | |
2983 | goto again; | |
fae7f21c | 2984 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2985 | ret = -ENOSPC; |
2986 | } | |
2b82032c YZ |
2987 | error: |
2988 | btrfs_free_path(path); | |
2989 | return ret; | |
8f18cf13 CM |
2990 | } |
2991 | ||
a6f93c71 LB |
2992 | /* |
2993 | * return 1 : allocate a data chunk successfully, | |
2994 | * return <0: errors during allocating a data chunk, | |
2995 | * return 0 : no need to allocate a data chunk. | |
2996 | */ | |
2997 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
2998 | u64 chunk_offset) | |
2999 | { | |
32da5386 | 3000 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3001 | u64 bytes_used; |
3002 | u64 chunk_type; | |
3003 | ||
3004 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3005 | ASSERT(cache); | |
3006 | chunk_type = cache->flags; | |
3007 | btrfs_put_block_group(cache); | |
3008 | ||
5ae21692 JT |
3009 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3010 | return 0; | |
3011 | ||
3012 | spin_lock(&fs_info->data_sinfo->lock); | |
3013 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3014 | spin_unlock(&fs_info->data_sinfo->lock); | |
3015 | ||
3016 | if (!bytes_used) { | |
3017 | struct btrfs_trans_handle *trans; | |
3018 | int ret; | |
3019 | ||
3020 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3021 | if (IS_ERR(trans)) | |
3022 | return PTR_ERR(trans); | |
3023 | ||
3024 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3025 | btrfs_end_transaction(trans); | |
3026 | if (ret < 0) | |
3027 | return ret; | |
3028 | return 1; | |
a6f93c71 | 3029 | } |
5ae21692 | 3030 | |
a6f93c71 LB |
3031 | return 0; |
3032 | } | |
3033 | ||
6bccf3ab | 3034 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3035 | struct btrfs_balance_control *bctl) |
3036 | { | |
6bccf3ab | 3037 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3038 | struct btrfs_trans_handle *trans; |
3039 | struct btrfs_balance_item *item; | |
3040 | struct btrfs_disk_balance_args disk_bargs; | |
3041 | struct btrfs_path *path; | |
3042 | struct extent_buffer *leaf; | |
3043 | struct btrfs_key key; | |
3044 | int ret, err; | |
3045 | ||
3046 | path = btrfs_alloc_path(); | |
3047 | if (!path) | |
3048 | return -ENOMEM; | |
3049 | ||
3050 | trans = btrfs_start_transaction(root, 0); | |
3051 | if (IS_ERR(trans)) { | |
3052 | btrfs_free_path(path); | |
3053 | return PTR_ERR(trans); | |
3054 | } | |
3055 | ||
3056 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3057 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3058 | key.offset = 0; |
3059 | ||
3060 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3061 | sizeof(*item)); | |
3062 | if (ret) | |
3063 | goto out; | |
3064 | ||
3065 | leaf = path->nodes[0]; | |
3066 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3067 | ||
b159fa28 | 3068 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3069 | |
3070 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3071 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3072 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3073 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3074 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3075 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3076 | ||
3077 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3078 | ||
3079 | btrfs_mark_buffer_dirty(leaf); | |
3080 | out: | |
3081 | btrfs_free_path(path); | |
3a45bb20 | 3082 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3083 | if (err && !ret) |
3084 | ret = err; | |
3085 | return ret; | |
3086 | } | |
3087 | ||
6bccf3ab | 3088 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3089 | { |
6bccf3ab | 3090 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3091 | struct btrfs_trans_handle *trans; |
3092 | struct btrfs_path *path; | |
3093 | struct btrfs_key key; | |
3094 | int ret, err; | |
3095 | ||
3096 | path = btrfs_alloc_path(); | |
3097 | if (!path) | |
3098 | return -ENOMEM; | |
3099 | ||
3100 | trans = btrfs_start_transaction(root, 0); | |
3101 | if (IS_ERR(trans)) { | |
3102 | btrfs_free_path(path); | |
3103 | return PTR_ERR(trans); | |
3104 | } | |
3105 | ||
3106 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3107 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3108 | key.offset = 0; |
3109 | ||
3110 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3111 | if (ret < 0) | |
3112 | goto out; | |
3113 | if (ret > 0) { | |
3114 | ret = -ENOENT; | |
3115 | goto out; | |
3116 | } | |
3117 | ||
3118 | ret = btrfs_del_item(trans, root, path); | |
3119 | out: | |
3120 | btrfs_free_path(path); | |
3a45bb20 | 3121 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3122 | if (err && !ret) |
3123 | ret = err; | |
3124 | return ret; | |
3125 | } | |
3126 | ||
59641015 ID |
3127 | /* |
3128 | * This is a heuristic used to reduce the number of chunks balanced on | |
3129 | * resume after balance was interrupted. | |
3130 | */ | |
3131 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3132 | { | |
3133 | /* | |
3134 | * Turn on soft mode for chunk types that were being converted. | |
3135 | */ | |
3136 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3137 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3138 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3139 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3140 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3141 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3142 | ||
3143 | /* | |
3144 | * Turn on usage filter if is not already used. The idea is | |
3145 | * that chunks that we have already balanced should be | |
3146 | * reasonably full. Don't do it for chunks that are being | |
3147 | * converted - that will keep us from relocating unconverted | |
3148 | * (albeit full) chunks. | |
3149 | */ | |
3150 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3151 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3152 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3153 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3154 | bctl->data.usage = 90; | |
3155 | } | |
3156 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3157 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3158 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3159 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3160 | bctl->sys.usage = 90; | |
3161 | } | |
3162 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3163 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3164 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3165 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3166 | bctl->meta.usage = 90; | |
3167 | } | |
3168 | } | |
3169 | ||
149196a2 DS |
3170 | /* |
3171 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3172 | */ | |
3173 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3174 | { |
3175 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3176 | int ret; |
c9e9f97b ID |
3177 | |
3178 | BUG_ON(!fs_info->balance_ctl); | |
3179 | ||
3180 | spin_lock(&fs_info->balance_lock); | |
3181 | fs_info->balance_ctl = NULL; | |
3182 | spin_unlock(&fs_info->balance_lock); | |
3183 | ||
3184 | kfree(bctl); | |
149196a2 DS |
3185 | ret = del_balance_item(fs_info); |
3186 | if (ret) | |
3187 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3188 | } |
3189 | ||
ed25e9b2 ID |
3190 | /* |
3191 | * Balance filters. Return 1 if chunk should be filtered out | |
3192 | * (should not be balanced). | |
3193 | */ | |
899c81ea | 3194 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3195 | struct btrfs_balance_args *bargs) |
3196 | { | |
899c81ea ID |
3197 | chunk_type = chunk_to_extended(chunk_type) & |
3198 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3199 | |
899c81ea | 3200 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3201 | return 0; |
3202 | ||
3203 | return 1; | |
3204 | } | |
3205 | ||
dba72cb3 | 3206 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3207 | struct btrfs_balance_args *bargs) |
bc309467 | 3208 | { |
32da5386 | 3209 | struct btrfs_block_group *cache; |
bc309467 DS |
3210 | u64 chunk_used; |
3211 | u64 user_thresh_min; | |
3212 | u64 user_thresh_max; | |
3213 | int ret = 1; | |
3214 | ||
3215 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3216 | chunk_used = cache->used; |
bc309467 DS |
3217 | |
3218 | if (bargs->usage_min == 0) | |
3219 | user_thresh_min = 0; | |
3220 | else | |
b3470b5d DS |
3221 | user_thresh_min = div_factor_fine(cache->length, |
3222 | bargs->usage_min); | |
bc309467 DS |
3223 | |
3224 | if (bargs->usage_max == 0) | |
3225 | user_thresh_max = 1; | |
3226 | else if (bargs->usage_max > 100) | |
b3470b5d | 3227 | user_thresh_max = cache->length; |
bc309467 | 3228 | else |
b3470b5d DS |
3229 | user_thresh_max = div_factor_fine(cache->length, |
3230 | bargs->usage_max); | |
bc309467 DS |
3231 | |
3232 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3233 | ret = 0; | |
3234 | ||
3235 | btrfs_put_block_group(cache); | |
3236 | return ret; | |
3237 | } | |
3238 | ||
dba72cb3 | 3239 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3240 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3241 | { |
32da5386 | 3242 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3243 | u64 chunk_used, user_thresh; |
3244 | int ret = 1; | |
3245 | ||
3246 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3247 | chunk_used = cache->used; |
5ce5b3c0 | 3248 | |
bc309467 | 3249 | if (bargs->usage_min == 0) |
3e39cea6 | 3250 | user_thresh = 1; |
a105bb88 | 3251 | else if (bargs->usage > 100) |
b3470b5d | 3252 | user_thresh = cache->length; |
a105bb88 | 3253 | else |
b3470b5d | 3254 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3255 | |
5ce5b3c0 ID |
3256 | if (chunk_used < user_thresh) |
3257 | ret = 0; | |
3258 | ||
3259 | btrfs_put_block_group(cache); | |
3260 | return ret; | |
3261 | } | |
3262 | ||
409d404b ID |
3263 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3264 | struct btrfs_chunk *chunk, | |
3265 | struct btrfs_balance_args *bargs) | |
3266 | { | |
3267 | struct btrfs_stripe *stripe; | |
3268 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3269 | int i; | |
3270 | ||
3271 | for (i = 0; i < num_stripes; i++) { | |
3272 | stripe = btrfs_stripe_nr(chunk, i); | |
3273 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3274 | return 0; | |
3275 | } | |
3276 | ||
3277 | return 1; | |
3278 | } | |
3279 | ||
946c9256 DS |
3280 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3281 | { | |
3282 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3283 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3284 | const int nparity = btrfs_raid_array[index].nparity; | |
3285 | ||
3286 | if (nparity) | |
3287 | return num_stripes - nparity; | |
3288 | else | |
3289 | return num_stripes / ncopies; | |
3290 | } | |
3291 | ||
94e60d5a ID |
3292 | /* [pstart, pend) */ |
3293 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3294 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3295 | struct btrfs_balance_args *bargs) |
3296 | { | |
3297 | struct btrfs_stripe *stripe; | |
3298 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3299 | u64 stripe_offset; | |
3300 | u64 stripe_length; | |
946c9256 | 3301 | u64 type; |
94e60d5a ID |
3302 | int factor; |
3303 | int i; | |
3304 | ||
3305 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3306 | return 0; | |
3307 | ||
946c9256 DS |
3308 | type = btrfs_chunk_type(leaf, chunk); |
3309 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3310 | |
3311 | for (i = 0; i < num_stripes; i++) { | |
3312 | stripe = btrfs_stripe_nr(chunk, i); | |
3313 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3314 | continue; | |
3315 | ||
3316 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3317 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3318 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3319 | |
3320 | if (stripe_offset < bargs->pend && | |
3321 | stripe_offset + stripe_length > bargs->pstart) | |
3322 | return 0; | |
3323 | } | |
3324 | ||
3325 | return 1; | |
3326 | } | |
3327 | ||
ea67176a ID |
3328 | /* [vstart, vend) */ |
3329 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3330 | struct btrfs_chunk *chunk, | |
3331 | u64 chunk_offset, | |
3332 | struct btrfs_balance_args *bargs) | |
3333 | { | |
3334 | if (chunk_offset < bargs->vend && | |
3335 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3336 | /* at least part of the chunk is inside this vrange */ | |
3337 | return 0; | |
3338 | ||
3339 | return 1; | |
3340 | } | |
3341 | ||
dee32d0a GAP |
3342 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3343 | struct btrfs_chunk *chunk, | |
3344 | struct btrfs_balance_args *bargs) | |
3345 | { | |
3346 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3347 | ||
3348 | if (bargs->stripes_min <= num_stripes | |
3349 | && num_stripes <= bargs->stripes_max) | |
3350 | return 0; | |
3351 | ||
3352 | return 1; | |
3353 | } | |
3354 | ||
899c81ea | 3355 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3356 | struct btrfs_balance_args *bargs) |
3357 | { | |
3358 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3359 | return 0; | |
3360 | ||
899c81ea ID |
3361 | chunk_type = chunk_to_extended(chunk_type) & |
3362 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3363 | |
899c81ea | 3364 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3365 | return 1; |
3366 | ||
3367 | return 0; | |
3368 | } | |
3369 | ||
6ec0896c | 3370 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3371 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3372 | { | |
6ec0896c | 3373 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3374 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3375 | struct btrfs_balance_args *bargs = NULL; |
3376 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3377 | ||
3378 | /* type filter */ | |
3379 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3380 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3381 | return 0; | |
3382 | } | |
3383 | ||
3384 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3385 | bargs = &bctl->data; | |
3386 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3387 | bargs = &bctl->sys; | |
3388 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3389 | bargs = &bctl->meta; | |
3390 | ||
ed25e9b2 ID |
3391 | /* profiles filter */ |
3392 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3393 | chunk_profiles_filter(chunk_type, bargs)) { | |
3394 | return 0; | |
5ce5b3c0 ID |
3395 | } |
3396 | ||
3397 | /* usage filter */ | |
3398 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3399 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3400 | return 0; |
bc309467 | 3401 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3402 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3403 | return 0; |
409d404b ID |
3404 | } |
3405 | ||
3406 | /* devid filter */ | |
3407 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3408 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3409 | return 0; | |
94e60d5a ID |
3410 | } |
3411 | ||
3412 | /* drange filter, makes sense only with devid filter */ | |
3413 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3414 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3415 | return 0; |
ea67176a ID |
3416 | } |
3417 | ||
3418 | /* vrange filter */ | |
3419 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3420 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3421 | return 0; | |
ed25e9b2 ID |
3422 | } |
3423 | ||
dee32d0a GAP |
3424 | /* stripes filter */ |
3425 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3426 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3427 | return 0; | |
3428 | } | |
3429 | ||
cfa4c961 ID |
3430 | /* soft profile changing mode */ |
3431 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3432 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3433 | return 0; | |
3434 | } | |
3435 | ||
7d824b6f DS |
3436 | /* |
3437 | * limited by count, must be the last filter | |
3438 | */ | |
3439 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3440 | if (bargs->limit == 0) | |
3441 | return 0; | |
3442 | else | |
3443 | bargs->limit--; | |
12907fc7 DS |
3444 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3445 | /* | |
3446 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3447 | * determined here because we do not have the global information |
12907fc7 DS |
3448 | * about the count of all chunks that satisfy the filters. |
3449 | */ | |
3450 | if (bargs->limit_max == 0) | |
3451 | return 0; | |
3452 | else | |
3453 | bargs->limit_max--; | |
7d824b6f DS |
3454 | } |
3455 | ||
f43ffb60 ID |
3456 | return 1; |
3457 | } | |
3458 | ||
c9e9f97b | 3459 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3460 | { |
19a39dce | 3461 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3462 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3463 | u64 chunk_type; |
f43ffb60 | 3464 | struct btrfs_chunk *chunk; |
5a488b9d | 3465 | struct btrfs_path *path = NULL; |
ec44a35c | 3466 | struct btrfs_key key; |
ec44a35c | 3467 | struct btrfs_key found_key; |
f43ffb60 ID |
3468 | struct extent_buffer *leaf; |
3469 | int slot; | |
c9e9f97b ID |
3470 | int ret; |
3471 | int enospc_errors = 0; | |
19a39dce | 3472 | bool counting = true; |
12907fc7 | 3473 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3474 | u64 limit_data = bctl->data.limit; |
3475 | u64 limit_meta = bctl->meta.limit; | |
3476 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3477 | u32 count_data = 0; |
3478 | u32 count_meta = 0; | |
3479 | u32 count_sys = 0; | |
2c9fe835 | 3480 | int chunk_reserved = 0; |
ec44a35c | 3481 | |
ec44a35c | 3482 | path = btrfs_alloc_path(); |
17e9f796 MF |
3483 | if (!path) { |
3484 | ret = -ENOMEM; | |
3485 | goto error; | |
3486 | } | |
19a39dce ID |
3487 | |
3488 | /* zero out stat counters */ | |
3489 | spin_lock(&fs_info->balance_lock); | |
3490 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3491 | spin_unlock(&fs_info->balance_lock); | |
3492 | again: | |
7d824b6f | 3493 | if (!counting) { |
12907fc7 DS |
3494 | /* |
3495 | * The single value limit and min/max limits use the same bytes | |
3496 | * in the | |
3497 | */ | |
7d824b6f DS |
3498 | bctl->data.limit = limit_data; |
3499 | bctl->meta.limit = limit_meta; | |
3500 | bctl->sys.limit = limit_sys; | |
3501 | } | |
ec44a35c CM |
3502 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3503 | key.offset = (u64)-1; | |
3504 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3505 | ||
d397712b | 3506 | while (1) { |
19a39dce | 3507 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3508 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3509 | ret = -ECANCELED; |
3510 | goto error; | |
3511 | } | |
3512 | ||
67c5e7d4 | 3513 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3514 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3515 | if (ret < 0) { |
3516 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3517 | goto error; |
67c5e7d4 | 3518 | } |
ec44a35c CM |
3519 | |
3520 | /* | |
3521 | * this shouldn't happen, it means the last relocate | |
3522 | * failed | |
3523 | */ | |
3524 | if (ret == 0) | |
c9e9f97b | 3525 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3526 | |
3527 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3528 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3529 | if (ret) { |
67c5e7d4 | 3530 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3531 | ret = 0; |
ec44a35c | 3532 | break; |
c9e9f97b | 3533 | } |
7d9eb12c | 3534 | |
f43ffb60 ID |
3535 | leaf = path->nodes[0]; |
3536 | slot = path->slots[0]; | |
3537 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3538 | |
67c5e7d4 FM |
3539 | if (found_key.objectid != key.objectid) { |
3540 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3541 | break; |
67c5e7d4 | 3542 | } |
7d9eb12c | 3543 | |
f43ffb60 | 3544 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3545 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3546 | |
19a39dce ID |
3547 | if (!counting) { |
3548 | spin_lock(&fs_info->balance_lock); | |
3549 | bctl->stat.considered++; | |
3550 | spin_unlock(&fs_info->balance_lock); | |
3551 | } | |
3552 | ||
6ec0896c | 3553 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3554 | |
b3b4aa74 | 3555 | btrfs_release_path(path); |
67c5e7d4 FM |
3556 | if (!ret) { |
3557 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3558 | goto loop; |
67c5e7d4 | 3559 | } |
f43ffb60 | 3560 | |
19a39dce | 3561 | if (counting) { |
67c5e7d4 | 3562 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3563 | spin_lock(&fs_info->balance_lock); |
3564 | bctl->stat.expected++; | |
3565 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3566 | |
3567 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3568 | count_data++; | |
3569 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3570 | count_sys++; | |
3571 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3572 | count_meta++; | |
3573 | ||
3574 | goto loop; | |
3575 | } | |
3576 | ||
3577 | /* | |
3578 | * Apply limit_min filter, no need to check if the LIMITS | |
3579 | * filter is used, limit_min is 0 by default | |
3580 | */ | |
3581 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3582 | count_data < bctl->data.limit_min) | |
3583 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3584 | count_meta < bctl->meta.limit_min) | |
3585 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3586 | count_sys < bctl->sys.limit_min)) { | |
3587 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3588 | goto loop; |
3589 | } | |
3590 | ||
a6f93c71 LB |
3591 | if (!chunk_reserved) { |
3592 | /* | |
3593 | * We may be relocating the only data chunk we have, | |
3594 | * which could potentially end up with losing data's | |
3595 | * raid profile, so lets allocate an empty one in | |
3596 | * advance. | |
3597 | */ | |
3598 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3599 | found_key.offset); | |
2c9fe835 ZL |
3600 | if (ret < 0) { |
3601 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3602 | goto error; | |
a6f93c71 LB |
3603 | } else if (ret == 1) { |
3604 | chunk_reserved = 1; | |
2c9fe835 | 3605 | } |
2c9fe835 ZL |
3606 | } |
3607 | ||
5b4aacef | 3608 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3609 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3610 | if (ret == -ENOSPC) { |
c9e9f97b | 3611 | enospc_errors++; |
eede2bf3 OS |
3612 | } else if (ret == -ETXTBSY) { |
3613 | btrfs_info(fs_info, | |
3614 | "skipping relocation of block group %llu due to active swapfile", | |
3615 | found_key.offset); | |
3616 | ret = 0; | |
3617 | } else if (ret) { | |
3618 | goto error; | |
19a39dce ID |
3619 | } else { |
3620 | spin_lock(&fs_info->balance_lock); | |
3621 | bctl->stat.completed++; | |
3622 | spin_unlock(&fs_info->balance_lock); | |
3623 | } | |
f43ffb60 | 3624 | loop: |
795a3321 ID |
3625 | if (found_key.offset == 0) |
3626 | break; | |
ba1bf481 | 3627 | key.offset = found_key.offset - 1; |
ec44a35c | 3628 | } |
c9e9f97b | 3629 | |
19a39dce ID |
3630 | if (counting) { |
3631 | btrfs_release_path(path); | |
3632 | counting = false; | |
3633 | goto again; | |
3634 | } | |
ec44a35c CM |
3635 | error: |
3636 | btrfs_free_path(path); | |
c9e9f97b | 3637 | if (enospc_errors) { |
efe120a0 | 3638 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3639 | enospc_errors); |
c9e9f97b ID |
3640 | if (!ret) |
3641 | ret = -ENOSPC; | |
3642 | } | |
3643 | ||
ec44a35c CM |
3644 | return ret; |
3645 | } | |
3646 | ||
0c460c0d ID |
3647 | /** |
3648 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3649 | * @flags: profile to validate | |
3650 | * @extended: if true @flags is treated as an extended profile | |
3651 | */ | |
3652 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3653 | { | |
3654 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3655 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3656 | ||
3657 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3658 | ||
3659 | /* 1) check that all other bits are zeroed */ | |
3660 | if (flags & ~mask) | |
3661 | return 0; | |
3662 | ||
3663 | /* 2) see if profile is reduced */ | |
3664 | if (flags == 0) | |
3665 | return !extended; /* "0" is valid for usual profiles */ | |
3666 | ||
c1499166 | 3667 | return has_single_bit_set(flags); |
0c460c0d ID |
3668 | } |
3669 | ||
837d5b6e ID |
3670 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3671 | { | |
a7e99c69 ID |
3672 | /* cancel requested || normal exit path */ |
3673 | return atomic_read(&fs_info->balance_cancel_req) || | |
3674 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3675 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3676 | } |
3677 | ||
bdcd3c97 AM |
3678 | /* Non-zero return value signifies invalidity */ |
3679 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3680 | u64 allowed) | |
3681 | { | |
3682 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3683 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3684 | (bctl_arg->target & ~allowed))); | |
3685 | } | |
3686 | ||
56fc37d9 AJ |
3687 | /* |
3688 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3689 | * @size_buf including the terminating null. The output may be trimmed if it | |
3690 | * does not fit into the provided buffer. | |
3691 | */ | |
3692 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3693 | u32 size_buf) | |
3694 | { | |
3695 | int ret; | |
3696 | u32 size_bp = size_buf; | |
3697 | char *bp = buf; | |
3698 | u64 flags = bargs->flags; | |
3699 | char tmp_buf[128] = {'\0'}; | |
3700 | ||
3701 | if (!flags) | |
3702 | return; | |
3703 | ||
3704 | #define CHECK_APPEND_NOARG(a) \ | |
3705 | do { \ | |
3706 | ret = snprintf(bp, size_bp, (a)); \ | |
3707 | if (ret < 0 || ret >= size_bp) \ | |
3708 | goto out_overflow; \ | |
3709 | size_bp -= ret; \ | |
3710 | bp += ret; \ | |
3711 | } while (0) | |
3712 | ||
3713 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3714 | do { \ | |
3715 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3716 | if (ret < 0 || ret >= size_bp) \ | |
3717 | goto out_overflow; \ | |
3718 | size_bp -= ret; \ | |
3719 | bp += ret; \ | |
3720 | } while (0) | |
3721 | ||
3722 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3723 | do { \ | |
3724 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3725 | if (ret < 0 || ret >= size_bp) \ | |
3726 | goto out_overflow; \ | |
3727 | size_bp -= ret; \ | |
3728 | bp += ret; \ | |
3729 | } while (0) | |
3730 | ||
158da513 DS |
3731 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3732 | CHECK_APPEND_1ARG("convert=%s,", | |
3733 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3734 | |
3735 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3736 | CHECK_APPEND_NOARG("soft,"); | |
3737 | ||
3738 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3739 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3740 | sizeof(tmp_buf)); | |
3741 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3742 | } | |
3743 | ||
3744 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3745 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3746 | ||
3747 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3748 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3749 | bargs->usage_min, bargs->usage_max); | |
3750 | ||
3751 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3752 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3753 | ||
3754 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3755 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3756 | bargs->pstart, bargs->pend); | |
3757 | ||
3758 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3759 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3760 | bargs->vstart, bargs->vend); | |
3761 | ||
3762 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3763 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3764 | ||
3765 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3766 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3767 | bargs->limit_min, bargs->limit_max); | |
3768 | ||
3769 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3770 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3771 | bargs->stripes_min, bargs->stripes_max); | |
3772 | ||
3773 | #undef CHECK_APPEND_2ARG | |
3774 | #undef CHECK_APPEND_1ARG | |
3775 | #undef CHECK_APPEND_NOARG | |
3776 | ||
3777 | out_overflow: | |
3778 | ||
3779 | if (size_bp < size_buf) | |
3780 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3781 | else | |
3782 | buf[0] = '\0'; | |
3783 | } | |
3784 | ||
3785 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3786 | { | |
3787 | u32 size_buf = 1024; | |
3788 | char tmp_buf[192] = {'\0'}; | |
3789 | char *buf; | |
3790 | char *bp; | |
3791 | u32 size_bp = size_buf; | |
3792 | int ret; | |
3793 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3794 | ||
3795 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3796 | if (!buf) | |
3797 | return; | |
3798 | ||
3799 | bp = buf; | |
3800 | ||
3801 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3802 | do { \ | |
3803 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3804 | if (ret < 0 || ret >= size_bp) \ | |
3805 | goto out_overflow; \ | |
3806 | size_bp -= ret; \ | |
3807 | bp += ret; \ | |
3808 | } while (0) | |
3809 | ||
3810 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3811 | CHECK_APPEND_1ARG("%s", "-f "); | |
3812 | ||
3813 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3814 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3815 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3816 | } | |
3817 | ||
3818 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
3819 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
3820 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
3821 | } | |
3822 | ||
3823 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
3824 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
3825 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
3826 | } | |
3827 | ||
3828 | #undef CHECK_APPEND_1ARG | |
3829 | ||
3830 | out_overflow: | |
3831 | ||
3832 | if (size_bp < size_buf) | |
3833 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
3834 | btrfs_info(fs_info, "balance: %s %s", | |
3835 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
3836 | "resume" : "start", buf); | |
3837 | ||
3838 | kfree(buf); | |
3839 | } | |
3840 | ||
c9e9f97b | 3841 | /* |
dccdb07b | 3842 | * Should be called with balance mutexe held |
c9e9f97b | 3843 | */ |
6fcf6e2b DS |
3844 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3845 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3846 | struct btrfs_ioctl_balance_args *bargs) |
3847 | { | |
14506127 | 3848 | u64 meta_target, data_target; |
f43ffb60 | 3849 | u64 allowed; |
e4837f8f | 3850 | int mixed = 0; |
c9e9f97b | 3851 | int ret; |
8dabb742 | 3852 | u64 num_devices; |
de98ced9 | 3853 | unsigned seq; |
e62869be | 3854 | bool reducing_redundancy; |
081db89b | 3855 | int i; |
c9e9f97b | 3856 | |
837d5b6e | 3857 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3858 | atomic_read(&fs_info->balance_pause_req) || |
3859 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3860 | ret = -EINVAL; |
3861 | goto out; | |
3862 | } | |
3863 | ||
e4837f8f ID |
3864 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3865 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3866 | mixed = 1; | |
3867 | ||
f43ffb60 ID |
3868 | /* |
3869 | * In case of mixed groups both data and meta should be picked, | |
3870 | * and identical options should be given for both of them. | |
3871 | */ | |
e4837f8f ID |
3872 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3873 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3874 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3875 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3876 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3877 | btrfs_err(fs_info, |
6dac13f8 | 3878 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3879 | ret = -EINVAL; |
3880 | goto out; | |
3881 | } | |
3882 | } | |
3883 | ||
b35cf1f0 JB |
3884 | /* |
3885 | * rw_devices will not change at the moment, device add/delete/replace | |
3886 | * are excluded by EXCL_OP | |
3887 | */ | |
3888 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
3889 | |
3890 | /* | |
3891 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
3892 | * special bit for it, to make it easier to distinguish. Thus we need | |
3893 | * to set it manually, or balance would refuse the profile. | |
3894 | */ | |
3895 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
3896 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
3897 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
3898 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 3899 | |
bdcd3c97 | 3900 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e | 3901 | btrfs_err(fs_info, |
6dac13f8 | 3902 | "balance: invalid convert data profile %s", |
158da513 | 3903 | btrfs_bg_type_to_raid_name(bctl->data.target)); |
e4d8ec0f ID |
3904 | ret = -EINVAL; |
3905 | goto out; | |
3906 | } | |
bdcd3c97 | 3907 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3908 | btrfs_err(fs_info, |
6dac13f8 | 3909 | "balance: invalid convert metadata profile %s", |
158da513 | 3910 | btrfs_bg_type_to_raid_name(bctl->meta.target)); |
e4d8ec0f ID |
3911 | ret = -EINVAL; |
3912 | goto out; | |
3913 | } | |
bdcd3c97 | 3914 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3915 | btrfs_err(fs_info, |
6dac13f8 | 3916 | "balance: invalid convert system profile %s", |
158da513 | 3917 | btrfs_bg_type_to_raid_name(bctl->sys.target)); |
e4d8ec0f ID |
3918 | ret = -EINVAL; |
3919 | goto out; | |
3920 | } | |
3921 | ||
6079e12c DS |
3922 | /* |
3923 | * Allow to reduce metadata or system integrity only if force set for | |
3924 | * profiles with redundancy (copies, parity) | |
3925 | */ | |
3926 | allowed = 0; | |
3927 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
3928 | if (btrfs_raid_array[i].ncopies >= 2 || | |
3929 | btrfs_raid_array[i].tolerated_failures >= 1) | |
3930 | allowed |= btrfs_raid_array[i].bg_flag; | |
3931 | } | |
de98ced9 MX |
3932 | do { |
3933 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3934 | ||
3935 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3936 | (fs_info->avail_system_alloc_bits & allowed) && | |
3937 | !(bctl->sys.target & allowed)) || | |
3938 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3939 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 3940 | !(bctl->meta.target & allowed))) |
e62869be | 3941 | reducing_redundancy = true; |
5a8067c0 | 3942 | else |
e62869be | 3943 | reducing_redundancy = false; |
5a8067c0 FM |
3944 | |
3945 | /* if we're not converting, the target field is uninitialized */ | |
3946 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3947 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3948 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3949 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 3950 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3951 | |
e62869be | 3952 | if (reducing_redundancy) { |
5a8067c0 FM |
3953 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
3954 | btrfs_info(fs_info, | |
e62869be | 3955 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
3956 | } else { |
3957 | btrfs_err(fs_info, | |
e62869be | 3958 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
3959 | ret = -EINVAL; |
3960 | goto out; | |
3961 | } | |
3962 | } | |
3963 | ||
14506127 AB |
3964 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
3965 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 3966 | btrfs_warn(fs_info, |
6dac13f8 | 3967 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
3968 | btrfs_bg_type_to_raid_name(meta_target), |
3969 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
3970 | } |
3971 | ||
9e967495 FM |
3972 | if (fs_info->send_in_progress) { |
3973 | btrfs_warn_rl(fs_info, | |
3974 | "cannot run balance while send operations are in progress (%d in progress)", | |
3975 | fs_info->send_in_progress); | |
3976 | ret = -EAGAIN; | |
3977 | goto out; | |
3978 | } | |
3979 | ||
6bccf3ab | 3980 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3981 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3982 | goto out; |
3983 | ||
59641015 ID |
3984 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3985 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
3986 | BUG_ON(fs_info->balance_ctl); |
3987 | spin_lock(&fs_info->balance_lock); | |
3988 | fs_info->balance_ctl = bctl; | |
3989 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
3990 | } else { |
3991 | BUG_ON(ret != -EEXIST); | |
3992 | spin_lock(&fs_info->balance_lock); | |
3993 | update_balance_args(bctl); | |
3994 | spin_unlock(&fs_info->balance_lock); | |
3995 | } | |
c9e9f97b | 3996 | |
3009a62f DS |
3997 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
3998 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 3999 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4000 | mutex_unlock(&fs_info->balance_mutex); |
4001 | ||
4002 | ret = __btrfs_balance(fs_info); | |
4003 | ||
4004 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4005 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4006 | btrfs_info(fs_info, "balance: paused"); | |
4007 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4008 | btrfs_info(fs_info, "balance: canceled"); | |
4009 | else | |
4010 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4011 | ||
3009a62f | 4012 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4013 | |
4014 | if (bargs) { | |
4015 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4016 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4017 | } |
4018 | ||
3a01aa7a ID |
4019 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4020 | balance_need_close(fs_info)) { | |
149196a2 | 4021 | reset_balance_state(fs_info); |
a17c95df | 4022 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4023 | } |
4024 | ||
837d5b6e | 4025 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4026 | |
4027 | return ret; | |
4028 | out: | |
59641015 | 4029 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4030 | reset_balance_state(fs_info); |
a17c95df | 4031 | else |
59641015 | 4032 | kfree(bctl); |
a17c95df DS |
4033 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4034 | ||
59641015 ID |
4035 | return ret; |
4036 | } | |
4037 | ||
4038 | static int balance_kthread(void *data) | |
4039 | { | |
2b6ba629 | 4040 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4041 | int ret = 0; |
59641015 | 4042 | |
59641015 | 4043 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4044 | if (fs_info->balance_ctl) |
6fcf6e2b | 4045 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4046 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4047 | |
59641015 ID |
4048 | return ret; |
4049 | } | |
4050 | ||
2b6ba629 ID |
4051 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4052 | { | |
4053 | struct task_struct *tsk; | |
4054 | ||
1354e1a1 | 4055 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4056 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4057 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4058 | return 0; |
4059 | } | |
1354e1a1 | 4060 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4061 | |
3cdde224 | 4062 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4063 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4064 | return 0; |
4065 | } | |
4066 | ||
02ee654d AJ |
4067 | /* |
4068 | * A ro->rw remount sequence should continue with the paused balance | |
4069 | * regardless of who pauses it, system or the user as of now, so set | |
4070 | * the resume flag. | |
4071 | */ | |
4072 | spin_lock(&fs_info->balance_lock); | |
4073 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4074 | spin_unlock(&fs_info->balance_lock); | |
4075 | ||
2b6ba629 | 4076 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4077 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4078 | } |
4079 | ||
68310a5e | 4080 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4081 | { |
59641015 ID |
4082 | struct btrfs_balance_control *bctl; |
4083 | struct btrfs_balance_item *item; | |
4084 | struct btrfs_disk_balance_args disk_bargs; | |
4085 | struct btrfs_path *path; | |
4086 | struct extent_buffer *leaf; | |
4087 | struct btrfs_key key; | |
4088 | int ret; | |
4089 | ||
4090 | path = btrfs_alloc_path(); | |
4091 | if (!path) | |
4092 | return -ENOMEM; | |
4093 | ||
59641015 | 4094 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4095 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4096 | key.offset = 0; |
4097 | ||
68310a5e | 4098 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4099 | if (ret < 0) |
68310a5e | 4100 | goto out; |
59641015 ID |
4101 | if (ret > 0) { /* ret = -ENOENT; */ |
4102 | ret = 0; | |
68310a5e ID |
4103 | goto out; |
4104 | } | |
4105 | ||
4106 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4107 | if (!bctl) { | |
4108 | ret = -ENOMEM; | |
4109 | goto out; | |
59641015 ID |
4110 | } |
4111 | ||
4112 | leaf = path->nodes[0]; | |
4113 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4114 | ||
68310a5e ID |
4115 | bctl->flags = btrfs_balance_flags(leaf, item); |
4116 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4117 | |
4118 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4119 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4120 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4121 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4122 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4123 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4124 | ||
eee95e3f DS |
4125 | /* |
4126 | * This should never happen, as the paused balance state is recovered | |
4127 | * during mount without any chance of other exclusive ops to collide. | |
4128 | * | |
4129 | * This gives the exclusive op status to balance and keeps in paused | |
4130 | * state until user intervention (cancel or umount). If the ownership | |
4131 | * cannot be assigned, show a message but do not fail. The balance | |
4132 | * is in a paused state and must have fs_info::balance_ctl properly | |
4133 | * set up. | |
4134 | */ | |
4135 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4136 | btrfs_warn(fs_info, | |
6dac13f8 | 4137 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4138 | |
68310a5e | 4139 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4140 | BUG_ON(fs_info->balance_ctl); |
4141 | spin_lock(&fs_info->balance_lock); | |
4142 | fs_info->balance_ctl = bctl; | |
4143 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4144 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4145 | out: |
4146 | btrfs_free_path(path); | |
ec44a35c CM |
4147 | return ret; |
4148 | } | |
4149 | ||
837d5b6e ID |
4150 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4151 | { | |
4152 | int ret = 0; | |
4153 | ||
4154 | mutex_lock(&fs_info->balance_mutex); | |
4155 | if (!fs_info->balance_ctl) { | |
4156 | mutex_unlock(&fs_info->balance_mutex); | |
4157 | return -ENOTCONN; | |
4158 | } | |
4159 | ||
3009a62f | 4160 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4161 | atomic_inc(&fs_info->balance_pause_req); |
4162 | mutex_unlock(&fs_info->balance_mutex); | |
4163 | ||
4164 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4165 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4166 | |
4167 | mutex_lock(&fs_info->balance_mutex); | |
4168 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4169 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4170 | atomic_dec(&fs_info->balance_pause_req); |
4171 | } else { | |
4172 | ret = -ENOTCONN; | |
4173 | } | |
4174 | ||
4175 | mutex_unlock(&fs_info->balance_mutex); | |
4176 | return ret; | |
4177 | } | |
4178 | ||
a7e99c69 ID |
4179 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4180 | { | |
4181 | mutex_lock(&fs_info->balance_mutex); | |
4182 | if (!fs_info->balance_ctl) { | |
4183 | mutex_unlock(&fs_info->balance_mutex); | |
4184 | return -ENOTCONN; | |
4185 | } | |
4186 | ||
cf7d20f4 DS |
4187 | /* |
4188 | * A paused balance with the item stored on disk can be resumed at | |
4189 | * mount time if the mount is read-write. Otherwise it's still paused | |
4190 | * and we must not allow cancelling as it deletes the item. | |
4191 | */ | |
4192 | if (sb_rdonly(fs_info->sb)) { | |
4193 | mutex_unlock(&fs_info->balance_mutex); | |
4194 | return -EROFS; | |
4195 | } | |
4196 | ||
a7e99c69 ID |
4197 | atomic_inc(&fs_info->balance_cancel_req); |
4198 | /* | |
4199 | * if we are running just wait and return, balance item is | |
4200 | * deleted in btrfs_balance in this case | |
4201 | */ | |
3009a62f | 4202 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4203 | mutex_unlock(&fs_info->balance_mutex); |
4204 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4205 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4206 | mutex_lock(&fs_info->balance_mutex); |
4207 | } else { | |
a7e99c69 | 4208 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4209 | /* |
4210 | * Lock released to allow other waiters to continue, we'll | |
4211 | * reexamine the status again. | |
4212 | */ | |
a7e99c69 ID |
4213 | mutex_lock(&fs_info->balance_mutex); |
4214 | ||
a17c95df | 4215 | if (fs_info->balance_ctl) { |
149196a2 | 4216 | reset_balance_state(fs_info); |
a17c95df | 4217 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4218 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4219 | } |
a7e99c69 ID |
4220 | } |
4221 | ||
3009a62f DS |
4222 | BUG_ON(fs_info->balance_ctl || |
4223 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4224 | atomic_dec(&fs_info->balance_cancel_req); |
4225 | mutex_unlock(&fs_info->balance_mutex); | |
4226 | return 0; | |
4227 | } | |
4228 | ||
803b2f54 SB |
4229 | static int btrfs_uuid_scan_kthread(void *data) |
4230 | { | |
4231 | struct btrfs_fs_info *fs_info = data; | |
4232 | struct btrfs_root *root = fs_info->tree_root; | |
4233 | struct btrfs_key key; | |
803b2f54 SB |
4234 | struct btrfs_path *path = NULL; |
4235 | int ret = 0; | |
4236 | struct extent_buffer *eb; | |
4237 | int slot; | |
4238 | struct btrfs_root_item root_item; | |
4239 | u32 item_size; | |
f45388f3 | 4240 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4241 | |
4242 | path = btrfs_alloc_path(); | |
4243 | if (!path) { | |
4244 | ret = -ENOMEM; | |
4245 | goto out; | |
4246 | } | |
4247 | ||
4248 | key.objectid = 0; | |
4249 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4250 | key.offset = 0; | |
4251 | ||
803b2f54 | 4252 | while (1) { |
7c829b72 AJ |
4253 | ret = btrfs_search_forward(root, &key, path, |
4254 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4255 | if (ret) { |
4256 | if (ret > 0) | |
4257 | ret = 0; | |
4258 | break; | |
4259 | } | |
4260 | ||
4261 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4262 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4263 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4264 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4265 | goto skip; | |
4266 | ||
4267 | eb = path->nodes[0]; | |
4268 | slot = path->slots[0]; | |
4269 | item_size = btrfs_item_size_nr(eb, slot); | |
4270 | if (item_size < sizeof(root_item)) | |
4271 | goto skip; | |
4272 | ||
803b2f54 SB |
4273 | read_extent_buffer(eb, &root_item, |
4274 | btrfs_item_ptr_offset(eb, slot), | |
4275 | (int)sizeof(root_item)); | |
4276 | if (btrfs_root_refs(&root_item) == 0) | |
4277 | goto skip; | |
f45388f3 FDBM |
4278 | |
4279 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4280 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4281 | if (trans) | |
4282 | goto update_tree; | |
4283 | ||
4284 | btrfs_release_path(path); | |
803b2f54 SB |
4285 | /* |
4286 | * 1 - subvol uuid item | |
4287 | * 1 - received_subvol uuid item | |
4288 | */ | |
4289 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4290 | if (IS_ERR(trans)) { | |
4291 | ret = PTR_ERR(trans); | |
4292 | break; | |
4293 | } | |
f45388f3 FDBM |
4294 | continue; |
4295 | } else { | |
4296 | goto skip; | |
4297 | } | |
4298 | update_tree: | |
4299 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4300 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4301 | BTRFS_UUID_KEY_SUBVOL, |
4302 | key.objectid); | |
4303 | if (ret < 0) { | |
efe120a0 | 4304 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4305 | ret); |
803b2f54 SB |
4306 | break; |
4307 | } | |
4308 | } | |
4309 | ||
4310 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4311 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4312 | root_item.received_uuid, |
4313 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4314 | key.objectid); | |
4315 | if (ret < 0) { | |
efe120a0 | 4316 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4317 | ret); |
803b2f54 SB |
4318 | break; |
4319 | } | |
4320 | } | |
4321 | ||
f45388f3 | 4322 | skip: |
803b2f54 | 4323 | if (trans) { |
3a45bb20 | 4324 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4325 | trans = NULL; |
803b2f54 SB |
4326 | if (ret) |
4327 | break; | |
4328 | } | |
4329 | ||
803b2f54 SB |
4330 | btrfs_release_path(path); |
4331 | if (key.offset < (u64)-1) { | |
4332 | key.offset++; | |
4333 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4334 | key.offset = 0; | |
4335 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4336 | } else if (key.objectid < (u64)-1) { | |
4337 | key.offset = 0; | |
4338 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4339 | key.objectid++; | |
4340 | } else { | |
4341 | break; | |
4342 | } | |
4343 | cond_resched(); | |
4344 | } | |
4345 | ||
4346 | out: | |
4347 | btrfs_free_path(path); | |
f45388f3 | 4348 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4349 | btrfs_end_transaction(trans); |
803b2f54 | 4350 | if (ret) |
efe120a0 | 4351 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4352 | else |
afcdd129 | 4353 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4354 | up(&fs_info->uuid_tree_rescan_sem); |
4355 | return 0; | |
4356 | } | |
4357 | ||
70f80175 SB |
4358 | /* |
4359 | * Callback for btrfs_uuid_tree_iterate(). | |
4360 | * returns: | |
4361 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4362 | * < 0 if an error occurred. |
70f80175 SB |
4363 | * > 0 if the check failed, which means the caller shall remove the entry. |
4364 | */ | |
4365 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4366 | u8 *uuid, u8 type, u64 subid) | |
4367 | { | |
4368 | struct btrfs_key key; | |
4369 | int ret = 0; | |
4370 | struct btrfs_root *subvol_root; | |
4371 | ||
4372 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4373 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4374 | goto out; | |
4375 | ||
4376 | key.objectid = subid; | |
4377 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4378 | key.offset = (u64)-1; | |
4379 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4380 | if (IS_ERR(subvol_root)) { | |
4381 | ret = PTR_ERR(subvol_root); | |
4382 | if (ret == -ENOENT) | |
4383 | ret = 1; | |
4384 | goto out; | |
4385 | } | |
4386 | ||
4387 | switch (type) { | |
4388 | case BTRFS_UUID_KEY_SUBVOL: | |
4389 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4390 | ret = 1; | |
4391 | break; | |
4392 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4393 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4394 | BTRFS_UUID_SIZE)) | |
4395 | ret = 1; | |
4396 | break; | |
4397 | } | |
4398 | ||
4399 | out: | |
4400 | return ret; | |
4401 | } | |
4402 | ||
4403 | static int btrfs_uuid_rescan_kthread(void *data) | |
4404 | { | |
4405 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4406 | int ret; | |
4407 | ||
4408 | /* | |
4409 | * 1st step is to iterate through the existing UUID tree and | |
4410 | * to delete all entries that contain outdated data. | |
4411 | * 2nd step is to add all missing entries to the UUID tree. | |
4412 | */ | |
4413 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4414 | if (ret < 0) { | |
efe120a0 | 4415 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4416 | up(&fs_info->uuid_tree_rescan_sem); |
4417 | return ret; | |
4418 | } | |
4419 | return btrfs_uuid_scan_kthread(data); | |
4420 | } | |
4421 | ||
f7a81ea4 SB |
4422 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4423 | { | |
4424 | struct btrfs_trans_handle *trans; | |
4425 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4426 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4427 | struct task_struct *task; |
4428 | int ret; | |
f7a81ea4 SB |
4429 | |
4430 | /* | |
4431 | * 1 - root node | |
4432 | * 1 - root item | |
4433 | */ | |
4434 | trans = btrfs_start_transaction(tree_root, 2); | |
4435 | if (IS_ERR(trans)) | |
4436 | return PTR_ERR(trans); | |
4437 | ||
9b7a2440 | 4438 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4439 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4440 | ret = PTR_ERR(uuid_root); |
66642832 | 4441 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4442 | btrfs_end_transaction(trans); |
6d13f549 | 4443 | return ret; |
f7a81ea4 SB |
4444 | } |
4445 | ||
4446 | fs_info->uuid_root = uuid_root; | |
4447 | ||
3a45bb20 | 4448 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4449 | if (ret) |
4450 | return ret; | |
4451 | ||
4452 | down(&fs_info->uuid_tree_rescan_sem); | |
4453 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4454 | if (IS_ERR(task)) { | |
70f80175 | 4455 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4456 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4457 | up(&fs_info->uuid_tree_rescan_sem); |
4458 | return PTR_ERR(task); | |
4459 | } | |
4460 | ||
4461 | return 0; | |
f7a81ea4 | 4462 | } |
803b2f54 | 4463 | |
70f80175 SB |
4464 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4465 | { | |
4466 | struct task_struct *task; | |
4467 | ||
4468 | down(&fs_info->uuid_tree_rescan_sem); | |
4469 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4470 | if (IS_ERR(task)) { | |
4471 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4472 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4473 | up(&fs_info->uuid_tree_rescan_sem); |
4474 | return PTR_ERR(task); | |
4475 | } | |
4476 | ||
4477 | return 0; | |
4478 | } | |
4479 | ||
8f18cf13 CM |
4480 | /* |
4481 | * shrinking a device means finding all of the device extents past | |
4482 | * the new size, and then following the back refs to the chunks. | |
4483 | * The chunk relocation code actually frees the device extent | |
4484 | */ | |
4485 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4486 | { | |
0b246afa JM |
4487 | struct btrfs_fs_info *fs_info = device->fs_info; |
4488 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4489 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4490 | struct btrfs_dev_extent *dev_extent = NULL; |
4491 | struct btrfs_path *path; | |
4492 | u64 length; | |
8f18cf13 CM |
4493 | u64 chunk_offset; |
4494 | int ret; | |
4495 | int slot; | |
ba1bf481 JB |
4496 | int failed = 0; |
4497 | bool retried = false; | |
8f18cf13 CM |
4498 | struct extent_buffer *l; |
4499 | struct btrfs_key key; | |
0b246afa | 4500 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4501 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4502 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4503 | u64 diff; |
61d0d0d2 | 4504 | u64 start; |
7dfb8be1 NB |
4505 | |
4506 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4507 | start = new_size; |
0e4324a4 | 4508 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4509 | |
401e29c1 | 4510 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4511 | return -EINVAL; |
4512 | ||
8f18cf13 CM |
4513 | path = btrfs_alloc_path(); |
4514 | if (!path) | |
4515 | return -ENOMEM; | |
4516 | ||
0338dff6 | 4517 | path->reada = READA_BACK; |
8f18cf13 | 4518 | |
61d0d0d2 NB |
4519 | trans = btrfs_start_transaction(root, 0); |
4520 | if (IS_ERR(trans)) { | |
4521 | btrfs_free_path(path); | |
4522 | return PTR_ERR(trans); | |
4523 | } | |
4524 | ||
34441361 | 4525 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4526 | |
7cc8e58d | 4527 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4528 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4529 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4530 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4531 | } |
61d0d0d2 NB |
4532 | |
4533 | /* | |
4534 | * Once the device's size has been set to the new size, ensure all | |
4535 | * in-memory chunks are synced to disk so that the loop below sees them | |
4536 | * and relocates them accordingly. | |
4537 | */ | |
1c11b63e | 4538 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4539 | mutex_unlock(&fs_info->chunk_mutex); |
4540 | ret = btrfs_commit_transaction(trans); | |
4541 | if (ret) | |
4542 | goto done; | |
4543 | } else { | |
4544 | mutex_unlock(&fs_info->chunk_mutex); | |
4545 | btrfs_end_transaction(trans); | |
4546 | } | |
8f18cf13 | 4547 | |
ba1bf481 | 4548 | again: |
8f18cf13 CM |
4549 | key.objectid = device->devid; |
4550 | key.offset = (u64)-1; | |
4551 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4552 | ||
213e64da | 4553 | do { |
0b246afa | 4554 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4555 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4556 | if (ret < 0) { |
0b246afa | 4557 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4558 | goto done; |
67c5e7d4 | 4559 | } |
8f18cf13 CM |
4560 | |
4561 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4562 | if (ret) |
0b246afa | 4563 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4564 | if (ret < 0) |
4565 | goto done; | |
4566 | if (ret) { | |
4567 | ret = 0; | |
b3b4aa74 | 4568 | btrfs_release_path(path); |
bf1fb512 | 4569 | break; |
8f18cf13 CM |
4570 | } |
4571 | ||
4572 | l = path->nodes[0]; | |
4573 | slot = path->slots[0]; | |
4574 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4575 | ||
ba1bf481 | 4576 | if (key.objectid != device->devid) { |
0b246afa | 4577 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4578 | btrfs_release_path(path); |
bf1fb512 | 4579 | break; |
ba1bf481 | 4580 | } |
8f18cf13 CM |
4581 | |
4582 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4583 | length = btrfs_dev_extent_length(l, dev_extent); | |
4584 | ||
ba1bf481 | 4585 | if (key.offset + length <= new_size) { |
0b246afa | 4586 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4587 | btrfs_release_path(path); |
d6397bae | 4588 | break; |
ba1bf481 | 4589 | } |
8f18cf13 | 4590 | |
8f18cf13 | 4591 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4592 | btrfs_release_path(path); |
8f18cf13 | 4593 | |
a6f93c71 LB |
4594 | /* |
4595 | * We may be relocating the only data chunk we have, | |
4596 | * which could potentially end up with losing data's | |
4597 | * raid profile, so lets allocate an empty one in | |
4598 | * advance. | |
4599 | */ | |
4600 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4601 | if (ret < 0) { | |
4602 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4603 | goto done; | |
4604 | } | |
4605 | ||
0b246afa JM |
4606 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4607 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4608 | if (ret == -ENOSPC) { |
ba1bf481 | 4609 | failed++; |
eede2bf3 OS |
4610 | } else if (ret) { |
4611 | if (ret == -ETXTBSY) { | |
4612 | btrfs_warn(fs_info, | |
4613 | "could not shrink block group %llu due to active swapfile", | |
4614 | chunk_offset); | |
4615 | } | |
4616 | goto done; | |
4617 | } | |
213e64da | 4618 | } while (key.offset-- > 0); |
ba1bf481 JB |
4619 | |
4620 | if (failed && !retried) { | |
4621 | failed = 0; | |
4622 | retried = true; | |
4623 | goto again; | |
4624 | } else if (failed && retried) { | |
4625 | ret = -ENOSPC; | |
ba1bf481 | 4626 | goto done; |
8f18cf13 CM |
4627 | } |
4628 | ||
d6397bae | 4629 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4630 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4631 | if (IS_ERR(trans)) { |
4632 | ret = PTR_ERR(trans); | |
4633 | goto done; | |
4634 | } | |
4635 | ||
34441361 | 4636 | mutex_lock(&fs_info->chunk_mutex); |
7cc8e58d | 4637 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4638 | if (list_empty(&device->post_commit_list)) |
4639 | list_add_tail(&device->post_commit_list, | |
4640 | &trans->transaction->dev_update_list); | |
d6397bae | 4641 | |
d6397bae | 4642 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4643 | btrfs_set_super_total_bytes(super_copy, |
4644 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4645 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4646 | |
4647 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4648 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4649 | if (ret < 0) { |
4650 | btrfs_abort_transaction(trans, ret); | |
4651 | btrfs_end_transaction(trans); | |
4652 | } else { | |
4653 | ret = btrfs_commit_transaction(trans); | |
4654 | } | |
8f18cf13 CM |
4655 | done: |
4656 | btrfs_free_path(path); | |
53e489bc | 4657 | if (ret) { |
34441361 | 4658 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4659 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4660 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4661 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4662 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4663 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4664 | } |
8f18cf13 CM |
4665 | return ret; |
4666 | } | |
4667 | ||
2ff7e61e | 4668 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4669 | struct btrfs_key *key, |
4670 | struct btrfs_chunk *chunk, int item_size) | |
4671 | { | |
0b246afa | 4672 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4673 | struct btrfs_disk_key disk_key; |
4674 | u32 array_size; | |
4675 | u8 *ptr; | |
4676 | ||
34441361 | 4677 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4678 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4679 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4680 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4681 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4682 | return -EFBIG; |
fe48a5c0 | 4683 | } |
0b86a832 CM |
4684 | |
4685 | ptr = super_copy->sys_chunk_array + array_size; | |
4686 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4687 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4688 | ptr += sizeof(disk_key); | |
4689 | memcpy(ptr, chunk, item_size); | |
4690 | item_size += sizeof(disk_key); | |
4691 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4692 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4693 | |
0b86a832 CM |
4694 | return 0; |
4695 | } | |
4696 | ||
73c5de00 AJ |
4697 | /* |
4698 | * sort the devices in descending order by max_avail, total_avail | |
4699 | */ | |
4700 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4701 | { |
73c5de00 AJ |
4702 | const struct btrfs_device_info *di_a = a; |
4703 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4704 | |
73c5de00 | 4705 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4706 | return -1; |
73c5de00 | 4707 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4708 | return 1; |
73c5de00 AJ |
4709 | if (di_a->total_avail > di_b->total_avail) |
4710 | return -1; | |
4711 | if (di_a->total_avail < di_b->total_avail) | |
4712 | return 1; | |
4713 | return 0; | |
b2117a39 | 4714 | } |
0b86a832 | 4715 | |
53b381b3 DW |
4716 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4717 | { | |
ffe2d203 | 4718 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4719 | return; |
4720 | ||
ceda0864 | 4721 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4722 | } |
4723 | ||
cfbb825c DS |
4724 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4725 | { | |
4726 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4727 | return; | |
4728 | ||
4729 | btrfs_set_fs_incompat(info, RAID1C34); | |
4730 | } | |
4731 | ||
73c5de00 | 4732 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4733 | u64 start, u64 type) |
b2117a39 | 4734 | { |
2ff7e61e | 4735 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4736 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4737 | struct btrfs_device *device; |
73c5de00 AJ |
4738 | struct map_lookup *map = NULL; |
4739 | struct extent_map_tree *em_tree; | |
4740 | struct extent_map *em; | |
4741 | struct btrfs_device_info *devices_info = NULL; | |
4742 | u64 total_avail; | |
4743 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4744 | int data_stripes; /* number of stripes that count for |
4745 | block group size */ | |
73c5de00 AJ |
4746 | int sub_stripes; /* sub_stripes info for map */ |
4747 | int dev_stripes; /* stripes per dev */ | |
4748 | int devs_max; /* max devs to use */ | |
4749 | int devs_min; /* min devs needed */ | |
4750 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4751 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4752 | int nparity; /* number of stripes worth of bytes to |
4753 | store parity information */ | |
73c5de00 AJ |
4754 | int ret; |
4755 | u64 max_stripe_size; | |
4756 | u64 max_chunk_size; | |
4757 | u64 stripe_size; | |
23f0ff1e | 4758 | u64 chunk_size; |
73c5de00 AJ |
4759 | int ndevs; |
4760 | int i; | |
4761 | int j; | |
31e50229 | 4762 | int index; |
593060d7 | 4763 | |
0c460c0d | 4764 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4765 | |
4117f207 QW |
4766 | if (list_empty(&fs_devices->alloc_list)) { |
4767 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4768 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4769 | return -ENOSPC; |
4117f207 | 4770 | } |
b2117a39 | 4771 | |
3e72ee88 | 4772 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4773 | |
31e50229 LB |
4774 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4775 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4776 | devs_max = btrfs_raid_array[index].devs_max; | |
e3ecdb3f DS |
4777 | if (!devs_max) |
4778 | devs_max = BTRFS_MAX_DEVS(info); | |
31e50229 LB |
4779 | devs_min = btrfs_raid_array[index].devs_min; |
4780 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4781 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4782 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4783 | |
9b3f68b9 | 4784 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4785 | max_stripe_size = SZ_1G; |
fce466ea | 4786 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
9b3f68b9 | 4787 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4788 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4789 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4790 | max_stripe_size = SZ_1G; | |
1100373f | 4791 | else |
ee22184b | 4792 | max_stripe_size = SZ_256M; |
73c5de00 | 4793 | max_chunk_size = max_stripe_size; |
a40a90a0 | 4794 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4795 | max_stripe_size = SZ_32M; |
73c5de00 | 4796 | max_chunk_size = 2 * max_stripe_size; |
c17add7a | 4797 | devs_max = min_t(int, devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); |
73c5de00 | 4798 | } else { |
351fd353 | 4799 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 4800 | type); |
290342f6 | 4801 | BUG(); |
9b3f68b9 CM |
4802 | } |
4803 | ||
52042d8e | 4804 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
4805 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
4806 | max_chunk_size); | |
9b3f68b9 | 4807 | |
31e818fe | 4808 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4809 | GFP_NOFS); |
4810 | if (!devices_info) | |
4811 | return -ENOMEM; | |
0cad8a11 | 4812 | |
9f680ce0 | 4813 | /* |
73c5de00 AJ |
4814 | * in the first pass through the devices list, we gather information |
4815 | * about the available holes on each device. | |
9f680ce0 | 4816 | */ |
73c5de00 | 4817 | ndevs = 0; |
ebcc9301 | 4818 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4819 | u64 max_avail; |
4820 | u64 dev_offset; | |
b2117a39 | 4821 | |
ebbede42 | 4822 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4823 | WARN(1, KERN_ERR |
efe120a0 | 4824 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4825 | continue; |
4826 | } | |
b2117a39 | 4827 | |
e12c9621 AJ |
4828 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4829 | &device->dev_state) || | |
401e29c1 | 4830 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4831 | continue; |
b2117a39 | 4832 | |
73c5de00 AJ |
4833 | if (device->total_bytes > device->bytes_used) |
4834 | total_avail = device->total_bytes - device->bytes_used; | |
4835 | else | |
4836 | total_avail = 0; | |
38c01b96 | 4837 | |
4838 | /* If there is no space on this device, skip it. */ | |
4839 | if (total_avail == 0) | |
4840 | continue; | |
b2117a39 | 4841 | |
60dfdf25 | 4842 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
4843 | max_stripe_size * dev_stripes, |
4844 | &dev_offset, &max_avail); | |
4845 | if (ret && ret != -ENOSPC) | |
4846 | goto error; | |
b2117a39 | 4847 | |
73c5de00 AJ |
4848 | if (ret == 0) |
4849 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4850 | |
4117f207 QW |
4851 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4852 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4853 | btrfs_debug(info, | |
4854 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4855 | __func__, device->devid, max_avail, | |
4856 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4857 | continue; |
4117f207 | 4858 | } |
b2117a39 | 4859 | |
063d006f ES |
4860 | if (ndevs == fs_devices->rw_devices) { |
4861 | WARN(1, "%s: found more than %llu devices\n", | |
4862 | __func__, fs_devices->rw_devices); | |
4863 | break; | |
4864 | } | |
73c5de00 AJ |
4865 | devices_info[ndevs].dev_offset = dev_offset; |
4866 | devices_info[ndevs].max_avail = max_avail; | |
4867 | devices_info[ndevs].total_avail = total_avail; | |
4868 | devices_info[ndevs].dev = device; | |
4869 | ++ndevs; | |
4870 | } | |
b2117a39 | 4871 | |
73c5de00 AJ |
4872 | /* |
4873 | * now sort the devices by hole size / available space | |
4874 | */ | |
4875 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4876 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4877 | |
47e6f742 DS |
4878 | /* |
4879 | * Round down to number of usable stripes, devs_increment can be any | |
4880 | * number so we can't use round_down() | |
4881 | */ | |
4882 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4883 | |
ba89b802 | 4884 | if (ndevs < devs_min) { |
73c5de00 | 4885 | ret = -ENOSPC; |
4117f207 QW |
4886 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4887 | btrfs_debug(info, | |
4888 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4889 | __func__, ndevs, devs_min); |
4117f207 | 4890 | } |
73c5de00 | 4891 | goto error; |
b2117a39 | 4892 | } |
9f680ce0 | 4893 | |
f148ef4d NB |
4894 | ndevs = min(ndevs, devs_max); |
4895 | ||
73c5de00 | 4896 | /* |
92e222df HK |
4897 | * The primary goal is to maximize the number of stripes, so use as |
4898 | * many devices as possible, even if the stripes are not maximum sized. | |
4899 | * | |
4900 | * The DUP profile stores more than one stripe per device, the | |
4901 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4902 | */ |
92e222df | 4903 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4904 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4905 | |
53b381b3 DW |
4906 | /* |
4907 | * this will have to be fixed for RAID1 and RAID10 over | |
4908 | * more drives | |
4909 | */ | |
b50836ed | 4910 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
4911 | |
4912 | /* | |
4913 | * Use the number of data stripes to figure out how big this chunk | |
4914 | * is really going to be in terms of logical address space, | |
baf92114 HK |
4915 | * and compare that answer with the max chunk size. If it's higher, |
4916 | * we try to reduce stripe_size. | |
86db2578 CM |
4917 | */ |
4918 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 4919 | /* |
baf92114 HK |
4920 | * Reduce stripe_size, round it up to a 16MB boundary again and |
4921 | * then use it, unless it ends up being even bigger than the | |
4922 | * previous value we had already. | |
86db2578 | 4923 | */ |
baf92114 HK |
4924 | stripe_size = min(round_up(div_u64(max_chunk_size, |
4925 | data_stripes), SZ_16M), | |
793ff2c8 | 4926 | stripe_size); |
86db2578 CM |
4927 | } |
4928 | ||
37db63a4 | 4929 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4930 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4931 | |
4932 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4933 | if (!map) { | |
4934 | ret = -ENOMEM; | |
4935 | goto error; | |
4936 | } | |
4937 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4938 | |
73c5de00 AJ |
4939 | for (i = 0; i < ndevs; ++i) { |
4940 | for (j = 0; j < dev_stripes; ++j) { | |
4941 | int s = i * dev_stripes + j; | |
4942 | map->stripes[s].dev = devices_info[i].dev; | |
4943 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4944 | j * stripe_size; | |
6324fbf3 | 4945 | } |
6324fbf3 | 4946 | } |
500ceed8 NB |
4947 | map->stripe_len = BTRFS_STRIPE_LEN; |
4948 | map->io_align = BTRFS_STRIPE_LEN; | |
4949 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4950 | map->type = type; |
2b82032c | 4951 | map->sub_stripes = sub_stripes; |
0b86a832 | 4952 | |
23f0ff1e | 4953 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 4954 | |
23f0ff1e | 4955 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 4956 | |
172ddd60 | 4957 | em = alloc_extent_map(); |
2b82032c | 4958 | if (!em) { |
298a8f9c | 4959 | kfree(map); |
b2117a39 MX |
4960 | ret = -ENOMEM; |
4961 | goto error; | |
593060d7 | 4962 | } |
298a8f9c | 4963 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4964 | em->map_lookup = map; |
2b82032c | 4965 | em->start = start; |
23f0ff1e | 4966 | em->len = chunk_size; |
2b82032c YZ |
4967 | em->block_start = 0; |
4968 | em->block_len = em->len; | |
6df9a95e | 4969 | em->orig_block_len = stripe_size; |
593060d7 | 4970 | |
c8bf1b67 | 4971 | em_tree = &info->mapping_tree; |
890871be | 4972 | write_lock(&em_tree->lock); |
09a2a8f9 | 4973 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 4974 | if (ret) { |
1efb72a3 | 4975 | write_unlock(&em_tree->lock); |
0f5d42b2 | 4976 | free_extent_map(em); |
1dd4602f | 4977 | goto error; |
0f5d42b2 | 4978 | } |
1efb72a3 NB |
4979 | write_unlock(&em_tree->lock); |
4980 | ||
23f0ff1e | 4981 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
4982 | if (ret) |
4983 | goto error_del_extent; | |
2b82032c | 4984 | |
bbbf7243 NB |
4985 | for (i = 0; i < map->num_stripes; i++) { |
4986 | struct btrfs_device *dev = map->stripes[i].dev; | |
4987 | ||
4988 | btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size); | |
4989 | if (list_empty(&dev->post_commit_list)) | |
4990 | list_add_tail(&dev->post_commit_list, | |
4991 | &trans->transaction->dev_update_list); | |
4992 | } | |
43530c46 | 4993 | |
a5ed45f8 | 4994 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 4995 | |
0f5d42b2 | 4996 | free_extent_map(em); |
0b246afa | 4997 | check_raid56_incompat_flag(info, type); |
cfbb825c | 4998 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 4999 | |
b2117a39 | 5000 | kfree(devices_info); |
2b82032c | 5001 | return 0; |
b2117a39 | 5002 | |
6df9a95e | 5003 | error_del_extent: |
0f5d42b2 JB |
5004 | write_lock(&em_tree->lock); |
5005 | remove_extent_mapping(em_tree, em); | |
5006 | write_unlock(&em_tree->lock); | |
5007 | ||
5008 | /* One for our allocation */ | |
5009 | free_extent_map(em); | |
5010 | /* One for the tree reference */ | |
5011 | free_extent_map(em); | |
b2117a39 | 5012 | error: |
b2117a39 MX |
5013 | kfree(devices_info); |
5014 | return ret; | |
2b82032c YZ |
5015 | } |
5016 | ||
6df9a95e | 5017 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5018 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5019 | { |
97aff912 | 5020 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5021 | struct btrfs_root *extent_root = fs_info->extent_root; |
5022 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5023 | struct btrfs_key key; |
2b82032c YZ |
5024 | struct btrfs_device *device; |
5025 | struct btrfs_chunk *chunk; | |
5026 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5027 | struct extent_map *em; |
5028 | struct map_lookup *map; | |
5029 | size_t item_size; | |
5030 | u64 dev_offset; | |
5031 | u64 stripe_size; | |
5032 | int i = 0; | |
140e639f | 5033 | int ret = 0; |
2b82032c | 5034 | |
60ca842e | 5035 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5036 | if (IS_ERR(em)) |
5037 | return PTR_ERR(em); | |
6df9a95e | 5038 | |
95617d69 | 5039 | map = em->map_lookup; |
6df9a95e JB |
5040 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5041 | stripe_size = em->orig_block_len; | |
5042 | ||
2b82032c | 5043 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5044 | if (!chunk) { |
5045 | ret = -ENOMEM; | |
5046 | goto out; | |
5047 | } | |
5048 | ||
50460e37 FM |
5049 | /* |
5050 | * Take the device list mutex to prevent races with the final phase of | |
5051 | * a device replace operation that replaces the device object associated | |
5052 | * with the map's stripes, because the device object's id can change | |
5053 | * at any time during that final phase of the device replace operation | |
5054 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5055 | */ | |
0b246afa | 5056 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5057 | for (i = 0; i < map->num_stripes; i++) { |
5058 | device = map->stripes[i].dev; | |
5059 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5060 | |
0b86a832 | 5061 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5062 | if (ret) |
50460e37 | 5063 | break; |
b5d9071c NB |
5064 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5065 | dev_offset, stripe_size); | |
6df9a95e | 5066 | if (ret) |
50460e37 FM |
5067 | break; |
5068 | } | |
5069 | if (ret) { | |
0b246afa | 5070 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5071 | goto out; |
2b82032c YZ |
5072 | } |
5073 | ||
2b82032c | 5074 | stripe = &chunk->stripe; |
6df9a95e JB |
5075 | for (i = 0; i < map->num_stripes; i++) { |
5076 | device = map->stripes[i].dev; | |
5077 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5078 | |
e17cade2 CM |
5079 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5080 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5081 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5082 | stripe++; |
0b86a832 | 5083 | } |
0b246afa | 5084 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5085 | |
2b82032c | 5086 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5087 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5088 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5089 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5090 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5091 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5092 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5093 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5094 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5095 | |
2b82032c YZ |
5096 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5097 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5098 | key.offset = chunk_offset; | |
0b86a832 | 5099 | |
2b82032c | 5100 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5101 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5102 | /* | |
5103 | * TODO: Cleanup of inserted chunk root in case of | |
5104 | * failure. | |
5105 | */ | |
2ff7e61e | 5106 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5107 | } |
1abe9b8a | 5108 | |
6df9a95e | 5109 | out: |
0b86a832 | 5110 | kfree(chunk); |
6df9a95e | 5111 | free_extent_map(em); |
4ed1d16e | 5112 | return ret; |
2b82032c | 5113 | } |
0b86a832 | 5114 | |
2b82032c | 5115 | /* |
52042d8e AG |
5116 | * Chunk allocation falls into two parts. The first part does work |
5117 | * that makes the new allocated chunk usable, but does not do any operation | |
5118 | * that modifies the chunk tree. The second part does the work that | |
5119 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5120 | * bootstrap process of adding storage to a seed btrfs. |
5121 | */ | |
c216b203 | 5122 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5123 | { |
5124 | u64 chunk_offset; | |
2b82032c | 5125 | |
c216b203 NB |
5126 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5127 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5128 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5129 | } |
5130 | ||
6f8e0fc7 | 5131 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5132 | { |
6f8e0fc7 | 5133 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c YZ |
5134 | u64 chunk_offset; |
5135 | u64 sys_chunk_offset; | |
2b82032c | 5136 | u64 alloc_profile; |
2b82032c YZ |
5137 | int ret; |
5138 | ||
6df9a95e | 5139 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5140 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5141 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5142 | if (ret) |
5143 | return ret; | |
2b82032c | 5144 | |
0b246afa | 5145 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5146 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5147 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5148 | return ret; |
2b82032c YZ |
5149 | } |
5150 | ||
d20983b4 MX |
5151 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5152 | { | |
fc9a2ac7 | 5153 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5154 | |
fc9a2ac7 | 5155 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5156 | } |
5157 | ||
2ff7e61e | 5158 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5159 | { |
5160 | struct extent_map *em; | |
5161 | struct map_lookup *map; | |
2b82032c | 5162 | int readonly = 0; |
d20983b4 | 5163 | int miss_ndevs = 0; |
2b82032c YZ |
5164 | int i; |
5165 | ||
60ca842e | 5166 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5167 | if (IS_ERR(em)) |
2b82032c YZ |
5168 | return 1; |
5169 | ||
95617d69 | 5170 | map = em->map_lookup; |
2b82032c | 5171 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5172 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5173 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5174 | miss_ndevs++; |
5175 | continue; | |
5176 | } | |
ebbede42 AJ |
5177 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5178 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5179 | readonly = 1; |
d20983b4 | 5180 | goto end; |
2b82032c YZ |
5181 | } |
5182 | } | |
d20983b4 MX |
5183 | |
5184 | /* | |
5185 | * If the number of missing devices is larger than max errors, | |
5186 | * we can not write the data into that chunk successfully, so | |
5187 | * set it readonly. | |
5188 | */ | |
5189 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5190 | readonly = 1; | |
5191 | end: | |
0b86a832 | 5192 | free_extent_map(em); |
2b82032c | 5193 | return readonly; |
0b86a832 CM |
5194 | } |
5195 | ||
c8bf1b67 | 5196 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5197 | { |
5198 | struct extent_map *em; | |
5199 | ||
d397712b | 5200 | while (1) { |
c8bf1b67 DS |
5201 | write_lock(&tree->lock); |
5202 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5203 | if (em) |
c8bf1b67 DS |
5204 | remove_extent_mapping(tree, em); |
5205 | write_unlock(&tree->lock); | |
0b86a832 CM |
5206 | if (!em) |
5207 | break; | |
0b86a832 CM |
5208 | /* once for us */ |
5209 | free_extent_map(em); | |
5210 | /* once for the tree */ | |
5211 | free_extent_map(em); | |
5212 | } | |
5213 | } | |
5214 | ||
5d964051 | 5215 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5216 | { |
5217 | struct extent_map *em; | |
5218 | struct map_lookup *map; | |
f188591e CM |
5219 | int ret; |
5220 | ||
60ca842e | 5221 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5222 | if (IS_ERR(em)) |
5223 | /* | |
5224 | * We could return errors for these cases, but that could get | |
5225 | * ugly and we'd probably do the same thing which is just not do | |
5226 | * anything else and exit, so return 1 so the callers don't try | |
5227 | * to use other copies. | |
5228 | */ | |
fb7669b5 | 5229 | return 1; |
fb7669b5 | 5230 | |
95617d69 | 5231 | map = em->map_lookup; |
c7369b3f | 5232 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5233 | ret = map->num_stripes; |
321aecc6 CM |
5234 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5235 | ret = map->sub_stripes; | |
53b381b3 DW |
5236 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5237 | ret = 2; | |
5238 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5239 | /* |
5240 | * There could be two corrupted data stripes, we need | |
5241 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5242 | * |
8810f751 LB |
5243 | * Fail a stripe at a time on every retry except the |
5244 | * stripe under reconstruction. | |
5245 | */ | |
5246 | ret = map->num_stripes; | |
f188591e CM |
5247 | else |
5248 | ret = 1; | |
5249 | free_extent_map(em); | |
ad6d620e | 5250 | |
cb5583dd | 5251 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5252 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5253 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5254 | ret++; |
cb5583dd | 5255 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5256 | |
f188591e CM |
5257 | return ret; |
5258 | } | |
5259 | ||
2ff7e61e | 5260 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5261 | u64 logical) |
5262 | { | |
5263 | struct extent_map *em; | |
5264 | struct map_lookup *map; | |
0b246afa | 5265 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5266 | |
60ca842e | 5267 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5268 | |
69f03f13 NB |
5269 | if (!WARN_ON(IS_ERR(em))) { |
5270 | map = em->map_lookup; | |
5271 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5272 | len = map->stripe_len * nr_data_stripes(map); | |
5273 | free_extent_map(em); | |
5274 | } | |
53b381b3 DW |
5275 | return len; |
5276 | } | |
5277 | ||
e4ff5fb5 | 5278 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5279 | { |
5280 | struct extent_map *em; | |
5281 | struct map_lookup *map; | |
53b381b3 DW |
5282 | int ret = 0; |
5283 | ||
60ca842e | 5284 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5285 | |
69f03f13 NB |
5286 | if(!WARN_ON(IS_ERR(em))) { |
5287 | map = em->map_lookup; | |
5288 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5289 | ret = 1; | |
5290 | free_extent_map(em); | |
5291 | } | |
53b381b3 DW |
5292 | return ret; |
5293 | } | |
5294 | ||
30d9861f | 5295 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5296 | struct map_lookup *map, int first, |
8ba0ae78 | 5297 | int dev_replace_is_ongoing) |
dfe25020 CM |
5298 | { |
5299 | int i; | |
99f92a7c | 5300 | int num_stripes; |
8ba0ae78 | 5301 | int preferred_mirror; |
30d9861f SB |
5302 | int tolerance; |
5303 | struct btrfs_device *srcdev; | |
5304 | ||
99f92a7c | 5305 | ASSERT((map->type & |
c7369b3f | 5306 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5307 | |
5308 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5309 | num_stripes = map->sub_stripes; | |
5310 | else | |
5311 | num_stripes = map->num_stripes; | |
5312 | ||
8ba0ae78 AJ |
5313 | preferred_mirror = first + current->pid % num_stripes; |
5314 | ||
30d9861f SB |
5315 | if (dev_replace_is_ongoing && |
5316 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5317 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5318 | srcdev = fs_info->dev_replace.srcdev; | |
5319 | else | |
5320 | srcdev = NULL; | |
5321 | ||
5322 | /* | |
5323 | * try to avoid the drive that is the source drive for a | |
5324 | * dev-replace procedure, only choose it if no other non-missing | |
5325 | * mirror is available | |
5326 | */ | |
5327 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5328 | if (map->stripes[preferred_mirror].dev->bdev && |
5329 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5330 | return preferred_mirror; | |
99f92a7c | 5331 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5332 | if (map->stripes[i].dev->bdev && |
5333 | (tolerance || map->stripes[i].dev != srcdev)) | |
5334 | return i; | |
5335 | } | |
dfe25020 | 5336 | } |
30d9861f | 5337 | |
dfe25020 CM |
5338 | /* we couldn't find one that doesn't fail. Just return something |
5339 | * and the io error handling code will clean up eventually | |
5340 | */ | |
8ba0ae78 | 5341 | return preferred_mirror; |
dfe25020 CM |
5342 | } |
5343 | ||
53b381b3 DW |
5344 | static inline int parity_smaller(u64 a, u64 b) |
5345 | { | |
5346 | return a > b; | |
5347 | } | |
5348 | ||
5349 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5350 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5351 | { |
5352 | struct btrfs_bio_stripe s; | |
5353 | int i; | |
5354 | u64 l; | |
5355 | int again = 1; | |
5356 | ||
5357 | while (again) { | |
5358 | again = 0; | |
cc7539ed | 5359 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5360 | if (parity_smaller(bbio->raid_map[i], |
5361 | bbio->raid_map[i+1])) { | |
53b381b3 | 5362 | s = bbio->stripes[i]; |
8e5cfb55 | 5363 | l = bbio->raid_map[i]; |
53b381b3 | 5364 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5365 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5366 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5367 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5368 | |
53b381b3 DW |
5369 | again = 1; |
5370 | } | |
5371 | } | |
5372 | } | |
5373 | } | |
5374 | ||
6e9606d2 ZL |
5375 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5376 | { | |
5377 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5378 | /* the size of the btrfs_bio */ |
6e9606d2 | 5379 | sizeof(struct btrfs_bio) + |
e57cf21e | 5380 | /* plus the variable array for the stripes */ |
6e9606d2 | 5381 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5382 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5383 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5384 | /* |
5385 | * plus the raid_map, which includes both the tgt dev | |
5386 | * and the stripes | |
5387 | */ | |
5388 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5389 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5390 | |
5391 | atomic_set(&bbio->error, 0); | |
140475ae | 5392 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5393 | |
5394 | return bbio; | |
5395 | } | |
5396 | ||
5397 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5398 | { | |
140475ae ER |
5399 | WARN_ON(!refcount_read(&bbio->refs)); |
5400 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5401 | } |
5402 | ||
5403 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5404 | { | |
5405 | if (!bbio) | |
5406 | return; | |
140475ae | 5407 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5408 | kfree(bbio); |
5409 | } | |
5410 | ||
0b3d4cd3 LB |
5411 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5412 | /* | |
5413 | * Please note that, discard won't be sent to target device of device | |
5414 | * replace. | |
5415 | */ | |
5416 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5417 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5418 | struct btrfs_bio **bbio_ret) |
5419 | { | |
5420 | struct extent_map *em; | |
5421 | struct map_lookup *map; | |
5422 | struct btrfs_bio *bbio; | |
6b7faadd | 5423 | u64 length = *length_ret; |
0b3d4cd3 LB |
5424 | u64 offset; |
5425 | u64 stripe_nr; | |
5426 | u64 stripe_nr_end; | |
5427 | u64 stripe_end_offset; | |
5428 | u64 stripe_cnt; | |
5429 | u64 stripe_len; | |
5430 | u64 stripe_offset; | |
5431 | u64 num_stripes; | |
5432 | u32 stripe_index; | |
5433 | u32 factor = 0; | |
5434 | u32 sub_stripes = 0; | |
5435 | u64 stripes_per_dev = 0; | |
5436 | u32 remaining_stripes = 0; | |
5437 | u32 last_stripe = 0; | |
5438 | int ret = 0; | |
5439 | int i; | |
5440 | ||
5441 | /* discard always return a bbio */ | |
5442 | ASSERT(bbio_ret); | |
5443 | ||
60ca842e | 5444 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5445 | if (IS_ERR(em)) |
5446 | return PTR_ERR(em); | |
5447 | ||
5448 | map = em->map_lookup; | |
5449 | /* we don't discard raid56 yet */ | |
5450 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5451 | ret = -EOPNOTSUPP; | |
5452 | goto out; | |
5453 | } | |
5454 | ||
5455 | offset = logical - em->start; | |
2d974619 | 5456 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5457 | *length_ret = length; |
0b3d4cd3 LB |
5458 | |
5459 | stripe_len = map->stripe_len; | |
5460 | /* | |
5461 | * stripe_nr counts the total number of stripes we have to stride | |
5462 | * to get to this block | |
5463 | */ | |
5464 | stripe_nr = div64_u64(offset, stripe_len); | |
5465 | ||
5466 | /* stripe_offset is the offset of this block in its stripe */ | |
5467 | stripe_offset = offset - stripe_nr * stripe_len; | |
5468 | ||
5469 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5470 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5471 | stripe_cnt = stripe_nr_end - stripe_nr; |
5472 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5473 | (offset + length); | |
5474 | /* | |
5475 | * after this, stripe_nr is the number of stripes on this | |
5476 | * device we have to walk to find the data, and stripe_index is | |
5477 | * the number of our device in the stripe array | |
5478 | */ | |
5479 | num_stripes = 1; | |
5480 | stripe_index = 0; | |
5481 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5482 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5483 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5484 | sub_stripes = 1; | |
5485 | else | |
5486 | sub_stripes = map->sub_stripes; | |
5487 | ||
5488 | factor = map->num_stripes / sub_stripes; | |
5489 | num_stripes = min_t(u64, map->num_stripes, | |
5490 | sub_stripes * stripe_cnt); | |
5491 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5492 | stripe_index *= sub_stripes; | |
5493 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5494 | &remaining_stripes); | |
5495 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5496 | last_stripe *= sub_stripes; | |
c7369b3f | 5497 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5498 | BTRFS_BLOCK_GROUP_DUP)) { |
5499 | num_stripes = map->num_stripes; | |
5500 | } else { | |
5501 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5502 | &stripe_index); | |
5503 | } | |
5504 | ||
5505 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5506 | if (!bbio) { | |
5507 | ret = -ENOMEM; | |
5508 | goto out; | |
5509 | } | |
5510 | ||
5511 | for (i = 0; i < num_stripes; i++) { | |
5512 | bbio->stripes[i].physical = | |
5513 | map->stripes[stripe_index].physical + | |
5514 | stripe_offset + stripe_nr * map->stripe_len; | |
5515 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5516 | ||
5517 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5518 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5519 | bbio->stripes[i].length = stripes_per_dev * | |
5520 | map->stripe_len; | |
5521 | ||
5522 | if (i / sub_stripes < remaining_stripes) | |
5523 | bbio->stripes[i].length += | |
5524 | map->stripe_len; | |
5525 | ||
5526 | /* | |
5527 | * Special for the first stripe and | |
5528 | * the last stripe: | |
5529 | * | |
5530 | * |-------|...|-------| | |
5531 | * |----------| | |
5532 | * off end_off | |
5533 | */ | |
5534 | if (i < sub_stripes) | |
5535 | bbio->stripes[i].length -= | |
5536 | stripe_offset; | |
5537 | ||
5538 | if (stripe_index >= last_stripe && | |
5539 | stripe_index <= (last_stripe + | |
5540 | sub_stripes - 1)) | |
5541 | bbio->stripes[i].length -= | |
5542 | stripe_end_offset; | |
5543 | ||
5544 | if (i == sub_stripes - 1) | |
5545 | stripe_offset = 0; | |
5546 | } else { | |
5547 | bbio->stripes[i].length = length; | |
5548 | } | |
5549 | ||
5550 | stripe_index++; | |
5551 | if (stripe_index == map->num_stripes) { | |
5552 | stripe_index = 0; | |
5553 | stripe_nr++; | |
5554 | } | |
5555 | } | |
5556 | ||
5557 | *bbio_ret = bbio; | |
5558 | bbio->map_type = map->type; | |
5559 | bbio->num_stripes = num_stripes; | |
5560 | out: | |
5561 | free_extent_map(em); | |
5562 | return ret; | |
5563 | } | |
5564 | ||
5ab56090 LB |
5565 | /* |
5566 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5567 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5568 | * left cursor can also be read from the target drive. | |
5569 | * | |
5570 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5571 | * array of stripes. | |
5572 | * For READ, it also needs to be supported using the same mirror number. | |
5573 | * | |
5574 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5575 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5576 | * case. | |
5577 | */ | |
5578 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5579 | u64 logical, u64 length, | |
5580 | u64 srcdev_devid, int *mirror_num, | |
5581 | u64 *physical) | |
5582 | { | |
5583 | struct btrfs_bio *bbio = NULL; | |
5584 | int num_stripes; | |
5585 | int index_srcdev = 0; | |
5586 | int found = 0; | |
5587 | u64 physical_of_found = 0; | |
5588 | int i; | |
5589 | int ret = 0; | |
5590 | ||
5591 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5592 | logical, &length, &bbio, 0, 0); | |
5593 | if (ret) { | |
5594 | ASSERT(bbio == NULL); | |
5595 | return ret; | |
5596 | } | |
5597 | ||
5598 | num_stripes = bbio->num_stripes; | |
5599 | if (*mirror_num > num_stripes) { | |
5600 | /* | |
5601 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5602 | * that means that the requested area is not left of the left | |
5603 | * cursor | |
5604 | */ | |
5605 | btrfs_put_bbio(bbio); | |
5606 | return -EIO; | |
5607 | } | |
5608 | ||
5609 | /* | |
5610 | * process the rest of the function using the mirror_num of the source | |
5611 | * drive. Therefore look it up first. At the end, patch the device | |
5612 | * pointer to the one of the target drive. | |
5613 | */ | |
5614 | for (i = 0; i < num_stripes; i++) { | |
5615 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5616 | continue; | |
5617 | ||
5618 | /* | |
5619 | * In case of DUP, in order to keep it simple, only add the | |
5620 | * mirror with the lowest physical address | |
5621 | */ | |
5622 | if (found && | |
5623 | physical_of_found <= bbio->stripes[i].physical) | |
5624 | continue; | |
5625 | ||
5626 | index_srcdev = i; | |
5627 | found = 1; | |
5628 | physical_of_found = bbio->stripes[i].physical; | |
5629 | } | |
5630 | ||
5631 | btrfs_put_bbio(bbio); | |
5632 | ||
5633 | ASSERT(found); | |
5634 | if (!found) | |
5635 | return -EIO; | |
5636 | ||
5637 | *mirror_num = index_srcdev + 1; | |
5638 | *physical = physical_of_found; | |
5639 | return ret; | |
5640 | } | |
5641 | ||
73c0f228 LB |
5642 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5643 | struct btrfs_bio **bbio_ret, | |
5644 | struct btrfs_dev_replace *dev_replace, | |
5645 | int *num_stripes_ret, int *max_errors_ret) | |
5646 | { | |
5647 | struct btrfs_bio *bbio = *bbio_ret; | |
5648 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5649 | int tgtdev_indexes = 0; | |
5650 | int num_stripes = *num_stripes_ret; | |
5651 | int max_errors = *max_errors_ret; | |
5652 | int i; | |
5653 | ||
5654 | if (op == BTRFS_MAP_WRITE) { | |
5655 | int index_where_to_add; | |
5656 | ||
5657 | /* | |
5658 | * duplicate the write operations while the dev replace | |
5659 | * procedure is running. Since the copying of the old disk to | |
5660 | * the new disk takes place at run time while the filesystem is | |
5661 | * mounted writable, the regular write operations to the old | |
5662 | * disk have to be duplicated to go to the new disk as well. | |
5663 | * | |
5664 | * Note that device->missing is handled by the caller, and that | |
5665 | * the write to the old disk is already set up in the stripes | |
5666 | * array. | |
5667 | */ | |
5668 | index_where_to_add = num_stripes; | |
5669 | for (i = 0; i < num_stripes; i++) { | |
5670 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5671 | /* write to new disk, too */ | |
5672 | struct btrfs_bio_stripe *new = | |
5673 | bbio->stripes + index_where_to_add; | |
5674 | struct btrfs_bio_stripe *old = | |
5675 | bbio->stripes + i; | |
5676 | ||
5677 | new->physical = old->physical; | |
5678 | new->length = old->length; | |
5679 | new->dev = dev_replace->tgtdev; | |
5680 | bbio->tgtdev_map[i] = index_where_to_add; | |
5681 | index_where_to_add++; | |
5682 | max_errors++; | |
5683 | tgtdev_indexes++; | |
5684 | } | |
5685 | } | |
5686 | num_stripes = index_where_to_add; | |
5687 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5688 | int index_srcdev = 0; | |
5689 | int found = 0; | |
5690 | u64 physical_of_found = 0; | |
5691 | ||
5692 | /* | |
5693 | * During the dev-replace procedure, the target drive can also | |
5694 | * be used to read data in case it is needed to repair a corrupt | |
5695 | * block elsewhere. This is possible if the requested area is | |
5696 | * left of the left cursor. In this area, the target drive is a | |
5697 | * full copy of the source drive. | |
5698 | */ | |
5699 | for (i = 0; i < num_stripes; i++) { | |
5700 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5701 | /* | |
5702 | * In case of DUP, in order to keep it simple, | |
5703 | * only add the mirror with the lowest physical | |
5704 | * address | |
5705 | */ | |
5706 | if (found && | |
5707 | physical_of_found <= | |
5708 | bbio->stripes[i].physical) | |
5709 | continue; | |
5710 | index_srcdev = i; | |
5711 | found = 1; | |
5712 | physical_of_found = bbio->stripes[i].physical; | |
5713 | } | |
5714 | } | |
5715 | if (found) { | |
5716 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5717 | bbio->stripes + num_stripes; | |
5718 | ||
5719 | tgtdev_stripe->physical = physical_of_found; | |
5720 | tgtdev_stripe->length = | |
5721 | bbio->stripes[index_srcdev].length; | |
5722 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5723 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5724 | ||
5725 | tgtdev_indexes++; | |
5726 | num_stripes++; | |
5727 | } | |
5728 | } | |
5729 | ||
5730 | *num_stripes_ret = num_stripes; | |
5731 | *max_errors_ret = max_errors; | |
5732 | bbio->num_tgtdevs = tgtdev_indexes; | |
5733 | *bbio_ret = bbio; | |
5734 | } | |
5735 | ||
2b19a1fe LB |
5736 | static bool need_full_stripe(enum btrfs_map_op op) |
5737 | { | |
5738 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5739 | } | |
5740 | ||
5f141126 NB |
5741 | /* |
5742 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5743 | * tuple. This information is used to calculate how big a | |
5744 | * particular bio can get before it straddles a stripe. | |
5745 | * | |
5746 | * @fs_info - the filesystem | |
5747 | * @logical - address that we want to figure out the geometry of | |
5748 | * @len - the length of IO we are going to perform, starting at @logical | |
5749 | * @op - type of operation - write or read | |
5750 | * @io_geom - pointer used to return values | |
5751 | * | |
5752 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5753 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5754 | */ | |
5755 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5756 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5757 | { |
5758 | struct extent_map *em; | |
5759 | struct map_lookup *map; | |
5760 | u64 offset; | |
5761 | u64 stripe_offset; | |
5762 | u64 stripe_nr; | |
5763 | u64 stripe_len; | |
5764 | u64 raid56_full_stripe_start = (u64)-1; | |
5765 | int data_stripes; | |
373c3b80 | 5766 | int ret = 0; |
5f141126 NB |
5767 | |
5768 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5769 | ||
5770 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5771 | if (IS_ERR(em)) | |
5772 | return PTR_ERR(em); | |
5773 | ||
5774 | map = em->map_lookup; | |
5775 | /* Offset of this logical address in the chunk */ | |
5776 | offset = logical - em->start; | |
5777 | /* Len of a stripe in a chunk */ | |
5778 | stripe_len = map->stripe_len; | |
5779 | /* Stripe wher this block falls in */ | |
5780 | stripe_nr = div64_u64(offset, stripe_len); | |
5781 | /* Offset of stripe in the chunk */ | |
5782 | stripe_offset = stripe_nr * stripe_len; | |
5783 | if (offset < stripe_offset) { | |
5784 | btrfs_crit(fs_info, | |
5785 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5786 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5787 | ret = -EINVAL; |
5788 | goto out; | |
5f141126 NB |
5789 | } |
5790 | ||
5791 | /* stripe_offset is the offset of this block in its stripe */ | |
5792 | stripe_offset = offset - stripe_offset; | |
5793 | data_stripes = nr_data_stripes(map); | |
5794 | ||
5795 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5796 | u64 max_len = stripe_len - stripe_offset; | |
5797 | ||
5798 | /* | |
5799 | * In case of raid56, we need to know the stripe aligned start | |
5800 | */ | |
5801 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5802 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5803 | raid56_full_stripe_start = offset; | |
5804 | ||
5805 | /* | |
5806 | * Allow a write of a full stripe, but make sure we | |
5807 | * don't allow straddling of stripes | |
5808 | */ | |
5809 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5810 | full_stripe_len); | |
5811 | raid56_full_stripe_start *= full_stripe_len; | |
5812 | ||
5813 | /* | |
5814 | * For writes to RAID[56], allow a full stripeset across | |
5815 | * all disks. For other RAID types and for RAID[56] | |
5816 | * reads, just allow a single stripe (on a single disk). | |
5817 | */ | |
5818 | if (op == BTRFS_MAP_WRITE) { | |
5819 | max_len = stripe_len * data_stripes - | |
5820 | (offset - raid56_full_stripe_start); | |
5821 | } | |
5822 | } | |
5823 | len = min_t(u64, em->len - offset, max_len); | |
5824 | } else { | |
5825 | len = em->len - offset; | |
5826 | } | |
5827 | ||
5828 | io_geom->len = len; | |
5829 | io_geom->offset = offset; | |
5830 | io_geom->stripe_len = stripe_len; | |
5831 | io_geom->stripe_nr = stripe_nr; | |
5832 | io_geom->stripe_offset = stripe_offset; | |
5833 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
5834 | ||
373c3b80 JT |
5835 | out: |
5836 | /* once for us */ | |
5837 | free_extent_map(em); | |
5838 | return ret; | |
5f141126 NB |
5839 | } |
5840 | ||
cf8cddd3 CH |
5841 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5842 | enum btrfs_map_op op, | |
f2d8d74d | 5843 | u64 logical, u64 *length, |
a1d3c478 | 5844 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5845 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5846 | { |
5847 | struct extent_map *em; | |
5848 | struct map_lookup *map; | |
593060d7 CM |
5849 | u64 stripe_offset; |
5850 | u64 stripe_nr; | |
53b381b3 | 5851 | u64 stripe_len; |
9d644a62 | 5852 | u32 stripe_index; |
cff82672 | 5853 | int data_stripes; |
cea9e445 | 5854 | int i; |
de11cc12 | 5855 | int ret = 0; |
f2d8d74d | 5856 | int num_stripes; |
a236aed1 | 5857 | int max_errors = 0; |
2c8cdd6e | 5858 | int tgtdev_indexes = 0; |
a1d3c478 | 5859 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5860 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5861 | int dev_replace_is_ongoing = 0; | |
5862 | int num_alloc_stripes; | |
ad6d620e SB |
5863 | int patch_the_first_stripe_for_dev_replace = 0; |
5864 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5865 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
5866 | struct btrfs_io_geometry geom; |
5867 | ||
5868 | ASSERT(bbio_ret); | |
0b86a832 | 5869 | |
0b3d4cd3 LB |
5870 | if (op == BTRFS_MAP_DISCARD) |
5871 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6b7faadd | 5872 | length, bbio_ret); |
0b3d4cd3 | 5873 | |
89b798ad NB |
5874 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
5875 | if (ret < 0) | |
5876 | return ret; | |
0b86a832 | 5877 | |
89b798ad | 5878 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 5879 | ASSERT(!IS_ERR(em)); |
95617d69 | 5880 | map = em->map_lookup; |
593060d7 | 5881 | |
89b798ad | 5882 | *length = geom.len; |
89b798ad NB |
5883 | stripe_len = geom.stripe_len; |
5884 | stripe_nr = geom.stripe_nr; | |
5885 | stripe_offset = geom.stripe_offset; | |
5886 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 5887 | data_stripes = nr_data_stripes(map); |
593060d7 | 5888 | |
cb5583dd | 5889 | down_read(&dev_replace->rwsem); |
472262f3 | 5890 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
5891 | /* |
5892 | * Hold the semaphore for read during the whole operation, write is | |
5893 | * requested at commit time but must wait. | |
5894 | */ | |
472262f3 | 5895 | if (!dev_replace_is_ongoing) |
cb5583dd | 5896 | up_read(&dev_replace->rwsem); |
472262f3 | 5897 | |
ad6d620e | 5898 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5899 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5900 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5901 | dev_replace->srcdev->devid, | |
5902 | &mirror_num, | |
5903 | &physical_to_patch_in_first_stripe); | |
5904 | if (ret) | |
ad6d620e | 5905 | goto out; |
5ab56090 LB |
5906 | else |
5907 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5908 | } else if (mirror_num > map->num_stripes) { |
5909 | mirror_num = 0; | |
5910 | } | |
5911 | ||
f2d8d74d | 5912 | num_stripes = 1; |
cea9e445 | 5913 | stripe_index = 0; |
fce3bb9a | 5914 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5915 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5916 | &stripe_index); | |
de483734 | 5917 | if (!need_full_stripe(op)) |
28e1cc7d | 5918 | mirror_num = 1; |
c7369b3f | 5919 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 5920 | if (need_full_stripe(op)) |
f2d8d74d | 5921 | num_stripes = map->num_stripes; |
2fff734f | 5922 | else if (mirror_num) |
f188591e | 5923 | stripe_index = mirror_num - 1; |
dfe25020 | 5924 | else { |
30d9861f | 5925 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5926 | dev_replace_is_ongoing); |
a1d3c478 | 5927 | mirror_num = stripe_index + 1; |
dfe25020 | 5928 | } |
2fff734f | 5929 | |
611f0e00 | 5930 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5931 | if (need_full_stripe(op)) { |
f2d8d74d | 5932 | num_stripes = map->num_stripes; |
a1d3c478 | 5933 | } else if (mirror_num) { |
f188591e | 5934 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5935 | } else { |
5936 | mirror_num = 1; | |
5937 | } | |
2fff734f | 5938 | |
321aecc6 | 5939 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5940 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5941 | |
47c5713f | 5942 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5943 | stripe_index *= map->sub_stripes; |
5944 | ||
de483734 | 5945 | if (need_full_stripe(op)) |
f2d8d74d | 5946 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5947 | else if (mirror_num) |
5948 | stripe_index += mirror_num - 1; | |
dfe25020 | 5949 | else { |
3e74317a | 5950 | int old_stripe_index = stripe_index; |
30d9861f SB |
5951 | stripe_index = find_live_mirror(fs_info, map, |
5952 | stripe_index, | |
30d9861f | 5953 | dev_replace_is_ongoing); |
3e74317a | 5954 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5955 | } |
53b381b3 | 5956 | |
ffe2d203 | 5957 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 5958 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 5959 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 5960 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 5961 | stripe_len * data_stripes); |
53b381b3 DW |
5962 | |
5963 | /* RAID[56] write or recovery. Return all stripes */ | |
5964 | num_stripes = map->num_stripes; | |
5965 | max_errors = nr_parity_stripes(map); | |
5966 | ||
53b381b3 DW |
5967 | *length = map->stripe_len; |
5968 | stripe_index = 0; | |
5969 | stripe_offset = 0; | |
5970 | } else { | |
5971 | /* | |
5972 | * Mirror #0 or #1 means the original data block. | |
5973 | * Mirror #2 is RAID5 parity block. | |
5974 | * Mirror #3 is RAID6 Q block. | |
5975 | */ | |
47c5713f | 5976 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 5977 | data_stripes, &stripe_index); |
53b381b3 | 5978 | if (mirror_num > 1) |
cff82672 | 5979 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
5980 | |
5981 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5982 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5983 | &stripe_index); | |
de483734 | 5984 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 5985 | mirror_num = 1; |
53b381b3 | 5986 | } |
8790d502 CM |
5987 | } else { |
5988 | /* | |
47c5713f DS |
5989 | * after this, stripe_nr is the number of stripes on this |
5990 | * device we have to walk to find the data, and stripe_index is | |
5991 | * the number of our device in the stripe array | |
8790d502 | 5992 | */ |
47c5713f DS |
5993 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5994 | &stripe_index); | |
a1d3c478 | 5995 | mirror_num = stripe_index + 1; |
8790d502 | 5996 | } |
e042d1ec | 5997 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5998 | btrfs_crit(fs_info, |
5999 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6000 | stripe_index, map->num_stripes); |
6001 | ret = -EINVAL; | |
6002 | goto out; | |
6003 | } | |
cea9e445 | 6004 | |
472262f3 | 6005 | num_alloc_stripes = num_stripes; |
6fad823f | 6006 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6007 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6008 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6009 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6010 | num_alloc_stripes++; |
2c8cdd6e | 6011 | tgtdev_indexes = num_stripes; |
ad6d620e | 6012 | } |
2c8cdd6e | 6013 | |
6e9606d2 | 6014 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6015 | if (!bbio) { |
6016 | ret = -ENOMEM; | |
6017 | goto out; | |
6018 | } | |
6fad823f | 6019 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6020 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6021 | |
8e5cfb55 | 6022 | /* build raid_map */ |
2b19a1fe LB |
6023 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6024 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6025 | u64 tmp; |
9d644a62 | 6026 | unsigned rot; |
8e5cfb55 ZL |
6027 | |
6028 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6029 | sizeof(struct btrfs_bio_stripe) * | |
6030 | num_alloc_stripes + | |
6031 | sizeof(int) * tgtdev_indexes); | |
6032 | ||
6033 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6034 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6035 | |
6036 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6037 | tmp = stripe_nr * data_stripes; |
6038 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6039 | bbio->raid_map[(i+rot) % num_stripes] = |
6040 | em->start + (tmp + i) * map->stripe_len; | |
6041 | ||
6042 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6043 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6044 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6045 | RAID6_Q_STRIPE; | |
6046 | } | |
6047 | ||
b89203f7 | 6048 | |
0b3d4cd3 LB |
6049 | for (i = 0; i < num_stripes; i++) { |
6050 | bbio->stripes[i].physical = | |
6051 | map->stripes[stripe_index].physical + | |
6052 | stripe_offset + | |
6053 | stripe_nr * map->stripe_len; | |
6054 | bbio->stripes[i].dev = | |
6055 | map->stripes[stripe_index].dev; | |
6056 | stripe_index++; | |
593060d7 | 6057 | } |
de11cc12 | 6058 | |
2b19a1fe | 6059 | if (need_full_stripe(op)) |
d20983b4 | 6060 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6061 | |
8e5cfb55 ZL |
6062 | if (bbio->raid_map) |
6063 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6064 | |
73c0f228 | 6065 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6066 | need_full_stripe(op)) { |
73c0f228 LB |
6067 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6068 | &max_errors); | |
472262f3 SB |
6069 | } |
6070 | ||
de11cc12 | 6071 | *bbio_ret = bbio; |
10f11900 | 6072 | bbio->map_type = map->type; |
de11cc12 LZ |
6073 | bbio->num_stripes = num_stripes; |
6074 | bbio->max_errors = max_errors; | |
6075 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6076 | |
6077 | /* | |
6078 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6079 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6080 | * available as a mirror | |
6081 | */ | |
6082 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6083 | WARN_ON(num_stripes > 1); | |
6084 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6085 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6086 | bbio->mirror_num = map->num_stripes + 1; | |
6087 | } | |
cea9e445 | 6088 | out: |
73beece9 | 6089 | if (dev_replace_is_ongoing) { |
53176dde DS |
6090 | lockdep_assert_held(&dev_replace->rwsem); |
6091 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6092 | up_read(&dev_replace->rwsem); |
73beece9 | 6093 | } |
0b86a832 | 6094 | free_extent_map(em); |
de11cc12 | 6095 | return ret; |
0b86a832 CM |
6096 | } |
6097 | ||
cf8cddd3 | 6098 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6099 | u64 logical, u64 *length, |
a1d3c478 | 6100 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6101 | { |
b3d3fa51 | 6102 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6103 | mirror_num, 0); |
f2d8d74d CM |
6104 | } |
6105 | ||
af8e2d1d | 6106 | /* For Scrub/replace */ |
cf8cddd3 | 6107 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6108 | u64 logical, u64 *length, |
825ad4c9 | 6109 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6110 | { |
825ad4c9 | 6111 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6112 | } |
6113 | ||
63a9c7b9 NB |
6114 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6115 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6116 | { |
a512bbf8 YZ |
6117 | struct extent_map *em; |
6118 | struct map_lookup *map; | |
6119 | u64 *buf; | |
6120 | u64 bytenr; | |
6121 | u64 length; | |
6122 | u64 stripe_nr; | |
53b381b3 | 6123 | u64 rmap_len; |
a512bbf8 YZ |
6124 | int i, j, nr = 0; |
6125 | ||
60ca842e | 6126 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6127 | if (IS_ERR(em)) |
835d974f | 6128 | return -EIO; |
835d974f | 6129 | |
95617d69 | 6130 | map = em->map_lookup; |
a512bbf8 | 6131 | length = em->len; |
53b381b3 DW |
6132 | rmap_len = map->stripe_len; |
6133 | ||
a512bbf8 | 6134 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6135 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6136 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6137 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6138 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6139 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6140 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6141 | } | |
a512bbf8 | 6142 | |
31e818fe | 6143 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6144 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6145 | |
6146 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6147 | if (map->stripes[i].physical > physical || |
6148 | map->stripes[i].physical + length <= physical) | |
6149 | continue; | |
6150 | ||
6151 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6152 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6153 | |
6154 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6155 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6156 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6157 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6158 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6159 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6160 | * Alternatively, just use rmap_len below instead of | |
6161 | * map->stripe_len */ | |
6162 | ||
6163 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6164 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6165 | for (j = 0; j < nr; j++) { |
6166 | if (buf[j] == bytenr) | |
6167 | break; | |
6168 | } | |
934d375b CM |
6169 | if (j == nr) { |
6170 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6171 | buf[nr++] = bytenr; |
934d375b | 6172 | } |
a512bbf8 YZ |
6173 | } |
6174 | ||
a512bbf8 YZ |
6175 | *logical = buf; |
6176 | *naddrs = nr; | |
53b381b3 | 6177 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6178 | |
6179 | free_extent_map(em); | |
6180 | return 0; | |
f2d8d74d CM |
6181 | } |
6182 | ||
4246a0b6 | 6183 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6184 | { |
326e1dbb MS |
6185 | bio->bi_private = bbio->private; |
6186 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6187 | bio_endio(bio); |
326e1dbb | 6188 | |
6e9606d2 | 6189 | btrfs_put_bbio(bbio); |
8408c716 MX |
6190 | } |
6191 | ||
4246a0b6 | 6192 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6193 | { |
9be3395b | 6194 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6195 | int is_orig_bio = 0; |
8790d502 | 6196 | |
4e4cbee9 | 6197 | if (bio->bi_status) { |
a1d3c478 | 6198 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6199 | if (bio->bi_status == BLK_STS_IOERR || |
6200 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6201 | unsigned int stripe_index = |
9be3395b | 6202 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6203 | struct btrfs_device *dev; |
442a4f63 SB |
6204 | |
6205 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6206 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6207 | if (dev->bdev) { |
37226b21 | 6208 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6209 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6210 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6211 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6212 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6213 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6214 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6215 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6216 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6217 | } |
442a4f63 SB |
6218 | } |
6219 | } | |
8790d502 | 6220 | |
a1d3c478 | 6221 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6222 | is_orig_bio = 1; |
6223 | ||
c404e0dc MX |
6224 | btrfs_bio_counter_dec(bbio->fs_info); |
6225 | ||
a1d3c478 | 6226 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6227 | if (!is_orig_bio) { |
6228 | bio_put(bio); | |
a1d3c478 | 6229 | bio = bbio->orig_bio; |
7d2b4daa | 6230 | } |
c7b22bb1 | 6231 | |
9be3395b | 6232 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6233 | /* only send an error to the higher layers if it is |
53b381b3 | 6234 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6235 | */ |
a1d3c478 | 6236 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6237 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6238 | } else { |
1259ab75 CM |
6239 | /* |
6240 | * this bio is actually up to date, we didn't | |
6241 | * go over the max number of errors | |
6242 | */ | |
2dbe0c77 | 6243 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6244 | } |
c55f1396 | 6245 | |
4246a0b6 | 6246 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6247 | } else if (!is_orig_bio) { |
8790d502 CM |
6248 | bio_put(bio); |
6249 | } | |
8790d502 CM |
6250 | } |
6251 | ||
2ff7e61e | 6252 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
08635bae | 6253 | u64 physical, int dev_nr) |
de1ee92a JB |
6254 | { |
6255 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6256 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6257 | |
6258 | bio->bi_private = bbio; | |
9be3395b | 6259 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6260 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6261 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6262 | btrfs_debug_in_rcu(fs_info, |
6263 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6264 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6265 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6266 | bio->bi_iter.bi_size); | |
74d46992 | 6267 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6268 | |
2ff7e61e | 6269 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6270 | |
08635bae | 6271 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6272 | } |
6273 | ||
de1ee92a JB |
6274 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6275 | { | |
6276 | atomic_inc(&bbio->error); | |
6277 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6278 | /* Should be the original bio. */ |
8408c716 MX |
6279 | WARN_ON(bio != bbio->orig_bio); |
6280 | ||
9be3395b | 6281 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6282 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6283 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6284 | bio->bi_status = BLK_STS_IOERR; | |
6285 | else | |
6286 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6287 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6288 | } |
6289 | } | |
6290 | ||
58efbc9f | 6291 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6292 | int mirror_num) |
0b86a832 | 6293 | { |
0b86a832 | 6294 | struct btrfs_device *dev; |
8790d502 | 6295 | struct bio *first_bio = bio; |
4f024f37 | 6296 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6297 | u64 length = 0; |
6298 | u64 map_length; | |
0b86a832 | 6299 | int ret; |
08da757d ZL |
6300 | int dev_nr; |
6301 | int total_devs; | |
a1d3c478 | 6302 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6303 | |
4f024f37 | 6304 | length = bio->bi_iter.bi_size; |
0b86a832 | 6305 | map_length = length; |
cea9e445 | 6306 | |
0b246afa | 6307 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6308 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6309 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6310 | if (ret) { |
0b246afa | 6311 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6312 | return errno_to_blk_status(ret); |
c404e0dc | 6313 | } |
cea9e445 | 6314 | |
a1d3c478 | 6315 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6316 | bbio->orig_bio = first_bio; |
6317 | bbio->private = first_bio->bi_private; | |
6318 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6319 | bbio->fs_info = fs_info; |
53b381b3 DW |
6320 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6321 | ||
ad1ba2a0 | 6322 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6323 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6324 | /* In this case, map_length has been set to the length of |
6325 | a single stripe; not the whole write */ | |
37226b21 | 6326 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6327 | ret = raid56_parity_write(fs_info, bio, bbio, |
6328 | map_length); | |
53b381b3 | 6329 | } else { |
2ff7e61e JM |
6330 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6331 | map_length, mirror_num, 1); | |
53b381b3 | 6332 | } |
4245215d | 6333 | |
0b246afa | 6334 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6335 | return errno_to_blk_status(ret); |
53b381b3 DW |
6336 | } |
6337 | ||
cea9e445 | 6338 | if (map_length < length) { |
0b246afa | 6339 | btrfs_crit(fs_info, |
5d163e0e JM |
6340 | "mapping failed logical %llu bio len %llu len %llu", |
6341 | logical, length, map_length); | |
cea9e445 CM |
6342 | BUG(); |
6343 | } | |
a1d3c478 | 6344 | |
08da757d | 6345 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6346 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6347 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6348 | &dev->dev_state) || | |
ebbede42 AJ |
6349 | (bio_op(first_bio) == REQ_OP_WRITE && |
6350 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6351 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6352 | continue; |
6353 | } | |
6354 | ||
3aa8e074 | 6355 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6356 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6357 | else |
a1d3c478 | 6358 | bio = first_bio; |
de1ee92a | 6359 | |
2ff7e61e | 6360 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
08635bae | 6361 | dev_nr); |
8790d502 | 6362 | } |
0b246afa | 6363 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6364 | return BLK_STS_OK; |
0b86a832 CM |
6365 | } |
6366 | ||
09ba3bc9 AJ |
6367 | /* |
6368 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6369 | * return NULL. | |
6370 | * | |
6371 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6372 | * only devid is used. | |
6373 | * | |
6374 | * If @seed is true, traverse through the seed devices. | |
6375 | */ | |
e4319cd9 | 6376 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6377 | u64 devid, u8 *uuid, u8 *fsid, |
6378 | bool seed) | |
0b86a832 | 6379 | { |
2b82032c | 6380 | struct btrfs_device *device; |
2b82032c | 6381 | |
e4319cd9 | 6382 | while (fs_devices) { |
2b82032c | 6383 | if (!fsid || |
e4319cd9 | 6384 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6385 | list_for_each_entry(device, &fs_devices->devices, |
6386 | dev_list) { | |
6387 | if (device->devid == devid && | |
6388 | (!uuid || memcmp(device->uuid, uuid, | |
6389 | BTRFS_UUID_SIZE) == 0)) | |
6390 | return device; | |
6391 | } | |
2b82032c | 6392 | } |
09ba3bc9 AJ |
6393 | if (seed) |
6394 | fs_devices = fs_devices->seed; | |
6395 | else | |
6396 | return NULL; | |
2b82032c YZ |
6397 | } |
6398 | return NULL; | |
0b86a832 CM |
6399 | } |
6400 | ||
2ff7e61e | 6401 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6402 | u64 devid, u8 *dev_uuid) |
6403 | { | |
6404 | struct btrfs_device *device; | |
dfe25020 | 6405 | |
12bd2fc0 ID |
6406 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6407 | if (IS_ERR(device)) | |
adfb69af | 6408 | return device; |
12bd2fc0 ID |
6409 | |
6410 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6411 | device->fs_devices = fs_devices; |
dfe25020 | 6412 | fs_devices->num_devices++; |
12bd2fc0 | 6413 | |
e6e674bd | 6414 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6415 | fs_devices->missing_devices++; |
12bd2fc0 | 6416 | |
dfe25020 CM |
6417 | return device; |
6418 | } | |
6419 | ||
12bd2fc0 ID |
6420 | /** |
6421 | * btrfs_alloc_device - allocate struct btrfs_device | |
6422 | * @fs_info: used only for generating a new devid, can be NULL if | |
6423 | * devid is provided (i.e. @devid != NULL). | |
6424 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6425 | * is generated. | |
6426 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6427 | * is generated. | |
6428 | * | |
6429 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6430 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6431 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6432 | */ |
6433 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6434 | const u64 *devid, | |
6435 | const u8 *uuid) | |
6436 | { | |
6437 | struct btrfs_device *dev; | |
6438 | u64 tmp; | |
6439 | ||
fae7f21c | 6440 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6441 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6442 | |
6443 | dev = __alloc_device(); | |
6444 | if (IS_ERR(dev)) | |
6445 | return dev; | |
6446 | ||
6447 | if (devid) | |
6448 | tmp = *devid; | |
6449 | else { | |
6450 | int ret; | |
6451 | ||
6452 | ret = find_next_devid(fs_info, &tmp); | |
6453 | if (ret) { | |
a425f9d4 | 6454 | btrfs_free_device(dev); |
12bd2fc0 ID |
6455 | return ERR_PTR(ret); |
6456 | } | |
6457 | } | |
6458 | dev->devid = tmp; | |
6459 | ||
6460 | if (uuid) | |
6461 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6462 | else | |
6463 | generate_random_uuid(dev->uuid); | |
6464 | ||
12bd2fc0 ID |
6465 | return dev; |
6466 | } | |
6467 | ||
5a2b8e60 | 6468 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6469 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6470 | { |
2b902dfc AJ |
6471 | if (error) |
6472 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6473 | devid, uuid); | |
6474 | else | |
6475 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6476 | devid, uuid); | |
5a2b8e60 AJ |
6477 | } |
6478 | ||
39e264a4 NB |
6479 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6480 | { | |
6481 | int index = btrfs_bg_flags_to_raid_index(type); | |
6482 | int ncopies = btrfs_raid_array[index].ncopies; | |
e4f6c6be | 6483 | const int nparity = btrfs_raid_array[index].nparity; |
39e264a4 NB |
6484 | int data_stripes; |
6485 | ||
e4f6c6be DS |
6486 | if (nparity) |
6487 | data_stripes = num_stripes - nparity; | |
6488 | else | |
39e264a4 | 6489 | data_stripes = num_stripes / ncopies; |
e4f6c6be | 6490 | |
39e264a4 NB |
6491 | return div_u64(chunk_len, data_stripes); |
6492 | } | |
6493 | ||
9690ac09 | 6494 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6495 | struct btrfs_chunk *chunk) |
6496 | { | |
9690ac09 | 6497 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6498 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6499 | struct map_lookup *map; |
6500 | struct extent_map *em; | |
6501 | u64 logical; | |
6502 | u64 length; | |
e06cd3dd LB |
6503 | u64 devid; |
6504 | u8 uuid[BTRFS_UUID_SIZE]; | |
6505 | int num_stripes; | |
6506 | int ret; | |
6507 | int i; | |
6508 | ||
6509 | logical = key->offset; | |
6510 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6511 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6512 | ||
075cb3c7 QW |
6513 | /* |
6514 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6515 | * as chunk item in tree block is already verified by tree-checker. | |
6516 | */ | |
6517 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6518 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6519 | if (ret) |
6520 | return ret; | |
6521 | } | |
a061fc8d | 6522 | |
c8bf1b67 DS |
6523 | read_lock(&map_tree->lock); |
6524 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6525 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6526 | |
6527 | /* already mapped? */ | |
6528 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6529 | free_extent_map(em); | |
0b86a832 CM |
6530 | return 0; |
6531 | } else if (em) { | |
6532 | free_extent_map(em); | |
6533 | } | |
0b86a832 | 6534 | |
172ddd60 | 6535 | em = alloc_extent_map(); |
0b86a832 CM |
6536 | if (!em) |
6537 | return -ENOMEM; | |
593060d7 | 6538 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6539 | if (!map) { |
6540 | free_extent_map(em); | |
6541 | return -ENOMEM; | |
6542 | } | |
6543 | ||
298a8f9c | 6544 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6545 | em->map_lookup = map; |
0b86a832 CM |
6546 | em->start = logical; |
6547 | em->len = length; | |
70c8a91c | 6548 | em->orig_start = 0; |
0b86a832 | 6549 | em->block_start = 0; |
c8b97818 | 6550 | em->block_len = em->len; |
0b86a832 | 6551 | |
593060d7 CM |
6552 | map->num_stripes = num_stripes; |
6553 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6554 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6555 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6556 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6557 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6558 | map->verified_stripes = 0; |
39e264a4 NB |
6559 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6560 | map->num_stripes); | |
593060d7 CM |
6561 | for (i = 0; i < num_stripes; i++) { |
6562 | map->stripes[i].physical = | |
6563 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6564 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6565 | read_extent_buffer(leaf, uuid, (unsigned long) |
6566 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6567 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6568 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6569 | devid, uuid, NULL, true); |
3cdde224 | 6570 | if (!map->stripes[i].dev && |
0b246afa | 6571 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6572 | free_extent_map(em); |
2b902dfc | 6573 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6574 | return -ENOENT; |
593060d7 | 6575 | } |
dfe25020 CM |
6576 | if (!map->stripes[i].dev) { |
6577 | map->stripes[i].dev = | |
2ff7e61e JM |
6578 | add_missing_dev(fs_info->fs_devices, devid, |
6579 | uuid); | |
adfb69af | 6580 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6581 | free_extent_map(em); |
adfb69af AJ |
6582 | btrfs_err(fs_info, |
6583 | "failed to init missing dev %llu: %ld", | |
6584 | devid, PTR_ERR(map->stripes[i].dev)); | |
6585 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6586 | } |
2b902dfc | 6587 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6588 | } |
e12c9621 AJ |
6589 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6590 | &(map->stripes[i].dev->dev_state)); | |
6591 | ||
0b86a832 CM |
6592 | } |
6593 | ||
c8bf1b67 DS |
6594 | write_lock(&map_tree->lock); |
6595 | ret = add_extent_mapping(map_tree, em, 0); | |
6596 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6597 | if (ret < 0) { |
6598 | btrfs_err(fs_info, | |
6599 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6600 | em->start, em->len, ret); | |
6601 | } | |
0b86a832 CM |
6602 | free_extent_map(em); |
6603 | ||
64f64f43 | 6604 | return ret; |
0b86a832 CM |
6605 | } |
6606 | ||
143bede5 | 6607 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6608 | struct btrfs_dev_item *dev_item, |
6609 | struct btrfs_device *device) | |
6610 | { | |
6611 | unsigned long ptr; | |
0b86a832 CM |
6612 | |
6613 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6614 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6615 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6616 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6617 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6618 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6619 | device->type = btrfs_device_type(leaf, dev_item); |
6620 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6621 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6622 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6623 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6624 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6625 | |
410ba3a2 | 6626 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6627 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6628 | } |
6629 | ||
2ff7e61e | 6630 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6631 | u8 *fsid) |
2b82032c YZ |
6632 | { |
6633 | struct btrfs_fs_devices *fs_devices; | |
6634 | int ret; | |
6635 | ||
a32bf9a3 | 6636 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6637 | ASSERT(fsid); |
2b82032c | 6638 | |
0b246afa | 6639 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6640 | while (fs_devices) { |
44880fdc | 6641 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6642 | return fs_devices; |
6643 | ||
2b82032c YZ |
6644 | fs_devices = fs_devices->seed; |
6645 | } | |
6646 | ||
7239ff4b | 6647 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6648 | if (!fs_devices) { |
0b246afa | 6649 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6650 | return ERR_PTR(-ENOENT); |
6651 | ||
7239ff4b | 6652 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6653 | if (IS_ERR(fs_devices)) |
6654 | return fs_devices; | |
6655 | ||
0395d84f | 6656 | fs_devices->seeding = true; |
5f375835 MX |
6657 | fs_devices->opened = 1; |
6658 | return fs_devices; | |
2b82032c | 6659 | } |
e4404d6e YZ |
6660 | |
6661 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6662 | if (IS_ERR(fs_devices)) |
6663 | return fs_devices; | |
2b82032c | 6664 | |
897fb573 | 6665 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6666 | if (ret) { |
6667 | free_fs_devices(fs_devices); | |
5f375835 | 6668 | fs_devices = ERR_PTR(ret); |
2b82032c | 6669 | goto out; |
48d28232 | 6670 | } |
2b82032c YZ |
6671 | |
6672 | if (!fs_devices->seeding) { | |
0226e0eb | 6673 | close_fs_devices(fs_devices); |
e4404d6e | 6674 | free_fs_devices(fs_devices); |
5f375835 | 6675 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6676 | goto out; |
6677 | } | |
6678 | ||
0b246afa JM |
6679 | fs_devices->seed = fs_info->fs_devices->seed; |
6680 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6681 | out: |
5f375835 | 6682 | return fs_devices; |
2b82032c YZ |
6683 | } |
6684 | ||
17850759 | 6685 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6686 | struct btrfs_dev_item *dev_item) |
6687 | { | |
17850759 | 6688 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6689 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6690 | struct btrfs_device *device; |
6691 | u64 devid; | |
6692 | int ret; | |
44880fdc | 6693 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6694 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6695 | ||
0b86a832 | 6696 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6697 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6698 | BTRFS_UUID_SIZE); |
1473b24e | 6699 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6700 | BTRFS_FSID_SIZE); |
2b82032c | 6701 | |
de37aa51 | 6702 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6703 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6704 | if (IS_ERR(fs_devices)) |
6705 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6706 | } |
6707 | ||
e4319cd9 | 6708 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6709 | fs_uuid, true); |
5f375835 | 6710 | if (!device) { |
c5502451 | 6711 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6712 | btrfs_report_missing_device(fs_info, devid, |
6713 | dev_uuid, true); | |
45dbdbc9 | 6714 | return -ENOENT; |
c5502451 | 6715 | } |
2b82032c | 6716 | |
2ff7e61e | 6717 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6718 | if (IS_ERR(device)) { |
6719 | btrfs_err(fs_info, | |
6720 | "failed to add missing dev %llu: %ld", | |
6721 | devid, PTR_ERR(device)); | |
6722 | return PTR_ERR(device); | |
6723 | } | |
2b902dfc | 6724 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6725 | } else { |
c5502451 | 6726 | if (!device->bdev) { |
2b902dfc AJ |
6727 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6728 | btrfs_report_missing_device(fs_info, | |
6729 | devid, dev_uuid, true); | |
45dbdbc9 | 6730 | return -ENOENT; |
2b902dfc AJ |
6731 | } |
6732 | btrfs_report_missing_device(fs_info, devid, | |
6733 | dev_uuid, false); | |
c5502451 | 6734 | } |
5f375835 | 6735 | |
e6e674bd AJ |
6736 | if (!device->bdev && |
6737 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6738 | /* |
6739 | * this happens when a device that was properly setup | |
6740 | * in the device info lists suddenly goes bad. | |
6741 | * device->bdev is NULL, and so we have to set | |
6742 | * device->missing to one here | |
6743 | */ | |
5f375835 | 6744 | device->fs_devices->missing_devices++; |
e6e674bd | 6745 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6746 | } |
5f375835 MX |
6747 | |
6748 | /* Move the device to its own fs_devices */ | |
6749 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6750 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6751 | &device->dev_state)); | |
5f375835 MX |
6752 | |
6753 | list_move(&device->dev_list, &fs_devices->devices); | |
6754 | device->fs_devices->num_devices--; | |
6755 | fs_devices->num_devices++; | |
6756 | ||
6757 | device->fs_devices->missing_devices--; | |
6758 | fs_devices->missing_devices++; | |
6759 | ||
6760 | device->fs_devices = fs_devices; | |
6761 | } | |
2b82032c YZ |
6762 | } |
6763 | ||
0b246afa | 6764 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6765 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6766 | if (device->generation != |
6767 | btrfs_device_generation(leaf, dev_item)) | |
6768 | return -EINVAL; | |
6324fbf3 | 6769 | } |
0b86a832 CM |
6770 | |
6771 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6772 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6773 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6774 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6775 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6776 | atomic64_add(device->total_bytes - device->bytes_used, |
6777 | &fs_info->free_chunk_space); | |
2bf64758 | 6778 | } |
0b86a832 | 6779 | ret = 0; |
0b86a832 CM |
6780 | return ret; |
6781 | } | |
6782 | ||
6bccf3ab | 6783 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6784 | { |
6bccf3ab | 6785 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6786 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6787 | struct extent_buffer *sb; |
0b86a832 | 6788 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6789 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6790 | u8 *array_ptr; |
6791 | unsigned long sb_array_offset; | |
84eed90f | 6792 | int ret = 0; |
0b86a832 CM |
6793 | u32 num_stripes; |
6794 | u32 array_size; | |
6795 | u32 len = 0; | |
1ffb22cf | 6796 | u32 cur_offset; |
e06cd3dd | 6797 | u64 type; |
84eed90f | 6798 | struct btrfs_key key; |
0b86a832 | 6799 | |
0b246afa | 6800 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6801 | /* |
6802 | * This will create extent buffer of nodesize, superblock size is | |
6803 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6804 | * overallocate but we can keep it as-is, only the first page is used. | |
6805 | */ | |
2ff7e61e | 6806 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6807 | if (IS_ERR(sb)) |
6808 | return PTR_ERR(sb); | |
4db8c528 | 6809 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6810 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6811 | /* |
01327610 | 6812 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6813 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6814 | * pages up-to-date when the page is larger: extent does not cover the |
6815 | * whole page and consequently check_page_uptodate does not find all | |
6816 | * the page's extents up-to-date (the hole beyond sb), | |
6817 | * write_extent_buffer then triggers a WARN_ON. | |
6818 | * | |
6819 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6820 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6821 | * to silence the warning eg. on PowerPC 64. | |
6822 | */ | |
09cbfeaf | 6823 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6824 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6825 | |
a061fc8d | 6826 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6827 | array_size = btrfs_super_sys_array_size(super_copy); |
6828 | ||
1ffb22cf DS |
6829 | array_ptr = super_copy->sys_chunk_array; |
6830 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6831 | cur_offset = 0; | |
0b86a832 | 6832 | |
1ffb22cf DS |
6833 | while (cur_offset < array_size) { |
6834 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6835 | len = sizeof(*disk_key); |
6836 | if (cur_offset + len > array_size) | |
6837 | goto out_short_read; | |
6838 | ||
0b86a832 CM |
6839 | btrfs_disk_key_to_cpu(&key, disk_key); |
6840 | ||
1ffb22cf DS |
6841 | array_ptr += len; |
6842 | sb_array_offset += len; | |
6843 | cur_offset += len; | |
0b86a832 | 6844 | |
32ab3d1b JT |
6845 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
6846 | btrfs_err(fs_info, | |
6847 | "unexpected item type %u in sys_array at offset %u", | |
6848 | (u32)key.type, cur_offset); | |
6849 | ret = -EIO; | |
6850 | break; | |
6851 | } | |
f5cdedd7 | 6852 | |
32ab3d1b JT |
6853 | chunk = (struct btrfs_chunk *)sb_array_offset; |
6854 | /* | |
6855 | * At least one btrfs_chunk with one stripe must be present, | |
6856 | * exact stripe count check comes afterwards | |
6857 | */ | |
6858 | len = btrfs_chunk_item_size(1); | |
6859 | if (cur_offset + len > array_size) | |
6860 | goto out_short_read; | |
e06cd3dd | 6861 | |
32ab3d1b JT |
6862 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
6863 | if (!num_stripes) { | |
6864 | btrfs_err(fs_info, | |
6865 | "invalid number of stripes %u in sys_array at offset %u", | |
6866 | num_stripes, cur_offset); | |
6867 | ret = -EIO; | |
6868 | break; | |
6869 | } | |
e3540eab | 6870 | |
32ab3d1b JT |
6871 | type = btrfs_chunk_type(sb, chunk); |
6872 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6873 | btrfs_err(fs_info, |
32ab3d1b JT |
6874 | "invalid chunk type %llu in sys_array at offset %u", |
6875 | type, cur_offset); | |
84eed90f CM |
6876 | ret = -EIO; |
6877 | break; | |
0b86a832 | 6878 | } |
32ab3d1b JT |
6879 | |
6880 | len = btrfs_chunk_item_size(num_stripes); | |
6881 | if (cur_offset + len > array_size) | |
6882 | goto out_short_read; | |
6883 | ||
6884 | ret = read_one_chunk(&key, sb, chunk); | |
6885 | if (ret) | |
6886 | break; | |
6887 | ||
1ffb22cf DS |
6888 | array_ptr += len; |
6889 | sb_array_offset += len; | |
6890 | cur_offset += len; | |
0b86a832 | 6891 | } |
d865177a | 6892 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6893 | free_extent_buffer_stale(sb); |
84eed90f | 6894 | return ret; |
e3540eab DS |
6895 | |
6896 | out_short_read: | |
ab8d0fc4 | 6897 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6898 | len, cur_offset); |
d865177a | 6899 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6900 | free_extent_buffer_stale(sb); |
e3540eab | 6901 | return -EIO; |
0b86a832 CM |
6902 | } |
6903 | ||
21634a19 QW |
6904 | /* |
6905 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6906 | * | |
6528b99d AJ |
6907 | * If the @failing_dev is specified, it's accounted as missing. |
6908 | * | |
21634a19 QW |
6909 | * Return true if all chunks meet the minimal RW mount requirements. |
6910 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6911 | */ | |
6528b99d AJ |
6912 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6913 | struct btrfs_device *failing_dev) | |
21634a19 | 6914 | { |
c8bf1b67 | 6915 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
6916 | struct extent_map *em; |
6917 | u64 next_start = 0; | |
6918 | bool ret = true; | |
6919 | ||
c8bf1b67 DS |
6920 | read_lock(&map_tree->lock); |
6921 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
6922 | read_unlock(&map_tree->lock); | |
21634a19 QW |
6923 | /* No chunk at all? Return false anyway */ |
6924 | if (!em) { | |
6925 | ret = false; | |
6926 | goto out; | |
6927 | } | |
6928 | while (em) { | |
6929 | struct map_lookup *map; | |
6930 | int missing = 0; | |
6931 | int max_tolerated; | |
6932 | int i; | |
6933 | ||
6934 | map = em->map_lookup; | |
6935 | max_tolerated = | |
6936 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6937 | map->type); | |
6938 | for (i = 0; i < map->num_stripes; i++) { | |
6939 | struct btrfs_device *dev = map->stripes[i].dev; | |
6940 | ||
e6e674bd AJ |
6941 | if (!dev || !dev->bdev || |
6942 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6943 | dev->last_flush_error) |
6944 | missing++; | |
6528b99d AJ |
6945 | else if (failing_dev && failing_dev == dev) |
6946 | missing++; | |
21634a19 QW |
6947 | } |
6948 | if (missing > max_tolerated) { | |
6528b99d AJ |
6949 | if (!failing_dev) |
6950 | btrfs_warn(fs_info, | |
52042d8e | 6951 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
6952 | em->start, missing, max_tolerated); |
6953 | free_extent_map(em); | |
6954 | ret = false; | |
6955 | goto out; | |
6956 | } | |
6957 | next_start = extent_map_end(em); | |
6958 | free_extent_map(em); | |
6959 | ||
c8bf1b67 DS |
6960 | read_lock(&map_tree->lock); |
6961 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 6962 | (u64)(-1) - next_start); |
c8bf1b67 | 6963 | read_unlock(&map_tree->lock); |
21634a19 QW |
6964 | } |
6965 | out: | |
6966 | return ret; | |
6967 | } | |
6968 | ||
5b4aacef | 6969 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6970 | { |
5b4aacef | 6971 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6972 | struct btrfs_path *path; |
6973 | struct extent_buffer *leaf; | |
6974 | struct btrfs_key key; | |
6975 | struct btrfs_key found_key; | |
6976 | int ret; | |
6977 | int slot; | |
99e3ecfc | 6978 | u64 total_dev = 0; |
0b86a832 | 6979 | |
0b86a832 CM |
6980 | path = btrfs_alloc_path(); |
6981 | if (!path) | |
6982 | return -ENOMEM; | |
6983 | ||
3dd0f7a3 AJ |
6984 | /* |
6985 | * uuid_mutex is needed only if we are mounting a sprout FS | |
6986 | * otherwise we don't need it. | |
6987 | */ | |
b367e47f | 6988 | mutex_lock(&uuid_mutex); |
34441361 | 6989 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6990 | |
395927a9 FDBM |
6991 | /* |
6992 | * Read all device items, and then all the chunk items. All | |
6993 | * device items are found before any chunk item (their object id | |
6994 | * is smaller than the lowest possible object id for a chunk | |
6995 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6996 | */ |
6997 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6998 | key.offset = 0; | |
6999 | key.type = 0; | |
0b86a832 | 7000 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7001 | if (ret < 0) |
7002 | goto error; | |
d397712b | 7003 | while (1) { |
0b86a832 CM |
7004 | leaf = path->nodes[0]; |
7005 | slot = path->slots[0]; | |
7006 | if (slot >= btrfs_header_nritems(leaf)) { | |
7007 | ret = btrfs_next_leaf(root, path); | |
7008 | if (ret == 0) | |
7009 | continue; | |
7010 | if (ret < 0) | |
7011 | goto error; | |
7012 | break; | |
7013 | } | |
7014 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7015 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7016 | struct btrfs_dev_item *dev_item; | |
7017 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7018 | struct btrfs_dev_item); |
17850759 | 7019 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7020 | if (ret) |
7021 | goto error; | |
99e3ecfc | 7022 | total_dev++; |
0b86a832 CM |
7023 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7024 | struct btrfs_chunk *chunk; | |
7025 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
9690ac09 | 7026 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7027 | if (ret) |
7028 | goto error; | |
0b86a832 CM |
7029 | } |
7030 | path->slots[0]++; | |
7031 | } | |
99e3ecfc LB |
7032 | |
7033 | /* | |
7034 | * After loading chunk tree, we've got all device information, | |
7035 | * do another round of validation checks. | |
7036 | */ | |
0b246afa JM |
7037 | if (total_dev != fs_info->fs_devices->total_devices) { |
7038 | btrfs_err(fs_info, | |
99e3ecfc | 7039 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7040 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7041 | total_dev); |
7042 | ret = -EINVAL; | |
7043 | goto error; | |
7044 | } | |
0b246afa JM |
7045 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7046 | fs_info->fs_devices->total_rw_bytes) { | |
7047 | btrfs_err(fs_info, | |
99e3ecfc | 7048 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7049 | btrfs_super_total_bytes(fs_info->super_copy), |
7050 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7051 | ret = -EINVAL; |
7052 | goto error; | |
7053 | } | |
0b86a832 CM |
7054 | ret = 0; |
7055 | error: | |
34441361 | 7056 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7057 | mutex_unlock(&uuid_mutex); |
7058 | ||
2b82032c | 7059 | btrfs_free_path(path); |
0b86a832 CM |
7060 | return ret; |
7061 | } | |
442a4f63 | 7062 | |
cb517eab MX |
7063 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7064 | { | |
7065 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7066 | struct btrfs_device *device; | |
7067 | ||
29cc83f6 LB |
7068 | while (fs_devices) { |
7069 | mutex_lock(&fs_devices->device_list_mutex); | |
7070 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7071 | device->fs_info = fs_info; |
29cc83f6 LB |
7072 | mutex_unlock(&fs_devices->device_list_mutex); |
7073 | ||
7074 | fs_devices = fs_devices->seed; | |
7075 | } | |
cb517eab MX |
7076 | } |
7077 | ||
1dc990df DS |
7078 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7079 | const struct btrfs_dev_stats_item *ptr, | |
7080 | int index) | |
7081 | { | |
7082 | u64 val; | |
7083 | ||
7084 | read_extent_buffer(eb, &val, | |
7085 | offsetof(struct btrfs_dev_stats_item, values) + | |
7086 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7087 | sizeof(val)); | |
7088 | return val; | |
7089 | } | |
7090 | ||
7091 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7092 | struct btrfs_dev_stats_item *ptr, | |
7093 | int index, u64 val) | |
7094 | { | |
7095 | write_extent_buffer(eb, &val, | |
7096 | offsetof(struct btrfs_dev_stats_item, values) + | |
7097 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7098 | sizeof(val)); | |
7099 | } | |
7100 | ||
733f4fbb SB |
7101 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7102 | { | |
7103 | struct btrfs_key key; | |
733f4fbb SB |
7104 | struct btrfs_root *dev_root = fs_info->dev_root; |
7105 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7106 | struct extent_buffer *eb; | |
7107 | int slot; | |
7108 | int ret = 0; | |
7109 | struct btrfs_device *device; | |
7110 | struct btrfs_path *path = NULL; | |
7111 | int i; | |
7112 | ||
7113 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7114 | if (!path) |
7115 | return -ENOMEM; | |
733f4fbb SB |
7116 | |
7117 | mutex_lock(&fs_devices->device_list_mutex); | |
7118 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7119 | int item_size; | |
7120 | struct btrfs_dev_stats_item *ptr; | |
7121 | ||
242e2956 DS |
7122 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7123 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7124 | key.offset = device->devid; |
7125 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7126 | if (ret) { | |
ae4b9b4c AJ |
7127 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7128 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7129 | device->dev_stats_valid = 1; |
7130 | btrfs_release_path(path); | |
7131 | continue; | |
7132 | } | |
7133 | slot = path->slots[0]; | |
7134 | eb = path->nodes[0]; | |
733f4fbb SB |
7135 | item_size = btrfs_item_size_nr(eb, slot); |
7136 | ||
7137 | ptr = btrfs_item_ptr(eb, slot, | |
7138 | struct btrfs_dev_stats_item); | |
7139 | ||
7140 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7141 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7142 | btrfs_dev_stat_set(device, i, | |
7143 | btrfs_dev_stats_value(eb, ptr, i)); | |
7144 | else | |
4e411a7d | 7145 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7146 | } |
7147 | ||
7148 | device->dev_stats_valid = 1; | |
7149 | btrfs_dev_stat_print_on_load(device); | |
7150 | btrfs_release_path(path); | |
7151 | } | |
7152 | mutex_unlock(&fs_devices->device_list_mutex); | |
7153 | ||
733f4fbb SB |
7154 | btrfs_free_path(path); |
7155 | return ret < 0 ? ret : 0; | |
7156 | } | |
7157 | ||
7158 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7159 | struct btrfs_device *device) |
7160 | { | |
5495f195 | 7161 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7162 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7163 | struct btrfs_path *path; |
7164 | struct btrfs_key key; | |
7165 | struct extent_buffer *eb; | |
7166 | struct btrfs_dev_stats_item *ptr; | |
7167 | int ret; | |
7168 | int i; | |
7169 | ||
242e2956 DS |
7170 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7171 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7172 | key.offset = device->devid; |
7173 | ||
7174 | path = btrfs_alloc_path(); | |
fa252992 DS |
7175 | if (!path) |
7176 | return -ENOMEM; | |
733f4fbb SB |
7177 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7178 | if (ret < 0) { | |
0b246afa | 7179 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7180 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7181 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7182 | goto out; |
7183 | } | |
7184 | ||
7185 | if (ret == 0 && | |
7186 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7187 | /* need to delete old one and insert a new one */ | |
7188 | ret = btrfs_del_item(trans, dev_root, path); | |
7189 | if (ret != 0) { | |
0b246afa | 7190 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7191 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7192 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7193 | goto out; |
7194 | } | |
7195 | ret = 1; | |
7196 | } | |
7197 | ||
7198 | if (ret == 1) { | |
7199 | /* need to insert a new item */ | |
7200 | btrfs_release_path(path); | |
7201 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7202 | &key, sizeof(*ptr)); | |
7203 | if (ret < 0) { | |
0b246afa | 7204 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7205 | "insert dev_stats item for device %s failed %d", |
7206 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7207 | goto out; |
7208 | } | |
7209 | } | |
7210 | ||
7211 | eb = path->nodes[0]; | |
7212 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7213 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7214 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7215 | btrfs_dev_stat_read(device, i)); | |
7216 | btrfs_mark_buffer_dirty(eb); | |
7217 | ||
7218 | out: | |
7219 | btrfs_free_path(path); | |
7220 | return ret; | |
7221 | } | |
7222 | ||
7223 | /* | |
7224 | * called from commit_transaction. Writes all changed device stats to disk. | |
7225 | */ | |
196c9d8d | 7226 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7227 | { |
196c9d8d | 7228 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7229 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7230 | struct btrfs_device *device; | |
addc3fa7 | 7231 | int stats_cnt; |
733f4fbb SB |
7232 | int ret = 0; |
7233 | ||
7234 | mutex_lock(&fs_devices->device_list_mutex); | |
7235 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7236 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7237 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7238 | continue; |
7239 | ||
9deae968 NB |
7240 | |
7241 | /* | |
7242 | * There is a LOAD-LOAD control dependency between the value of | |
7243 | * dev_stats_ccnt and updating the on-disk values which requires | |
7244 | * reading the in-memory counters. Such control dependencies | |
7245 | * require explicit read memory barriers. | |
7246 | * | |
7247 | * This memory barriers pairs with smp_mb__before_atomic in | |
7248 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7249 | * barrier implied by atomic_xchg in | |
7250 | * btrfs_dev_stats_read_and_reset | |
7251 | */ | |
7252 | smp_rmb(); | |
7253 | ||
5495f195 | 7254 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7255 | if (!ret) |
addc3fa7 | 7256 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7257 | } |
7258 | mutex_unlock(&fs_devices->device_list_mutex); | |
7259 | ||
7260 | return ret; | |
7261 | } | |
7262 | ||
442a4f63 SB |
7263 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7264 | { | |
7265 | btrfs_dev_stat_inc(dev, index); | |
7266 | btrfs_dev_stat_print_on_error(dev); | |
7267 | } | |
7268 | ||
48a3b636 | 7269 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7270 | { |
733f4fbb SB |
7271 | if (!dev->dev_stats_valid) |
7272 | return; | |
fb456252 | 7273 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7274 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7275 | rcu_str_deref(dev->name), |
442a4f63 SB |
7276 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7277 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7278 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7279 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7280 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7281 | } |
c11d2c23 | 7282 | |
733f4fbb SB |
7283 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7284 | { | |
a98cdb85 SB |
7285 | int i; |
7286 | ||
7287 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7288 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7289 | break; | |
7290 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7291 | return; /* all values == 0, suppress message */ | |
7292 | ||
fb456252 | 7293 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7294 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7295 | rcu_str_deref(dev->name), |
733f4fbb SB |
7296 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7297 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7298 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7299 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7300 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7301 | } | |
7302 | ||
2ff7e61e | 7303 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7304 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7305 | { |
7306 | struct btrfs_device *dev; | |
0b246afa | 7307 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7308 | int i; |
7309 | ||
7310 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7311 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7312 | true); | |
c11d2c23 SB |
7313 | mutex_unlock(&fs_devices->device_list_mutex); |
7314 | ||
7315 | if (!dev) { | |
0b246afa | 7316 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7317 | return -ENODEV; |
733f4fbb | 7318 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7319 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7320 | return -ENODEV; |
b27f7c0c | 7321 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7322 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7323 | if (stats->nr_items > i) | |
7324 | stats->values[i] = | |
7325 | btrfs_dev_stat_read_and_reset(dev, i); | |
7326 | else | |
4e411a7d | 7327 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 SB |
7328 | } |
7329 | } else { | |
7330 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7331 | if (stats->nr_items > i) | |
7332 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7333 | } | |
7334 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7335 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7336 | return 0; | |
7337 | } | |
a8a6dab7 | 7338 | |
da353f6b | 7339 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7340 | { |
7341 | struct buffer_head *bh; | |
7342 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7343 | int copy_num; |
a8a6dab7 | 7344 | |
12b1c263 AJ |
7345 | if (!bdev) |
7346 | return; | |
a8a6dab7 | 7347 | |
12b1c263 AJ |
7348 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7349 | copy_num++) { | |
a8a6dab7 | 7350 | |
12b1c263 AJ |
7351 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7352 | continue; | |
7353 | ||
7354 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7355 | ||
7356 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7357 | set_buffer_dirty(bh); | |
7358 | sync_dirty_buffer(bh); | |
7359 | brelse(bh); | |
7360 | } | |
7361 | ||
7362 | /* Notify udev that device has changed */ | |
7363 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7364 | ||
7365 | /* Update ctime/mtime for device path for libblkid */ | |
7366 | update_dev_time(device_path); | |
a8a6dab7 | 7367 | } |
935e5cc9 MX |
7368 | |
7369 | /* | |
bbbf7243 NB |
7370 | * Update the size and bytes used for each device where it changed. This is |
7371 | * delayed since we would otherwise get errors while writing out the | |
7372 | * superblocks. | |
7373 | * | |
7374 | * Must be invoked during transaction commit. | |
935e5cc9 | 7375 | */ |
bbbf7243 | 7376 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7377 | { |
935e5cc9 MX |
7378 | struct btrfs_device *curr, *next; |
7379 | ||
bbbf7243 | 7380 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7381 | |
bbbf7243 | 7382 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7383 | return; |
7384 | ||
bbbf7243 NB |
7385 | /* |
7386 | * We don't need the device_list_mutex here. This list is owned by the | |
7387 | * transaction and the transaction must complete before the device is | |
7388 | * released. | |
7389 | */ | |
7390 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7391 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7392 | post_commit_list) { | |
7393 | list_del_init(&curr->post_commit_list); | |
7394 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7395 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7396 | } |
bbbf7243 | 7397 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7398 | } |
5a13f430 AJ |
7399 | |
7400 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7401 | { | |
7402 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7403 | while (fs_devices) { | |
7404 | fs_devices->fs_info = fs_info; | |
7405 | fs_devices = fs_devices->seed; | |
7406 | } | |
7407 | } | |
7408 | ||
7409 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7410 | { | |
7411 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7412 | while (fs_devices) { | |
7413 | fs_devices->fs_info = NULL; | |
7414 | fs_devices = fs_devices->seed; | |
7415 | } | |
7416 | } | |
46df06b8 DS |
7417 | |
7418 | /* | |
7419 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7420 | */ | |
7421 | int btrfs_bg_type_to_factor(u64 flags) | |
7422 | { | |
44b28ada DS |
7423 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7424 | ||
7425 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7426 | } |
cf90d884 QW |
7427 | |
7428 | ||
cf90d884 QW |
7429 | |
7430 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7431 | u64 chunk_offset, u64 devid, | |
7432 | u64 physical_offset, u64 physical_len) | |
7433 | { | |
c8bf1b67 | 7434 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7435 | struct extent_map *em; |
7436 | struct map_lookup *map; | |
05a37c48 | 7437 | struct btrfs_device *dev; |
cf90d884 QW |
7438 | u64 stripe_len; |
7439 | bool found = false; | |
7440 | int ret = 0; | |
7441 | int i; | |
7442 | ||
7443 | read_lock(&em_tree->lock); | |
7444 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7445 | read_unlock(&em_tree->lock); | |
7446 | ||
7447 | if (!em) { | |
7448 | btrfs_err(fs_info, | |
7449 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7450 | physical_offset, devid); | |
7451 | ret = -EUCLEAN; | |
7452 | goto out; | |
7453 | } | |
7454 | ||
7455 | map = em->map_lookup; | |
7456 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7457 | if (physical_len != stripe_len) { | |
7458 | btrfs_err(fs_info, | |
7459 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7460 | physical_offset, devid, em->start, physical_len, | |
7461 | stripe_len); | |
7462 | ret = -EUCLEAN; | |
7463 | goto out; | |
7464 | } | |
7465 | ||
7466 | for (i = 0; i < map->num_stripes; i++) { | |
7467 | if (map->stripes[i].dev->devid == devid && | |
7468 | map->stripes[i].physical == physical_offset) { | |
7469 | found = true; | |
7470 | if (map->verified_stripes >= map->num_stripes) { | |
7471 | btrfs_err(fs_info, | |
7472 | "too many dev extents for chunk %llu found", | |
7473 | em->start); | |
7474 | ret = -EUCLEAN; | |
7475 | goto out; | |
7476 | } | |
7477 | map->verified_stripes++; | |
7478 | break; | |
7479 | } | |
7480 | } | |
7481 | if (!found) { | |
7482 | btrfs_err(fs_info, | |
7483 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7484 | physical_offset, devid); | |
7485 | ret = -EUCLEAN; | |
7486 | } | |
05a37c48 QW |
7487 | |
7488 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7489 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7490 | if (!dev) { |
7491 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7492 | ret = -EUCLEAN; | |
7493 | goto out; | |
7494 | } | |
1b3922a8 QW |
7495 | |
7496 | /* It's possible this device is a dummy for seed device */ | |
7497 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7498 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7499 | NULL, false); | |
1b3922a8 QW |
7500 | if (!dev) { |
7501 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7502 | devid); | |
7503 | ret = -EUCLEAN; | |
7504 | goto out; | |
7505 | } | |
7506 | } | |
7507 | ||
05a37c48 QW |
7508 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7509 | btrfs_err(fs_info, | |
7510 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7511 | devid, physical_offset, physical_len, | |
7512 | dev->disk_total_bytes); | |
7513 | ret = -EUCLEAN; | |
7514 | goto out; | |
7515 | } | |
cf90d884 QW |
7516 | out: |
7517 | free_extent_map(em); | |
7518 | return ret; | |
7519 | } | |
7520 | ||
7521 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7522 | { | |
c8bf1b67 | 7523 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7524 | struct extent_map *em; |
7525 | struct rb_node *node; | |
7526 | int ret = 0; | |
7527 | ||
7528 | read_lock(&em_tree->lock); | |
07e1ce09 | 7529 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7530 | em = rb_entry(node, struct extent_map, rb_node); |
7531 | if (em->map_lookup->num_stripes != | |
7532 | em->map_lookup->verified_stripes) { | |
7533 | btrfs_err(fs_info, | |
7534 | "chunk %llu has missing dev extent, have %d expect %d", | |
7535 | em->start, em->map_lookup->verified_stripes, | |
7536 | em->map_lookup->num_stripes); | |
7537 | ret = -EUCLEAN; | |
7538 | goto out; | |
7539 | } | |
7540 | } | |
7541 | out: | |
7542 | read_unlock(&em_tree->lock); | |
7543 | return ret; | |
7544 | } | |
7545 | ||
7546 | /* | |
7547 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7548 | * later chunk allocation/free would cause unexpected behavior. | |
7549 | * | |
7550 | * NOTE: This will iterate through the whole device tree, which should be of | |
7551 | * the same size level as the chunk tree. This slightly increases mount time. | |
7552 | */ | |
7553 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7554 | { | |
7555 | struct btrfs_path *path; | |
7556 | struct btrfs_root *root = fs_info->dev_root; | |
7557 | struct btrfs_key key; | |
5eb19381 QW |
7558 | u64 prev_devid = 0; |
7559 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7560 | int ret = 0; |
7561 | ||
7562 | key.objectid = 1; | |
7563 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7564 | key.offset = 0; | |
7565 | ||
7566 | path = btrfs_alloc_path(); | |
7567 | if (!path) | |
7568 | return -ENOMEM; | |
7569 | ||
7570 | path->reada = READA_FORWARD; | |
7571 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7572 | if (ret < 0) | |
7573 | goto out; | |
7574 | ||
7575 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7576 | ret = btrfs_next_item(root, path); | |
7577 | if (ret < 0) | |
7578 | goto out; | |
7579 | /* No dev extents at all? Not good */ | |
7580 | if (ret > 0) { | |
7581 | ret = -EUCLEAN; | |
7582 | goto out; | |
7583 | } | |
7584 | } | |
7585 | while (1) { | |
7586 | struct extent_buffer *leaf = path->nodes[0]; | |
7587 | struct btrfs_dev_extent *dext; | |
7588 | int slot = path->slots[0]; | |
7589 | u64 chunk_offset; | |
7590 | u64 physical_offset; | |
7591 | u64 physical_len; | |
7592 | u64 devid; | |
7593 | ||
7594 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7595 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7596 | break; | |
7597 | devid = key.objectid; | |
7598 | physical_offset = key.offset; | |
7599 | ||
7600 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7601 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7602 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7603 | ||
5eb19381 QW |
7604 | /* Check if this dev extent overlaps with the previous one */ |
7605 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7606 | btrfs_err(fs_info, | |
7607 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7608 | devid, physical_offset, prev_dev_ext_end); | |
7609 | ret = -EUCLEAN; | |
7610 | goto out; | |
7611 | } | |
7612 | ||
cf90d884 QW |
7613 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7614 | physical_offset, physical_len); | |
7615 | if (ret < 0) | |
7616 | goto out; | |
5eb19381 QW |
7617 | prev_devid = devid; |
7618 | prev_dev_ext_end = physical_offset + physical_len; | |
7619 | ||
cf90d884 QW |
7620 | ret = btrfs_next_item(root, path); |
7621 | if (ret < 0) | |
7622 | goto out; | |
7623 | if (ret > 0) { | |
7624 | ret = 0; | |
7625 | break; | |
7626 | } | |
7627 | } | |
7628 | ||
7629 | /* Ensure all chunks have corresponding dev extents */ | |
7630 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7631 | out: | |
7632 | btrfs_free_path(path); | |
7633 | return ret; | |
7634 | } | |
eede2bf3 OS |
7635 | |
7636 | /* | |
7637 | * Check whether the given block group or device is pinned by any inode being | |
7638 | * used as a swapfile. | |
7639 | */ | |
7640 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7641 | { | |
7642 | struct btrfs_swapfile_pin *sp; | |
7643 | struct rb_node *node; | |
7644 | ||
7645 | spin_lock(&fs_info->swapfile_pins_lock); | |
7646 | node = fs_info->swapfile_pins.rb_node; | |
7647 | while (node) { | |
7648 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7649 | if (ptr < sp->ptr) | |
7650 | node = node->rb_left; | |
7651 | else if (ptr > sp->ptr) | |
7652 | node = node->rb_right; | |
7653 | else | |
7654 | break; | |
7655 | } | |
7656 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7657 | return node != NULL; | |
7658 | } |