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 | 6 | #include <linux/sched.h> |
fccc0007 | 7 | #include <linux/sched/mm.h> |
0b86a832 | 8 | #include <linux/bio.h> |
5a0e3ad6 | 9 | #include <linux/slab.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" |
b0643e59 | 33 | #include "discard.h" |
0b86a832 | 34 | |
af902047 ZL |
35 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
36 | [BTRFS_RAID_RAID10] = { | |
37 | .sub_stripes = 2, | |
38 | .dev_stripes = 1, | |
39 | .devs_max = 0, /* 0 == as many as possible */ | |
40 | .devs_min = 4, | |
8789f4fe | 41 | .tolerated_failures = 1, |
af902047 ZL |
42 | .devs_increment = 2, |
43 | .ncopies = 2, | |
b50836ed | 44 | .nparity = 0, |
ed23467b | 45 | .raid_name = "raid10", |
41a6e891 | 46 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 47 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
48 | }, |
49 | [BTRFS_RAID_RAID1] = { | |
50 | .sub_stripes = 1, | |
51 | .dev_stripes = 1, | |
52 | .devs_max = 2, | |
53 | .devs_min = 2, | |
8789f4fe | 54 | .tolerated_failures = 1, |
af902047 ZL |
55 | .devs_increment = 2, |
56 | .ncopies = 2, | |
b50836ed | 57 | .nparity = 0, |
ed23467b | 58 | .raid_name = "raid1", |
41a6e891 | 59 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 60 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 61 | }, |
47e6f742 DS |
62 | [BTRFS_RAID_RAID1C3] = { |
63 | .sub_stripes = 1, | |
64 | .dev_stripes = 1, | |
cf93e15e | 65 | .devs_max = 3, |
47e6f742 DS |
66 | .devs_min = 3, |
67 | .tolerated_failures = 2, | |
68 | .devs_increment = 3, | |
69 | .ncopies = 3, | |
db26a024 | 70 | .nparity = 0, |
47e6f742 DS |
71 | .raid_name = "raid1c3", |
72 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
73 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
74 | }, | |
8d6fac00 DS |
75 | [BTRFS_RAID_RAID1C4] = { |
76 | .sub_stripes = 1, | |
77 | .dev_stripes = 1, | |
cf93e15e | 78 | .devs_max = 4, |
8d6fac00 DS |
79 | .devs_min = 4, |
80 | .tolerated_failures = 3, | |
81 | .devs_increment = 4, | |
82 | .ncopies = 4, | |
db26a024 | 83 | .nparity = 0, |
8d6fac00 DS |
84 | .raid_name = "raid1c4", |
85 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
86 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
87 | }, | |
af902047 ZL |
88 | [BTRFS_RAID_DUP] = { |
89 | .sub_stripes = 1, | |
90 | .dev_stripes = 2, | |
91 | .devs_max = 1, | |
92 | .devs_min = 1, | |
8789f4fe | 93 | .tolerated_failures = 0, |
af902047 ZL |
94 | .devs_increment = 1, |
95 | .ncopies = 2, | |
b50836ed | 96 | .nparity = 0, |
ed23467b | 97 | .raid_name = "dup", |
41a6e891 | 98 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 99 | .mindev_error = 0, |
af902047 ZL |
100 | }, |
101 | [BTRFS_RAID_RAID0] = { | |
102 | .sub_stripes = 1, | |
103 | .dev_stripes = 1, | |
104 | .devs_max = 0, | |
105 | .devs_min = 2, | |
8789f4fe | 106 | .tolerated_failures = 0, |
af902047 ZL |
107 | .devs_increment = 1, |
108 | .ncopies = 1, | |
b50836ed | 109 | .nparity = 0, |
ed23467b | 110 | .raid_name = "raid0", |
41a6e891 | 111 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 112 | .mindev_error = 0, |
af902047 ZL |
113 | }, |
114 | [BTRFS_RAID_SINGLE] = { | |
115 | .sub_stripes = 1, | |
116 | .dev_stripes = 1, | |
117 | .devs_max = 1, | |
118 | .devs_min = 1, | |
8789f4fe | 119 | .tolerated_failures = 0, |
af902047 ZL |
120 | .devs_increment = 1, |
121 | .ncopies = 1, | |
b50836ed | 122 | .nparity = 0, |
ed23467b | 123 | .raid_name = "single", |
41a6e891 | 124 | .bg_flag = 0, |
f9fbcaa2 | 125 | .mindev_error = 0, |
af902047 ZL |
126 | }, |
127 | [BTRFS_RAID_RAID5] = { | |
128 | .sub_stripes = 1, | |
129 | .dev_stripes = 1, | |
130 | .devs_max = 0, | |
131 | .devs_min = 2, | |
8789f4fe | 132 | .tolerated_failures = 1, |
af902047 | 133 | .devs_increment = 1, |
da612e31 | 134 | .ncopies = 1, |
b50836ed | 135 | .nparity = 1, |
ed23467b | 136 | .raid_name = "raid5", |
41a6e891 | 137 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 138 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
139 | }, |
140 | [BTRFS_RAID_RAID6] = { | |
141 | .sub_stripes = 1, | |
142 | .dev_stripes = 1, | |
143 | .devs_max = 0, | |
144 | .devs_min = 3, | |
8789f4fe | 145 | .tolerated_failures = 2, |
af902047 | 146 | .devs_increment = 1, |
da612e31 | 147 | .ncopies = 1, |
b50836ed | 148 | .nparity = 2, |
ed23467b | 149 | .raid_name = "raid6", |
41a6e891 | 150 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 151 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
152 | }, |
153 | }; | |
154 | ||
158da513 | 155 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 156 | { |
158da513 DS |
157 | const int index = btrfs_bg_flags_to_raid_index(flags); |
158 | ||
159 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
160 | return NULL; |
161 | ||
158da513 | 162 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
163 | } |
164 | ||
f89e09cf AJ |
165 | /* |
166 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
167 | * bytes including terminating null byte. | |
168 | */ | |
169 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
170 | { | |
171 | int i; | |
172 | int ret; | |
173 | char *bp = buf; | |
174 | u64 flags = bg_flags; | |
175 | u32 size_bp = size_buf; | |
176 | ||
177 | if (!flags) { | |
178 | strcpy(bp, "NONE"); | |
179 | return; | |
180 | } | |
181 | ||
182 | #define DESCRIBE_FLAG(flag, desc) \ | |
183 | do { \ | |
184 | if (flags & (flag)) { \ | |
185 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
186 | if (ret < 0 || ret >= size_bp) \ | |
187 | goto out_overflow; \ | |
188 | size_bp -= ret; \ | |
189 | bp += ret; \ | |
190 | flags &= ~(flag); \ | |
191 | } \ | |
192 | } while (0) | |
193 | ||
194 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
195 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
196 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
197 | ||
198 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
199 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
200 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
201 | btrfs_raid_array[i].raid_name); | |
202 | #undef DESCRIBE_FLAG | |
203 | ||
204 | if (flags) { | |
205 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
206 | size_bp -= ret; | |
207 | } | |
208 | ||
209 | if (size_bp < size_buf) | |
210 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
211 | ||
212 | /* | |
213 | * The text is trimmed, it's up to the caller to provide sufficiently | |
214 | * large buffer | |
215 | */ | |
216 | out_overflow:; | |
217 | } | |
218 | ||
6f8e0fc7 | 219 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 220 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 221 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 222 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
223 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
224 | enum btrfs_map_op op, | |
225 | u64 logical, u64 *length, | |
226 | struct btrfs_bio **bbio_ret, | |
227 | int mirror_num, int need_raid_map); | |
2b82032c | 228 | |
9c6b1c4d DS |
229 | /* |
230 | * Device locking | |
231 | * ============== | |
232 | * | |
233 | * There are several mutexes that protect manipulation of devices and low-level | |
234 | * structures like chunks but not block groups, extents or files | |
235 | * | |
236 | * uuid_mutex (global lock) | |
237 | * ------------------------ | |
238 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
239 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
240 | * device) or requested by the device= mount option | |
241 | * | |
242 | * the mutex can be very coarse and can cover long-running operations | |
243 | * | |
244 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 245 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
246 | * |
247 | * global::fs_devs - add, remove, updates to the global list | |
248 | * | |
18c850fd JB |
249 | * does not protect: manipulation of the fs_devices::devices list in general |
250 | * but in mount context it could be used to exclude list modifications by eg. | |
251 | * scan ioctl | |
9c6b1c4d DS |
252 | * |
253 | * btrfs_device::name - renames (write side), read is RCU | |
254 | * | |
255 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
256 | * ------------------------------------------------ | |
257 | * protects updates to fs_devices::devices, ie. adding and deleting | |
258 | * | |
259 | * simple list traversal with read-only actions can be done with RCU protection | |
260 | * | |
261 | * may be used to exclude some operations from running concurrently without any | |
262 | * modifications to the list (see write_all_supers) | |
263 | * | |
18c850fd JB |
264 | * Is not required at mount and close times, because our device list is |
265 | * protected by the uuid_mutex at that point. | |
266 | * | |
9c6b1c4d DS |
267 | * balance_mutex |
268 | * ------------- | |
269 | * protects balance structures (status, state) and context accessed from | |
270 | * several places (internally, ioctl) | |
271 | * | |
272 | * chunk_mutex | |
273 | * ----------- | |
274 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
275 | * device is added/removed. Additionally it also protects post_commit_list of |
276 | * individual devices, since they can be added to the transaction's | |
277 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
278 | * |
279 | * cleaner_mutex | |
280 | * ------------- | |
281 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
282 | * cleaning together with relocation or delayed iputs | |
283 | * | |
284 | * | |
285 | * Lock nesting | |
286 | * ============ | |
287 | * | |
288 | * uuid_mutex | |
ae3e715f AJ |
289 | * device_list_mutex |
290 | * chunk_mutex | |
291 | * balance_mutex | |
89595e80 AJ |
292 | * |
293 | * | |
c3e1f96c GR |
294 | * Exclusive operations |
295 | * ==================== | |
89595e80 AJ |
296 | * |
297 | * Maintains the exclusivity of the following operations that apply to the | |
298 | * whole filesystem and cannot run in parallel. | |
299 | * | |
300 | * - Balance (*) | |
301 | * - Device add | |
302 | * - Device remove | |
303 | * - Device replace (*) | |
304 | * - Resize | |
305 | * | |
306 | * The device operations (as above) can be in one of the following states: | |
307 | * | |
308 | * - Running state | |
309 | * - Paused state | |
310 | * - Completed state | |
311 | * | |
312 | * Only device operations marked with (*) can go into the Paused state for the | |
313 | * following reasons: | |
314 | * | |
315 | * - ioctl (only Balance can be Paused through ioctl) | |
316 | * - filesystem remounted as read-only | |
317 | * - filesystem unmounted and mounted as read-only | |
318 | * - system power-cycle and filesystem mounted as read-only | |
319 | * - filesystem or device errors leading to forced read-only | |
320 | * | |
c3e1f96c GR |
321 | * The status of exclusive operation is set and cleared atomically. |
322 | * During the course of Paused state, fs_info::exclusive_operation remains set. | |
89595e80 AJ |
323 | * A device operation in Paused or Running state can be canceled or resumed |
324 | * either by ioctl (Balance only) or when remounted as read-write. | |
c3e1f96c | 325 | * The exclusive status is cleared when the device operation is canceled or |
89595e80 | 326 | * completed. |
9c6b1c4d DS |
327 | */ |
328 | ||
67a2c45e | 329 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 330 | static LIST_HEAD(fs_uuids); |
4143cb8b | 331 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
332 | { |
333 | return &fs_uuids; | |
334 | } | |
8a4b83cc | 335 | |
2dfeca9b DS |
336 | /* |
337 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
338 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
339 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
340 | * |
341 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
342 | * The returned struct is not linked onto any lists and can be destroyed with | |
343 | * kfree() right away. | |
344 | */ | |
7239ff4b NB |
345 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
346 | const u8 *metadata_fsid) | |
2208a378 ID |
347 | { |
348 | struct btrfs_fs_devices *fs_devs; | |
349 | ||
78f2c9e6 | 350 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
351 | if (!fs_devs) |
352 | return ERR_PTR(-ENOMEM); | |
353 | ||
354 | mutex_init(&fs_devs->device_list_mutex); | |
355 | ||
356 | INIT_LIST_HEAD(&fs_devs->devices); | |
357 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 358 | INIT_LIST_HEAD(&fs_devs->fs_list); |
944d3f9f | 359 | INIT_LIST_HEAD(&fs_devs->seed_list); |
2208a378 ID |
360 | if (fsid) |
361 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 362 | |
7239ff4b NB |
363 | if (metadata_fsid) |
364 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
365 | else if (fsid) | |
366 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
367 | ||
2208a378 ID |
368 | return fs_devs; |
369 | } | |
370 | ||
a425f9d4 | 371 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 372 | { |
bbbf7243 | 373 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 374 | rcu_string_free(device->name); |
1c11b63e | 375 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
376 | bio_put(device->flush_bio); |
377 | kfree(device); | |
378 | } | |
379 | ||
e4404d6e YZ |
380 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
381 | { | |
382 | struct btrfs_device *device; | |
383 | WARN_ON(fs_devices->opened); | |
384 | while (!list_empty(&fs_devices->devices)) { | |
385 | device = list_entry(fs_devices->devices.next, | |
386 | struct btrfs_device, dev_list); | |
387 | list_del(&device->dev_list); | |
a425f9d4 | 388 | btrfs_free_device(device); |
e4404d6e YZ |
389 | } |
390 | kfree(fs_devices); | |
391 | } | |
392 | ||
ffc5a379 | 393 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
394 | { |
395 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 396 | |
2b82032c YZ |
397 | while (!list_empty(&fs_uuids)) { |
398 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
399 | struct btrfs_fs_devices, fs_list); |
400 | list_del(&fs_devices->fs_list); | |
e4404d6e | 401 | free_fs_devices(fs_devices); |
8a4b83cc | 402 | } |
8a4b83cc CM |
403 | } |
404 | ||
48dae9cf DS |
405 | /* |
406 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
407 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 408 | * btrfs_free_device. |
48dae9cf | 409 | */ |
154f7cb8 | 410 | static struct btrfs_device *__alloc_device(struct btrfs_fs_info *fs_info) |
12bd2fc0 ID |
411 | { |
412 | struct btrfs_device *dev; | |
413 | ||
78f2c9e6 | 414 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
415 | if (!dev) |
416 | return ERR_PTR(-ENOMEM); | |
417 | ||
e0ae9994 DS |
418 | /* |
419 | * Preallocate a bio that's always going to be used for flushing device | |
420 | * barriers and matches the device lifespan | |
421 | */ | |
422 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
423 | if (!dev->flush_bio) { | |
424 | kfree(dev); | |
425 | return ERR_PTR(-ENOMEM); | |
426 | } | |
e0ae9994 | 427 | |
12bd2fc0 ID |
428 | INIT_LIST_HEAD(&dev->dev_list); |
429 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 430 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 | 431 | |
12bd2fc0 | 432 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 433 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 434 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 435 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 436 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
154f7cb8 QW |
437 | extent_io_tree_init(fs_info, &dev->alloc_state, |
438 | IO_TREE_DEVICE_ALLOC_STATE, NULL); | |
12bd2fc0 ID |
439 | |
440 | return dev; | |
441 | } | |
442 | ||
7239ff4b NB |
443 | static noinline struct btrfs_fs_devices *find_fsid( |
444 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 445 | { |
8a4b83cc CM |
446 | struct btrfs_fs_devices *fs_devices; |
447 | ||
7239ff4b NB |
448 | ASSERT(fsid); |
449 | ||
7a62d0f0 | 450 | /* Handle non-split brain cases */ |
c4babc5e | 451 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
452 | if (metadata_fsid) { |
453 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
454 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
455 | BTRFS_FSID_SIZE) == 0) | |
456 | return fs_devices; | |
457 | } else { | |
458 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
459 | return fs_devices; | |
460 | } | |
8a4b83cc CM |
461 | } |
462 | return NULL; | |
463 | } | |
464 | ||
c6730a0e SY |
465 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
466 | struct btrfs_super_block *disk_super) | |
467 | { | |
468 | ||
469 | struct btrfs_fs_devices *fs_devices; | |
470 | ||
471 | /* | |
472 | * Handle scanned device having completed its fsid change but | |
473 | * belonging to a fs_devices that was created by first scanning | |
474 | * a device which didn't have its fsid/metadata_uuid changed | |
475 | * at all and the CHANGING_FSID_V2 flag set. | |
476 | */ | |
477 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
478 | if (fs_devices->fsid_change && | |
479 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
480 | BTRFS_FSID_SIZE) == 0 && | |
481 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
482 | BTRFS_FSID_SIZE) == 0) { | |
483 | return fs_devices; | |
484 | } | |
485 | } | |
486 | /* | |
487 | * Handle scanned device having completed its fsid change but | |
488 | * belonging to a fs_devices that was created by a device that | |
489 | * has an outdated pair of fsid/metadata_uuid and | |
490 | * CHANGING_FSID_V2 flag set. | |
491 | */ | |
492 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
493 | if (fs_devices->fsid_change && | |
494 | memcmp(fs_devices->metadata_uuid, | |
495 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
496 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
497 | BTRFS_FSID_SIZE) == 0) { | |
498 | return fs_devices; | |
499 | } | |
500 | } | |
501 | ||
502 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
503 | } | |
504 | ||
505 | ||
beaf8ab3 SB |
506 | static int |
507 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
508 | int flush, struct block_device **bdev, | |
8f32380d | 509 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
510 | { |
511 | int ret; | |
512 | ||
513 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
514 | ||
515 | if (IS_ERR(*bdev)) { | |
516 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
517 | goto error; |
518 | } | |
519 | ||
520 | if (flush) | |
521 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 522 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
523 | if (ret) { |
524 | blkdev_put(*bdev, flags); | |
525 | goto error; | |
526 | } | |
527 | invalidate_bdev(*bdev); | |
8f32380d JT |
528 | *disk_super = btrfs_read_dev_super(*bdev); |
529 | if (IS_ERR(*disk_super)) { | |
530 | ret = PTR_ERR(*disk_super); | |
beaf8ab3 SB |
531 | blkdev_put(*bdev, flags); |
532 | goto error; | |
533 | } | |
534 | ||
535 | return 0; | |
536 | ||
537 | error: | |
538 | *bdev = NULL; | |
beaf8ab3 SB |
539 | return ret; |
540 | } | |
541 | ||
70bc7088 AJ |
542 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
543 | { | |
544 | int found; | |
545 | ||
546 | rcu_read_lock(); | |
547 | found = strcmp(rcu_str_deref(device->name), path); | |
548 | rcu_read_unlock(); | |
549 | ||
550 | return found == 0; | |
551 | } | |
552 | ||
d8367db3 AJ |
553 | /* |
554 | * Search and remove all stale (devices which are not mounted) devices. | |
555 | * When both inputs are NULL, it will search and release all stale devices. | |
556 | * path: Optional. When provided will it release all unmounted devices | |
557 | * matching this path only. | |
558 | * skip_dev: Optional. Will skip this device when searching for the stale | |
559 | * devices. | |
70bc7088 AJ |
560 | * Return: 0 for success or if @path is NULL. |
561 | * -EBUSY if @path is a mounted device. | |
562 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 563 | */ |
70bc7088 | 564 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 565 | struct btrfs_device *skip_device) |
4fde46f0 | 566 | { |
fa6d2ae5 AJ |
567 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
568 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
569 | int ret = 0; |
570 | ||
571 | if (path) | |
572 | ret = -ENOENT; | |
4fde46f0 | 573 | |
fa6d2ae5 | 574 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 575 | |
70bc7088 | 576 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
577 | list_for_each_entry_safe(device, tmp_device, |
578 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 579 | if (skip_device && skip_device == device) |
d8367db3 | 580 | continue; |
fa6d2ae5 | 581 | if (path && !device->name) |
4fde46f0 | 582 | continue; |
70bc7088 | 583 | if (path && !device_path_matched(path, device)) |
38cf665d | 584 | continue; |
70bc7088 AJ |
585 | if (fs_devices->opened) { |
586 | /* for an already deleted device return 0 */ | |
587 | if (path && ret != 0) | |
588 | ret = -EBUSY; | |
589 | break; | |
590 | } | |
4fde46f0 | 591 | |
4fde46f0 | 592 | /* delete the stale device */ |
7bcb8164 AJ |
593 | fs_devices->num_devices--; |
594 | list_del(&device->dev_list); | |
595 | btrfs_free_device(device); | |
596 | ||
70bc7088 | 597 | ret = 0; |
7bcb8164 AJ |
598 | } |
599 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 600 | |
7bcb8164 AJ |
601 | if (fs_devices->num_devices == 0) { |
602 | btrfs_sysfs_remove_fsid(fs_devices); | |
603 | list_del(&fs_devices->fs_list); | |
604 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
605 | } |
606 | } | |
70bc7088 AJ |
607 | |
608 | return ret; | |
4fde46f0 AJ |
609 | } |
610 | ||
18c850fd JB |
611 | /* |
612 | * This is only used on mount, and we are protected from competing things | |
613 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
614 | * fs_devices->device_list_mutex here. | |
615 | */ | |
0fb08bcc AJ |
616 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
617 | struct btrfs_device *device, fmode_t flags, | |
618 | void *holder) | |
619 | { | |
620 | struct request_queue *q; | |
621 | struct block_device *bdev; | |
0fb08bcc AJ |
622 | struct btrfs_super_block *disk_super; |
623 | u64 devid; | |
624 | int ret; | |
625 | ||
626 | if (device->bdev) | |
627 | return -EINVAL; | |
628 | if (!device->name) | |
629 | return -EINVAL; | |
630 | ||
631 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 632 | &bdev, &disk_super); |
0fb08bcc AJ |
633 | if (ret) |
634 | return ret; | |
635 | ||
0fb08bcc AJ |
636 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
637 | if (devid != device->devid) | |
8f32380d | 638 | goto error_free_page; |
0fb08bcc AJ |
639 | |
640 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 641 | goto error_free_page; |
0fb08bcc AJ |
642 | |
643 | device->generation = btrfs_super_generation(disk_super); | |
644 | ||
645 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
646 | if (btrfs_super_incompat_flags(disk_super) & |
647 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
648 | pr_err( | |
649 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 650 | goto error_free_page; |
7239ff4b NB |
651 | } |
652 | ||
ebbede42 | 653 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 654 | fs_devices->seeding = true; |
0fb08bcc | 655 | } else { |
ebbede42 AJ |
656 | if (bdev_read_only(bdev)) |
657 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
658 | else | |
659 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
660 | } |
661 | ||
662 | q = bdev_get_queue(bdev); | |
0fb08bcc | 663 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 664 | fs_devices->rotating = true; |
0fb08bcc AJ |
665 | |
666 | device->bdev = bdev; | |
e12c9621 | 667 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
668 | device->mode = flags; |
669 | ||
670 | fs_devices->open_devices++; | |
ebbede42 AJ |
671 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
672 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 673 | fs_devices->rw_devices++; |
b1b8e386 | 674 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 675 | } |
8f32380d | 676 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
677 | |
678 | return 0; | |
679 | ||
8f32380d JT |
680 | error_free_page: |
681 | btrfs_release_disk_super(disk_super); | |
0fb08bcc AJ |
682 | blkdev_put(bdev, flags); |
683 | ||
684 | return -EINVAL; | |
685 | } | |
686 | ||
7a62d0f0 NB |
687 | /* |
688 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
689 | * being created with a disk that has already completed its fsid change. Such |
690 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
691 | * Handle both cases here. | |
7a62d0f0 NB |
692 | */ |
693 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
694 | struct btrfs_super_block *disk_super) | |
695 | { | |
696 | struct btrfs_fs_devices *fs_devices; | |
697 | ||
698 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
699 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
700 | BTRFS_FSID_SIZE) != 0 && | |
701 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
702 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
703 | return fs_devices; | |
704 | } | |
705 | } | |
706 | ||
c0d81c7c | 707 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
708 | } |
709 | ||
cc5de4e7 NB |
710 | |
711 | static struct btrfs_fs_devices *find_fsid_changed( | |
712 | struct btrfs_super_block *disk_super) | |
713 | { | |
714 | struct btrfs_fs_devices *fs_devices; | |
715 | ||
716 | /* | |
717 | * Handles the case where scanned device is part of an fs that had | |
718 | * multiple successful changes of FSID but curently device didn't | |
05840710 NB |
719 | * observe it. Meaning our fsid will be different than theirs. We need |
720 | * to handle two subcases : | |
721 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
722 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
723 | * are equal). | |
cc5de4e7 NB |
724 | */ |
725 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 726 | /* Changed UUIDs */ |
cc5de4e7 NB |
727 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
728 | BTRFS_FSID_SIZE) != 0 && | |
729 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
730 | BTRFS_FSID_SIZE) == 0 && | |
731 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
732 | BTRFS_FSID_SIZE) != 0) |
733 | return fs_devices; | |
734 | ||
735 | /* Unchanged UUIDs */ | |
736 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
737 | BTRFS_FSID_SIZE) == 0 && | |
738 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
739 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 740 | return fs_devices; |
cc5de4e7 NB |
741 | } |
742 | ||
743 | return NULL; | |
744 | } | |
1362089d NB |
745 | |
746 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
747 | struct btrfs_super_block *disk_super) | |
748 | { | |
749 | struct btrfs_fs_devices *fs_devices; | |
750 | ||
751 | /* | |
752 | * Handle the case where the scanned device is part of an fs whose last | |
753 | * metadata UUID change reverted it to the original FSID. At the same | |
754 | * time * fs_devices was first created by another constitutent device | |
755 | * which didn't fully observe the operation. This results in an | |
756 | * btrfs_fs_devices created with metadata/fsid different AND | |
757 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
758 | * fs_devices equal to the FSID of the disk. | |
759 | */ | |
760 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
761 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
762 | BTRFS_FSID_SIZE) != 0 && | |
763 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
764 | BTRFS_FSID_SIZE) == 0 && | |
765 | fs_devices->fsid_change) | |
766 | return fs_devices; | |
767 | } | |
768 | ||
769 | return NULL; | |
770 | } | |
60999ca4 DS |
771 | /* |
772 | * Add new device to list of registered devices | |
773 | * | |
774 | * Returns: | |
e124ece5 AJ |
775 | * device pointer which was just added or updated when successful |
776 | * error pointer when failed | |
60999ca4 | 777 | */ |
e124ece5 | 778 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
779 | struct btrfs_super_block *disk_super, |
780 | bool *new_device_added) | |
8a4b83cc CM |
781 | { |
782 | struct btrfs_device *device; | |
7a62d0f0 | 783 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 784 | struct rcu_string *name; |
8a4b83cc | 785 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 786 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
787 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
788 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
789 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
790 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 791 | |
cc5de4e7 | 792 | if (fsid_change_in_progress) { |
c0d81c7c | 793 | if (!has_metadata_uuid) |
cc5de4e7 | 794 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 795 | else |
cc5de4e7 | 796 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 797 | } else if (has_metadata_uuid) { |
c6730a0e | 798 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 799 | } else { |
1362089d NB |
800 | fs_devices = find_fsid_reverted_metadata(disk_super); |
801 | if (!fs_devices) | |
802 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
803 | } |
804 | ||
8a4b83cc | 805 | |
8a4b83cc | 806 | if (!fs_devices) { |
7239ff4b NB |
807 | if (has_metadata_uuid) |
808 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
809 | disk_super->metadata_uuid); | |
810 | else | |
811 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
812 | ||
2208a378 | 813 | if (IS_ERR(fs_devices)) |
e124ece5 | 814 | return ERR_CAST(fs_devices); |
2208a378 | 815 | |
92900e51 AV |
816 | fs_devices->fsid_change = fsid_change_in_progress; |
817 | ||
9c6d173e | 818 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 819 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 820 | |
8a4b83cc CM |
821 | device = NULL; |
822 | } else { | |
9c6d173e | 823 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
824 | device = btrfs_find_device(fs_devices, devid, |
825 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
826 | |
827 | /* | |
828 | * If this disk has been pulled into an fs devices created by | |
829 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
830 | * metadata_uuid/fsid values of the fs_devices. | |
831 | */ | |
1362089d | 832 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
833 | found_transid > fs_devices->latest_generation) { |
834 | memcpy(fs_devices->fsid, disk_super->fsid, | |
835 | BTRFS_FSID_SIZE); | |
1362089d NB |
836 | |
837 | if (has_metadata_uuid) | |
838 | memcpy(fs_devices->metadata_uuid, | |
839 | disk_super->metadata_uuid, | |
840 | BTRFS_FSID_SIZE); | |
841 | else | |
842 | memcpy(fs_devices->metadata_uuid, | |
843 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
844 | |
845 | fs_devices->fsid_change = false; | |
846 | } | |
8a4b83cc | 847 | } |
443f24fe | 848 | |
8a4b83cc | 849 | if (!device) { |
9c6d173e AJ |
850 | if (fs_devices->opened) { |
851 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 852 | return ERR_PTR(-EBUSY); |
9c6d173e | 853 | } |
2b82032c | 854 | |
12bd2fc0 ID |
855 | device = btrfs_alloc_device(NULL, &devid, |
856 | disk_super->dev_item.uuid); | |
857 | if (IS_ERR(device)) { | |
9c6d173e | 858 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 859 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 860 | return device; |
8a4b83cc | 861 | } |
606686ee JB |
862 | |
863 | name = rcu_string_strdup(path, GFP_NOFS); | |
864 | if (!name) { | |
a425f9d4 | 865 | btrfs_free_device(device); |
9c6d173e | 866 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 867 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 868 | } |
606686ee | 869 | rcu_assign_pointer(device->name, name); |
90519d66 | 870 | |
1f78160c | 871 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 872 | fs_devices->num_devices++; |
e5e9a520 | 873 | |
2b82032c | 874 | device->fs_devices = fs_devices; |
4306a974 | 875 | *new_device_added = true; |
327f18cc AJ |
876 | |
877 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
878 | pr_info( |
879 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
880 | disk_super->label, devid, found_transid, path, | |
881 | current->comm, task_pid_nr(current)); | |
327f18cc | 882 | else |
aa6c0df7 AJ |
883 | pr_info( |
884 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
885 | disk_super->fsid, devid, found_transid, path, | |
886 | current->comm, task_pid_nr(current)); | |
327f18cc | 887 | |
606686ee | 888 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
889 | /* |
890 | * When FS is already mounted. | |
891 | * 1. If you are here and if the device->name is NULL that | |
892 | * means this device was missing at time of FS mount. | |
893 | * 2. If you are here and if the device->name is different | |
894 | * from 'path' that means either | |
895 | * a. The same device disappeared and reappeared with | |
896 | * different name. or | |
897 | * b. The missing-disk-which-was-replaced, has | |
898 | * reappeared now. | |
899 | * | |
900 | * We must allow 1 and 2a above. But 2b would be a spurious | |
901 | * and unintentional. | |
902 | * | |
903 | * Further in case of 1 and 2a above, the disk at 'path' | |
904 | * would have missed some transaction when it was away and | |
905 | * in case of 2a the stale bdev has to be updated as well. | |
906 | * 2b must not be allowed at all time. | |
907 | */ | |
908 | ||
909 | /* | |
0f23ae74 CM |
910 | * For now, we do allow update to btrfs_fs_device through the |
911 | * btrfs dev scan cli after FS has been mounted. We're still | |
912 | * tracking a problem where systems fail mount by subvolume id | |
913 | * when we reject replacement on a mounted FS. | |
b96de000 | 914 | */ |
0f23ae74 | 915 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
916 | /* |
917 | * That is if the FS is _not_ mounted and if you | |
918 | * are here, that means there is more than one | |
919 | * disk with same uuid and devid.We keep the one | |
920 | * with larger generation number or the last-in if | |
921 | * generation are equal. | |
922 | */ | |
9c6d173e | 923 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 924 | return ERR_PTR(-EEXIST); |
77bdae4d | 925 | } |
b96de000 | 926 | |
a9261d41 AJ |
927 | /* |
928 | * We are going to replace the device path for a given devid, | |
929 | * make sure it's the same device if the device is mounted | |
930 | */ | |
931 | if (device->bdev) { | |
932 | struct block_device *path_bdev; | |
933 | ||
934 | path_bdev = lookup_bdev(path); | |
935 | if (IS_ERR(path_bdev)) { | |
936 | mutex_unlock(&fs_devices->device_list_mutex); | |
937 | return ERR_CAST(path_bdev); | |
938 | } | |
939 | ||
940 | if (device->bdev != path_bdev) { | |
941 | bdput(path_bdev); | |
942 | mutex_unlock(&fs_devices->device_list_mutex); | |
943 | btrfs_warn_in_rcu(device->fs_info, | |
79dae17d AJ |
944 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
945 | path, devid, found_transid, | |
946 | current->comm, | |
947 | task_pid_nr(current)); | |
a9261d41 AJ |
948 | return ERR_PTR(-EEXIST); |
949 | } | |
950 | bdput(path_bdev); | |
951 | btrfs_info_in_rcu(device->fs_info, | |
79dae17d AJ |
952 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
953 | devid, rcu_str_deref(device->name), | |
954 | path, current->comm, | |
955 | task_pid_nr(current)); | |
a9261d41 AJ |
956 | } |
957 | ||
606686ee | 958 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
959 | if (!name) { |
960 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 961 | return ERR_PTR(-ENOMEM); |
9c6d173e | 962 | } |
606686ee JB |
963 | rcu_string_free(device->name); |
964 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 965 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 966 | fs_devices->missing_devices--; |
e6e674bd | 967 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 968 | } |
8a4b83cc CM |
969 | } |
970 | ||
77bdae4d AJ |
971 | /* |
972 | * Unmount does not free the btrfs_device struct but would zero | |
973 | * generation along with most of the other members. So just update | |
974 | * it back. We need it to pick the disk with largest generation | |
975 | * (as above). | |
976 | */ | |
d1a63002 | 977 | if (!fs_devices->opened) { |
77bdae4d | 978 | device->generation = found_transid; |
d1a63002 NB |
979 | fs_devices->latest_generation = max_t(u64, found_transid, |
980 | fs_devices->latest_generation); | |
981 | } | |
77bdae4d | 982 | |
f2788d2f AJ |
983 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
984 | ||
9c6d173e | 985 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 986 | return device; |
8a4b83cc CM |
987 | } |
988 | ||
e4404d6e YZ |
989 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
990 | { | |
991 | struct btrfs_fs_devices *fs_devices; | |
992 | struct btrfs_device *device; | |
993 | struct btrfs_device *orig_dev; | |
d2979aa2 | 994 | int ret = 0; |
e4404d6e | 995 | |
7239ff4b | 996 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
997 | if (IS_ERR(fs_devices)) |
998 | return fs_devices; | |
e4404d6e | 999 | |
adbbb863 | 1000 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1001 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1002 | |
1003 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1004 | struct rcu_string *name; |
1005 | ||
12bd2fc0 ID |
1006 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1007 | orig_dev->uuid); | |
d2979aa2 AJ |
1008 | if (IS_ERR(device)) { |
1009 | ret = PTR_ERR(device); | |
e4404d6e | 1010 | goto error; |
d2979aa2 | 1011 | } |
e4404d6e | 1012 | |
606686ee JB |
1013 | /* |
1014 | * This is ok to do without rcu read locked because we hold the | |
1015 | * uuid mutex so nothing we touch in here is going to disappear. | |
1016 | */ | |
e755f780 | 1017 | if (orig_dev->name) { |
78f2c9e6 DS |
1018 | name = rcu_string_strdup(orig_dev->name->str, |
1019 | GFP_KERNEL); | |
e755f780 | 1020 | if (!name) { |
a425f9d4 | 1021 | btrfs_free_device(device); |
d2979aa2 | 1022 | ret = -ENOMEM; |
e755f780 AJ |
1023 | goto error; |
1024 | } | |
1025 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1026 | } |
e4404d6e | 1027 | |
e4404d6e YZ |
1028 | list_add(&device->dev_list, &fs_devices->devices); |
1029 | device->fs_devices = fs_devices; | |
1030 | fs_devices->num_devices++; | |
1031 | } | |
adbbb863 | 1032 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1033 | return fs_devices; |
1034 | error: | |
adbbb863 | 1035 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1036 | free_fs_devices(fs_devices); |
d2979aa2 | 1037 | return ERR_PTR(ret); |
e4404d6e YZ |
1038 | } |
1039 | ||
3712ccb7 NB |
1040 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
1041 | int step, struct btrfs_device **latest_dev) | |
dfe25020 | 1042 | { |
c6e30871 | 1043 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1044 | |
46224705 | 1045 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1046 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1047 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1048 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1049 | &device->dev_state) && |
998a0671 AJ |
1050 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1051 | &device->dev_state) && | |
3712ccb7 NB |
1052 | (!*latest_dev || |
1053 | device->generation > (*latest_dev)->generation)) { | |
1054 | *latest_dev = device; | |
a6b0d5c8 | 1055 | } |
2b82032c | 1056 | continue; |
a6b0d5c8 | 1057 | } |
2b82032c | 1058 | |
8dabb742 SB |
1059 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1060 | /* | |
1061 | * In the first step, keep the device which has | |
1062 | * the correct fsid and the devid that is used | |
1063 | * for the dev_replace procedure. | |
1064 | * In the second step, the dev_replace state is | |
1065 | * read from the device tree and it is known | |
1066 | * whether the procedure is really active or | |
1067 | * not, which means whether this device is | |
1068 | * used or whether it should be removed. | |
1069 | */ | |
401e29c1 AJ |
1070 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1071 | &device->dev_state)) { | |
8dabb742 SB |
1072 | continue; |
1073 | } | |
1074 | } | |
2b82032c | 1075 | if (device->bdev) { |
d4d77629 | 1076 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1077 | device->bdev = NULL; |
1078 | fs_devices->open_devices--; | |
1079 | } | |
ebbede42 | 1080 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1081 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1082 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1083 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1084 | &device->dev_state)) | |
8dabb742 | 1085 | fs_devices->rw_devices--; |
2b82032c | 1086 | } |
e4404d6e YZ |
1087 | list_del_init(&device->dev_list); |
1088 | fs_devices->num_devices--; | |
a425f9d4 | 1089 | btrfs_free_device(device); |
dfe25020 | 1090 | } |
2b82032c | 1091 | |
3712ccb7 NB |
1092 | } |
1093 | ||
1094 | /* | |
1095 | * After we have read the system tree and know devids belonging to this | |
1096 | * filesystem, remove the device which does not belong there. | |
1097 | */ | |
1098 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
1099 | { | |
1100 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1101 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1102 | |
1103 | mutex_lock(&uuid_mutex); | |
3712ccb7 | 1104 | __btrfs_free_extra_devids(fs_devices, step, &latest_dev); |
944d3f9f NB |
1105 | |
1106 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
1107 | __btrfs_free_extra_devids(seed_dev, step, &latest_dev); | |
2b82032c | 1108 | |
443f24fe | 1109 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1110 | |
dfe25020 | 1111 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1112 | } |
a0af469b | 1113 | |
14238819 AJ |
1114 | static void btrfs_close_bdev(struct btrfs_device *device) |
1115 | { | |
08ffcae8 DS |
1116 | if (!device->bdev) |
1117 | return; | |
1118 | ||
ebbede42 | 1119 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1120 | sync_blockdev(device->bdev); |
1121 | invalidate_bdev(device->bdev); | |
1122 | } | |
1123 | ||
08ffcae8 | 1124 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1125 | } |
1126 | ||
959b1c04 | 1127 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1128 | { |
1129 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1130 | |
ebbede42 | 1131 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1132 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1133 | list_del_init(&device->dev_alloc_list); | |
1134 | fs_devices->rw_devices--; | |
1135 | } | |
1136 | ||
e6e674bd | 1137 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1138 | fs_devices->missing_devices--; |
1139 | ||
959b1c04 | 1140 | btrfs_close_bdev(device); |
321f69f8 | 1141 | if (device->bdev) { |
3fff3975 | 1142 | fs_devices->open_devices--; |
321f69f8 | 1143 | device->bdev = NULL; |
f448341a | 1144 | } |
321f69f8 | 1145 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
f448341a | 1146 | |
321f69f8 JT |
1147 | device->fs_info = NULL; |
1148 | atomic_set(&device->dev_stats_ccnt, 0); | |
1149 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1150 | |
321f69f8 JT |
1151 | /* Verify the device is back in a pristine state */ |
1152 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1153 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1154 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1155 | ASSERT(list_empty(&device->post_commit_list)); | |
1156 | ASSERT(atomic_read(&device->reada_in_flight) == 0); | |
f448341a AJ |
1157 | } |
1158 | ||
54eed6ae | 1159 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1160 | { |
2037a093 | 1161 | struct btrfs_device *device, *tmp; |
e4404d6e | 1162 | |
425c6ed6 JB |
1163 | lockdep_assert_held(&uuid_mutex); |
1164 | ||
2b82032c | 1165 | if (--fs_devices->opened > 0) |
54eed6ae | 1166 | return; |
8a4b83cc | 1167 | |
425c6ed6 | 1168 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1169 | btrfs_close_one_device(device); |
c9513edb | 1170 | |
e4404d6e YZ |
1171 | WARN_ON(fs_devices->open_devices); |
1172 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1173 | fs_devices->opened = 0; |
0395d84f | 1174 | fs_devices->seeding = false; |
c4989c2f | 1175 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1176 | } |
1177 | ||
54eed6ae | 1178 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1179 | { |
944d3f9f NB |
1180 | LIST_HEAD(list); |
1181 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1182 | |
1183 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1184 | close_fs_devices(fs_devices); |
944d3f9f NB |
1185 | if (!fs_devices->opened) |
1186 | list_splice_init(&fs_devices->seed_list, &list); | |
e4404d6e | 1187 | |
944d3f9f | 1188 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
0226e0eb | 1189 | close_fs_devices(fs_devices); |
944d3f9f | 1190 | list_del(&fs_devices->seed_list); |
e4404d6e YZ |
1191 | free_fs_devices(fs_devices); |
1192 | } | |
425c6ed6 | 1193 | mutex_unlock(&uuid_mutex); |
2b82032c YZ |
1194 | } |
1195 | ||
897fb573 | 1196 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1197 | fmode_t flags, void *holder) |
8a4b83cc | 1198 | { |
8a4b83cc | 1199 | struct btrfs_device *device; |
443f24fe | 1200 | struct btrfs_device *latest_dev = NULL; |
8a4b83cc | 1201 | |
d4d77629 TH |
1202 | flags |= FMODE_EXCL; |
1203 | ||
f117e290 | 1204 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1205 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1206 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1207 | continue; |
a0af469b | 1208 | |
9f050db4 AJ |
1209 | if (!latest_dev || |
1210 | device->generation > latest_dev->generation) | |
1211 | latest_dev = device; | |
8a4b83cc | 1212 | } |
1ed802c9 AJ |
1213 | if (fs_devices->open_devices == 0) |
1214 | return -EINVAL; | |
1215 | ||
2b82032c | 1216 | fs_devices->opened = 1; |
443f24fe | 1217 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1218 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1219 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
1ed802c9 AJ |
1220 | |
1221 | return 0; | |
2b82032c YZ |
1222 | } |
1223 | ||
f8e10cd3 AJ |
1224 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1225 | { | |
1226 | struct btrfs_device *dev1, *dev2; | |
1227 | ||
1228 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1229 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1230 | ||
1231 | if (dev1->devid < dev2->devid) | |
1232 | return -1; | |
1233 | else if (dev1->devid > dev2->devid) | |
1234 | return 1; | |
1235 | return 0; | |
1236 | } | |
1237 | ||
2b82032c | 1238 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1239 | fmode_t flags, void *holder) |
2b82032c YZ |
1240 | { |
1241 | int ret; | |
1242 | ||
f5194e34 | 1243 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1244 | /* |
1245 | * The device_list_mutex cannot be taken here in case opening the | |
1246 | * underlying device takes further locks like bd_mutex. | |
1247 | * | |
1248 | * We also don't need the lock here as this is called during mount and | |
1249 | * exclusion is provided by uuid_mutex | |
1250 | */ | |
f5194e34 | 1251 | |
2b82032c | 1252 | if (fs_devices->opened) { |
e4404d6e YZ |
1253 | fs_devices->opened++; |
1254 | ret = 0; | |
2b82032c | 1255 | } else { |
f8e10cd3 | 1256 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1257 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1258 | } |
542c5908 | 1259 | |
8a4b83cc CM |
1260 | return ret; |
1261 | } | |
1262 | ||
8f32380d | 1263 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1264 | { |
8f32380d JT |
1265 | struct page *page = virt_to_page(super); |
1266 | ||
6cf86a00 AJ |
1267 | put_page(page); |
1268 | } | |
1269 | ||
b335eab8 NB |
1270 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
1271 | u64 bytenr) | |
6cf86a00 | 1272 | { |
b335eab8 NB |
1273 | struct btrfs_super_block *disk_super; |
1274 | struct page *page; | |
6cf86a00 AJ |
1275 | void *p; |
1276 | pgoff_t index; | |
1277 | ||
1278 | /* make sure our super fits in the device */ | |
1279 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
b335eab8 | 1280 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1281 | |
1282 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1283 | if (sizeof(*disk_super) > PAGE_SIZE) |
1284 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1285 | |
1286 | /* make sure our super doesn't straddle pages on disk */ | |
1287 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1288 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1289 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1290 | |
1291 | /* pull in the page with our super */ | |
b335eab8 | 1292 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1293 | |
b335eab8 NB |
1294 | if (IS_ERR(page)) |
1295 | return ERR_CAST(page); | |
6cf86a00 | 1296 | |
b335eab8 | 1297 | p = page_address(page); |
6cf86a00 AJ |
1298 | |
1299 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1300 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1301 | |
b335eab8 NB |
1302 | if (btrfs_super_bytenr(disk_super) != bytenr || |
1303 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
8f32380d | 1304 | btrfs_release_disk_super(p); |
b335eab8 | 1305 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1306 | } |
1307 | ||
b335eab8 NB |
1308 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1309 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1310 | |
b335eab8 | 1311 | return disk_super; |
6cf86a00 AJ |
1312 | } |
1313 | ||
228a73ab AJ |
1314 | int btrfs_forget_devices(const char *path) |
1315 | { | |
1316 | int ret; | |
1317 | ||
1318 | mutex_lock(&uuid_mutex); | |
1319 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1320 | mutex_unlock(&uuid_mutex); | |
1321 | ||
1322 | return ret; | |
1323 | } | |
1324 | ||
6f60cbd3 DS |
1325 | /* |
1326 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1327 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1328 | * is read via pagecache | |
1329 | */ | |
36350e95 GJ |
1330 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1331 | void *holder) | |
8a4b83cc CM |
1332 | { |
1333 | struct btrfs_super_block *disk_super; | |
4306a974 | 1334 | bool new_device_added = false; |
36350e95 | 1335 | struct btrfs_device *device = NULL; |
8a4b83cc | 1336 | struct block_device *bdev; |
6f60cbd3 | 1337 | u64 bytenr; |
8a4b83cc | 1338 | |
899f9307 DS |
1339 | lockdep_assert_held(&uuid_mutex); |
1340 | ||
6f60cbd3 DS |
1341 | /* |
1342 | * we would like to check all the supers, but that would make | |
1343 | * a btrfs mount succeed after a mkfs from a different FS. | |
1344 | * So, we need to add a special mount option to scan for | |
1345 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1346 | */ | |
1347 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1348 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1349 | |
1350 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1351 | if (IS_ERR(bdev)) |
36350e95 | 1352 | return ERR_CAST(bdev); |
6f60cbd3 | 1353 | |
b335eab8 NB |
1354 | disk_super = btrfs_read_disk_super(bdev, bytenr); |
1355 | if (IS_ERR(disk_super)) { | |
1356 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1357 | goto error_bdev_put; |
05a5c55d | 1358 | } |
6f60cbd3 | 1359 | |
4306a974 | 1360 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1361 | if (!IS_ERR(device)) { |
4306a974 AJ |
1362 | if (new_device_added) |
1363 | btrfs_free_stale_devices(path, device); | |
1364 | } | |
6f60cbd3 | 1365 | |
8f32380d | 1366 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1367 | |
1368 | error_bdev_put: | |
d4d77629 | 1369 | blkdev_put(bdev, flags); |
b6ed73bc | 1370 | |
36350e95 | 1371 | return device; |
8a4b83cc | 1372 | } |
0b86a832 | 1373 | |
1c11b63e JM |
1374 | /* |
1375 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1376 | * such is found, record the end of it in *start | |
1377 | */ | |
1c11b63e JM |
1378 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1379 | u64 len) | |
6df9a95e | 1380 | { |
1c11b63e | 1381 | u64 physical_start, physical_end; |
6df9a95e | 1382 | |
1c11b63e | 1383 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1384 | |
1c11b63e JM |
1385 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1386 | &physical_start, &physical_end, | |
1387 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1388 | |
1c11b63e JM |
1389 | if (in_range(physical_start, *start, len) || |
1390 | in_range(*start, physical_start, | |
1391 | physical_end - physical_start)) { | |
1392 | *start = physical_end + 1; | |
1393 | return true; | |
6df9a95e JB |
1394 | } |
1395 | } | |
1c11b63e | 1396 | return false; |
6df9a95e JB |
1397 | } |
1398 | ||
3b4ffa40 NA |
1399 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1400 | { | |
1401 | switch (device->fs_devices->chunk_alloc_policy) { | |
1402 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1403 | /* | |
1404 | * We don't want to overwrite the superblock on the drive nor | |
1405 | * any area used by the boot loader (grub for example), so we | |
1406 | * make sure to start at an offset of at least 1MB. | |
1407 | */ | |
1408 | return max_t(u64, start, SZ_1M); | |
1409 | default: | |
1410 | BUG(); | |
1411 | } | |
1412 | } | |
1413 | ||
1414 | /** | |
1415 | * dev_extent_hole_check - check if specified hole is suitable for allocation | |
1416 | * @device: the device which we have the hole | |
1417 | * @hole_start: starting position of the hole | |
1418 | * @hole_size: the size of the hole | |
1419 | * @num_bytes: the size of the free space that we need | |
1420 | * | |
1421 | * This function may modify @hole_start and @hole_end to reflect the suitable | |
1422 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. | |
1423 | */ | |
1424 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1425 | u64 *hole_size, u64 num_bytes) | |
1426 | { | |
1427 | bool changed = false; | |
1428 | u64 hole_end = *hole_start + *hole_size; | |
1429 | ||
1430 | /* | |
1431 | * Check before we set max_hole_start, otherwise we could end up | |
1432 | * sending back this offset anyway. | |
1433 | */ | |
1434 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1435 | if (hole_end >= *hole_start) | |
1436 | *hole_size = hole_end - *hole_start; | |
1437 | else | |
1438 | *hole_size = 0; | |
1439 | changed = true; | |
1440 | } | |
1441 | ||
1442 | switch (device->fs_devices->chunk_alloc_policy) { | |
1443 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1444 | /* No extra check */ | |
1445 | break; | |
1446 | default: | |
1447 | BUG(); | |
1448 | } | |
1449 | ||
1450 | return changed; | |
1451 | } | |
6df9a95e | 1452 | |
0b86a832 | 1453 | /* |
499f377f JM |
1454 | * find_free_dev_extent_start - find free space in the specified device |
1455 | * @device: the device which we search the free space in | |
1456 | * @num_bytes: the size of the free space that we need | |
1457 | * @search_start: the position from which to begin the search | |
1458 | * @start: store the start of the free space. | |
1459 | * @len: the size of the free space. that we find, or the size | |
1460 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1461 | * |
0b86a832 CM |
1462 | * this uses a pretty simple search, the expectation is that it is |
1463 | * called very infrequently and that a given device has a small number | |
1464 | * of extents | |
7bfc837d MX |
1465 | * |
1466 | * @start is used to store the start of the free space if we find. But if we | |
1467 | * don't find suitable free space, it will be used to store the start position | |
1468 | * of the max free space. | |
1469 | * | |
1470 | * @len is used to store the size of the free space that we find. | |
1471 | * But if we don't find suitable free space, it is used to store the size of | |
1472 | * the max free space. | |
135da976 QW |
1473 | * |
1474 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1475 | * check to ensure dev extents are not double allocated. | |
1476 | * This makes the function safe to allocate dev extents but may not report | |
1477 | * correct usable device space, as device extent freed in current transaction | |
1478 | * is not reported as avaiable. | |
0b86a832 | 1479 | */ |
9e3246a5 QW |
1480 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1481 | u64 num_bytes, u64 search_start, u64 *start, | |
1482 | u64 *len) | |
0b86a832 | 1483 | { |
0b246afa JM |
1484 | struct btrfs_fs_info *fs_info = device->fs_info; |
1485 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1486 | struct btrfs_key key; |
7bfc837d | 1487 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1488 | struct btrfs_path *path; |
7bfc837d MX |
1489 | u64 hole_size; |
1490 | u64 max_hole_start; | |
1491 | u64 max_hole_size; | |
1492 | u64 extent_end; | |
0b86a832 CM |
1493 | u64 search_end = device->total_bytes; |
1494 | int ret; | |
7bfc837d | 1495 | int slot; |
0b86a832 | 1496 | struct extent_buffer *l; |
8cdc7c5b | 1497 | |
3b4ffa40 | 1498 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1499 | |
6df9a95e JB |
1500 | path = btrfs_alloc_path(); |
1501 | if (!path) | |
1502 | return -ENOMEM; | |
f2ab7618 | 1503 | |
7bfc837d MX |
1504 | max_hole_start = search_start; |
1505 | max_hole_size = 0; | |
1506 | ||
f2ab7618 | 1507 | again: |
401e29c1 AJ |
1508 | if (search_start >= search_end || |
1509 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1510 | ret = -ENOSPC; |
6df9a95e | 1511 | goto out; |
7bfc837d MX |
1512 | } |
1513 | ||
e4058b54 | 1514 | path->reada = READA_FORWARD; |
6df9a95e JB |
1515 | path->search_commit_root = 1; |
1516 | path->skip_locking = 1; | |
7bfc837d | 1517 | |
0b86a832 CM |
1518 | key.objectid = device->devid; |
1519 | key.offset = search_start; | |
1520 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1521 | |
125ccb0a | 1522 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1523 | if (ret < 0) |
7bfc837d | 1524 | goto out; |
1fcbac58 YZ |
1525 | if (ret > 0) { |
1526 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1527 | if (ret < 0) | |
7bfc837d | 1528 | goto out; |
1fcbac58 | 1529 | } |
7bfc837d | 1530 | |
0b86a832 CM |
1531 | while (1) { |
1532 | l = path->nodes[0]; | |
1533 | slot = path->slots[0]; | |
1534 | if (slot >= btrfs_header_nritems(l)) { | |
1535 | ret = btrfs_next_leaf(root, path); | |
1536 | if (ret == 0) | |
1537 | continue; | |
1538 | if (ret < 0) | |
7bfc837d MX |
1539 | goto out; |
1540 | ||
1541 | break; | |
0b86a832 CM |
1542 | } |
1543 | btrfs_item_key_to_cpu(l, &key, slot); | |
1544 | ||
1545 | if (key.objectid < device->devid) | |
1546 | goto next; | |
1547 | ||
1548 | if (key.objectid > device->devid) | |
7bfc837d | 1549 | break; |
0b86a832 | 1550 | |
962a298f | 1551 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1552 | goto next; |
9779b72f | 1553 | |
7bfc837d MX |
1554 | if (key.offset > search_start) { |
1555 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1556 | dev_extent_hole_check(device, &search_start, &hole_size, |
1557 | num_bytes); | |
6df9a95e | 1558 | |
7bfc837d MX |
1559 | if (hole_size > max_hole_size) { |
1560 | max_hole_start = search_start; | |
1561 | max_hole_size = hole_size; | |
1562 | } | |
9779b72f | 1563 | |
7bfc837d MX |
1564 | /* |
1565 | * If this free space is greater than which we need, | |
1566 | * it must be the max free space that we have found | |
1567 | * until now, so max_hole_start must point to the start | |
1568 | * of this free space and the length of this free space | |
1569 | * is stored in max_hole_size. Thus, we return | |
1570 | * max_hole_start and max_hole_size and go back to the | |
1571 | * caller. | |
1572 | */ | |
1573 | if (hole_size >= num_bytes) { | |
1574 | ret = 0; | |
1575 | goto out; | |
0b86a832 CM |
1576 | } |
1577 | } | |
0b86a832 | 1578 | |
0b86a832 | 1579 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1580 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1581 | dev_extent); | |
1582 | if (extent_end > search_start) | |
1583 | search_start = extent_end; | |
0b86a832 CM |
1584 | next: |
1585 | path->slots[0]++; | |
1586 | cond_resched(); | |
1587 | } | |
0b86a832 | 1588 | |
38c01b96 | 1589 | /* |
1590 | * At this point, search_start should be the end of | |
1591 | * allocated dev extents, and when shrinking the device, | |
1592 | * search_end may be smaller than search_start. | |
1593 | */ | |
f2ab7618 | 1594 | if (search_end > search_start) { |
38c01b96 | 1595 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1596 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1597 | num_bytes)) { | |
f2ab7618 ZL |
1598 | btrfs_release_path(path); |
1599 | goto again; | |
1600 | } | |
0b86a832 | 1601 | |
f2ab7618 ZL |
1602 | if (hole_size > max_hole_size) { |
1603 | max_hole_start = search_start; | |
1604 | max_hole_size = hole_size; | |
1605 | } | |
6df9a95e JB |
1606 | } |
1607 | ||
7bfc837d | 1608 | /* See above. */ |
f2ab7618 | 1609 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1610 | ret = -ENOSPC; |
1611 | else | |
1612 | ret = 0; | |
1613 | ||
1614 | out: | |
2b82032c | 1615 | btrfs_free_path(path); |
7bfc837d | 1616 | *start = max_hole_start; |
b2117a39 | 1617 | if (len) |
7bfc837d | 1618 | *len = max_hole_size; |
0b86a832 CM |
1619 | return ret; |
1620 | } | |
1621 | ||
60dfdf25 | 1622 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1623 | u64 *start, u64 *len) |
1624 | { | |
499f377f | 1625 | /* FIXME use last free of some kind */ |
60dfdf25 | 1626 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1627 | } |
1628 | ||
b2950863 | 1629 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1630 | struct btrfs_device *device, |
2196d6e8 | 1631 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1632 | { |
0b246afa JM |
1633 | struct btrfs_fs_info *fs_info = device->fs_info; |
1634 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1635 | int ret; |
1636 | struct btrfs_path *path; | |
8f18cf13 | 1637 | struct btrfs_key key; |
a061fc8d CM |
1638 | struct btrfs_key found_key; |
1639 | struct extent_buffer *leaf = NULL; | |
1640 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1641 | |
1642 | path = btrfs_alloc_path(); | |
1643 | if (!path) | |
1644 | return -ENOMEM; | |
1645 | ||
1646 | key.objectid = device->devid; | |
1647 | key.offset = start; | |
1648 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1649 | again: |
8f18cf13 | 1650 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1651 | if (ret > 0) { |
1652 | ret = btrfs_previous_item(root, path, key.objectid, | |
1653 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1654 | if (ret) |
1655 | goto out; | |
a061fc8d CM |
1656 | leaf = path->nodes[0]; |
1657 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1658 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1659 | struct btrfs_dev_extent); | |
1660 | BUG_ON(found_key.offset > start || found_key.offset + | |
1661 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1662 | key = found_key; |
1663 | btrfs_release_path(path); | |
1664 | goto again; | |
a061fc8d CM |
1665 | } else if (ret == 0) { |
1666 | leaf = path->nodes[0]; | |
1667 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1668 | struct btrfs_dev_extent); | |
79787eaa | 1669 | } else { |
0b246afa | 1670 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1671 | goto out; |
a061fc8d | 1672 | } |
8f18cf13 | 1673 | |
2196d6e8 MX |
1674 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1675 | ||
8f18cf13 | 1676 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1677 | if (ret) { |
0b246afa JM |
1678 | btrfs_handle_fs_error(fs_info, ret, |
1679 | "Failed to remove dev extent item"); | |
13212b54 | 1680 | } else { |
3204d33c | 1681 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1682 | } |
b0b802d7 | 1683 | out: |
8f18cf13 CM |
1684 | btrfs_free_path(path); |
1685 | return ret; | |
1686 | } | |
1687 | ||
48a3b636 ES |
1688 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1689 | struct btrfs_device *device, | |
48a3b636 | 1690 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1691 | { |
1692 | int ret; | |
1693 | struct btrfs_path *path; | |
0b246afa JM |
1694 | struct btrfs_fs_info *fs_info = device->fs_info; |
1695 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1696 | struct btrfs_dev_extent *extent; |
1697 | struct extent_buffer *leaf; | |
1698 | struct btrfs_key key; | |
1699 | ||
e12c9621 | 1700 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1701 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1702 | path = btrfs_alloc_path(); |
1703 | if (!path) | |
1704 | return -ENOMEM; | |
1705 | ||
0b86a832 | 1706 | key.objectid = device->devid; |
2b82032c | 1707 | key.offset = start; |
0b86a832 CM |
1708 | key.type = BTRFS_DEV_EXTENT_KEY; |
1709 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1710 | sizeof(*extent)); | |
2cdcecbc MF |
1711 | if (ret) |
1712 | goto out; | |
0b86a832 CM |
1713 | |
1714 | leaf = path->nodes[0]; | |
1715 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1716 | struct btrfs_dev_extent); | |
b5d9071c NB |
1717 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1718 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1719 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1720 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1721 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1722 | ||
0b86a832 CM |
1723 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1724 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1725 | out: |
0b86a832 CM |
1726 | btrfs_free_path(path); |
1727 | return ret; | |
1728 | } | |
1729 | ||
6df9a95e | 1730 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1731 | { |
6df9a95e JB |
1732 | struct extent_map_tree *em_tree; |
1733 | struct extent_map *em; | |
1734 | struct rb_node *n; | |
1735 | u64 ret = 0; | |
0b86a832 | 1736 | |
c8bf1b67 | 1737 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1738 | read_lock(&em_tree->lock); |
07e1ce09 | 1739 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1740 | if (n) { |
1741 | em = rb_entry(n, struct extent_map, rb_node); | |
1742 | ret = em->start + em->len; | |
0b86a832 | 1743 | } |
6df9a95e JB |
1744 | read_unlock(&em_tree->lock); |
1745 | ||
0b86a832 CM |
1746 | return ret; |
1747 | } | |
1748 | ||
53f10659 ID |
1749 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1750 | u64 *devid_ret) | |
0b86a832 CM |
1751 | { |
1752 | int ret; | |
1753 | struct btrfs_key key; | |
1754 | struct btrfs_key found_key; | |
2b82032c YZ |
1755 | struct btrfs_path *path; |
1756 | ||
2b82032c YZ |
1757 | path = btrfs_alloc_path(); |
1758 | if (!path) | |
1759 | return -ENOMEM; | |
0b86a832 CM |
1760 | |
1761 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1762 | key.type = BTRFS_DEV_ITEM_KEY; | |
1763 | key.offset = (u64)-1; | |
1764 | ||
53f10659 | 1765 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1766 | if (ret < 0) |
1767 | goto error; | |
1768 | ||
a06dee4d AJ |
1769 | if (ret == 0) { |
1770 | /* Corruption */ | |
1771 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1772 | ret = -EUCLEAN; | |
1773 | goto error; | |
1774 | } | |
0b86a832 | 1775 | |
53f10659 ID |
1776 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1777 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1778 | BTRFS_DEV_ITEM_KEY); |
1779 | if (ret) { | |
53f10659 | 1780 | *devid_ret = 1; |
0b86a832 CM |
1781 | } else { |
1782 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1783 | path->slots[0]); | |
53f10659 | 1784 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1785 | } |
1786 | ret = 0; | |
1787 | error: | |
2b82032c | 1788 | btrfs_free_path(path); |
0b86a832 CM |
1789 | return ret; |
1790 | } | |
1791 | ||
1792 | /* | |
1793 | * the device information is stored in the chunk root | |
1794 | * the btrfs_device struct should be fully filled in | |
1795 | */ | |
c74a0b02 | 1796 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1797 | struct btrfs_device *device) |
0b86a832 CM |
1798 | { |
1799 | int ret; | |
1800 | struct btrfs_path *path; | |
1801 | struct btrfs_dev_item *dev_item; | |
1802 | struct extent_buffer *leaf; | |
1803 | struct btrfs_key key; | |
1804 | unsigned long ptr; | |
0b86a832 | 1805 | |
0b86a832 CM |
1806 | path = btrfs_alloc_path(); |
1807 | if (!path) | |
1808 | return -ENOMEM; | |
1809 | ||
0b86a832 CM |
1810 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1811 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1812 | key.offset = device->devid; |
0b86a832 | 1813 | |
8e87e856 NB |
1814 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1815 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1816 | if (ret) |
1817 | goto out; | |
1818 | ||
1819 | leaf = path->nodes[0]; | |
1820 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1821 | ||
1822 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1823 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1824 | btrfs_set_device_type(leaf, dev_item, device->type); |
1825 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1826 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1827 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1828 | btrfs_set_device_total_bytes(leaf, dev_item, |
1829 | btrfs_device_get_disk_total_bytes(device)); | |
1830 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1831 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1832 | btrfs_set_device_group(leaf, dev_item, 0); |
1833 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1834 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1835 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1836 | |
410ba3a2 | 1837 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1838 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1839 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1840 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1841 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1842 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1843 | |
2b82032c | 1844 | ret = 0; |
0b86a832 CM |
1845 | out: |
1846 | btrfs_free_path(path); | |
1847 | return ret; | |
1848 | } | |
8f18cf13 | 1849 | |
5a1972bd QW |
1850 | /* |
1851 | * Function to update ctime/mtime for a given device path. | |
1852 | * Mainly used for ctime/mtime based probe like libblkid. | |
1853 | */ | |
da353f6b | 1854 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1855 | { |
1856 | struct file *filp; | |
1857 | ||
1858 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1859 | if (IS_ERR(filp)) |
5a1972bd QW |
1860 | return; |
1861 | file_update_time(filp); | |
1862 | filp_close(filp, NULL); | |
5a1972bd QW |
1863 | } |
1864 | ||
f331a952 | 1865 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1866 | { |
f331a952 | 1867 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1868 | int ret; |
1869 | struct btrfs_path *path; | |
a061fc8d | 1870 | struct btrfs_key key; |
a061fc8d CM |
1871 | struct btrfs_trans_handle *trans; |
1872 | ||
a061fc8d CM |
1873 | path = btrfs_alloc_path(); |
1874 | if (!path) | |
1875 | return -ENOMEM; | |
1876 | ||
a22285a6 | 1877 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1878 | if (IS_ERR(trans)) { |
1879 | btrfs_free_path(path); | |
1880 | return PTR_ERR(trans); | |
1881 | } | |
a061fc8d CM |
1882 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1883 | key.type = BTRFS_DEV_ITEM_KEY; | |
1884 | key.offset = device->devid; | |
1885 | ||
1886 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1887 | if (ret) { |
1888 | if (ret > 0) | |
1889 | ret = -ENOENT; | |
1890 | btrfs_abort_transaction(trans, ret); | |
1891 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1892 | goto out; |
1893 | } | |
1894 | ||
1895 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1896 | if (ret) { |
1897 | btrfs_abort_transaction(trans, ret); | |
1898 | btrfs_end_transaction(trans); | |
1899 | } | |
1900 | ||
a061fc8d CM |
1901 | out: |
1902 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1903 | if (!ret) |
1904 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1905 | return ret; |
1906 | } | |
1907 | ||
3cc31a0d DS |
1908 | /* |
1909 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1910 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1911 | * replace. | |
1912 | */ | |
1913 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1914 | u64 num_devices) | |
a061fc8d | 1915 | { |
a061fc8d | 1916 | u64 all_avail; |
de98ced9 | 1917 | unsigned seq; |
418775a2 | 1918 | int i; |
a061fc8d | 1919 | |
de98ced9 | 1920 | do { |
bd45ffbc | 1921 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1922 | |
bd45ffbc AJ |
1923 | all_avail = fs_info->avail_data_alloc_bits | |
1924 | fs_info->avail_system_alloc_bits | | |
1925 | fs_info->avail_metadata_alloc_bits; | |
1926 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1927 | |
418775a2 | 1928 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1929 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1930 | continue; |
a061fc8d | 1931 | |
418775a2 | 1932 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1933 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1934 | |
418775a2 DS |
1935 | if (ret) |
1936 | return ret; | |
1937 | } | |
53b381b3 DW |
1938 | } |
1939 | ||
bd45ffbc | 1940 | return 0; |
f1fa7f26 AJ |
1941 | } |
1942 | ||
c9162bdf OS |
1943 | static struct btrfs_device * btrfs_find_next_active_device( |
1944 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1945 | { |
2b82032c | 1946 | struct btrfs_device *next_device; |
88acff64 AJ |
1947 | |
1948 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1949 | if (next_device != device && | |
e6e674bd AJ |
1950 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1951 | && next_device->bdev) | |
88acff64 AJ |
1952 | return next_device; |
1953 | } | |
1954 | ||
1955 | return NULL; | |
1956 | } | |
1957 | ||
1958 | /* | |
1959 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1960 | * and replace it with the provided or the next active device, in the context | |
1961 | * where this function called, there should be always be another device (or | |
1962 | * this_dev) which is active. | |
1963 | */ | |
b105e927 | 1964 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
d6507cf1 | 1965 | struct btrfs_device *this_dev) |
88acff64 | 1966 | { |
d6507cf1 | 1967 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1968 | struct btrfs_device *next_device; |
1969 | ||
1970 | if (this_dev) | |
1971 | next_device = this_dev; | |
1972 | else | |
1973 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1974 | device); | |
1975 | ASSERT(next_device); | |
1976 | ||
1977 | if (fs_info->sb->s_bdev && | |
1978 | (fs_info->sb->s_bdev == device->bdev)) | |
1979 | fs_info->sb->s_bdev = next_device->bdev; | |
1980 | ||
1981 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1982 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1983 | } | |
1984 | ||
1da73967 AJ |
1985 | /* |
1986 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1987 | * currently replaced. | |
1988 | */ | |
1989 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1990 | { | |
1991 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1992 | ||
cb5583dd | 1993 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1994 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
1995 | ASSERT(num_devices > 1); | |
1996 | num_devices--; | |
1997 | } | |
cb5583dd | 1998 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1999 | |
2000 | return num_devices; | |
2001 | } | |
2002 | ||
313b0858 JB |
2003 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2004 | struct block_device *bdev, | |
2005 | const char *device_path) | |
6fbceb9f | 2006 | { |
6fbceb9f JT |
2007 | struct btrfs_super_block *disk_super; |
2008 | int copy_num; | |
2009 | ||
2010 | if (!bdev) | |
2011 | return; | |
2012 | ||
2013 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
8f32380d JT |
2014 | struct page *page; |
2015 | int ret; | |
6fbceb9f | 2016 | |
8f32380d JT |
2017 | disk_super = btrfs_read_dev_one_super(bdev, copy_num); |
2018 | if (IS_ERR(disk_super)) | |
2019 | continue; | |
6fbceb9f JT |
2020 | |
2021 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
8f32380d JT |
2022 | |
2023 | page = virt_to_page(disk_super); | |
2024 | set_page_dirty(page); | |
2025 | lock_page(page); | |
2026 | /* write_on_page() unlocks the page */ | |
2027 | ret = write_one_page(page); | |
2028 | if (ret) | |
2029 | btrfs_warn(fs_info, | |
2030 | "error clearing superblock number %d (%d)", | |
2031 | copy_num, ret); | |
2032 | btrfs_release_disk_super(disk_super); | |
2033 | ||
6fbceb9f JT |
2034 | } |
2035 | ||
2036 | /* Notify udev that device has changed */ | |
2037 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2038 | ||
2039 | /* Update ctime/mtime for device path for libblkid */ | |
2040 | update_dev_time(device_path); | |
2041 | } | |
2042 | ||
da353f6b | 2043 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
53f8a74c | 2044 | u64 devid) |
f1fa7f26 AJ |
2045 | { |
2046 | struct btrfs_device *device; | |
1f78160c | 2047 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2048 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2049 | u64 num_devices; |
a061fc8d CM |
2050 | int ret = 0; |
2051 | ||
a061fc8d CM |
2052 | mutex_lock(&uuid_mutex); |
2053 | ||
1da73967 | 2054 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2055 | |
0b246afa | 2056 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2057 | if (ret) |
a061fc8d | 2058 | goto out; |
a061fc8d | 2059 | |
a27a94c2 NB |
2060 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2061 | ||
2062 | if (IS_ERR(device)) { | |
2063 | if (PTR_ERR(device) == -ENOENT && | |
2064 | strcmp(device_path, "missing") == 0) | |
2065 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2066 | else | |
2067 | ret = PTR_ERR(device); | |
53b381b3 | 2068 | goto out; |
a27a94c2 | 2069 | } |
dfe25020 | 2070 | |
eede2bf3 OS |
2071 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2072 | btrfs_warn_in_rcu(fs_info, | |
2073 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2074 | rcu_str_deref(device->name), device->devid); | |
2075 | ret = -ETXTBSY; | |
2076 | goto out; | |
2077 | } | |
2078 | ||
401e29c1 | 2079 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2080 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2081 | goto out; |
63a212ab SB |
2082 | } |
2083 | ||
ebbede42 AJ |
2084 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2085 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2086 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2087 | goto out; |
2b82032c YZ |
2088 | } |
2089 | ||
ebbede42 | 2090 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2091 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2092 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2093 | device->fs_devices->rw_devices--; |
34441361 | 2094 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2095 | } |
a061fc8d | 2096 | |
d7901554 | 2097 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2098 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2099 | mutex_lock(&uuid_mutex); |
a061fc8d | 2100 | if (ret) |
9b3517e9 | 2101 | goto error_undo; |
a061fc8d | 2102 | |
63a212ab SB |
2103 | /* |
2104 | * TODO: the superblock still includes this device in its num_devices | |
2105 | * counter although write_all_supers() is not locked out. This | |
2106 | * could give a filesystem state which requires a degraded mount. | |
2107 | */ | |
f331a952 | 2108 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2109 | if (ret) |
9b3517e9 | 2110 | goto error_undo; |
a061fc8d | 2111 | |
e12c9621 | 2112 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2113 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2114 | |
2115 | /* | |
2116 | * the device list mutex makes sure that we don't change | |
2117 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2118 | * the device supers. Whoever is writing all supers, should |
2119 | * lock the device list mutex before getting the number of | |
2120 | * devices in the super block (super_copy). Conversely, | |
2121 | * whoever updates the number of devices in the super block | |
2122 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2123 | */ |
1f78160c | 2124 | |
41a52a0f AJ |
2125 | /* |
2126 | * In normal cases the cur_devices == fs_devices. But in case | |
2127 | * of deleting a seed device, the cur_devices should point to | |
2128 | * its own fs_devices listed under the fs_devices->seed. | |
2129 | */ | |
1f78160c | 2130 | cur_devices = device->fs_devices; |
b5185197 | 2131 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2132 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2133 | |
41a52a0f AJ |
2134 | cur_devices->num_devices--; |
2135 | cur_devices->total_devices--; | |
b4993e64 AJ |
2136 | /* Update total_devices of the parent fs_devices if it's seed */ |
2137 | if (cur_devices != fs_devices) | |
2138 | fs_devices->total_devices--; | |
2b82032c | 2139 | |
e6e674bd | 2140 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2141 | cur_devices->missing_devices--; |
cd02dca5 | 2142 | |
d6507cf1 | 2143 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2144 | |
0bfaa9c5 | 2145 | if (device->bdev) { |
41a52a0f | 2146 | cur_devices->open_devices--; |
0bfaa9c5 | 2147 | /* remove sysfs entry */ |
53f8a74c | 2148 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2149 | } |
99994cde | 2150 | |
0b246afa JM |
2151 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2152 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2153 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2154 | |
cea67ab9 JM |
2155 | /* |
2156 | * at this point, the device is zero sized and detached from | |
2157 | * the devices list. All that's left is to zero out the old | |
2158 | * supers and free the device. | |
2159 | */ | |
ebbede42 | 2160 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
8f32380d JT |
2161 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2162 | device->name->str); | |
cea67ab9 JM |
2163 | |
2164 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2165 | synchronize_rcu(); |
2166 | btrfs_free_device(device); | |
cea67ab9 | 2167 | |
1f78160c | 2168 | if (cur_devices->open_devices == 0) { |
944d3f9f | 2169 | list_del_init(&cur_devices->seed_list); |
0226e0eb | 2170 | close_fs_devices(cur_devices); |
1f78160c | 2171 | free_fs_devices(cur_devices); |
2b82032c YZ |
2172 | } |
2173 | ||
a061fc8d CM |
2174 | out: |
2175 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2176 | return ret; |
24fc572f | 2177 | |
9b3517e9 | 2178 | error_undo: |
ebbede42 | 2179 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2180 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2181 | list_add(&device->dev_alloc_list, |
b5185197 | 2182 | &fs_devices->alloc_list); |
c3929c36 | 2183 | device->fs_devices->rw_devices++; |
34441361 | 2184 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2185 | } |
24fc572f | 2186 | goto out; |
a061fc8d CM |
2187 | } |
2188 | ||
68a9db5f | 2189 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2190 | { |
d51908ce AJ |
2191 | struct btrfs_fs_devices *fs_devices; |
2192 | ||
68a9db5f | 2193 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2194 | |
25e8e911 AJ |
2195 | /* |
2196 | * in case of fs with no seed, srcdev->fs_devices will point | |
2197 | * to fs_devices of fs_info. However when the dev being replaced is | |
2198 | * a seed dev it will point to the seed's local fs_devices. In short | |
2199 | * srcdev will have its correct fs_devices in both the cases. | |
2200 | */ | |
2201 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2202 | |
e93c89c1 | 2203 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2204 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2205 | fs_devices->num_devices--; |
e6e674bd | 2206 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2207 | fs_devices->missing_devices--; |
e93c89c1 | 2208 | |
ebbede42 | 2209 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2210 | fs_devices->rw_devices--; |
1357272f | 2211 | |
82372bc8 | 2212 | if (srcdev->bdev) |
d51908ce | 2213 | fs_devices->open_devices--; |
084b6e7c QW |
2214 | } |
2215 | ||
65237ee3 | 2216 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2217 | { |
2218 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2219 | |
a466c85e JB |
2220 | mutex_lock(&uuid_mutex); |
2221 | ||
14238819 | 2222 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2223 | synchronize_rcu(); |
2224 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2225 | |
94d5f0c2 AJ |
2226 | /* if this is no devs we rather delete the fs_devices */ |
2227 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2228 | /* |
2229 | * On a mounted FS, num_devices can't be zero unless it's a | |
2230 | * seed. In case of a seed device being replaced, the replace | |
2231 | * target added to the sprout FS, so there will be no more | |
2232 | * device left under the seed FS. | |
2233 | */ | |
2234 | ASSERT(fs_devices->seeding); | |
2235 | ||
944d3f9f | 2236 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2237 | close_fs_devices(fs_devices); |
8bef8401 | 2238 | free_fs_devices(fs_devices); |
94d5f0c2 | 2239 | } |
a466c85e | 2240 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2241 | } |
2242 | ||
4f5ad7bd | 2243 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2244 | { |
4f5ad7bd | 2245 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2246 | |
d9a071f0 | 2247 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2248 | |
53f8a74c | 2249 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2250 | |
779bf3fe | 2251 | if (tgtdev->bdev) |
d9a071f0 | 2252 | fs_devices->open_devices--; |
779bf3fe | 2253 | |
d9a071f0 | 2254 | fs_devices->num_devices--; |
e93c89c1 | 2255 | |
d6507cf1 | 2256 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2257 | |
e93c89c1 | 2258 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2259 | |
d9a071f0 | 2260 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2261 | |
2262 | /* | |
2263 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2264 | * may lead to a call to btrfs_show_devname() which will try | |
2265 | * to hold device_list_mutex. And here this device | |
2266 | * is already out of device list, so we don't have to hold | |
2267 | * the device_list_mutex lock. | |
2268 | */ | |
8f32380d JT |
2269 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2270 | tgtdev->name->str); | |
14238819 AJ |
2271 | |
2272 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2273 | synchronize_rcu(); |
2274 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2275 | } |
2276 | ||
b444ad46 NB |
2277 | static struct btrfs_device *btrfs_find_device_by_path( |
2278 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2279 | { |
2280 | int ret = 0; | |
2281 | struct btrfs_super_block *disk_super; | |
2282 | u64 devid; | |
2283 | u8 *dev_uuid; | |
2284 | struct block_device *bdev; | |
b444ad46 | 2285 | struct btrfs_device *device; |
7ba15b7d | 2286 | |
7ba15b7d | 2287 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
8f32380d | 2288 | fs_info->bdev_holder, 0, &bdev, &disk_super); |
7ba15b7d | 2289 | if (ret) |
b444ad46 | 2290 | return ERR_PTR(ret); |
8f32380d | 2291 | |
7ba15b7d SB |
2292 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2293 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2294 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2295 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2296 | disk_super->metadata_uuid, true); |
7239ff4b | 2297 | else |
e4319cd9 | 2298 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2299 | disk_super->fsid, true); |
7239ff4b | 2300 | |
8f32380d | 2301 | btrfs_release_disk_super(disk_super); |
b444ad46 NB |
2302 | if (!device) |
2303 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2304 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2305 | return device; |
7ba15b7d SB |
2306 | } |
2307 | ||
5c5c0df0 DS |
2308 | /* |
2309 | * Lookup a device given by device id, or the path if the id is 0. | |
2310 | */ | |
a27a94c2 | 2311 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2312 | struct btrfs_fs_info *fs_info, u64 devid, |
2313 | const char *device_path) | |
24e0474b | 2314 | { |
a27a94c2 | 2315 | struct btrfs_device *device; |
24e0474b | 2316 | |
5c5c0df0 | 2317 | if (devid) { |
e4319cd9 | 2318 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2319 | NULL, true); |
a27a94c2 NB |
2320 | if (!device) |
2321 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2322 | return device; |
2323 | } | |
2324 | ||
2325 | if (!device_path || !device_path[0]) | |
2326 | return ERR_PTR(-EINVAL); | |
2327 | ||
2328 | if (strcmp(device_path, "missing") == 0) { | |
2329 | /* Find first missing device */ | |
2330 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2331 | dev_list) { | |
2332 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2333 | &device->dev_state) && !device->bdev) | |
2334 | return device; | |
d95a830c | 2335 | } |
6e927ceb | 2336 | return ERR_PTR(-ENOENT); |
24e0474b | 2337 | } |
6e927ceb AJ |
2338 | |
2339 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2340 | } |
2341 | ||
2b82032c YZ |
2342 | /* |
2343 | * does all the dirty work required for changing file system's UUID. | |
2344 | */ | |
2ff7e61e | 2345 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2346 | { |
0b246afa | 2347 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2348 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2349 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2350 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2351 | struct btrfs_device *device; |
2352 | u64 super_flags; | |
2353 | ||
a32bf9a3 | 2354 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2355 | if (!fs_devices->seeding) |
2b82032c YZ |
2356 | return -EINVAL; |
2357 | ||
427c8fdd NB |
2358 | /* |
2359 | * Private copy of the seed devices, anchored at | |
2360 | * fs_info->fs_devices->seed_list | |
2361 | */ | |
7239ff4b | 2362 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2363 | if (IS_ERR(seed_devices)) |
2364 | return PTR_ERR(seed_devices); | |
2b82032c | 2365 | |
427c8fdd NB |
2366 | /* |
2367 | * It's necessary to retain a copy of the original seed fs_devices in | |
2368 | * fs_uuids so that filesystems which have been seeded can successfully | |
2369 | * reference the seed device from open_seed_devices. This also supports | |
2370 | * multiple fs seed. | |
2371 | */ | |
e4404d6e YZ |
2372 | old_devices = clone_fs_devices(fs_devices); |
2373 | if (IS_ERR(old_devices)) { | |
2374 | kfree(seed_devices); | |
2375 | return PTR_ERR(old_devices); | |
2b82032c | 2376 | } |
e4404d6e | 2377 | |
c4babc5e | 2378 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2379 | |
e4404d6e YZ |
2380 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2381 | seed_devices->opened = 1; | |
2382 | INIT_LIST_HEAD(&seed_devices->devices); | |
2383 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2384 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2385 | |
321a4bf7 | 2386 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2387 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2388 | synchronize_rcu); | |
2196d6e8 MX |
2389 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2390 | device->fs_devices = seed_devices; | |
c9513edb | 2391 | |
0395d84f | 2392 | fs_devices->seeding = false; |
2b82032c YZ |
2393 | fs_devices->num_devices = 0; |
2394 | fs_devices->open_devices = 0; | |
69611ac8 | 2395 | fs_devices->missing_devices = 0; |
7f0432d0 | 2396 | fs_devices->rotating = false; |
944d3f9f | 2397 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2398 | |
2399 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2400 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2401 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2402 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2403 | |
2b82032c YZ |
2404 | super_flags = btrfs_super_flags(disk_super) & |
2405 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2406 | btrfs_set_super_flags(disk_super, super_flags); | |
2407 | ||
2408 | return 0; | |
2409 | } | |
2410 | ||
2411 | /* | |
01327610 | 2412 | * Store the expected generation for seed devices in device items. |
2b82032c | 2413 | */ |
5c466629 | 2414 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2415 | { |
5c466629 | 2416 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2417 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2418 | struct btrfs_path *path; |
2419 | struct extent_buffer *leaf; | |
2420 | struct btrfs_dev_item *dev_item; | |
2421 | struct btrfs_device *device; | |
2422 | struct btrfs_key key; | |
44880fdc | 2423 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2424 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2425 | u64 devid; | |
2426 | int ret; | |
2427 | ||
2428 | path = btrfs_alloc_path(); | |
2429 | if (!path) | |
2430 | return -ENOMEM; | |
2431 | ||
2b82032c YZ |
2432 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2433 | key.offset = 0; | |
2434 | key.type = BTRFS_DEV_ITEM_KEY; | |
2435 | ||
2436 | while (1) { | |
2437 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2438 | if (ret < 0) | |
2439 | goto error; | |
2440 | ||
2441 | leaf = path->nodes[0]; | |
2442 | next_slot: | |
2443 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2444 | ret = btrfs_next_leaf(root, path); | |
2445 | if (ret > 0) | |
2446 | break; | |
2447 | if (ret < 0) | |
2448 | goto error; | |
2449 | leaf = path->nodes[0]; | |
2450 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2451 | btrfs_release_path(path); |
2b82032c YZ |
2452 | continue; |
2453 | } | |
2454 | ||
2455 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2456 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2457 | key.type != BTRFS_DEV_ITEM_KEY) | |
2458 | break; | |
2459 | ||
2460 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2461 | struct btrfs_dev_item); | |
2462 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2463 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2464 | BTRFS_UUID_SIZE); |
1473b24e | 2465 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2466 | BTRFS_FSID_SIZE); |
e4319cd9 | 2467 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2468 | fs_uuid, true); |
79787eaa | 2469 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2470 | |
2471 | if (device->fs_devices->seeding) { | |
2472 | btrfs_set_device_generation(leaf, dev_item, | |
2473 | device->generation); | |
2474 | btrfs_mark_buffer_dirty(leaf); | |
2475 | } | |
2476 | ||
2477 | path->slots[0]++; | |
2478 | goto next_slot; | |
2479 | } | |
2480 | ret = 0; | |
2481 | error: | |
2482 | btrfs_free_path(path); | |
2483 | return ret; | |
2484 | } | |
2485 | ||
da353f6b | 2486 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2487 | { |
5112febb | 2488 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2489 | struct request_queue *q; |
788f20eb CM |
2490 | struct btrfs_trans_handle *trans; |
2491 | struct btrfs_device *device; | |
2492 | struct block_device *bdev; | |
0b246afa | 2493 | struct super_block *sb = fs_info->sb; |
606686ee | 2494 | struct rcu_string *name; |
5da54bc1 | 2495 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2496 | u64 orig_super_total_bytes; |
2497 | u64 orig_super_num_devices; | |
2b82032c | 2498 | int seeding_dev = 0; |
788f20eb | 2499 | int ret = 0; |
44cab9ba | 2500 | bool locked = false; |
788f20eb | 2501 | |
5da54bc1 | 2502 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2503 | return -EROFS; |
788f20eb | 2504 | |
a5d16333 | 2505 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2506 | fs_info->bdev_holder); |
7f59203a JB |
2507 | if (IS_ERR(bdev)) |
2508 | return PTR_ERR(bdev); | |
a2135011 | 2509 | |
5da54bc1 | 2510 | if (fs_devices->seeding) { |
2b82032c YZ |
2511 | seeding_dev = 1; |
2512 | down_write(&sb->s_umount); | |
2513 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2514 | locked = true; |
2b82032c YZ |
2515 | } |
2516 | ||
b9ba017f | 2517 | sync_blockdev(bdev); |
a2135011 | 2518 | |
f4cfa9bd NB |
2519 | rcu_read_lock(); |
2520 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
788f20eb CM |
2521 | if (device->bdev == bdev) { |
2522 | ret = -EEXIST; | |
f4cfa9bd | 2523 | rcu_read_unlock(); |
2b82032c | 2524 | goto error; |
788f20eb CM |
2525 | } |
2526 | } | |
f4cfa9bd | 2527 | rcu_read_unlock(); |
788f20eb | 2528 | |
0b246afa | 2529 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2530 | if (IS_ERR(device)) { |
788f20eb | 2531 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2532 | ret = PTR_ERR(device); |
2b82032c | 2533 | goto error; |
788f20eb CM |
2534 | } |
2535 | ||
78f2c9e6 | 2536 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2537 | if (!name) { |
2b82032c | 2538 | ret = -ENOMEM; |
5c4cf6c9 | 2539 | goto error_free_device; |
788f20eb | 2540 | } |
606686ee | 2541 | rcu_assign_pointer(device->name, name); |
2b82032c | 2542 | |
a22285a6 | 2543 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2544 | if (IS_ERR(trans)) { |
98d5dc13 | 2545 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2546 | goto error_free_device; |
98d5dc13 TI |
2547 | } |
2548 | ||
d5e2003c | 2549 | q = bdev_get_queue(bdev); |
ebbede42 | 2550 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2551 | device->generation = trans->transid; |
0b246afa JM |
2552 | device->io_width = fs_info->sectorsize; |
2553 | device->io_align = fs_info->sectorsize; | |
2554 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2555 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2556 | fs_info->sectorsize); | |
2cc3c559 | 2557 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2558 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2559 | device->fs_info = fs_info; |
788f20eb | 2560 | device->bdev = bdev; |
e12c9621 | 2561 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2562 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2563 | device->mode = FMODE_EXCL; |
27087f37 | 2564 | device->dev_stats_valid = 1; |
9f6d2510 | 2565 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2566 | |
2b82032c | 2567 | if (seeding_dev) { |
1751e8a6 | 2568 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2569 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2570 | if (ret) { |
2571 | btrfs_abort_transaction(trans, ret); | |
2572 | goto error_trans; | |
2573 | } | |
2b82032c | 2574 | } |
788f20eb | 2575 | |
5da54bc1 | 2576 | device->fs_devices = fs_devices; |
e5e9a520 | 2577 | |
5da54bc1 | 2578 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2579 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2580 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2581 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2582 | fs_devices->num_devices++; | |
2583 | fs_devices->open_devices++; | |
2584 | fs_devices->rw_devices++; | |
2585 | fs_devices->total_devices++; | |
2586 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2587 | |
a5ed45f8 | 2588 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2589 | |
e884f4f0 | 2590 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2591 | fs_devices->rotating = true; |
c289811c | 2592 | |
39379faa | 2593 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2594 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2595 | round_down(orig_super_total_bytes + device->total_bytes, |
2596 | fs_info->sectorsize)); | |
788f20eb | 2597 | |
39379faa NA |
2598 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2599 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2600 | orig_super_num_devices + 1); | |
0d39376a | 2601 | |
2196d6e8 MX |
2602 | /* |
2603 | * we've got more storage, clear any full flags on the space | |
2604 | * infos | |
2605 | */ | |
0b246afa | 2606 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2607 | |
34441361 | 2608 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2609 | |
2610 | /* Add sysfs device entry */ | |
cd36da2e | 2611 | btrfs_sysfs_add_device(device); |
ca10845a | 2612 | |
5da54bc1 | 2613 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2614 | |
2b82032c | 2615 | if (seeding_dev) { |
34441361 | 2616 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2617 | ret = init_first_rw_device(trans); |
34441361 | 2618 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2619 | if (ret) { |
66642832 | 2620 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2621 | goto error_sysfs; |
005d6427 | 2622 | } |
2196d6e8 MX |
2623 | } |
2624 | ||
8e87e856 | 2625 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2626 | if (ret) { |
66642832 | 2627 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2628 | goto error_sysfs; |
2196d6e8 MX |
2629 | } |
2630 | ||
2631 | if (seeding_dev) { | |
5c466629 | 2632 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2633 | if (ret) { |
66642832 | 2634 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2635 | goto error_sysfs; |
005d6427 | 2636 | } |
b2373f25 | 2637 | |
8e560081 NB |
2638 | /* |
2639 | * fs_devices now represents the newly sprouted filesystem and | |
2640 | * its fsid has been changed by btrfs_prepare_sprout | |
2641 | */ | |
2642 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2643 | } |
2644 | ||
3a45bb20 | 2645 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2646 | |
2b82032c YZ |
2647 | if (seeding_dev) { |
2648 | mutex_unlock(&uuid_mutex); | |
2649 | up_write(&sb->s_umount); | |
44cab9ba | 2650 | locked = false; |
788f20eb | 2651 | |
79787eaa JM |
2652 | if (ret) /* transaction commit */ |
2653 | return ret; | |
2654 | ||
2ff7e61e | 2655 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2656 | if (ret < 0) |
0b246afa | 2657 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2658 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2659 | trans = btrfs_attach_transaction(root); |
2660 | if (IS_ERR(trans)) { | |
2661 | if (PTR_ERR(trans) == -ENOENT) | |
2662 | return 0; | |
7132a262 AJ |
2663 | ret = PTR_ERR(trans); |
2664 | trans = NULL; | |
2665 | goto error_sysfs; | |
671415b7 | 2666 | } |
3a45bb20 | 2667 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2668 | } |
c9e9f97b | 2669 | |
7f551d96 AJ |
2670 | /* |
2671 | * Now that we have written a new super block to this device, check all | |
2672 | * other fs_devices list if device_path alienates any other scanned | |
2673 | * device. | |
2674 | * We can ignore the return value as it typically returns -EINVAL and | |
2675 | * only succeeds if the device was an alien. | |
2676 | */ | |
2677 | btrfs_forget_devices(device_path); | |
2678 | ||
2679 | /* Update ctime/mtime for blkid or udev */ | |
5a1972bd | 2680 | update_dev_time(device_path); |
7f551d96 | 2681 | |
2b82032c | 2682 | return ret; |
79787eaa | 2683 | |
d31c32f6 | 2684 | error_sysfs: |
53f8a74c | 2685 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2686 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2687 | mutex_lock(&fs_info->chunk_mutex); | |
2688 | list_del_rcu(&device->dev_list); | |
2689 | list_del(&device->dev_alloc_list); | |
2690 | fs_info->fs_devices->num_devices--; | |
2691 | fs_info->fs_devices->open_devices--; | |
2692 | fs_info->fs_devices->rw_devices--; | |
2693 | fs_info->fs_devices->total_devices--; | |
2694 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2695 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2696 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2697 | orig_super_total_bytes); | |
2698 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2699 | orig_super_num_devices); | |
2700 | mutex_unlock(&fs_info->chunk_mutex); | |
2701 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2702 | error_trans: |
0af2c4bf | 2703 | if (seeding_dev) |
1751e8a6 | 2704 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2705 | if (trans) |
2706 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2707 | error_free_device: |
a425f9d4 | 2708 | btrfs_free_device(device); |
2b82032c | 2709 | error: |
e525fd89 | 2710 | blkdev_put(bdev, FMODE_EXCL); |
44cab9ba | 2711 | if (locked) { |
2b82032c YZ |
2712 | mutex_unlock(&uuid_mutex); |
2713 | up_write(&sb->s_umount); | |
2714 | } | |
c9e9f97b | 2715 | return ret; |
788f20eb CM |
2716 | } |
2717 | ||
d397712b CM |
2718 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2719 | struct btrfs_device *device) | |
0b86a832 CM |
2720 | { |
2721 | int ret; | |
2722 | struct btrfs_path *path; | |
0b246afa | 2723 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2724 | struct btrfs_dev_item *dev_item; |
2725 | struct extent_buffer *leaf; | |
2726 | struct btrfs_key key; | |
2727 | ||
0b86a832 CM |
2728 | path = btrfs_alloc_path(); |
2729 | if (!path) | |
2730 | return -ENOMEM; | |
2731 | ||
2732 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2733 | key.type = BTRFS_DEV_ITEM_KEY; | |
2734 | key.offset = device->devid; | |
2735 | ||
2736 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2737 | if (ret < 0) | |
2738 | goto out; | |
2739 | ||
2740 | if (ret > 0) { | |
2741 | ret = -ENOENT; | |
2742 | goto out; | |
2743 | } | |
2744 | ||
2745 | leaf = path->nodes[0]; | |
2746 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2747 | ||
2748 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2749 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2750 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2751 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2752 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2753 | btrfs_set_device_total_bytes(leaf, dev_item, |
2754 | btrfs_device_get_disk_total_bytes(device)); | |
2755 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2756 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2757 | btrfs_mark_buffer_dirty(leaf); |
2758 | ||
2759 | out: | |
2760 | btrfs_free_path(path); | |
2761 | return ret; | |
2762 | } | |
2763 | ||
2196d6e8 | 2764 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2765 | struct btrfs_device *device, u64 new_size) |
2766 | { | |
0b246afa JM |
2767 | struct btrfs_fs_info *fs_info = device->fs_info; |
2768 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2769 | u64 old_total; |
2770 | u64 diff; | |
8f18cf13 | 2771 | |
ebbede42 | 2772 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2773 | return -EACCES; |
2196d6e8 | 2774 | |
7dfb8be1 NB |
2775 | new_size = round_down(new_size, fs_info->sectorsize); |
2776 | ||
34441361 | 2777 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2778 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2779 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2780 | |
63a212ab | 2781 | if (new_size <= device->total_bytes || |
401e29c1 | 2782 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2783 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2784 | return -EINVAL; |
2196d6e8 | 2785 | } |
2b82032c | 2786 | |
7dfb8be1 NB |
2787 | btrfs_set_super_total_bytes(super_copy, |
2788 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2789 | device->fs_devices->total_rw_bytes += diff; |
2790 | ||
7cc8e58d MX |
2791 | btrfs_device_set_total_bytes(device, new_size); |
2792 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2793 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2794 | if (list_empty(&device->post_commit_list)) |
2795 | list_add_tail(&device->post_commit_list, | |
2796 | &trans->transaction->dev_update_list); | |
34441361 | 2797 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2798 | |
8f18cf13 CM |
2799 | return btrfs_update_device(trans, device); |
2800 | } | |
2801 | ||
f4208794 | 2802 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2803 | { |
f4208794 | 2804 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2805 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2806 | int ret; |
2807 | struct btrfs_path *path; | |
2808 | struct btrfs_key key; | |
2809 | ||
8f18cf13 CM |
2810 | path = btrfs_alloc_path(); |
2811 | if (!path) | |
2812 | return -ENOMEM; | |
2813 | ||
408fbf19 | 2814 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2815 | key.offset = chunk_offset; |
2816 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2817 | ||
2818 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2819 | if (ret < 0) |
2820 | goto out; | |
2821 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2822 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2823 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2824 | ret = -ENOENT; |
2825 | goto out; | |
2826 | } | |
8f18cf13 CM |
2827 | |
2828 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2829 | if (ret < 0) |
0b246afa JM |
2830 | btrfs_handle_fs_error(fs_info, ret, |
2831 | "Failed to delete chunk item."); | |
79787eaa | 2832 | out: |
8f18cf13 | 2833 | btrfs_free_path(path); |
65a246c5 | 2834 | return ret; |
8f18cf13 CM |
2835 | } |
2836 | ||
408fbf19 | 2837 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2838 | { |
0b246afa | 2839 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2840 | struct btrfs_disk_key *disk_key; |
2841 | struct btrfs_chunk *chunk; | |
2842 | u8 *ptr; | |
2843 | int ret = 0; | |
2844 | u32 num_stripes; | |
2845 | u32 array_size; | |
2846 | u32 len = 0; | |
2847 | u32 cur; | |
2848 | struct btrfs_key key; | |
2849 | ||
34441361 | 2850 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2851 | array_size = btrfs_super_sys_array_size(super_copy); |
2852 | ||
2853 | ptr = super_copy->sys_chunk_array; | |
2854 | cur = 0; | |
2855 | ||
2856 | while (cur < array_size) { | |
2857 | disk_key = (struct btrfs_disk_key *)ptr; | |
2858 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2859 | ||
2860 | len = sizeof(*disk_key); | |
2861 | ||
2862 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2863 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2864 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2865 | len += btrfs_chunk_item_size(num_stripes); | |
2866 | } else { | |
2867 | ret = -EIO; | |
2868 | break; | |
2869 | } | |
408fbf19 | 2870 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2871 | key.offset == chunk_offset) { |
2872 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2873 | array_size -= len; | |
2874 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2875 | } else { | |
2876 | ptr += len; | |
2877 | cur += len; | |
2878 | } | |
2879 | } | |
34441361 | 2880 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2881 | return ret; |
2882 | } | |
2883 | ||
60ca842e OS |
2884 | /* |
2885 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2886 | * @logical: Logical block offset in bytes. | |
2887 | * @length: Length of extent in bytes. | |
2888 | * | |
2889 | * Return: Chunk mapping or ERR_PTR. | |
2890 | */ | |
2891 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2892 | u64 logical, u64 length) | |
592d92ee LB |
2893 | { |
2894 | struct extent_map_tree *em_tree; | |
2895 | struct extent_map *em; | |
2896 | ||
c8bf1b67 | 2897 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2898 | read_lock(&em_tree->lock); |
2899 | em = lookup_extent_mapping(em_tree, logical, length); | |
2900 | read_unlock(&em_tree->lock); | |
2901 | ||
2902 | if (!em) { | |
2903 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2904 | logical, length); | |
2905 | return ERR_PTR(-EINVAL); | |
2906 | } | |
2907 | ||
2908 | if (em->start > logical || em->start + em->len < logical) { | |
2909 | btrfs_crit(fs_info, | |
2910 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2911 | logical, length, em->start, em->start + em->len); | |
2912 | free_extent_map(em); | |
2913 | return ERR_PTR(-EINVAL); | |
2914 | } | |
2915 | ||
2916 | /* callers are responsible for dropping em's ref. */ | |
2917 | return em; | |
2918 | } | |
2919 | ||
97aff912 | 2920 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2921 | { |
97aff912 | 2922 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2923 | struct extent_map *em; |
2924 | struct map_lookup *map; | |
2196d6e8 | 2925 | u64 dev_extent_len = 0; |
47ab2a6c | 2926 | int i, ret = 0; |
0b246afa | 2927 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2928 | |
60ca842e | 2929 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2930 | if (IS_ERR(em)) { |
47ab2a6c JB |
2931 | /* |
2932 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2933 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2934 | * do anything we still error out. |
2935 | */ | |
2936 | ASSERT(0); | |
592d92ee | 2937 | return PTR_ERR(em); |
47ab2a6c | 2938 | } |
95617d69 | 2939 | map = em->map_lookup; |
34441361 | 2940 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2941 | check_system_chunk(trans, map->type); |
34441361 | 2942 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2943 | |
57ba4cb8 FM |
2944 | /* |
2945 | * Take the device list mutex to prevent races with the final phase of | |
2946 | * a device replace operation that replaces the device object associated | |
2947 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2948 | */ | |
2949 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2950 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2951 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2952 | ret = btrfs_free_dev_extent(trans, device, |
2953 | map->stripes[i].physical, | |
2954 | &dev_extent_len); | |
47ab2a6c | 2955 | if (ret) { |
57ba4cb8 | 2956 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2957 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2958 | goto out; |
2959 | } | |
a061fc8d | 2960 | |
2196d6e8 | 2961 | if (device->bytes_used > 0) { |
34441361 | 2962 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2963 | btrfs_device_set_bytes_used(device, |
2964 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2965 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2966 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2967 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2968 | } |
a061fc8d | 2969 | |
64bc6c2a NB |
2970 | ret = btrfs_update_device(trans, device); |
2971 | if (ret) { | |
2972 | mutex_unlock(&fs_devices->device_list_mutex); | |
2973 | btrfs_abort_transaction(trans, ret); | |
2974 | goto out; | |
dfe25020 | 2975 | } |
8f18cf13 | 2976 | } |
57ba4cb8 FM |
2977 | mutex_unlock(&fs_devices->device_list_mutex); |
2978 | ||
f4208794 | 2979 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2980 | if (ret) { |
66642832 | 2981 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2982 | goto out; |
2983 | } | |
8f18cf13 | 2984 | |
6bccf3ab | 2985 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2986 | |
8f18cf13 | 2987 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2988 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2989 | if (ret) { |
66642832 | 2990 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2991 | goto out; |
2992 | } | |
8f18cf13 CM |
2993 | } |
2994 | ||
5a98ec01 | 2995 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2996 | if (ret) { |
66642832 | 2997 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2998 | goto out; |
2999 | } | |
2b82032c | 3000 | |
47ab2a6c | 3001 | out: |
2b82032c YZ |
3002 | /* once for us */ |
3003 | free_extent_map(em); | |
47ab2a6c JB |
3004 | return ret; |
3005 | } | |
2b82032c | 3006 | |
5b4aacef | 3007 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3008 | { |
5b4aacef | 3009 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3010 | struct btrfs_trans_handle *trans; |
b0643e59 | 3011 | struct btrfs_block_group *block_group; |
47ab2a6c | 3012 | int ret; |
2b82032c | 3013 | |
67c5e7d4 FM |
3014 | /* |
3015 | * Prevent races with automatic removal of unused block groups. | |
3016 | * After we relocate and before we remove the chunk with offset | |
3017 | * chunk_offset, automatic removal of the block group can kick in, | |
3018 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3019 | * | |
3020 | * Make sure to acquire this mutex before doing a tree search (dev | |
3021 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3022 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3023 | * we release the path used to search the chunk/dev tree and before | |
3024 | * the current task acquires this mutex and calls us. | |
3025 | */ | |
a32bf9a3 | 3026 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3027 | |
47ab2a6c | 3028 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3029 | btrfs_scrub_pause(fs_info); |
0b246afa | 3030 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3031 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3032 | if (ret) |
3033 | return ret; | |
3034 | ||
b0643e59 DZ |
3035 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3036 | if (!block_group) | |
3037 | return -ENOENT; | |
3038 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
3039 | btrfs_put_block_group(block_group); | |
3040 | ||
19c4d2f9 CM |
3041 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3042 | chunk_offset); | |
3043 | if (IS_ERR(trans)) { | |
3044 | ret = PTR_ERR(trans); | |
3045 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3046 | return ret; | |
3047 | } | |
3048 | ||
47ab2a6c | 3049 | /* |
19c4d2f9 CM |
3050 | * step two, delete the device extents and the |
3051 | * chunk tree entries | |
47ab2a6c | 3052 | */ |
97aff912 | 3053 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3054 | btrfs_end_transaction(trans); |
19c4d2f9 | 3055 | return ret; |
2b82032c YZ |
3056 | } |
3057 | ||
2ff7e61e | 3058 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3059 | { |
0b246afa | 3060 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3061 | struct btrfs_path *path; |
3062 | struct extent_buffer *leaf; | |
3063 | struct btrfs_chunk *chunk; | |
3064 | struct btrfs_key key; | |
3065 | struct btrfs_key found_key; | |
2b82032c | 3066 | u64 chunk_type; |
ba1bf481 JB |
3067 | bool retried = false; |
3068 | int failed = 0; | |
2b82032c YZ |
3069 | int ret; |
3070 | ||
3071 | path = btrfs_alloc_path(); | |
3072 | if (!path) | |
3073 | return -ENOMEM; | |
3074 | ||
ba1bf481 | 3075 | again: |
2b82032c YZ |
3076 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3077 | key.offset = (u64)-1; | |
3078 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3079 | ||
3080 | while (1) { | |
0b246afa | 3081 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3082 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3083 | if (ret < 0) { |
0b246afa | 3084 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3085 | goto error; |
67c5e7d4 | 3086 | } |
79787eaa | 3087 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3088 | |
3089 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3090 | key.type); | |
67c5e7d4 | 3091 | if (ret) |
0b246afa | 3092 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3093 | if (ret < 0) |
3094 | goto error; | |
3095 | if (ret > 0) | |
3096 | break; | |
1a40e23b | 3097 | |
2b82032c YZ |
3098 | leaf = path->nodes[0]; |
3099 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3100 | |
2b82032c YZ |
3101 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3102 | struct btrfs_chunk); | |
3103 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3104 | btrfs_release_path(path); |
8f18cf13 | 3105 | |
2b82032c | 3106 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3107 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3108 | if (ret == -ENOSPC) |
3109 | failed++; | |
14586651 HS |
3110 | else |
3111 | BUG_ON(ret); | |
2b82032c | 3112 | } |
0b246afa | 3113 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3114 | |
2b82032c YZ |
3115 | if (found_key.offset == 0) |
3116 | break; | |
3117 | key.offset = found_key.offset - 1; | |
3118 | } | |
3119 | ret = 0; | |
ba1bf481 JB |
3120 | if (failed && !retried) { |
3121 | failed = 0; | |
3122 | retried = true; | |
3123 | goto again; | |
fae7f21c | 3124 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3125 | ret = -ENOSPC; |
3126 | } | |
2b82032c YZ |
3127 | error: |
3128 | btrfs_free_path(path); | |
3129 | return ret; | |
8f18cf13 CM |
3130 | } |
3131 | ||
a6f93c71 LB |
3132 | /* |
3133 | * return 1 : allocate a data chunk successfully, | |
3134 | * return <0: errors during allocating a data chunk, | |
3135 | * return 0 : no need to allocate a data chunk. | |
3136 | */ | |
3137 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3138 | u64 chunk_offset) | |
3139 | { | |
32da5386 | 3140 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3141 | u64 bytes_used; |
3142 | u64 chunk_type; | |
3143 | ||
3144 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3145 | ASSERT(cache); | |
3146 | chunk_type = cache->flags; | |
3147 | btrfs_put_block_group(cache); | |
3148 | ||
5ae21692 JT |
3149 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3150 | return 0; | |
3151 | ||
3152 | spin_lock(&fs_info->data_sinfo->lock); | |
3153 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3154 | spin_unlock(&fs_info->data_sinfo->lock); | |
3155 | ||
3156 | if (!bytes_used) { | |
3157 | struct btrfs_trans_handle *trans; | |
3158 | int ret; | |
3159 | ||
3160 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3161 | if (IS_ERR(trans)) | |
3162 | return PTR_ERR(trans); | |
3163 | ||
3164 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3165 | btrfs_end_transaction(trans); | |
3166 | if (ret < 0) | |
3167 | return ret; | |
3168 | return 1; | |
a6f93c71 | 3169 | } |
5ae21692 | 3170 | |
a6f93c71 LB |
3171 | return 0; |
3172 | } | |
3173 | ||
6bccf3ab | 3174 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3175 | struct btrfs_balance_control *bctl) |
3176 | { | |
6bccf3ab | 3177 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3178 | struct btrfs_trans_handle *trans; |
3179 | struct btrfs_balance_item *item; | |
3180 | struct btrfs_disk_balance_args disk_bargs; | |
3181 | struct btrfs_path *path; | |
3182 | struct extent_buffer *leaf; | |
3183 | struct btrfs_key key; | |
3184 | int ret, err; | |
3185 | ||
3186 | path = btrfs_alloc_path(); | |
3187 | if (!path) | |
3188 | return -ENOMEM; | |
3189 | ||
3190 | trans = btrfs_start_transaction(root, 0); | |
3191 | if (IS_ERR(trans)) { | |
3192 | btrfs_free_path(path); | |
3193 | return PTR_ERR(trans); | |
3194 | } | |
3195 | ||
3196 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3197 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3198 | key.offset = 0; |
3199 | ||
3200 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3201 | sizeof(*item)); | |
3202 | if (ret) | |
3203 | goto out; | |
3204 | ||
3205 | leaf = path->nodes[0]; | |
3206 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3207 | ||
b159fa28 | 3208 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3209 | |
3210 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3211 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3212 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3213 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3214 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3215 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3216 | ||
3217 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3218 | ||
3219 | btrfs_mark_buffer_dirty(leaf); | |
3220 | out: | |
3221 | btrfs_free_path(path); | |
3a45bb20 | 3222 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3223 | if (err && !ret) |
3224 | ret = err; | |
3225 | return ret; | |
3226 | } | |
3227 | ||
6bccf3ab | 3228 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3229 | { |
6bccf3ab | 3230 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3231 | struct btrfs_trans_handle *trans; |
3232 | struct btrfs_path *path; | |
3233 | struct btrfs_key key; | |
3234 | int ret, err; | |
3235 | ||
3236 | path = btrfs_alloc_path(); | |
3237 | if (!path) | |
3238 | return -ENOMEM; | |
3239 | ||
3502a8c0 | 3240 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3241 | if (IS_ERR(trans)) { |
3242 | btrfs_free_path(path); | |
3243 | return PTR_ERR(trans); | |
3244 | } | |
3245 | ||
3246 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3247 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3248 | key.offset = 0; |
3249 | ||
3250 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3251 | if (ret < 0) | |
3252 | goto out; | |
3253 | if (ret > 0) { | |
3254 | ret = -ENOENT; | |
3255 | goto out; | |
3256 | } | |
3257 | ||
3258 | ret = btrfs_del_item(trans, root, path); | |
3259 | out: | |
3260 | btrfs_free_path(path); | |
3a45bb20 | 3261 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3262 | if (err && !ret) |
3263 | ret = err; | |
3264 | return ret; | |
3265 | } | |
3266 | ||
59641015 ID |
3267 | /* |
3268 | * This is a heuristic used to reduce the number of chunks balanced on | |
3269 | * resume after balance was interrupted. | |
3270 | */ | |
3271 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3272 | { | |
3273 | /* | |
3274 | * Turn on soft mode for chunk types that were being converted. | |
3275 | */ | |
3276 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3277 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3278 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3279 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3280 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3281 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3282 | ||
3283 | /* | |
3284 | * Turn on usage filter if is not already used. The idea is | |
3285 | * that chunks that we have already balanced should be | |
3286 | * reasonably full. Don't do it for chunks that are being | |
3287 | * converted - that will keep us from relocating unconverted | |
3288 | * (albeit full) chunks. | |
3289 | */ | |
3290 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3291 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3292 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3293 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3294 | bctl->data.usage = 90; | |
3295 | } | |
3296 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3297 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3298 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3299 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3300 | bctl->sys.usage = 90; | |
3301 | } | |
3302 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3303 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3304 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3305 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3306 | bctl->meta.usage = 90; | |
3307 | } | |
3308 | } | |
3309 | ||
149196a2 DS |
3310 | /* |
3311 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3312 | */ | |
3313 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3314 | { |
3315 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3316 | int ret; |
c9e9f97b ID |
3317 | |
3318 | BUG_ON(!fs_info->balance_ctl); | |
3319 | ||
3320 | spin_lock(&fs_info->balance_lock); | |
3321 | fs_info->balance_ctl = NULL; | |
3322 | spin_unlock(&fs_info->balance_lock); | |
3323 | ||
3324 | kfree(bctl); | |
149196a2 DS |
3325 | ret = del_balance_item(fs_info); |
3326 | if (ret) | |
3327 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3328 | } |
3329 | ||
ed25e9b2 ID |
3330 | /* |
3331 | * Balance filters. Return 1 if chunk should be filtered out | |
3332 | * (should not be balanced). | |
3333 | */ | |
899c81ea | 3334 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3335 | struct btrfs_balance_args *bargs) |
3336 | { | |
899c81ea ID |
3337 | chunk_type = chunk_to_extended(chunk_type) & |
3338 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3339 | |
899c81ea | 3340 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3341 | return 0; |
3342 | ||
3343 | return 1; | |
3344 | } | |
3345 | ||
dba72cb3 | 3346 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3347 | struct btrfs_balance_args *bargs) |
bc309467 | 3348 | { |
32da5386 | 3349 | struct btrfs_block_group *cache; |
bc309467 DS |
3350 | u64 chunk_used; |
3351 | u64 user_thresh_min; | |
3352 | u64 user_thresh_max; | |
3353 | int ret = 1; | |
3354 | ||
3355 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3356 | chunk_used = cache->used; |
bc309467 DS |
3357 | |
3358 | if (bargs->usage_min == 0) | |
3359 | user_thresh_min = 0; | |
3360 | else | |
b3470b5d DS |
3361 | user_thresh_min = div_factor_fine(cache->length, |
3362 | bargs->usage_min); | |
bc309467 DS |
3363 | |
3364 | if (bargs->usage_max == 0) | |
3365 | user_thresh_max = 1; | |
3366 | else if (bargs->usage_max > 100) | |
b3470b5d | 3367 | user_thresh_max = cache->length; |
bc309467 | 3368 | else |
b3470b5d DS |
3369 | user_thresh_max = div_factor_fine(cache->length, |
3370 | bargs->usage_max); | |
bc309467 DS |
3371 | |
3372 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3373 | ret = 0; | |
3374 | ||
3375 | btrfs_put_block_group(cache); | |
3376 | return ret; | |
3377 | } | |
3378 | ||
dba72cb3 | 3379 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3380 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3381 | { |
32da5386 | 3382 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3383 | u64 chunk_used, user_thresh; |
3384 | int ret = 1; | |
3385 | ||
3386 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3387 | chunk_used = cache->used; |
5ce5b3c0 | 3388 | |
bc309467 | 3389 | if (bargs->usage_min == 0) |
3e39cea6 | 3390 | user_thresh = 1; |
a105bb88 | 3391 | else if (bargs->usage > 100) |
b3470b5d | 3392 | user_thresh = cache->length; |
a105bb88 | 3393 | else |
b3470b5d | 3394 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3395 | |
5ce5b3c0 ID |
3396 | if (chunk_used < user_thresh) |
3397 | ret = 0; | |
3398 | ||
3399 | btrfs_put_block_group(cache); | |
3400 | return ret; | |
3401 | } | |
3402 | ||
409d404b ID |
3403 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3404 | struct btrfs_chunk *chunk, | |
3405 | struct btrfs_balance_args *bargs) | |
3406 | { | |
3407 | struct btrfs_stripe *stripe; | |
3408 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3409 | int i; | |
3410 | ||
3411 | for (i = 0; i < num_stripes; i++) { | |
3412 | stripe = btrfs_stripe_nr(chunk, i); | |
3413 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3414 | return 0; | |
3415 | } | |
3416 | ||
3417 | return 1; | |
3418 | } | |
3419 | ||
946c9256 DS |
3420 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3421 | { | |
3422 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3423 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3424 | const int nparity = btrfs_raid_array[index].nparity; | |
3425 | ||
3426 | if (nparity) | |
3427 | return num_stripes - nparity; | |
3428 | else | |
3429 | return num_stripes / ncopies; | |
3430 | } | |
3431 | ||
94e60d5a ID |
3432 | /* [pstart, pend) */ |
3433 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3434 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3435 | struct btrfs_balance_args *bargs) |
3436 | { | |
3437 | struct btrfs_stripe *stripe; | |
3438 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3439 | u64 stripe_offset; | |
3440 | u64 stripe_length; | |
946c9256 | 3441 | u64 type; |
94e60d5a ID |
3442 | int factor; |
3443 | int i; | |
3444 | ||
3445 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3446 | return 0; | |
3447 | ||
946c9256 DS |
3448 | type = btrfs_chunk_type(leaf, chunk); |
3449 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3450 | |
3451 | for (i = 0; i < num_stripes; i++) { | |
3452 | stripe = btrfs_stripe_nr(chunk, i); | |
3453 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3454 | continue; | |
3455 | ||
3456 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3457 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3458 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3459 | |
3460 | if (stripe_offset < bargs->pend && | |
3461 | stripe_offset + stripe_length > bargs->pstart) | |
3462 | return 0; | |
3463 | } | |
3464 | ||
3465 | return 1; | |
3466 | } | |
3467 | ||
ea67176a ID |
3468 | /* [vstart, vend) */ |
3469 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3470 | struct btrfs_chunk *chunk, | |
3471 | u64 chunk_offset, | |
3472 | struct btrfs_balance_args *bargs) | |
3473 | { | |
3474 | if (chunk_offset < bargs->vend && | |
3475 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3476 | /* at least part of the chunk is inside this vrange */ | |
3477 | return 0; | |
3478 | ||
3479 | return 1; | |
3480 | } | |
3481 | ||
dee32d0a GAP |
3482 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3483 | struct btrfs_chunk *chunk, | |
3484 | struct btrfs_balance_args *bargs) | |
3485 | { | |
3486 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3487 | ||
3488 | if (bargs->stripes_min <= num_stripes | |
3489 | && num_stripes <= bargs->stripes_max) | |
3490 | return 0; | |
3491 | ||
3492 | return 1; | |
3493 | } | |
3494 | ||
899c81ea | 3495 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3496 | struct btrfs_balance_args *bargs) |
3497 | { | |
3498 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3499 | return 0; | |
3500 | ||
899c81ea ID |
3501 | chunk_type = chunk_to_extended(chunk_type) & |
3502 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3503 | |
899c81ea | 3504 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3505 | return 1; |
3506 | ||
3507 | return 0; | |
3508 | } | |
3509 | ||
6ec0896c | 3510 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3511 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3512 | { | |
6ec0896c | 3513 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3514 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3515 | struct btrfs_balance_args *bargs = NULL; |
3516 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3517 | ||
3518 | /* type filter */ | |
3519 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3520 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3521 | return 0; | |
3522 | } | |
3523 | ||
3524 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3525 | bargs = &bctl->data; | |
3526 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3527 | bargs = &bctl->sys; | |
3528 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3529 | bargs = &bctl->meta; | |
3530 | ||
ed25e9b2 ID |
3531 | /* profiles filter */ |
3532 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3533 | chunk_profiles_filter(chunk_type, bargs)) { | |
3534 | return 0; | |
5ce5b3c0 ID |
3535 | } |
3536 | ||
3537 | /* usage filter */ | |
3538 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3539 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3540 | return 0; |
bc309467 | 3541 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3542 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3543 | return 0; |
409d404b ID |
3544 | } |
3545 | ||
3546 | /* devid filter */ | |
3547 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3548 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3549 | return 0; | |
94e60d5a ID |
3550 | } |
3551 | ||
3552 | /* drange filter, makes sense only with devid filter */ | |
3553 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3554 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3555 | return 0; |
ea67176a ID |
3556 | } |
3557 | ||
3558 | /* vrange filter */ | |
3559 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3560 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3561 | return 0; | |
ed25e9b2 ID |
3562 | } |
3563 | ||
dee32d0a GAP |
3564 | /* stripes filter */ |
3565 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3566 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3567 | return 0; | |
3568 | } | |
3569 | ||
cfa4c961 ID |
3570 | /* soft profile changing mode */ |
3571 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3572 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3573 | return 0; | |
3574 | } | |
3575 | ||
7d824b6f DS |
3576 | /* |
3577 | * limited by count, must be the last filter | |
3578 | */ | |
3579 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3580 | if (bargs->limit == 0) | |
3581 | return 0; | |
3582 | else | |
3583 | bargs->limit--; | |
12907fc7 DS |
3584 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3585 | /* | |
3586 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3587 | * determined here because we do not have the global information |
12907fc7 DS |
3588 | * about the count of all chunks that satisfy the filters. |
3589 | */ | |
3590 | if (bargs->limit_max == 0) | |
3591 | return 0; | |
3592 | else | |
3593 | bargs->limit_max--; | |
7d824b6f DS |
3594 | } |
3595 | ||
f43ffb60 ID |
3596 | return 1; |
3597 | } | |
3598 | ||
c9e9f97b | 3599 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3600 | { |
19a39dce | 3601 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3602 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3603 | u64 chunk_type; |
f43ffb60 | 3604 | struct btrfs_chunk *chunk; |
5a488b9d | 3605 | struct btrfs_path *path = NULL; |
ec44a35c | 3606 | struct btrfs_key key; |
ec44a35c | 3607 | struct btrfs_key found_key; |
f43ffb60 ID |
3608 | struct extent_buffer *leaf; |
3609 | int slot; | |
c9e9f97b ID |
3610 | int ret; |
3611 | int enospc_errors = 0; | |
19a39dce | 3612 | bool counting = true; |
12907fc7 | 3613 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3614 | u64 limit_data = bctl->data.limit; |
3615 | u64 limit_meta = bctl->meta.limit; | |
3616 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3617 | u32 count_data = 0; |
3618 | u32 count_meta = 0; | |
3619 | u32 count_sys = 0; | |
2c9fe835 | 3620 | int chunk_reserved = 0; |
ec44a35c | 3621 | |
ec44a35c | 3622 | path = btrfs_alloc_path(); |
17e9f796 MF |
3623 | if (!path) { |
3624 | ret = -ENOMEM; | |
3625 | goto error; | |
3626 | } | |
19a39dce ID |
3627 | |
3628 | /* zero out stat counters */ | |
3629 | spin_lock(&fs_info->balance_lock); | |
3630 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3631 | spin_unlock(&fs_info->balance_lock); | |
3632 | again: | |
7d824b6f | 3633 | if (!counting) { |
12907fc7 DS |
3634 | /* |
3635 | * The single value limit and min/max limits use the same bytes | |
3636 | * in the | |
3637 | */ | |
7d824b6f DS |
3638 | bctl->data.limit = limit_data; |
3639 | bctl->meta.limit = limit_meta; | |
3640 | bctl->sys.limit = limit_sys; | |
3641 | } | |
ec44a35c CM |
3642 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3643 | key.offset = (u64)-1; | |
3644 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3645 | ||
d397712b | 3646 | while (1) { |
19a39dce | 3647 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3648 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3649 | ret = -ECANCELED; |
3650 | goto error; | |
3651 | } | |
3652 | ||
67c5e7d4 | 3653 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3654 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3655 | if (ret < 0) { |
3656 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3657 | goto error; |
67c5e7d4 | 3658 | } |
ec44a35c CM |
3659 | |
3660 | /* | |
3661 | * this shouldn't happen, it means the last relocate | |
3662 | * failed | |
3663 | */ | |
3664 | if (ret == 0) | |
c9e9f97b | 3665 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3666 | |
3667 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3668 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3669 | if (ret) { |
67c5e7d4 | 3670 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3671 | ret = 0; |
ec44a35c | 3672 | break; |
c9e9f97b | 3673 | } |
7d9eb12c | 3674 | |
f43ffb60 ID |
3675 | leaf = path->nodes[0]; |
3676 | slot = path->slots[0]; | |
3677 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3678 | |
67c5e7d4 FM |
3679 | if (found_key.objectid != key.objectid) { |
3680 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3681 | break; |
67c5e7d4 | 3682 | } |
7d9eb12c | 3683 | |
f43ffb60 | 3684 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3685 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3686 | |
19a39dce ID |
3687 | if (!counting) { |
3688 | spin_lock(&fs_info->balance_lock); | |
3689 | bctl->stat.considered++; | |
3690 | spin_unlock(&fs_info->balance_lock); | |
3691 | } | |
3692 | ||
6ec0896c | 3693 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3694 | |
b3b4aa74 | 3695 | btrfs_release_path(path); |
67c5e7d4 FM |
3696 | if (!ret) { |
3697 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3698 | goto loop; |
67c5e7d4 | 3699 | } |
f43ffb60 | 3700 | |
19a39dce | 3701 | if (counting) { |
67c5e7d4 | 3702 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3703 | spin_lock(&fs_info->balance_lock); |
3704 | bctl->stat.expected++; | |
3705 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3706 | |
3707 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3708 | count_data++; | |
3709 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3710 | count_sys++; | |
3711 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3712 | count_meta++; | |
3713 | ||
3714 | goto loop; | |
3715 | } | |
3716 | ||
3717 | /* | |
3718 | * Apply limit_min filter, no need to check if the LIMITS | |
3719 | * filter is used, limit_min is 0 by default | |
3720 | */ | |
3721 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3722 | count_data < bctl->data.limit_min) | |
3723 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3724 | count_meta < bctl->meta.limit_min) | |
3725 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3726 | count_sys < bctl->sys.limit_min)) { | |
3727 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3728 | goto loop; |
3729 | } | |
3730 | ||
a6f93c71 LB |
3731 | if (!chunk_reserved) { |
3732 | /* | |
3733 | * We may be relocating the only data chunk we have, | |
3734 | * which could potentially end up with losing data's | |
3735 | * raid profile, so lets allocate an empty one in | |
3736 | * advance. | |
3737 | */ | |
3738 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3739 | found_key.offset); | |
2c9fe835 ZL |
3740 | if (ret < 0) { |
3741 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3742 | goto error; | |
a6f93c71 LB |
3743 | } else if (ret == 1) { |
3744 | chunk_reserved = 1; | |
2c9fe835 | 3745 | } |
2c9fe835 ZL |
3746 | } |
3747 | ||
5b4aacef | 3748 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3749 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3750 | if (ret == -ENOSPC) { |
c9e9f97b | 3751 | enospc_errors++; |
eede2bf3 OS |
3752 | } else if (ret == -ETXTBSY) { |
3753 | btrfs_info(fs_info, | |
3754 | "skipping relocation of block group %llu due to active swapfile", | |
3755 | found_key.offset); | |
3756 | ret = 0; | |
3757 | } else if (ret) { | |
3758 | goto error; | |
19a39dce ID |
3759 | } else { |
3760 | spin_lock(&fs_info->balance_lock); | |
3761 | bctl->stat.completed++; | |
3762 | spin_unlock(&fs_info->balance_lock); | |
3763 | } | |
f43ffb60 | 3764 | loop: |
795a3321 ID |
3765 | if (found_key.offset == 0) |
3766 | break; | |
ba1bf481 | 3767 | key.offset = found_key.offset - 1; |
ec44a35c | 3768 | } |
c9e9f97b | 3769 | |
19a39dce ID |
3770 | if (counting) { |
3771 | btrfs_release_path(path); | |
3772 | counting = false; | |
3773 | goto again; | |
3774 | } | |
ec44a35c CM |
3775 | error: |
3776 | btrfs_free_path(path); | |
c9e9f97b | 3777 | if (enospc_errors) { |
efe120a0 | 3778 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3779 | enospc_errors); |
c9e9f97b ID |
3780 | if (!ret) |
3781 | ret = -ENOSPC; | |
3782 | } | |
3783 | ||
ec44a35c CM |
3784 | return ret; |
3785 | } | |
3786 | ||
0c460c0d ID |
3787 | /** |
3788 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3789 | * @flags: profile to validate | |
3790 | * @extended: if true @flags is treated as an extended profile | |
3791 | */ | |
3792 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3793 | { | |
3794 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3795 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3796 | ||
3797 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3798 | ||
3799 | /* 1) check that all other bits are zeroed */ | |
3800 | if (flags & ~mask) | |
3801 | return 0; | |
3802 | ||
3803 | /* 2) see if profile is reduced */ | |
3804 | if (flags == 0) | |
3805 | return !extended; /* "0" is valid for usual profiles */ | |
3806 | ||
c1499166 | 3807 | return has_single_bit_set(flags); |
0c460c0d ID |
3808 | } |
3809 | ||
837d5b6e ID |
3810 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3811 | { | |
a7e99c69 ID |
3812 | /* cancel requested || normal exit path */ |
3813 | return atomic_read(&fs_info->balance_cancel_req) || | |
3814 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3815 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3816 | } |
3817 | ||
5ba366c3 DS |
3818 | /* |
3819 | * Validate target profile against allowed profiles and return true if it's OK. | |
3820 | * Otherwise print the error message and return false. | |
3821 | */ | |
3822 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
3823 | const struct btrfs_balance_args *bargs, | |
3824 | u64 allowed, const char *type) | |
bdcd3c97 | 3825 | { |
5ba366c3 DS |
3826 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
3827 | return true; | |
3828 | ||
3829 | /* Profile is valid and does not have bits outside of the allowed set */ | |
3830 | if (alloc_profile_is_valid(bargs->target, 1) && | |
3831 | (bargs->target & ~allowed) == 0) | |
3832 | return true; | |
3833 | ||
3834 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
3835 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
3836 | return false; | |
bdcd3c97 AM |
3837 | } |
3838 | ||
56fc37d9 AJ |
3839 | /* |
3840 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3841 | * @size_buf including the terminating null. The output may be trimmed if it | |
3842 | * does not fit into the provided buffer. | |
3843 | */ | |
3844 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3845 | u32 size_buf) | |
3846 | { | |
3847 | int ret; | |
3848 | u32 size_bp = size_buf; | |
3849 | char *bp = buf; | |
3850 | u64 flags = bargs->flags; | |
3851 | char tmp_buf[128] = {'\0'}; | |
3852 | ||
3853 | if (!flags) | |
3854 | return; | |
3855 | ||
3856 | #define CHECK_APPEND_NOARG(a) \ | |
3857 | do { \ | |
3858 | ret = snprintf(bp, size_bp, (a)); \ | |
3859 | if (ret < 0 || ret >= size_bp) \ | |
3860 | goto out_overflow; \ | |
3861 | size_bp -= ret; \ | |
3862 | bp += ret; \ | |
3863 | } while (0) | |
3864 | ||
3865 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3866 | do { \ | |
3867 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3868 | if (ret < 0 || ret >= size_bp) \ | |
3869 | goto out_overflow; \ | |
3870 | size_bp -= ret; \ | |
3871 | bp += ret; \ | |
3872 | } while (0) | |
3873 | ||
3874 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3875 | do { \ | |
3876 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3877 | if (ret < 0 || ret >= size_bp) \ | |
3878 | goto out_overflow; \ | |
3879 | size_bp -= ret; \ | |
3880 | bp += ret; \ | |
3881 | } while (0) | |
3882 | ||
158da513 DS |
3883 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3884 | CHECK_APPEND_1ARG("convert=%s,", | |
3885 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3886 | |
3887 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3888 | CHECK_APPEND_NOARG("soft,"); | |
3889 | ||
3890 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3891 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3892 | sizeof(tmp_buf)); | |
3893 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3894 | } | |
3895 | ||
3896 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3897 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3898 | ||
3899 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3900 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3901 | bargs->usage_min, bargs->usage_max); | |
3902 | ||
3903 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3904 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3905 | ||
3906 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3907 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3908 | bargs->pstart, bargs->pend); | |
3909 | ||
3910 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3911 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3912 | bargs->vstart, bargs->vend); | |
3913 | ||
3914 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3915 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3916 | ||
3917 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3918 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3919 | bargs->limit_min, bargs->limit_max); | |
3920 | ||
3921 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3922 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3923 | bargs->stripes_min, bargs->stripes_max); | |
3924 | ||
3925 | #undef CHECK_APPEND_2ARG | |
3926 | #undef CHECK_APPEND_1ARG | |
3927 | #undef CHECK_APPEND_NOARG | |
3928 | ||
3929 | out_overflow: | |
3930 | ||
3931 | if (size_bp < size_buf) | |
3932 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3933 | else | |
3934 | buf[0] = '\0'; | |
3935 | } | |
3936 | ||
3937 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3938 | { | |
3939 | u32 size_buf = 1024; | |
3940 | char tmp_buf[192] = {'\0'}; | |
3941 | char *buf; | |
3942 | char *bp; | |
3943 | u32 size_bp = size_buf; | |
3944 | int ret; | |
3945 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3946 | ||
3947 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3948 | if (!buf) | |
3949 | return; | |
3950 | ||
3951 | bp = buf; | |
3952 | ||
3953 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3954 | do { \ | |
3955 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3956 | if (ret < 0 || ret >= size_bp) \ | |
3957 | goto out_overflow; \ | |
3958 | size_bp -= ret; \ | |
3959 | bp += ret; \ | |
3960 | } while (0) | |
3961 | ||
3962 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3963 | CHECK_APPEND_1ARG("%s", "-f "); | |
3964 | ||
3965 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3966 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3967 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3968 | } | |
3969 | ||
3970 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
3971 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
3972 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
3973 | } | |
3974 | ||
3975 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
3976 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
3977 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
3978 | } | |
3979 | ||
3980 | #undef CHECK_APPEND_1ARG | |
3981 | ||
3982 | out_overflow: | |
3983 | ||
3984 | if (size_bp < size_buf) | |
3985 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
3986 | btrfs_info(fs_info, "balance: %s %s", | |
3987 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
3988 | "resume" : "start", buf); | |
3989 | ||
3990 | kfree(buf); | |
3991 | } | |
3992 | ||
c9e9f97b | 3993 | /* |
dccdb07b | 3994 | * Should be called with balance mutexe held |
c9e9f97b | 3995 | */ |
6fcf6e2b DS |
3996 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3997 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3998 | struct btrfs_ioctl_balance_args *bargs) |
3999 | { | |
14506127 | 4000 | u64 meta_target, data_target; |
f43ffb60 | 4001 | u64 allowed; |
e4837f8f | 4002 | int mixed = 0; |
c9e9f97b | 4003 | int ret; |
8dabb742 | 4004 | u64 num_devices; |
de98ced9 | 4005 | unsigned seq; |
e62869be | 4006 | bool reducing_redundancy; |
081db89b | 4007 | int i; |
c9e9f97b | 4008 | |
837d5b6e | 4009 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4010 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4011 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4012 | ret = -EINVAL; |
4013 | goto out; | |
4014 | } | |
4015 | ||
e4837f8f ID |
4016 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4017 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4018 | mixed = 1; | |
4019 | ||
f43ffb60 ID |
4020 | /* |
4021 | * In case of mixed groups both data and meta should be picked, | |
4022 | * and identical options should be given for both of them. | |
4023 | */ | |
e4837f8f ID |
4024 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4025 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4026 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4027 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4028 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4029 | btrfs_err(fs_info, |
6dac13f8 | 4030 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4031 | ret = -EINVAL; |
4032 | goto out; | |
4033 | } | |
4034 | } | |
4035 | ||
b35cf1f0 JB |
4036 | /* |
4037 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4038 | * are exclusive |
b35cf1f0 JB |
4039 | */ |
4040 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4041 | |
4042 | /* | |
4043 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4044 | * special bit for it, to make it easier to distinguish. Thus we need | |
4045 | * to set it manually, or balance would refuse the profile. | |
4046 | */ | |
4047 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4048 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4049 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4050 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4051 | |
5ba366c3 DS |
4052 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4053 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4054 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4055 | ret = -EINVAL; |
4056 | goto out; | |
4057 | } | |
4058 | ||
6079e12c DS |
4059 | /* |
4060 | * Allow to reduce metadata or system integrity only if force set for | |
4061 | * profiles with redundancy (copies, parity) | |
4062 | */ | |
4063 | allowed = 0; | |
4064 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4065 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4066 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4067 | allowed |= btrfs_raid_array[i].bg_flag; | |
4068 | } | |
de98ced9 MX |
4069 | do { |
4070 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4071 | ||
4072 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4073 | (fs_info->avail_system_alloc_bits & allowed) && | |
4074 | !(bctl->sys.target & allowed)) || | |
4075 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4076 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4077 | !(bctl->meta.target & allowed))) |
e62869be | 4078 | reducing_redundancy = true; |
5a8067c0 | 4079 | else |
e62869be | 4080 | reducing_redundancy = false; |
5a8067c0 FM |
4081 | |
4082 | /* if we're not converting, the target field is uninitialized */ | |
4083 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4084 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4085 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4086 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4087 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4088 | |
e62869be | 4089 | if (reducing_redundancy) { |
5a8067c0 FM |
4090 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4091 | btrfs_info(fs_info, | |
e62869be | 4092 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4093 | } else { |
4094 | btrfs_err(fs_info, | |
e62869be | 4095 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4096 | ret = -EINVAL; |
4097 | goto out; | |
4098 | } | |
4099 | } | |
4100 | ||
14506127 AB |
4101 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4102 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4103 | btrfs_warn(fs_info, |
6dac13f8 | 4104 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4105 | btrfs_bg_type_to_raid_name(meta_target), |
4106 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4107 | } |
4108 | ||
9e967495 FM |
4109 | if (fs_info->send_in_progress) { |
4110 | btrfs_warn_rl(fs_info, | |
4111 | "cannot run balance while send operations are in progress (%d in progress)", | |
4112 | fs_info->send_in_progress); | |
4113 | ret = -EAGAIN; | |
4114 | goto out; | |
4115 | } | |
4116 | ||
6bccf3ab | 4117 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4118 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4119 | goto out; |
4120 | ||
59641015 ID |
4121 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4122 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4123 | BUG_ON(fs_info->balance_ctl); |
4124 | spin_lock(&fs_info->balance_lock); | |
4125 | fs_info->balance_ctl = bctl; | |
4126 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4127 | } else { |
4128 | BUG_ON(ret != -EEXIST); | |
4129 | spin_lock(&fs_info->balance_lock); | |
4130 | update_balance_args(bctl); | |
4131 | spin_unlock(&fs_info->balance_lock); | |
4132 | } | |
c9e9f97b | 4133 | |
3009a62f DS |
4134 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4135 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4136 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4137 | mutex_unlock(&fs_info->balance_mutex); |
4138 | ||
4139 | ret = __btrfs_balance(fs_info); | |
4140 | ||
4141 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4142 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4143 | btrfs_info(fs_info, "balance: paused"); | |
44d354ab QW |
4144 | /* |
4145 | * Balance can be canceled by: | |
4146 | * | |
4147 | * - Regular cancel request | |
4148 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4149 | * | |
4150 | * - Fatal signal to "btrfs" process | |
4151 | * Either the signal caught by wait_reserve_ticket() and callers | |
4152 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4153 | * got -ECANCELED. | |
4154 | * Either way, in this case balance_cancel_req = 0, and | |
4155 | * ret == -EINTR or ret == -ECANCELED. | |
4156 | * | |
4157 | * So here we only check the return value to catch canceled balance. | |
4158 | */ | |
4159 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4160 | btrfs_info(fs_info, "balance: canceled"); |
4161 | else | |
4162 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4163 | ||
3009a62f | 4164 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4165 | |
4166 | if (bargs) { | |
4167 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4168 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4169 | } |
4170 | ||
3a01aa7a ID |
4171 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4172 | balance_need_close(fs_info)) { | |
149196a2 | 4173 | reset_balance_state(fs_info); |
c3e1f96c | 4174 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4175 | } |
4176 | ||
837d5b6e | 4177 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4178 | |
4179 | return ret; | |
4180 | out: | |
59641015 | 4181 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4182 | reset_balance_state(fs_info); |
a17c95df | 4183 | else |
59641015 | 4184 | kfree(bctl); |
c3e1f96c | 4185 | btrfs_exclop_finish(fs_info); |
a17c95df | 4186 | |
59641015 ID |
4187 | return ret; |
4188 | } | |
4189 | ||
4190 | static int balance_kthread(void *data) | |
4191 | { | |
2b6ba629 | 4192 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4193 | int ret = 0; |
59641015 | 4194 | |
59641015 | 4195 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4196 | if (fs_info->balance_ctl) |
6fcf6e2b | 4197 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4198 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4199 | |
59641015 ID |
4200 | return ret; |
4201 | } | |
4202 | ||
2b6ba629 ID |
4203 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4204 | { | |
4205 | struct task_struct *tsk; | |
4206 | ||
1354e1a1 | 4207 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4208 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4209 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4210 | return 0; |
4211 | } | |
1354e1a1 | 4212 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4213 | |
3cdde224 | 4214 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4215 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4216 | return 0; |
4217 | } | |
4218 | ||
02ee654d AJ |
4219 | /* |
4220 | * A ro->rw remount sequence should continue with the paused balance | |
4221 | * regardless of who pauses it, system or the user as of now, so set | |
4222 | * the resume flag. | |
4223 | */ | |
4224 | spin_lock(&fs_info->balance_lock); | |
4225 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4226 | spin_unlock(&fs_info->balance_lock); | |
4227 | ||
2b6ba629 | 4228 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4229 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4230 | } |
4231 | ||
68310a5e | 4232 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4233 | { |
59641015 ID |
4234 | struct btrfs_balance_control *bctl; |
4235 | struct btrfs_balance_item *item; | |
4236 | struct btrfs_disk_balance_args disk_bargs; | |
4237 | struct btrfs_path *path; | |
4238 | struct extent_buffer *leaf; | |
4239 | struct btrfs_key key; | |
4240 | int ret; | |
4241 | ||
4242 | path = btrfs_alloc_path(); | |
4243 | if (!path) | |
4244 | return -ENOMEM; | |
4245 | ||
59641015 | 4246 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4247 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4248 | key.offset = 0; |
4249 | ||
68310a5e | 4250 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4251 | if (ret < 0) |
68310a5e | 4252 | goto out; |
59641015 ID |
4253 | if (ret > 0) { /* ret = -ENOENT; */ |
4254 | ret = 0; | |
68310a5e ID |
4255 | goto out; |
4256 | } | |
4257 | ||
4258 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4259 | if (!bctl) { | |
4260 | ret = -ENOMEM; | |
4261 | goto out; | |
59641015 ID |
4262 | } |
4263 | ||
4264 | leaf = path->nodes[0]; | |
4265 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4266 | ||
68310a5e ID |
4267 | bctl->flags = btrfs_balance_flags(leaf, item); |
4268 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4269 | |
4270 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4271 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4272 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4273 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4274 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4275 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4276 | ||
eee95e3f DS |
4277 | /* |
4278 | * This should never happen, as the paused balance state is recovered | |
4279 | * during mount without any chance of other exclusive ops to collide. | |
4280 | * | |
4281 | * This gives the exclusive op status to balance and keeps in paused | |
4282 | * state until user intervention (cancel or umount). If the ownership | |
4283 | * cannot be assigned, show a message but do not fail. The balance | |
4284 | * is in a paused state and must have fs_info::balance_ctl properly | |
4285 | * set up. | |
4286 | */ | |
c3e1f96c | 4287 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) |
eee95e3f | 4288 | btrfs_warn(fs_info, |
6dac13f8 | 4289 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4290 | |
68310a5e | 4291 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4292 | BUG_ON(fs_info->balance_ctl); |
4293 | spin_lock(&fs_info->balance_lock); | |
4294 | fs_info->balance_ctl = bctl; | |
4295 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4296 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4297 | out: |
4298 | btrfs_free_path(path); | |
ec44a35c CM |
4299 | return ret; |
4300 | } | |
4301 | ||
837d5b6e ID |
4302 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4303 | { | |
4304 | int ret = 0; | |
4305 | ||
4306 | mutex_lock(&fs_info->balance_mutex); | |
4307 | if (!fs_info->balance_ctl) { | |
4308 | mutex_unlock(&fs_info->balance_mutex); | |
4309 | return -ENOTCONN; | |
4310 | } | |
4311 | ||
3009a62f | 4312 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4313 | atomic_inc(&fs_info->balance_pause_req); |
4314 | mutex_unlock(&fs_info->balance_mutex); | |
4315 | ||
4316 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4317 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4318 | |
4319 | mutex_lock(&fs_info->balance_mutex); | |
4320 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4321 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4322 | atomic_dec(&fs_info->balance_pause_req); |
4323 | } else { | |
4324 | ret = -ENOTCONN; | |
4325 | } | |
4326 | ||
4327 | mutex_unlock(&fs_info->balance_mutex); | |
4328 | return ret; | |
4329 | } | |
4330 | ||
a7e99c69 ID |
4331 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4332 | { | |
4333 | mutex_lock(&fs_info->balance_mutex); | |
4334 | if (!fs_info->balance_ctl) { | |
4335 | mutex_unlock(&fs_info->balance_mutex); | |
4336 | return -ENOTCONN; | |
4337 | } | |
4338 | ||
cf7d20f4 DS |
4339 | /* |
4340 | * A paused balance with the item stored on disk can be resumed at | |
4341 | * mount time if the mount is read-write. Otherwise it's still paused | |
4342 | * and we must not allow cancelling as it deletes the item. | |
4343 | */ | |
4344 | if (sb_rdonly(fs_info->sb)) { | |
4345 | mutex_unlock(&fs_info->balance_mutex); | |
4346 | return -EROFS; | |
4347 | } | |
4348 | ||
a7e99c69 ID |
4349 | atomic_inc(&fs_info->balance_cancel_req); |
4350 | /* | |
4351 | * if we are running just wait and return, balance item is | |
4352 | * deleted in btrfs_balance in this case | |
4353 | */ | |
3009a62f | 4354 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4355 | mutex_unlock(&fs_info->balance_mutex); |
4356 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4357 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4358 | mutex_lock(&fs_info->balance_mutex); |
4359 | } else { | |
a7e99c69 | 4360 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4361 | /* |
4362 | * Lock released to allow other waiters to continue, we'll | |
4363 | * reexamine the status again. | |
4364 | */ | |
a7e99c69 ID |
4365 | mutex_lock(&fs_info->balance_mutex); |
4366 | ||
a17c95df | 4367 | if (fs_info->balance_ctl) { |
149196a2 | 4368 | reset_balance_state(fs_info); |
c3e1f96c | 4369 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4370 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4371 | } |
a7e99c69 ID |
4372 | } |
4373 | ||
3009a62f DS |
4374 | BUG_ON(fs_info->balance_ctl || |
4375 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4376 | atomic_dec(&fs_info->balance_cancel_req); |
4377 | mutex_unlock(&fs_info->balance_mutex); | |
4378 | return 0; | |
4379 | } | |
4380 | ||
97f4dd09 | 4381 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4382 | { |
4383 | struct btrfs_fs_info *fs_info = data; | |
4384 | struct btrfs_root *root = fs_info->tree_root; | |
4385 | struct btrfs_key key; | |
803b2f54 SB |
4386 | struct btrfs_path *path = NULL; |
4387 | int ret = 0; | |
4388 | struct extent_buffer *eb; | |
4389 | int slot; | |
4390 | struct btrfs_root_item root_item; | |
4391 | u32 item_size; | |
f45388f3 | 4392 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4393 | bool closing = false; |
803b2f54 SB |
4394 | |
4395 | path = btrfs_alloc_path(); | |
4396 | if (!path) { | |
4397 | ret = -ENOMEM; | |
4398 | goto out; | |
4399 | } | |
4400 | ||
4401 | key.objectid = 0; | |
4402 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4403 | key.offset = 0; | |
4404 | ||
803b2f54 | 4405 | while (1) { |
c94bec2c JB |
4406 | if (btrfs_fs_closing(fs_info)) { |
4407 | closing = true; | |
4408 | break; | |
4409 | } | |
7c829b72 AJ |
4410 | ret = btrfs_search_forward(root, &key, path, |
4411 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4412 | if (ret) { |
4413 | if (ret > 0) | |
4414 | ret = 0; | |
4415 | break; | |
4416 | } | |
4417 | ||
4418 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4419 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4420 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4421 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4422 | goto skip; | |
4423 | ||
4424 | eb = path->nodes[0]; | |
4425 | slot = path->slots[0]; | |
4426 | item_size = btrfs_item_size_nr(eb, slot); | |
4427 | if (item_size < sizeof(root_item)) | |
4428 | goto skip; | |
4429 | ||
803b2f54 SB |
4430 | read_extent_buffer(eb, &root_item, |
4431 | btrfs_item_ptr_offset(eb, slot), | |
4432 | (int)sizeof(root_item)); | |
4433 | if (btrfs_root_refs(&root_item) == 0) | |
4434 | goto skip; | |
f45388f3 FDBM |
4435 | |
4436 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4437 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4438 | if (trans) | |
4439 | goto update_tree; | |
4440 | ||
4441 | btrfs_release_path(path); | |
803b2f54 SB |
4442 | /* |
4443 | * 1 - subvol uuid item | |
4444 | * 1 - received_subvol uuid item | |
4445 | */ | |
4446 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4447 | if (IS_ERR(trans)) { | |
4448 | ret = PTR_ERR(trans); | |
4449 | break; | |
4450 | } | |
f45388f3 FDBM |
4451 | continue; |
4452 | } else { | |
4453 | goto skip; | |
4454 | } | |
4455 | update_tree: | |
9771a5cf | 4456 | btrfs_release_path(path); |
f45388f3 | 4457 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4458 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4459 | BTRFS_UUID_KEY_SUBVOL, |
4460 | key.objectid); | |
4461 | if (ret < 0) { | |
efe120a0 | 4462 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4463 | ret); |
803b2f54 SB |
4464 | break; |
4465 | } | |
4466 | } | |
4467 | ||
4468 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4469 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4470 | root_item.received_uuid, |
4471 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4472 | key.objectid); | |
4473 | if (ret < 0) { | |
efe120a0 | 4474 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4475 | ret); |
803b2f54 SB |
4476 | break; |
4477 | } | |
4478 | } | |
4479 | ||
f45388f3 | 4480 | skip: |
9771a5cf | 4481 | btrfs_release_path(path); |
803b2f54 | 4482 | if (trans) { |
3a45bb20 | 4483 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4484 | trans = NULL; |
803b2f54 SB |
4485 | if (ret) |
4486 | break; | |
4487 | } | |
4488 | ||
803b2f54 SB |
4489 | if (key.offset < (u64)-1) { |
4490 | key.offset++; | |
4491 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4492 | key.offset = 0; | |
4493 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4494 | } else if (key.objectid < (u64)-1) { | |
4495 | key.offset = 0; | |
4496 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4497 | key.objectid++; | |
4498 | } else { | |
4499 | break; | |
4500 | } | |
4501 | cond_resched(); | |
4502 | } | |
4503 | ||
4504 | out: | |
4505 | btrfs_free_path(path); | |
f45388f3 | 4506 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4507 | btrfs_end_transaction(trans); |
803b2f54 | 4508 | if (ret) |
efe120a0 | 4509 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4510 | else if (!closing) |
afcdd129 | 4511 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4512 | up(&fs_info->uuid_tree_rescan_sem); |
4513 | return 0; | |
4514 | } | |
4515 | ||
f7a81ea4 SB |
4516 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4517 | { | |
4518 | struct btrfs_trans_handle *trans; | |
4519 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4520 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4521 | struct task_struct *task; |
4522 | int ret; | |
f7a81ea4 SB |
4523 | |
4524 | /* | |
4525 | * 1 - root node | |
4526 | * 1 - root item | |
4527 | */ | |
4528 | trans = btrfs_start_transaction(tree_root, 2); | |
4529 | if (IS_ERR(trans)) | |
4530 | return PTR_ERR(trans); | |
4531 | ||
9b7a2440 | 4532 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4533 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4534 | ret = PTR_ERR(uuid_root); |
66642832 | 4535 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4536 | btrfs_end_transaction(trans); |
6d13f549 | 4537 | return ret; |
f7a81ea4 SB |
4538 | } |
4539 | ||
4540 | fs_info->uuid_root = uuid_root; | |
4541 | ||
3a45bb20 | 4542 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4543 | if (ret) |
4544 | return ret; | |
4545 | ||
4546 | down(&fs_info->uuid_tree_rescan_sem); | |
4547 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4548 | if (IS_ERR(task)) { | |
70f80175 | 4549 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4550 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4551 | up(&fs_info->uuid_tree_rescan_sem); |
4552 | return PTR_ERR(task); | |
4553 | } | |
4554 | ||
4555 | return 0; | |
f7a81ea4 | 4556 | } |
803b2f54 | 4557 | |
8f18cf13 CM |
4558 | /* |
4559 | * shrinking a device means finding all of the device extents past | |
4560 | * the new size, and then following the back refs to the chunks. | |
4561 | * The chunk relocation code actually frees the device extent | |
4562 | */ | |
4563 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4564 | { | |
0b246afa JM |
4565 | struct btrfs_fs_info *fs_info = device->fs_info; |
4566 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4567 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4568 | struct btrfs_dev_extent *dev_extent = NULL; |
4569 | struct btrfs_path *path; | |
4570 | u64 length; | |
8f18cf13 CM |
4571 | u64 chunk_offset; |
4572 | int ret; | |
4573 | int slot; | |
ba1bf481 JB |
4574 | int failed = 0; |
4575 | bool retried = false; | |
8f18cf13 CM |
4576 | struct extent_buffer *l; |
4577 | struct btrfs_key key; | |
0b246afa | 4578 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4579 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4580 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4581 | u64 diff; |
61d0d0d2 | 4582 | u64 start; |
7dfb8be1 NB |
4583 | |
4584 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4585 | start = new_size; |
0e4324a4 | 4586 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4587 | |
401e29c1 | 4588 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4589 | return -EINVAL; |
4590 | ||
8f18cf13 CM |
4591 | path = btrfs_alloc_path(); |
4592 | if (!path) | |
4593 | return -ENOMEM; | |
4594 | ||
0338dff6 | 4595 | path->reada = READA_BACK; |
8f18cf13 | 4596 | |
61d0d0d2 NB |
4597 | trans = btrfs_start_transaction(root, 0); |
4598 | if (IS_ERR(trans)) { | |
4599 | btrfs_free_path(path); | |
4600 | return PTR_ERR(trans); | |
4601 | } | |
4602 | ||
34441361 | 4603 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4604 | |
7cc8e58d | 4605 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4606 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4607 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4608 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4609 | } |
61d0d0d2 NB |
4610 | |
4611 | /* | |
4612 | * Once the device's size has been set to the new size, ensure all | |
4613 | * in-memory chunks are synced to disk so that the loop below sees them | |
4614 | * and relocates them accordingly. | |
4615 | */ | |
1c11b63e | 4616 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4617 | mutex_unlock(&fs_info->chunk_mutex); |
4618 | ret = btrfs_commit_transaction(trans); | |
4619 | if (ret) | |
4620 | goto done; | |
4621 | } else { | |
4622 | mutex_unlock(&fs_info->chunk_mutex); | |
4623 | btrfs_end_transaction(trans); | |
4624 | } | |
8f18cf13 | 4625 | |
ba1bf481 | 4626 | again: |
8f18cf13 CM |
4627 | key.objectid = device->devid; |
4628 | key.offset = (u64)-1; | |
4629 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4630 | ||
213e64da | 4631 | do { |
0b246afa | 4632 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4633 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4634 | if (ret < 0) { |
0b246afa | 4635 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4636 | goto done; |
67c5e7d4 | 4637 | } |
8f18cf13 CM |
4638 | |
4639 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4640 | if (ret) |
0b246afa | 4641 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4642 | if (ret < 0) |
4643 | goto done; | |
4644 | if (ret) { | |
4645 | ret = 0; | |
b3b4aa74 | 4646 | btrfs_release_path(path); |
bf1fb512 | 4647 | break; |
8f18cf13 CM |
4648 | } |
4649 | ||
4650 | l = path->nodes[0]; | |
4651 | slot = path->slots[0]; | |
4652 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4653 | ||
ba1bf481 | 4654 | if (key.objectid != device->devid) { |
0b246afa | 4655 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4656 | btrfs_release_path(path); |
bf1fb512 | 4657 | break; |
ba1bf481 | 4658 | } |
8f18cf13 CM |
4659 | |
4660 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4661 | length = btrfs_dev_extent_length(l, dev_extent); | |
4662 | ||
ba1bf481 | 4663 | if (key.offset + length <= new_size) { |
0b246afa | 4664 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4665 | btrfs_release_path(path); |
d6397bae | 4666 | break; |
ba1bf481 | 4667 | } |
8f18cf13 | 4668 | |
8f18cf13 | 4669 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4670 | btrfs_release_path(path); |
8f18cf13 | 4671 | |
a6f93c71 LB |
4672 | /* |
4673 | * We may be relocating the only data chunk we have, | |
4674 | * which could potentially end up with losing data's | |
4675 | * raid profile, so lets allocate an empty one in | |
4676 | * advance. | |
4677 | */ | |
4678 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4679 | if (ret < 0) { | |
4680 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4681 | goto done; | |
4682 | } | |
4683 | ||
0b246afa JM |
4684 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4685 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4686 | if (ret == -ENOSPC) { |
ba1bf481 | 4687 | failed++; |
eede2bf3 OS |
4688 | } else if (ret) { |
4689 | if (ret == -ETXTBSY) { | |
4690 | btrfs_warn(fs_info, | |
4691 | "could not shrink block group %llu due to active swapfile", | |
4692 | chunk_offset); | |
4693 | } | |
4694 | goto done; | |
4695 | } | |
213e64da | 4696 | } while (key.offset-- > 0); |
ba1bf481 JB |
4697 | |
4698 | if (failed && !retried) { | |
4699 | failed = 0; | |
4700 | retried = true; | |
4701 | goto again; | |
4702 | } else if (failed && retried) { | |
4703 | ret = -ENOSPC; | |
ba1bf481 | 4704 | goto done; |
8f18cf13 CM |
4705 | } |
4706 | ||
d6397bae | 4707 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4708 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4709 | if (IS_ERR(trans)) { |
4710 | ret = PTR_ERR(trans); | |
4711 | goto done; | |
4712 | } | |
4713 | ||
34441361 | 4714 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4715 | /* Clear all state bits beyond the shrunk device size */ |
4716 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4717 | CHUNK_STATE_MASK); | |
4718 | ||
7cc8e58d | 4719 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4720 | if (list_empty(&device->post_commit_list)) |
4721 | list_add_tail(&device->post_commit_list, | |
4722 | &trans->transaction->dev_update_list); | |
d6397bae | 4723 | |
d6397bae | 4724 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4725 | btrfs_set_super_total_bytes(super_copy, |
4726 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4727 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4728 | |
4729 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4730 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4731 | if (ret < 0) { |
4732 | btrfs_abort_transaction(trans, ret); | |
4733 | btrfs_end_transaction(trans); | |
4734 | } else { | |
4735 | ret = btrfs_commit_transaction(trans); | |
4736 | } | |
8f18cf13 CM |
4737 | done: |
4738 | btrfs_free_path(path); | |
53e489bc | 4739 | if (ret) { |
34441361 | 4740 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4741 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4742 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4743 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4744 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4745 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4746 | } |
8f18cf13 CM |
4747 | return ret; |
4748 | } | |
4749 | ||
2ff7e61e | 4750 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4751 | struct btrfs_key *key, |
4752 | struct btrfs_chunk *chunk, int item_size) | |
4753 | { | |
0b246afa | 4754 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4755 | struct btrfs_disk_key disk_key; |
4756 | u32 array_size; | |
4757 | u8 *ptr; | |
4758 | ||
34441361 | 4759 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4760 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4761 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4762 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4763 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4764 | return -EFBIG; |
fe48a5c0 | 4765 | } |
0b86a832 CM |
4766 | |
4767 | ptr = super_copy->sys_chunk_array + array_size; | |
4768 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4769 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4770 | ptr += sizeof(disk_key); | |
4771 | memcpy(ptr, chunk, item_size); | |
4772 | item_size += sizeof(disk_key); | |
4773 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4774 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4775 | |
0b86a832 CM |
4776 | return 0; |
4777 | } | |
4778 | ||
73c5de00 AJ |
4779 | /* |
4780 | * sort the devices in descending order by max_avail, total_avail | |
4781 | */ | |
4782 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4783 | { |
73c5de00 AJ |
4784 | const struct btrfs_device_info *di_a = a; |
4785 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4786 | |
73c5de00 | 4787 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4788 | return -1; |
73c5de00 | 4789 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4790 | return 1; |
73c5de00 AJ |
4791 | if (di_a->total_avail > di_b->total_avail) |
4792 | return -1; | |
4793 | if (di_a->total_avail < di_b->total_avail) | |
4794 | return 1; | |
4795 | return 0; | |
b2117a39 | 4796 | } |
0b86a832 | 4797 | |
53b381b3 DW |
4798 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4799 | { | |
ffe2d203 | 4800 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4801 | return; |
4802 | ||
ceda0864 | 4803 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4804 | } |
4805 | ||
cfbb825c DS |
4806 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4807 | { | |
4808 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4809 | return; | |
4810 | ||
4811 | btrfs_set_fs_incompat(info, RAID1C34); | |
4812 | } | |
4813 | ||
4f2bafe8 NA |
4814 | /* |
4815 | * Structure used internally for __btrfs_alloc_chunk() function. | |
4816 | * Wraps needed parameters. | |
4817 | */ | |
4818 | struct alloc_chunk_ctl { | |
4819 | u64 start; | |
4820 | u64 type; | |
4821 | /* Total number of stripes to allocate */ | |
4822 | int num_stripes; | |
4823 | /* sub_stripes info for map */ | |
4824 | int sub_stripes; | |
4825 | /* Stripes per device */ | |
4826 | int dev_stripes; | |
4827 | /* Maximum number of devices to use */ | |
4828 | int devs_max; | |
4829 | /* Minimum number of devices to use */ | |
4830 | int devs_min; | |
4831 | /* ndevs has to be a multiple of this */ | |
4832 | int devs_increment; | |
4833 | /* Number of copies */ | |
4834 | int ncopies; | |
4835 | /* Number of stripes worth of bytes to store parity information */ | |
4836 | int nparity; | |
4837 | u64 max_stripe_size; | |
4838 | u64 max_chunk_size; | |
6aafb303 | 4839 | u64 dev_extent_min; |
4f2bafe8 NA |
4840 | u64 stripe_size; |
4841 | u64 chunk_size; | |
4842 | int ndevs; | |
4843 | }; | |
4844 | ||
27c314d5 NA |
4845 | static void init_alloc_chunk_ctl_policy_regular( |
4846 | struct btrfs_fs_devices *fs_devices, | |
4847 | struct alloc_chunk_ctl *ctl) | |
4848 | { | |
4849 | u64 type = ctl->type; | |
4850 | ||
4851 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4852 | ctl->max_stripe_size = SZ_1G; | |
4853 | ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; | |
4854 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4855 | /* For larger filesystems, use larger metadata chunks */ | |
4856 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) | |
4857 | ctl->max_stripe_size = SZ_1G; | |
4858 | else | |
4859 | ctl->max_stripe_size = SZ_256M; | |
4860 | ctl->max_chunk_size = ctl->max_stripe_size; | |
4861 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4862 | ctl->max_stripe_size = SZ_32M; | |
4863 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
4864 | ctl->devs_max = min_t(int, ctl->devs_max, | |
4865 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
4866 | } else { | |
4867 | BUG(); | |
4868 | } | |
4869 | ||
4870 | /* We don't want a chunk larger than 10% of writable space */ | |
4871 | ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4872 | ctl->max_chunk_size); | |
6aafb303 | 4873 | ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; |
27c314d5 NA |
4874 | } |
4875 | ||
4876 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, | |
4877 | struct alloc_chunk_ctl *ctl) | |
4878 | { | |
4879 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
4880 | ||
4881 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
4882 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4883 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
4884 | if (!ctl->devs_max) | |
4885 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
4886 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
4887 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
4888 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
4889 | ctl->nparity = btrfs_raid_array[index].nparity; | |
4890 | ctl->ndevs = 0; | |
4891 | ||
4892 | switch (fs_devices->chunk_alloc_policy) { | |
4893 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
4894 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
4895 | break; | |
4896 | default: | |
4897 | BUG(); | |
4898 | } | |
4899 | } | |
4900 | ||
560156cb NA |
4901 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
4902 | struct alloc_chunk_ctl *ctl, | |
4903 | struct btrfs_device_info *devices_info) | |
b2117a39 | 4904 | { |
560156cb | 4905 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 4906 | struct btrfs_device *device; |
73c5de00 | 4907 | u64 total_avail; |
560156cb | 4908 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 4909 | int ret; |
560156cb NA |
4910 | int ndevs = 0; |
4911 | u64 max_avail; | |
4912 | u64 dev_offset; | |
0cad8a11 | 4913 | |
9f680ce0 | 4914 | /* |
73c5de00 AJ |
4915 | * in the first pass through the devices list, we gather information |
4916 | * about the available holes on each device. | |
9f680ce0 | 4917 | */ |
ebcc9301 | 4918 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 4919 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4920 | WARN(1, KERN_ERR |
efe120a0 | 4921 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4922 | continue; |
4923 | } | |
b2117a39 | 4924 | |
e12c9621 AJ |
4925 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4926 | &device->dev_state) || | |
401e29c1 | 4927 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4928 | continue; |
b2117a39 | 4929 | |
73c5de00 AJ |
4930 | if (device->total_bytes > device->bytes_used) |
4931 | total_avail = device->total_bytes - device->bytes_used; | |
4932 | else | |
4933 | total_avail = 0; | |
38c01b96 | 4934 | |
4935 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 4936 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 4937 | continue; |
b2117a39 | 4938 | |
560156cb NA |
4939 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
4940 | &max_avail); | |
73c5de00 | 4941 | if (ret && ret != -ENOSPC) |
560156cb | 4942 | return ret; |
b2117a39 | 4943 | |
73c5de00 | 4944 | if (ret == 0) |
560156cb | 4945 | max_avail = dev_extent_want; |
b2117a39 | 4946 | |
6aafb303 | 4947 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
4948 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
4949 | btrfs_debug(info, | |
560156cb | 4950 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 4951 | __func__, device->devid, max_avail, |
6aafb303 | 4952 | ctl->dev_extent_min); |
73c5de00 | 4953 | continue; |
4117f207 | 4954 | } |
b2117a39 | 4955 | |
063d006f ES |
4956 | if (ndevs == fs_devices->rw_devices) { |
4957 | WARN(1, "%s: found more than %llu devices\n", | |
4958 | __func__, fs_devices->rw_devices); | |
4959 | break; | |
4960 | } | |
73c5de00 AJ |
4961 | devices_info[ndevs].dev_offset = dev_offset; |
4962 | devices_info[ndevs].max_avail = max_avail; | |
4963 | devices_info[ndevs].total_avail = total_avail; | |
4964 | devices_info[ndevs].dev = device; | |
4965 | ++ndevs; | |
4966 | } | |
560156cb | 4967 | ctl->ndevs = ndevs; |
b2117a39 | 4968 | |
73c5de00 AJ |
4969 | /* |
4970 | * now sort the devices by hole size / available space | |
4971 | */ | |
560156cb | 4972 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 4973 | btrfs_cmp_device_info, NULL); |
b2117a39 | 4974 | |
560156cb NA |
4975 | return 0; |
4976 | } | |
4977 | ||
5badf512 NA |
4978 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
4979 | struct btrfs_device_info *devices_info) | |
4980 | { | |
4981 | /* Number of stripes that count for block group size */ | |
4982 | int data_stripes; | |
4983 | ||
4984 | /* | |
4985 | * The primary goal is to maximize the number of stripes, so use as | |
4986 | * many devices as possible, even if the stripes are not maximum sized. | |
4987 | * | |
4988 | * The DUP profile stores more than one stripe per device, the | |
4989 | * max_avail is the total size so we have to adjust. | |
4990 | */ | |
4991 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
4992 | ctl->dev_stripes); | |
4993 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
4994 | ||
4995 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
4996 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
4997 | ||
4998 | /* | |
4999 | * Use the number of data stripes to figure out how big this chunk is | |
5000 | * really going to be in terms of logical address space, and compare | |
5001 | * that answer with the max chunk size. If it's higher, we try to | |
5002 | * reduce stripe_size. | |
5003 | */ | |
5004 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5005 | /* | |
5006 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5007 | * then use it, unless it ends up being even bigger than the | |
5008 | * previous value we had already. | |
5009 | */ | |
5010 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5011 | data_stripes), SZ_16M), | |
5012 | ctl->stripe_size); | |
5013 | } | |
5014 | ||
5015 | /* Align to BTRFS_STRIPE_LEN */ | |
5016 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5017 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5018 | ||
5019 | return 0; | |
5020 | } | |
5021 | ||
5022 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, | |
5023 | struct alloc_chunk_ctl *ctl, | |
5024 | struct btrfs_device_info *devices_info) | |
5025 | { | |
5026 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5027 | ||
5028 | /* | |
5029 | * Round down to number of usable stripes, devs_increment can be any | |
5030 | * number so we can't use round_down() that requires power of 2, while | |
5031 | * rounddown is safe. | |
5032 | */ | |
5033 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5034 | ||
5035 | if (ctl->ndevs < ctl->devs_min) { | |
5036 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5037 | btrfs_debug(info, | |
5038 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5039 | __func__, ctl->ndevs, ctl->devs_min); | |
5040 | } | |
5041 | return -ENOSPC; | |
5042 | } | |
5043 | ||
5044 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5045 | ||
5046 | switch (fs_devices->chunk_alloc_policy) { | |
5047 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5048 | return decide_stripe_size_regular(ctl, devices_info); | |
5049 | default: | |
5050 | BUG(); | |
5051 | } | |
5052 | } | |
5053 | ||
dce580ca NA |
5054 | static int create_chunk(struct btrfs_trans_handle *trans, |
5055 | struct alloc_chunk_ctl *ctl, | |
5056 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5057 | { |
5058 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5059 | struct map_lookup *map = NULL; |
5060 | struct extent_map_tree *em_tree; | |
5061 | struct extent_map *em; | |
dce580ca NA |
5062 | u64 start = ctl->start; |
5063 | u64 type = ctl->type; | |
560156cb NA |
5064 | int ret; |
5065 | int i; | |
5066 | int j; | |
5067 | ||
dce580ca NA |
5068 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5069 | if (!map) | |
560156cb | 5070 | return -ENOMEM; |
dce580ca | 5071 | map->num_stripes = ctl->num_stripes; |
560156cb | 5072 | |
dce580ca NA |
5073 | for (i = 0; i < ctl->ndevs; ++i) { |
5074 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5075 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5076 | map->stripes[s].dev = devices_info[i].dev; |
5077 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5078 | j * ctl->stripe_size; |
6324fbf3 | 5079 | } |
6324fbf3 | 5080 | } |
500ceed8 NB |
5081 | map->stripe_len = BTRFS_STRIPE_LEN; |
5082 | map->io_align = BTRFS_STRIPE_LEN; | |
5083 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5084 | map->type = type; |
dce580ca | 5085 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5086 | |
dce580ca | 5087 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5088 | |
172ddd60 | 5089 | em = alloc_extent_map(); |
2b82032c | 5090 | if (!em) { |
298a8f9c | 5091 | kfree(map); |
dce580ca | 5092 | return -ENOMEM; |
593060d7 | 5093 | } |
298a8f9c | 5094 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5095 | em->map_lookup = map; |
2b82032c | 5096 | em->start = start; |
dce580ca | 5097 | em->len = ctl->chunk_size; |
2b82032c YZ |
5098 | em->block_start = 0; |
5099 | em->block_len = em->len; | |
dce580ca | 5100 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5101 | |
c8bf1b67 | 5102 | em_tree = &info->mapping_tree; |
890871be | 5103 | write_lock(&em_tree->lock); |
09a2a8f9 | 5104 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5105 | if (ret) { |
1efb72a3 | 5106 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5107 | free_extent_map(em); |
dce580ca | 5108 | return ret; |
0f5d42b2 | 5109 | } |
1efb72a3 NB |
5110 | write_unlock(&em_tree->lock); |
5111 | ||
dce580ca | 5112 | ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); |
6df9a95e JB |
5113 | if (ret) |
5114 | goto error_del_extent; | |
2b82032c | 5115 | |
bbbf7243 NB |
5116 | for (i = 0; i < map->num_stripes; i++) { |
5117 | struct btrfs_device *dev = map->stripes[i].dev; | |
5118 | ||
4f2bafe8 | 5119 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5120 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5121 | if (list_empty(&dev->post_commit_list)) |
5122 | list_add_tail(&dev->post_commit_list, | |
5123 | &trans->transaction->dev_update_list); | |
5124 | } | |
43530c46 | 5125 | |
dce580ca | 5126 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5127 | &info->free_chunk_space); |
1c116187 | 5128 | |
0f5d42b2 | 5129 | free_extent_map(em); |
0b246afa | 5130 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5131 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5132 | |
2b82032c | 5133 | return 0; |
b2117a39 | 5134 | |
6df9a95e | 5135 | error_del_extent: |
0f5d42b2 JB |
5136 | write_lock(&em_tree->lock); |
5137 | remove_extent_mapping(em_tree, em); | |
5138 | write_unlock(&em_tree->lock); | |
5139 | ||
5140 | /* One for our allocation */ | |
5141 | free_extent_map(em); | |
5142 | /* One for the tree reference */ | |
5143 | free_extent_map(em); | |
dce580ca NA |
5144 | |
5145 | return ret; | |
5146 | } | |
5147 | ||
11c67b1a | 5148 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
dce580ca NA |
5149 | { |
5150 | struct btrfs_fs_info *info = trans->fs_info; | |
5151 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5152 | struct btrfs_device_info *devices_info = NULL; | |
5153 | struct alloc_chunk_ctl ctl; | |
5154 | int ret; | |
5155 | ||
11c67b1a NB |
5156 | lockdep_assert_held(&info->chunk_mutex); |
5157 | ||
dce580ca NA |
5158 | if (!alloc_profile_is_valid(type, 0)) { |
5159 | ASSERT(0); | |
5160 | return -EINVAL; | |
5161 | } | |
5162 | ||
5163 | if (list_empty(&fs_devices->alloc_list)) { | |
5164 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5165 | btrfs_debug(info, "%s: no writable device", __func__); | |
5166 | return -ENOSPC; | |
5167 | } | |
5168 | ||
5169 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5170 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5171 | ASSERT(0); | |
5172 | return -EINVAL; | |
5173 | } | |
5174 | ||
11c67b1a | 5175 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5176 | ctl.type = type; |
5177 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5178 | ||
5179 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5180 | GFP_NOFS); | |
5181 | if (!devices_info) | |
5182 | return -ENOMEM; | |
5183 | ||
5184 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
5185 | if (ret < 0) | |
5186 | goto out; | |
5187 | ||
5188 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
5189 | if (ret < 0) | |
5190 | goto out; | |
5191 | ||
5192 | ret = create_chunk(trans, &ctl, devices_info); | |
5193 | ||
5194 | out: | |
b2117a39 MX |
5195 | kfree(devices_info); |
5196 | return ret; | |
2b82032c YZ |
5197 | } |
5198 | ||
11c67b1a NB |
5199 | /* |
5200 | * Chunk allocation falls into two parts. The first part does work | |
5201 | * that makes the new allocated chunk usable, but does not do any operation | |
5202 | * that modifies the chunk tree. The second part does the work that | |
5203 | * requires modifying the chunk tree. This division is important for the | |
5204 | * bootstrap process of adding storage to a seed btrfs. | |
5205 | */ | |
6df9a95e | 5206 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5207 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5208 | { |
97aff912 | 5209 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5210 | struct btrfs_root *extent_root = fs_info->extent_root; |
5211 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5212 | struct btrfs_key key; |
2b82032c YZ |
5213 | struct btrfs_device *device; |
5214 | struct btrfs_chunk *chunk; | |
5215 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5216 | struct extent_map *em; |
5217 | struct map_lookup *map; | |
5218 | size_t item_size; | |
5219 | u64 dev_offset; | |
5220 | u64 stripe_size; | |
5221 | int i = 0; | |
140e639f | 5222 | int ret = 0; |
2b82032c | 5223 | |
60ca842e | 5224 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5225 | if (IS_ERR(em)) |
5226 | return PTR_ERR(em); | |
6df9a95e | 5227 | |
95617d69 | 5228 | map = em->map_lookup; |
6df9a95e JB |
5229 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5230 | stripe_size = em->orig_block_len; | |
5231 | ||
2b82032c | 5232 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5233 | if (!chunk) { |
5234 | ret = -ENOMEM; | |
5235 | goto out; | |
5236 | } | |
5237 | ||
50460e37 FM |
5238 | /* |
5239 | * Take the device list mutex to prevent races with the final phase of | |
5240 | * a device replace operation that replaces the device object associated | |
5241 | * with the map's stripes, because the device object's id can change | |
5242 | * at any time during that final phase of the device replace operation | |
5243 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5244 | */ | |
0b246afa | 5245 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5246 | for (i = 0; i < map->num_stripes; i++) { |
5247 | device = map->stripes[i].dev; | |
5248 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5249 | |
0b86a832 | 5250 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5251 | if (ret) |
50460e37 | 5252 | break; |
b5d9071c NB |
5253 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5254 | dev_offset, stripe_size); | |
6df9a95e | 5255 | if (ret) |
50460e37 FM |
5256 | break; |
5257 | } | |
5258 | if (ret) { | |
0b246afa | 5259 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5260 | goto out; |
2b82032c YZ |
5261 | } |
5262 | ||
2b82032c | 5263 | stripe = &chunk->stripe; |
6df9a95e JB |
5264 | for (i = 0; i < map->num_stripes; i++) { |
5265 | device = map->stripes[i].dev; | |
5266 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5267 | |
e17cade2 CM |
5268 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5269 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5270 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5271 | stripe++; |
0b86a832 | 5272 | } |
0b246afa | 5273 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5274 | |
2b82032c | 5275 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5276 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5277 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5278 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5279 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5280 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5281 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5282 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5283 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5284 | |
2b82032c YZ |
5285 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5286 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5287 | key.offset = chunk_offset; | |
0b86a832 | 5288 | |
2b82032c | 5289 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5290 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5291 | /* | |
5292 | * TODO: Cleanup of inserted chunk root in case of | |
5293 | * failure. | |
5294 | */ | |
2ff7e61e | 5295 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5296 | } |
1abe9b8a | 5297 | |
6df9a95e | 5298 | out: |
0b86a832 | 5299 | kfree(chunk); |
6df9a95e | 5300 | free_extent_map(em); |
4ed1d16e | 5301 | return ret; |
2b82032c | 5302 | } |
0b86a832 | 5303 | |
6f8e0fc7 | 5304 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5305 | { |
6f8e0fc7 | 5306 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5307 | u64 alloc_profile; |
2b82032c YZ |
5308 | int ret; |
5309 | ||
1b86826d | 5310 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
11c67b1a | 5311 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa JM |
5312 | if (ret) |
5313 | return ret; | |
2b82032c | 5314 | |
1b86826d | 5315 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
11c67b1a | 5316 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa | 5317 | return ret; |
2b82032c YZ |
5318 | } |
5319 | ||
d20983b4 MX |
5320 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5321 | { | |
fc9a2ac7 | 5322 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5323 | |
fc9a2ac7 | 5324 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5325 | } |
5326 | ||
2ff7e61e | 5327 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5328 | { |
5329 | struct extent_map *em; | |
5330 | struct map_lookup *map; | |
2b82032c | 5331 | int readonly = 0; |
d20983b4 | 5332 | int miss_ndevs = 0; |
2b82032c YZ |
5333 | int i; |
5334 | ||
60ca842e | 5335 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5336 | if (IS_ERR(em)) |
2b82032c YZ |
5337 | return 1; |
5338 | ||
95617d69 | 5339 | map = em->map_lookup; |
2b82032c | 5340 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5341 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5342 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5343 | miss_ndevs++; |
5344 | continue; | |
5345 | } | |
ebbede42 AJ |
5346 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5347 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5348 | readonly = 1; |
d20983b4 | 5349 | goto end; |
2b82032c YZ |
5350 | } |
5351 | } | |
d20983b4 MX |
5352 | |
5353 | /* | |
5354 | * If the number of missing devices is larger than max errors, | |
5355 | * we can not write the data into that chunk successfully, so | |
5356 | * set it readonly. | |
5357 | */ | |
5358 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5359 | readonly = 1; | |
5360 | end: | |
0b86a832 | 5361 | free_extent_map(em); |
2b82032c | 5362 | return readonly; |
0b86a832 CM |
5363 | } |
5364 | ||
c8bf1b67 | 5365 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5366 | { |
5367 | struct extent_map *em; | |
5368 | ||
d397712b | 5369 | while (1) { |
c8bf1b67 DS |
5370 | write_lock(&tree->lock); |
5371 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5372 | if (em) |
c8bf1b67 DS |
5373 | remove_extent_mapping(tree, em); |
5374 | write_unlock(&tree->lock); | |
0b86a832 CM |
5375 | if (!em) |
5376 | break; | |
0b86a832 CM |
5377 | /* once for us */ |
5378 | free_extent_map(em); | |
5379 | /* once for the tree */ | |
5380 | free_extent_map(em); | |
5381 | } | |
5382 | } | |
5383 | ||
5d964051 | 5384 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5385 | { |
5386 | struct extent_map *em; | |
5387 | struct map_lookup *map; | |
f188591e CM |
5388 | int ret; |
5389 | ||
60ca842e | 5390 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5391 | if (IS_ERR(em)) |
5392 | /* | |
5393 | * We could return errors for these cases, but that could get | |
5394 | * ugly and we'd probably do the same thing which is just not do | |
5395 | * anything else and exit, so return 1 so the callers don't try | |
5396 | * to use other copies. | |
5397 | */ | |
fb7669b5 | 5398 | return 1; |
fb7669b5 | 5399 | |
95617d69 | 5400 | map = em->map_lookup; |
c7369b3f | 5401 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5402 | ret = map->num_stripes; |
321aecc6 CM |
5403 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5404 | ret = map->sub_stripes; | |
53b381b3 DW |
5405 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5406 | ret = 2; | |
5407 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5408 | /* |
5409 | * There could be two corrupted data stripes, we need | |
5410 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5411 | * |
8810f751 LB |
5412 | * Fail a stripe at a time on every retry except the |
5413 | * stripe under reconstruction. | |
5414 | */ | |
5415 | ret = map->num_stripes; | |
f188591e CM |
5416 | else |
5417 | ret = 1; | |
5418 | free_extent_map(em); | |
ad6d620e | 5419 | |
cb5583dd | 5420 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5421 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5422 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5423 | ret++; |
cb5583dd | 5424 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5425 | |
f188591e CM |
5426 | return ret; |
5427 | } | |
5428 | ||
2ff7e61e | 5429 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5430 | u64 logical) |
5431 | { | |
5432 | struct extent_map *em; | |
5433 | struct map_lookup *map; | |
0b246afa | 5434 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5435 | |
60ca842e | 5436 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5437 | |
69f03f13 NB |
5438 | if (!WARN_ON(IS_ERR(em))) { |
5439 | map = em->map_lookup; | |
5440 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5441 | len = map->stripe_len * nr_data_stripes(map); | |
5442 | free_extent_map(em); | |
5443 | } | |
53b381b3 DW |
5444 | return len; |
5445 | } | |
5446 | ||
e4ff5fb5 | 5447 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5448 | { |
5449 | struct extent_map *em; | |
5450 | struct map_lookup *map; | |
53b381b3 DW |
5451 | int ret = 0; |
5452 | ||
60ca842e | 5453 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5454 | |
69f03f13 NB |
5455 | if(!WARN_ON(IS_ERR(em))) { |
5456 | map = em->map_lookup; | |
5457 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5458 | ret = 1; | |
5459 | free_extent_map(em); | |
5460 | } | |
53b381b3 DW |
5461 | return ret; |
5462 | } | |
5463 | ||
30d9861f | 5464 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5465 | struct map_lookup *map, int first, |
8ba0ae78 | 5466 | int dev_replace_is_ongoing) |
dfe25020 CM |
5467 | { |
5468 | int i; | |
99f92a7c | 5469 | int num_stripes; |
8ba0ae78 | 5470 | int preferred_mirror; |
30d9861f SB |
5471 | int tolerance; |
5472 | struct btrfs_device *srcdev; | |
5473 | ||
99f92a7c | 5474 | ASSERT((map->type & |
c7369b3f | 5475 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5476 | |
5477 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5478 | num_stripes = map->sub_stripes; | |
5479 | else | |
5480 | num_stripes = map->num_stripes; | |
5481 | ||
8ba0ae78 AJ |
5482 | preferred_mirror = first + current->pid % num_stripes; |
5483 | ||
30d9861f SB |
5484 | if (dev_replace_is_ongoing && |
5485 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5486 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5487 | srcdev = fs_info->dev_replace.srcdev; | |
5488 | else | |
5489 | srcdev = NULL; | |
5490 | ||
5491 | /* | |
5492 | * try to avoid the drive that is the source drive for a | |
5493 | * dev-replace procedure, only choose it if no other non-missing | |
5494 | * mirror is available | |
5495 | */ | |
5496 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5497 | if (map->stripes[preferred_mirror].dev->bdev && |
5498 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5499 | return preferred_mirror; | |
99f92a7c | 5500 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5501 | if (map->stripes[i].dev->bdev && |
5502 | (tolerance || map->stripes[i].dev != srcdev)) | |
5503 | return i; | |
5504 | } | |
dfe25020 | 5505 | } |
30d9861f | 5506 | |
dfe25020 CM |
5507 | /* we couldn't find one that doesn't fail. Just return something |
5508 | * and the io error handling code will clean up eventually | |
5509 | */ | |
8ba0ae78 | 5510 | return preferred_mirror; |
dfe25020 CM |
5511 | } |
5512 | ||
53b381b3 | 5513 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ |
8e5cfb55 | 5514 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 | 5515 | { |
53b381b3 | 5516 | int i; |
53b381b3 DW |
5517 | int again = 1; |
5518 | ||
5519 | while (again) { | |
5520 | again = 0; | |
cc7539ed | 5521 | for (i = 0; i < num_stripes - 1; i++) { |
eeb6f172 DS |
5522 | /* Swap if parity is on a smaller index */ |
5523 | if (bbio->raid_map[i] > bbio->raid_map[i + 1]) { | |
5524 | swap(bbio->stripes[i], bbio->stripes[i + 1]); | |
5525 | swap(bbio->raid_map[i], bbio->raid_map[i + 1]); | |
53b381b3 DW |
5526 | again = 1; |
5527 | } | |
5528 | } | |
5529 | } | |
5530 | } | |
5531 | ||
6e9606d2 ZL |
5532 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5533 | { | |
5534 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5535 | /* the size of the btrfs_bio */ |
6e9606d2 | 5536 | sizeof(struct btrfs_bio) + |
e57cf21e | 5537 | /* plus the variable array for the stripes */ |
6e9606d2 | 5538 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5539 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5540 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5541 | /* |
5542 | * plus the raid_map, which includes both the tgt dev | |
5543 | * and the stripes | |
5544 | */ | |
5545 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5546 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5547 | |
5548 | atomic_set(&bbio->error, 0); | |
140475ae | 5549 | refcount_set(&bbio->refs, 1); |
6e9606d2 | 5550 | |
608769a4 NB |
5551 | bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes); |
5552 | bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes); | |
5553 | ||
6e9606d2 ZL |
5554 | return bbio; |
5555 | } | |
5556 | ||
5557 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5558 | { | |
140475ae ER |
5559 | WARN_ON(!refcount_read(&bbio->refs)); |
5560 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5561 | } |
5562 | ||
5563 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5564 | { | |
5565 | if (!bbio) | |
5566 | return; | |
140475ae | 5567 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5568 | kfree(bbio); |
5569 | } | |
5570 | ||
0b3d4cd3 LB |
5571 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5572 | /* | |
5573 | * Please note that, discard won't be sent to target device of device | |
5574 | * replace. | |
5575 | */ | |
5576 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5577 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5578 | struct btrfs_bio **bbio_ret) |
5579 | { | |
5580 | struct extent_map *em; | |
5581 | struct map_lookup *map; | |
5582 | struct btrfs_bio *bbio; | |
6b7faadd | 5583 | u64 length = *length_ret; |
0b3d4cd3 LB |
5584 | u64 offset; |
5585 | u64 stripe_nr; | |
5586 | u64 stripe_nr_end; | |
5587 | u64 stripe_end_offset; | |
5588 | u64 stripe_cnt; | |
5589 | u64 stripe_len; | |
5590 | u64 stripe_offset; | |
5591 | u64 num_stripes; | |
5592 | u32 stripe_index; | |
5593 | u32 factor = 0; | |
5594 | u32 sub_stripes = 0; | |
5595 | u64 stripes_per_dev = 0; | |
5596 | u32 remaining_stripes = 0; | |
5597 | u32 last_stripe = 0; | |
5598 | int ret = 0; | |
5599 | int i; | |
5600 | ||
5601 | /* discard always return a bbio */ | |
5602 | ASSERT(bbio_ret); | |
5603 | ||
60ca842e | 5604 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5605 | if (IS_ERR(em)) |
5606 | return PTR_ERR(em); | |
5607 | ||
5608 | map = em->map_lookup; | |
5609 | /* we don't discard raid56 yet */ | |
5610 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5611 | ret = -EOPNOTSUPP; | |
5612 | goto out; | |
5613 | } | |
5614 | ||
5615 | offset = logical - em->start; | |
2d974619 | 5616 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5617 | *length_ret = length; |
0b3d4cd3 LB |
5618 | |
5619 | stripe_len = map->stripe_len; | |
5620 | /* | |
5621 | * stripe_nr counts the total number of stripes we have to stride | |
5622 | * to get to this block | |
5623 | */ | |
5624 | stripe_nr = div64_u64(offset, stripe_len); | |
5625 | ||
5626 | /* stripe_offset is the offset of this block in its stripe */ | |
5627 | stripe_offset = offset - stripe_nr * stripe_len; | |
5628 | ||
5629 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5630 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5631 | stripe_cnt = stripe_nr_end - stripe_nr; |
5632 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5633 | (offset + length); | |
5634 | /* | |
5635 | * after this, stripe_nr is the number of stripes on this | |
5636 | * device we have to walk to find the data, and stripe_index is | |
5637 | * the number of our device in the stripe array | |
5638 | */ | |
5639 | num_stripes = 1; | |
5640 | stripe_index = 0; | |
5641 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5642 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5643 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5644 | sub_stripes = 1; | |
5645 | else | |
5646 | sub_stripes = map->sub_stripes; | |
5647 | ||
5648 | factor = map->num_stripes / sub_stripes; | |
5649 | num_stripes = min_t(u64, map->num_stripes, | |
5650 | sub_stripes * stripe_cnt); | |
5651 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5652 | stripe_index *= sub_stripes; | |
5653 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5654 | &remaining_stripes); | |
5655 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5656 | last_stripe *= sub_stripes; | |
c7369b3f | 5657 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5658 | BTRFS_BLOCK_GROUP_DUP)) { |
5659 | num_stripes = map->num_stripes; | |
5660 | } else { | |
5661 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5662 | &stripe_index); | |
5663 | } | |
5664 | ||
5665 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5666 | if (!bbio) { | |
5667 | ret = -ENOMEM; | |
5668 | goto out; | |
5669 | } | |
5670 | ||
5671 | for (i = 0; i < num_stripes; i++) { | |
5672 | bbio->stripes[i].physical = | |
5673 | map->stripes[stripe_index].physical + | |
5674 | stripe_offset + stripe_nr * map->stripe_len; | |
5675 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5676 | ||
5677 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5678 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5679 | bbio->stripes[i].length = stripes_per_dev * | |
5680 | map->stripe_len; | |
5681 | ||
5682 | if (i / sub_stripes < remaining_stripes) | |
5683 | bbio->stripes[i].length += | |
5684 | map->stripe_len; | |
5685 | ||
5686 | /* | |
5687 | * Special for the first stripe and | |
5688 | * the last stripe: | |
5689 | * | |
5690 | * |-------|...|-------| | |
5691 | * |----------| | |
5692 | * off end_off | |
5693 | */ | |
5694 | if (i < sub_stripes) | |
5695 | bbio->stripes[i].length -= | |
5696 | stripe_offset; | |
5697 | ||
5698 | if (stripe_index >= last_stripe && | |
5699 | stripe_index <= (last_stripe + | |
5700 | sub_stripes - 1)) | |
5701 | bbio->stripes[i].length -= | |
5702 | stripe_end_offset; | |
5703 | ||
5704 | if (i == sub_stripes - 1) | |
5705 | stripe_offset = 0; | |
5706 | } else { | |
5707 | bbio->stripes[i].length = length; | |
5708 | } | |
5709 | ||
5710 | stripe_index++; | |
5711 | if (stripe_index == map->num_stripes) { | |
5712 | stripe_index = 0; | |
5713 | stripe_nr++; | |
5714 | } | |
5715 | } | |
5716 | ||
5717 | *bbio_ret = bbio; | |
5718 | bbio->map_type = map->type; | |
5719 | bbio->num_stripes = num_stripes; | |
5720 | out: | |
5721 | free_extent_map(em); | |
5722 | return ret; | |
5723 | } | |
5724 | ||
5ab56090 LB |
5725 | /* |
5726 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5727 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5728 | * left cursor can also be read from the target drive. | |
5729 | * | |
5730 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5731 | * array of stripes. | |
5732 | * For READ, it also needs to be supported using the same mirror number. | |
5733 | * | |
5734 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5735 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5736 | * case. | |
5737 | */ | |
5738 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5739 | u64 logical, u64 length, | |
5740 | u64 srcdev_devid, int *mirror_num, | |
5741 | u64 *physical) | |
5742 | { | |
5743 | struct btrfs_bio *bbio = NULL; | |
5744 | int num_stripes; | |
5745 | int index_srcdev = 0; | |
5746 | int found = 0; | |
5747 | u64 physical_of_found = 0; | |
5748 | int i; | |
5749 | int ret = 0; | |
5750 | ||
5751 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5752 | logical, &length, &bbio, 0, 0); | |
5753 | if (ret) { | |
5754 | ASSERT(bbio == NULL); | |
5755 | return ret; | |
5756 | } | |
5757 | ||
5758 | num_stripes = bbio->num_stripes; | |
5759 | if (*mirror_num > num_stripes) { | |
5760 | /* | |
5761 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5762 | * that means that the requested area is not left of the left | |
5763 | * cursor | |
5764 | */ | |
5765 | btrfs_put_bbio(bbio); | |
5766 | return -EIO; | |
5767 | } | |
5768 | ||
5769 | /* | |
5770 | * process the rest of the function using the mirror_num of the source | |
5771 | * drive. Therefore look it up first. At the end, patch the device | |
5772 | * pointer to the one of the target drive. | |
5773 | */ | |
5774 | for (i = 0; i < num_stripes; i++) { | |
5775 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5776 | continue; | |
5777 | ||
5778 | /* | |
5779 | * In case of DUP, in order to keep it simple, only add the | |
5780 | * mirror with the lowest physical address | |
5781 | */ | |
5782 | if (found && | |
5783 | physical_of_found <= bbio->stripes[i].physical) | |
5784 | continue; | |
5785 | ||
5786 | index_srcdev = i; | |
5787 | found = 1; | |
5788 | physical_of_found = bbio->stripes[i].physical; | |
5789 | } | |
5790 | ||
5791 | btrfs_put_bbio(bbio); | |
5792 | ||
5793 | ASSERT(found); | |
5794 | if (!found) | |
5795 | return -EIO; | |
5796 | ||
5797 | *mirror_num = index_srcdev + 1; | |
5798 | *physical = physical_of_found; | |
5799 | return ret; | |
5800 | } | |
5801 | ||
73c0f228 LB |
5802 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5803 | struct btrfs_bio **bbio_ret, | |
5804 | struct btrfs_dev_replace *dev_replace, | |
5805 | int *num_stripes_ret, int *max_errors_ret) | |
5806 | { | |
5807 | struct btrfs_bio *bbio = *bbio_ret; | |
5808 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5809 | int tgtdev_indexes = 0; | |
5810 | int num_stripes = *num_stripes_ret; | |
5811 | int max_errors = *max_errors_ret; | |
5812 | int i; | |
5813 | ||
5814 | if (op == BTRFS_MAP_WRITE) { | |
5815 | int index_where_to_add; | |
5816 | ||
5817 | /* | |
5818 | * duplicate the write operations while the dev replace | |
5819 | * procedure is running. Since the copying of the old disk to | |
5820 | * the new disk takes place at run time while the filesystem is | |
5821 | * mounted writable, the regular write operations to the old | |
5822 | * disk have to be duplicated to go to the new disk as well. | |
5823 | * | |
5824 | * Note that device->missing is handled by the caller, and that | |
5825 | * the write to the old disk is already set up in the stripes | |
5826 | * array. | |
5827 | */ | |
5828 | index_where_to_add = num_stripes; | |
5829 | for (i = 0; i < num_stripes; i++) { | |
5830 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5831 | /* write to new disk, too */ | |
5832 | struct btrfs_bio_stripe *new = | |
5833 | bbio->stripes + index_where_to_add; | |
5834 | struct btrfs_bio_stripe *old = | |
5835 | bbio->stripes + i; | |
5836 | ||
5837 | new->physical = old->physical; | |
5838 | new->length = old->length; | |
5839 | new->dev = dev_replace->tgtdev; | |
5840 | bbio->tgtdev_map[i] = index_where_to_add; | |
5841 | index_where_to_add++; | |
5842 | max_errors++; | |
5843 | tgtdev_indexes++; | |
5844 | } | |
5845 | } | |
5846 | num_stripes = index_where_to_add; | |
5847 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5848 | int index_srcdev = 0; | |
5849 | int found = 0; | |
5850 | u64 physical_of_found = 0; | |
5851 | ||
5852 | /* | |
5853 | * During the dev-replace procedure, the target drive can also | |
5854 | * be used to read data in case it is needed to repair a corrupt | |
5855 | * block elsewhere. This is possible if the requested area is | |
5856 | * left of the left cursor. In this area, the target drive is a | |
5857 | * full copy of the source drive. | |
5858 | */ | |
5859 | for (i = 0; i < num_stripes; i++) { | |
5860 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5861 | /* | |
5862 | * In case of DUP, in order to keep it simple, | |
5863 | * only add the mirror with the lowest physical | |
5864 | * address | |
5865 | */ | |
5866 | if (found && | |
5867 | physical_of_found <= | |
5868 | bbio->stripes[i].physical) | |
5869 | continue; | |
5870 | index_srcdev = i; | |
5871 | found = 1; | |
5872 | physical_of_found = bbio->stripes[i].physical; | |
5873 | } | |
5874 | } | |
5875 | if (found) { | |
5876 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5877 | bbio->stripes + num_stripes; | |
5878 | ||
5879 | tgtdev_stripe->physical = physical_of_found; | |
5880 | tgtdev_stripe->length = | |
5881 | bbio->stripes[index_srcdev].length; | |
5882 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5883 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5884 | ||
5885 | tgtdev_indexes++; | |
5886 | num_stripes++; | |
5887 | } | |
5888 | } | |
5889 | ||
5890 | *num_stripes_ret = num_stripes; | |
5891 | *max_errors_ret = max_errors; | |
5892 | bbio->num_tgtdevs = tgtdev_indexes; | |
5893 | *bbio_ret = bbio; | |
5894 | } | |
5895 | ||
2b19a1fe LB |
5896 | static bool need_full_stripe(enum btrfs_map_op op) |
5897 | { | |
5898 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5899 | } | |
5900 | ||
5f141126 NB |
5901 | /* |
5902 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5903 | * tuple. This information is used to calculate how big a | |
5904 | * particular bio can get before it straddles a stripe. | |
5905 | * | |
5906 | * @fs_info - the filesystem | |
5907 | * @logical - address that we want to figure out the geometry of | |
5908 | * @len - the length of IO we are going to perform, starting at @logical | |
5909 | * @op - type of operation - write or read | |
5910 | * @io_geom - pointer used to return values | |
5911 | * | |
5912 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5913 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5914 | */ | |
5915 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5916 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5917 | { |
5918 | struct extent_map *em; | |
5919 | struct map_lookup *map; | |
5920 | u64 offset; | |
5921 | u64 stripe_offset; | |
5922 | u64 stripe_nr; | |
5923 | u64 stripe_len; | |
5924 | u64 raid56_full_stripe_start = (u64)-1; | |
5925 | int data_stripes; | |
373c3b80 | 5926 | int ret = 0; |
5f141126 NB |
5927 | |
5928 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5929 | ||
5930 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5931 | if (IS_ERR(em)) | |
5932 | return PTR_ERR(em); | |
5933 | ||
5934 | map = em->map_lookup; | |
5935 | /* Offset of this logical address in the chunk */ | |
5936 | offset = logical - em->start; | |
5937 | /* Len of a stripe in a chunk */ | |
5938 | stripe_len = map->stripe_len; | |
5939 | /* Stripe wher this block falls in */ | |
5940 | stripe_nr = div64_u64(offset, stripe_len); | |
5941 | /* Offset of stripe in the chunk */ | |
5942 | stripe_offset = stripe_nr * stripe_len; | |
5943 | if (offset < stripe_offset) { | |
5944 | btrfs_crit(fs_info, | |
5945 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5946 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5947 | ret = -EINVAL; |
5948 | goto out; | |
5f141126 NB |
5949 | } |
5950 | ||
5951 | /* stripe_offset is the offset of this block in its stripe */ | |
5952 | stripe_offset = offset - stripe_offset; | |
5953 | data_stripes = nr_data_stripes(map); | |
5954 | ||
5955 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5956 | u64 max_len = stripe_len - stripe_offset; | |
5957 | ||
5958 | /* | |
5959 | * In case of raid56, we need to know the stripe aligned start | |
5960 | */ | |
5961 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5962 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5963 | raid56_full_stripe_start = offset; | |
5964 | ||
5965 | /* | |
5966 | * Allow a write of a full stripe, but make sure we | |
5967 | * don't allow straddling of stripes | |
5968 | */ | |
5969 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5970 | full_stripe_len); | |
5971 | raid56_full_stripe_start *= full_stripe_len; | |
5972 | ||
5973 | /* | |
5974 | * For writes to RAID[56], allow a full stripeset across | |
5975 | * all disks. For other RAID types and for RAID[56] | |
5976 | * reads, just allow a single stripe (on a single disk). | |
5977 | */ | |
5978 | if (op == BTRFS_MAP_WRITE) { | |
5979 | max_len = stripe_len * data_stripes - | |
5980 | (offset - raid56_full_stripe_start); | |
5981 | } | |
5982 | } | |
5983 | len = min_t(u64, em->len - offset, max_len); | |
5984 | } else { | |
5985 | len = em->len - offset; | |
5986 | } | |
5987 | ||
5988 | io_geom->len = len; | |
5989 | io_geom->offset = offset; | |
5990 | io_geom->stripe_len = stripe_len; | |
5991 | io_geom->stripe_nr = stripe_nr; | |
5992 | io_geom->stripe_offset = stripe_offset; | |
5993 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
5994 | ||
373c3b80 JT |
5995 | out: |
5996 | /* once for us */ | |
5997 | free_extent_map(em); | |
5998 | return ret; | |
5f141126 NB |
5999 | } |
6000 | ||
cf8cddd3 CH |
6001 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
6002 | enum btrfs_map_op op, | |
f2d8d74d | 6003 | u64 logical, u64 *length, |
a1d3c478 | 6004 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6005 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6006 | { |
6007 | struct extent_map *em; | |
6008 | struct map_lookup *map; | |
593060d7 CM |
6009 | u64 stripe_offset; |
6010 | u64 stripe_nr; | |
53b381b3 | 6011 | u64 stripe_len; |
9d644a62 | 6012 | u32 stripe_index; |
cff82672 | 6013 | int data_stripes; |
cea9e445 | 6014 | int i; |
de11cc12 | 6015 | int ret = 0; |
f2d8d74d | 6016 | int num_stripes; |
a236aed1 | 6017 | int max_errors = 0; |
2c8cdd6e | 6018 | int tgtdev_indexes = 0; |
a1d3c478 | 6019 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6020 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6021 | int dev_replace_is_ongoing = 0; | |
6022 | int num_alloc_stripes; | |
ad6d620e SB |
6023 | int patch_the_first_stripe_for_dev_replace = 0; |
6024 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6025 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6026 | struct btrfs_io_geometry geom; |
6027 | ||
6028 | ASSERT(bbio_ret); | |
75fb2e9e | 6029 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6030 | |
89b798ad NB |
6031 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
6032 | if (ret < 0) | |
6033 | return ret; | |
0b86a832 | 6034 | |
89b798ad | 6035 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 6036 | ASSERT(!IS_ERR(em)); |
95617d69 | 6037 | map = em->map_lookup; |
593060d7 | 6038 | |
89b798ad | 6039 | *length = geom.len; |
89b798ad NB |
6040 | stripe_len = geom.stripe_len; |
6041 | stripe_nr = geom.stripe_nr; | |
6042 | stripe_offset = geom.stripe_offset; | |
6043 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6044 | data_stripes = nr_data_stripes(map); |
593060d7 | 6045 | |
cb5583dd | 6046 | down_read(&dev_replace->rwsem); |
472262f3 | 6047 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6048 | /* |
6049 | * Hold the semaphore for read during the whole operation, write is | |
6050 | * requested at commit time but must wait. | |
6051 | */ | |
472262f3 | 6052 | if (!dev_replace_is_ongoing) |
cb5583dd | 6053 | up_read(&dev_replace->rwsem); |
472262f3 | 6054 | |
ad6d620e | 6055 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6056 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6057 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6058 | dev_replace->srcdev->devid, | |
6059 | &mirror_num, | |
6060 | &physical_to_patch_in_first_stripe); | |
6061 | if (ret) | |
ad6d620e | 6062 | goto out; |
5ab56090 LB |
6063 | else |
6064 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6065 | } else if (mirror_num > map->num_stripes) { |
6066 | mirror_num = 0; | |
6067 | } | |
6068 | ||
f2d8d74d | 6069 | num_stripes = 1; |
cea9e445 | 6070 | stripe_index = 0; |
fce3bb9a | 6071 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6072 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6073 | &stripe_index); | |
de483734 | 6074 | if (!need_full_stripe(op)) |
28e1cc7d | 6075 | mirror_num = 1; |
c7369b3f | 6076 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6077 | if (need_full_stripe(op)) |
f2d8d74d | 6078 | num_stripes = map->num_stripes; |
2fff734f | 6079 | else if (mirror_num) |
f188591e | 6080 | stripe_index = mirror_num - 1; |
dfe25020 | 6081 | else { |
30d9861f | 6082 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6083 | dev_replace_is_ongoing); |
a1d3c478 | 6084 | mirror_num = stripe_index + 1; |
dfe25020 | 6085 | } |
2fff734f | 6086 | |
611f0e00 | 6087 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6088 | if (need_full_stripe(op)) { |
f2d8d74d | 6089 | num_stripes = map->num_stripes; |
a1d3c478 | 6090 | } else if (mirror_num) { |
f188591e | 6091 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6092 | } else { |
6093 | mirror_num = 1; | |
6094 | } | |
2fff734f | 6095 | |
321aecc6 | 6096 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6097 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6098 | |
47c5713f | 6099 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6100 | stripe_index *= map->sub_stripes; |
6101 | ||
de483734 | 6102 | if (need_full_stripe(op)) |
f2d8d74d | 6103 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6104 | else if (mirror_num) |
6105 | stripe_index += mirror_num - 1; | |
dfe25020 | 6106 | else { |
3e74317a | 6107 | int old_stripe_index = stripe_index; |
30d9861f SB |
6108 | stripe_index = find_live_mirror(fs_info, map, |
6109 | stripe_index, | |
30d9861f | 6110 | dev_replace_is_ongoing); |
3e74317a | 6111 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6112 | } |
53b381b3 | 6113 | |
ffe2d203 | 6114 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6115 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6116 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6117 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6118 | stripe_len * data_stripes); |
53b381b3 DW |
6119 | |
6120 | /* RAID[56] write or recovery. Return all stripes */ | |
6121 | num_stripes = map->num_stripes; | |
6122 | max_errors = nr_parity_stripes(map); | |
6123 | ||
53b381b3 DW |
6124 | *length = map->stripe_len; |
6125 | stripe_index = 0; | |
6126 | stripe_offset = 0; | |
6127 | } else { | |
6128 | /* | |
6129 | * Mirror #0 or #1 means the original data block. | |
6130 | * Mirror #2 is RAID5 parity block. | |
6131 | * Mirror #3 is RAID6 Q block. | |
6132 | */ | |
47c5713f | 6133 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6134 | data_stripes, &stripe_index); |
53b381b3 | 6135 | if (mirror_num > 1) |
cff82672 | 6136 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6137 | |
6138 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6139 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6140 | &stripe_index); | |
de483734 | 6141 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6142 | mirror_num = 1; |
53b381b3 | 6143 | } |
8790d502 CM |
6144 | } else { |
6145 | /* | |
47c5713f DS |
6146 | * after this, stripe_nr is the number of stripes on this |
6147 | * device we have to walk to find the data, and stripe_index is | |
6148 | * the number of our device in the stripe array | |
8790d502 | 6149 | */ |
47c5713f DS |
6150 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6151 | &stripe_index); | |
a1d3c478 | 6152 | mirror_num = stripe_index + 1; |
8790d502 | 6153 | } |
e042d1ec | 6154 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6155 | btrfs_crit(fs_info, |
6156 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6157 | stripe_index, map->num_stripes); |
6158 | ret = -EINVAL; | |
6159 | goto out; | |
6160 | } | |
cea9e445 | 6161 | |
472262f3 | 6162 | num_alloc_stripes = num_stripes; |
6fad823f | 6163 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6164 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6165 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6166 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6167 | num_alloc_stripes++; |
2c8cdd6e | 6168 | tgtdev_indexes = num_stripes; |
ad6d620e | 6169 | } |
2c8cdd6e | 6170 | |
6e9606d2 | 6171 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6172 | if (!bbio) { |
6173 | ret = -ENOMEM; | |
6174 | goto out; | |
6175 | } | |
608769a4 NB |
6176 | |
6177 | for (i = 0; i < num_stripes; i++) { | |
6178 | bbio->stripes[i].physical = map->stripes[stripe_index].physical + | |
6179 | stripe_offset + stripe_nr * map->stripe_len; | |
6180 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
6181 | stripe_index++; | |
6182 | } | |
de11cc12 | 6183 | |
8e5cfb55 | 6184 | /* build raid_map */ |
2b19a1fe LB |
6185 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6186 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6187 | u64 tmp; |
9d644a62 | 6188 | unsigned rot; |
8e5cfb55 | 6189 | |
8e5cfb55 | 6190 | /* Work out the disk rotation on this stripe-set */ |
47c5713f | 6191 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6192 | |
6193 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6194 | tmp = stripe_nr * data_stripes; |
6195 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6196 | bbio->raid_map[(i+rot) % num_stripes] = |
6197 | em->start + (tmp + i) * map->stripe_len; | |
6198 | ||
6199 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6200 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6201 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6202 | RAID6_Q_STRIPE; | |
8e5cfb55 | 6203 | |
608769a4 | 6204 | sort_parity_stripes(bbio, num_stripes); |
593060d7 | 6205 | } |
de11cc12 | 6206 | |
2b19a1fe | 6207 | if (need_full_stripe(op)) |
d20983b4 | 6208 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6209 | |
73c0f228 | 6210 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6211 | need_full_stripe(op)) { |
73c0f228 LB |
6212 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6213 | &max_errors); | |
472262f3 SB |
6214 | } |
6215 | ||
de11cc12 | 6216 | *bbio_ret = bbio; |
10f11900 | 6217 | bbio->map_type = map->type; |
de11cc12 LZ |
6218 | bbio->num_stripes = num_stripes; |
6219 | bbio->max_errors = max_errors; | |
6220 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6221 | |
6222 | /* | |
6223 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6224 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6225 | * available as a mirror | |
6226 | */ | |
6227 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6228 | WARN_ON(num_stripes > 1); | |
6229 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6230 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6231 | bbio->mirror_num = map->num_stripes + 1; | |
6232 | } | |
cea9e445 | 6233 | out: |
73beece9 | 6234 | if (dev_replace_is_ongoing) { |
53176dde DS |
6235 | lockdep_assert_held(&dev_replace->rwsem); |
6236 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6237 | up_read(&dev_replace->rwsem); |
73beece9 | 6238 | } |
0b86a832 | 6239 | free_extent_map(em); |
de11cc12 | 6240 | return ret; |
0b86a832 CM |
6241 | } |
6242 | ||
cf8cddd3 | 6243 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6244 | u64 logical, u64 *length, |
a1d3c478 | 6245 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6246 | { |
75fb2e9e DS |
6247 | if (op == BTRFS_MAP_DISCARD) |
6248 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6249 | length, bbio_ret); | |
6250 | ||
b3d3fa51 | 6251 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6252 | mirror_num, 0); |
f2d8d74d CM |
6253 | } |
6254 | ||
af8e2d1d | 6255 | /* For Scrub/replace */ |
cf8cddd3 | 6256 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6257 | u64 logical, u64 *length, |
825ad4c9 | 6258 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6259 | { |
825ad4c9 | 6260 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6261 | } |
6262 | ||
4246a0b6 | 6263 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6264 | { |
326e1dbb MS |
6265 | bio->bi_private = bbio->private; |
6266 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6267 | bio_endio(bio); |
326e1dbb | 6268 | |
6e9606d2 | 6269 | btrfs_put_bbio(bbio); |
8408c716 MX |
6270 | } |
6271 | ||
4246a0b6 | 6272 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6273 | { |
9be3395b | 6274 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6275 | int is_orig_bio = 0; |
8790d502 | 6276 | |
4e4cbee9 | 6277 | if (bio->bi_status) { |
a1d3c478 | 6278 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6279 | if (bio->bi_status == BLK_STS_IOERR || |
6280 | bio->bi_status == BLK_STS_TARGET) { | |
c31efbdf | 6281 | struct btrfs_device *dev = btrfs_io_bio(bio)->device; |
442a4f63 | 6282 | |
3eee86c8 NB |
6283 | ASSERT(dev->bdev); |
6284 | if (bio_op(bio) == REQ_OP_WRITE) | |
6285 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6286 | BTRFS_DEV_STAT_WRITE_ERRS); |
3eee86c8 NB |
6287 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
6288 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6289 | BTRFS_DEV_STAT_READ_ERRS); |
3eee86c8 NB |
6290 | if (bio->bi_opf & REQ_PREFLUSH) |
6291 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6292 | BTRFS_DEV_STAT_FLUSH_ERRS); |
442a4f63 SB |
6293 | } |
6294 | } | |
8790d502 | 6295 | |
a1d3c478 | 6296 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6297 | is_orig_bio = 1; |
6298 | ||
c404e0dc MX |
6299 | btrfs_bio_counter_dec(bbio->fs_info); |
6300 | ||
a1d3c478 | 6301 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6302 | if (!is_orig_bio) { |
6303 | bio_put(bio); | |
a1d3c478 | 6304 | bio = bbio->orig_bio; |
7d2b4daa | 6305 | } |
c7b22bb1 | 6306 | |
9be3395b | 6307 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6308 | /* only send an error to the higher layers if it is |
53b381b3 | 6309 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6310 | */ |
a1d3c478 | 6311 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6312 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6313 | } else { |
1259ab75 CM |
6314 | /* |
6315 | * this bio is actually up to date, we didn't | |
6316 | * go over the max number of errors | |
6317 | */ | |
2dbe0c77 | 6318 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6319 | } |
c55f1396 | 6320 | |
4246a0b6 | 6321 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6322 | } else if (!is_orig_bio) { |
8790d502 CM |
6323 | bio_put(bio); |
6324 | } | |
8790d502 CM |
6325 | } |
6326 | ||
2ff7e61e | 6327 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
c31efbdf | 6328 | u64 physical, struct btrfs_device *dev) |
de1ee92a | 6329 | { |
2ff7e61e | 6330 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6331 | |
6332 | bio->bi_private = bbio; | |
c31efbdf | 6333 | btrfs_io_bio(bio)->device = dev; |
de1ee92a | 6334 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6335 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6336 | btrfs_debug_in_rcu(fs_info, |
6337 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6338 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
1db45a35 DS |
6339 | (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), |
6340 | dev->devid, bio->bi_iter.bi_size); | |
74d46992 | 6341 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6342 | |
2ff7e61e | 6343 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6344 | |
08635bae | 6345 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6346 | } |
6347 | ||
de1ee92a JB |
6348 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6349 | { | |
6350 | atomic_inc(&bbio->error); | |
6351 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6352 | /* Should be the original bio. */ |
8408c716 MX |
6353 | WARN_ON(bio != bbio->orig_bio); |
6354 | ||
9be3395b | 6355 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6356 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6357 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6358 | bio->bi_status = BLK_STS_IOERR; | |
6359 | else | |
6360 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6361 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6362 | } |
6363 | } | |
6364 | ||
58efbc9f | 6365 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6366 | int mirror_num) |
0b86a832 | 6367 | { |
0b86a832 | 6368 | struct btrfs_device *dev; |
8790d502 | 6369 | struct bio *first_bio = bio; |
4f024f37 | 6370 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6371 | u64 length = 0; |
6372 | u64 map_length; | |
0b86a832 | 6373 | int ret; |
08da757d ZL |
6374 | int dev_nr; |
6375 | int total_devs; | |
a1d3c478 | 6376 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6377 | |
4f024f37 | 6378 | length = bio->bi_iter.bi_size; |
0b86a832 | 6379 | map_length = length; |
cea9e445 | 6380 | |
0b246afa | 6381 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6382 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6383 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6384 | if (ret) { |
0b246afa | 6385 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6386 | return errno_to_blk_status(ret); |
c404e0dc | 6387 | } |
cea9e445 | 6388 | |
a1d3c478 | 6389 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6390 | bbio->orig_bio = first_bio; |
6391 | bbio->private = first_bio->bi_private; | |
6392 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6393 | bbio->fs_info = fs_info; |
53b381b3 DW |
6394 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6395 | ||
ad1ba2a0 | 6396 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6397 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6398 | /* In this case, map_length has been set to the length of |
6399 | a single stripe; not the whole write */ | |
37226b21 | 6400 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6401 | ret = raid56_parity_write(fs_info, bio, bbio, |
6402 | map_length); | |
53b381b3 | 6403 | } else { |
2ff7e61e JM |
6404 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6405 | map_length, mirror_num, 1); | |
53b381b3 | 6406 | } |
4245215d | 6407 | |
0b246afa | 6408 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6409 | return errno_to_blk_status(ret); |
53b381b3 DW |
6410 | } |
6411 | ||
cea9e445 | 6412 | if (map_length < length) { |
0b246afa | 6413 | btrfs_crit(fs_info, |
5d163e0e JM |
6414 | "mapping failed logical %llu bio len %llu len %llu", |
6415 | logical, length, map_length); | |
cea9e445 CM |
6416 | BUG(); |
6417 | } | |
a1d3c478 | 6418 | |
08da757d | 6419 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6420 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6421 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6422 | &dev->dev_state) || | |
ebbede42 AJ |
6423 | (bio_op(first_bio) == REQ_OP_WRITE && |
6424 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6425 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6426 | continue; |
6427 | } | |
6428 | ||
3aa8e074 | 6429 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6430 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6431 | else |
a1d3c478 | 6432 | bio = first_bio; |
de1ee92a | 6433 | |
c31efbdf | 6434 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev); |
8790d502 | 6435 | } |
0b246afa | 6436 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6437 | return BLK_STS_OK; |
0b86a832 CM |
6438 | } |
6439 | ||
09ba3bc9 AJ |
6440 | /* |
6441 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6442 | * return NULL. | |
6443 | * | |
6444 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6445 | * only devid is used. | |
6446 | * | |
6447 | * If @seed is true, traverse through the seed devices. | |
6448 | */ | |
e4319cd9 | 6449 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6450 | u64 devid, u8 *uuid, u8 *fsid, |
6451 | bool seed) | |
0b86a832 | 6452 | { |
2b82032c | 6453 | struct btrfs_device *device; |
944d3f9f NB |
6454 | struct btrfs_fs_devices *seed_devs; |
6455 | ||
6456 | if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { | |
6457 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6458 | if (device->devid == devid && | |
6459 | (!uuid || memcmp(device->uuid, uuid, | |
6460 | BTRFS_UUID_SIZE) == 0)) | |
6461 | return device; | |
6462 | } | |
6463 | } | |
2b82032c | 6464 | |
944d3f9f | 6465 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
2b82032c | 6466 | if (!fsid || |
944d3f9f NB |
6467 | !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
6468 | list_for_each_entry(device, &seed_devs->devices, | |
09ba3bc9 AJ |
6469 | dev_list) { |
6470 | if (device->devid == devid && | |
6471 | (!uuid || memcmp(device->uuid, uuid, | |
6472 | BTRFS_UUID_SIZE) == 0)) | |
6473 | return device; | |
6474 | } | |
2b82032c | 6475 | } |
2b82032c | 6476 | } |
944d3f9f | 6477 | |
2b82032c | 6478 | return NULL; |
0b86a832 CM |
6479 | } |
6480 | ||
2ff7e61e | 6481 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6482 | u64 devid, u8 *dev_uuid) |
6483 | { | |
6484 | struct btrfs_device *device; | |
fccc0007 | 6485 | unsigned int nofs_flag; |
dfe25020 | 6486 | |
fccc0007 JB |
6487 | /* |
6488 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6489 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6490 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6491 | * places. | |
6492 | */ | |
6493 | nofs_flag = memalloc_nofs_save(); | |
12bd2fc0 | 6494 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
fccc0007 | 6495 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6496 | if (IS_ERR(device)) |
adfb69af | 6497 | return device; |
12bd2fc0 ID |
6498 | |
6499 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6500 | device->fs_devices = fs_devices; |
dfe25020 | 6501 | fs_devices->num_devices++; |
12bd2fc0 | 6502 | |
e6e674bd | 6503 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6504 | fs_devices->missing_devices++; |
12bd2fc0 | 6505 | |
dfe25020 CM |
6506 | return device; |
6507 | } | |
6508 | ||
12bd2fc0 ID |
6509 | /** |
6510 | * btrfs_alloc_device - allocate struct btrfs_device | |
6511 | * @fs_info: used only for generating a new devid, can be NULL if | |
6512 | * devid is provided (i.e. @devid != NULL). | |
6513 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6514 | * is generated. | |
6515 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6516 | * is generated. | |
6517 | * | |
6518 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6519 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6520 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6521 | */ |
6522 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6523 | const u64 *devid, | |
6524 | const u8 *uuid) | |
6525 | { | |
6526 | struct btrfs_device *dev; | |
6527 | u64 tmp; | |
6528 | ||
fae7f21c | 6529 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6530 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6531 | |
154f7cb8 | 6532 | dev = __alloc_device(fs_info); |
12bd2fc0 ID |
6533 | if (IS_ERR(dev)) |
6534 | return dev; | |
6535 | ||
6536 | if (devid) | |
6537 | tmp = *devid; | |
6538 | else { | |
6539 | int ret; | |
6540 | ||
6541 | ret = find_next_devid(fs_info, &tmp); | |
6542 | if (ret) { | |
a425f9d4 | 6543 | btrfs_free_device(dev); |
12bd2fc0 ID |
6544 | return ERR_PTR(ret); |
6545 | } | |
6546 | } | |
6547 | dev->devid = tmp; | |
6548 | ||
6549 | if (uuid) | |
6550 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6551 | else | |
6552 | generate_random_uuid(dev->uuid); | |
6553 | ||
12bd2fc0 ID |
6554 | return dev; |
6555 | } | |
6556 | ||
5a2b8e60 | 6557 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6558 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6559 | { |
2b902dfc AJ |
6560 | if (error) |
6561 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6562 | devid, uuid); | |
6563 | else | |
6564 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6565 | devid, uuid); | |
5a2b8e60 AJ |
6566 | } |
6567 | ||
39e264a4 NB |
6568 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6569 | { | |
6570 | int index = btrfs_bg_flags_to_raid_index(type); | |
6571 | int ncopies = btrfs_raid_array[index].ncopies; | |
e4f6c6be | 6572 | const int nparity = btrfs_raid_array[index].nparity; |
39e264a4 NB |
6573 | int data_stripes; |
6574 | ||
e4f6c6be DS |
6575 | if (nparity) |
6576 | data_stripes = num_stripes - nparity; | |
6577 | else | |
39e264a4 | 6578 | data_stripes = num_stripes / ncopies; |
e4f6c6be | 6579 | |
39e264a4 NB |
6580 | return div_u64(chunk_len, data_stripes); |
6581 | } | |
6582 | ||
9690ac09 | 6583 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6584 | struct btrfs_chunk *chunk) |
6585 | { | |
9690ac09 | 6586 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6587 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6588 | struct map_lookup *map; |
6589 | struct extent_map *em; | |
6590 | u64 logical; | |
6591 | u64 length; | |
e06cd3dd LB |
6592 | u64 devid; |
6593 | u8 uuid[BTRFS_UUID_SIZE]; | |
6594 | int num_stripes; | |
6595 | int ret; | |
6596 | int i; | |
6597 | ||
6598 | logical = key->offset; | |
6599 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6600 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6601 | ||
075cb3c7 QW |
6602 | /* |
6603 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6604 | * as chunk item in tree block is already verified by tree-checker. | |
6605 | */ | |
6606 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6607 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6608 | if (ret) |
6609 | return ret; | |
6610 | } | |
a061fc8d | 6611 | |
c8bf1b67 DS |
6612 | read_lock(&map_tree->lock); |
6613 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6614 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6615 | |
6616 | /* already mapped? */ | |
6617 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6618 | free_extent_map(em); | |
0b86a832 CM |
6619 | return 0; |
6620 | } else if (em) { | |
6621 | free_extent_map(em); | |
6622 | } | |
0b86a832 | 6623 | |
172ddd60 | 6624 | em = alloc_extent_map(); |
0b86a832 CM |
6625 | if (!em) |
6626 | return -ENOMEM; | |
593060d7 | 6627 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6628 | if (!map) { |
6629 | free_extent_map(em); | |
6630 | return -ENOMEM; | |
6631 | } | |
6632 | ||
298a8f9c | 6633 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6634 | em->map_lookup = map; |
0b86a832 CM |
6635 | em->start = logical; |
6636 | em->len = length; | |
70c8a91c | 6637 | em->orig_start = 0; |
0b86a832 | 6638 | em->block_start = 0; |
c8b97818 | 6639 | em->block_len = em->len; |
0b86a832 | 6640 | |
593060d7 CM |
6641 | map->num_stripes = num_stripes; |
6642 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6643 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6644 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6645 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6646 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6647 | map->verified_stripes = 0; |
39e264a4 NB |
6648 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6649 | map->num_stripes); | |
593060d7 CM |
6650 | for (i = 0; i < num_stripes; i++) { |
6651 | map->stripes[i].physical = | |
6652 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6653 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6654 | read_extent_buffer(leaf, uuid, (unsigned long) |
6655 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6656 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6657 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6658 | devid, uuid, NULL, true); |
3cdde224 | 6659 | if (!map->stripes[i].dev && |
0b246afa | 6660 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6661 | free_extent_map(em); |
2b902dfc | 6662 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6663 | return -ENOENT; |
593060d7 | 6664 | } |
dfe25020 CM |
6665 | if (!map->stripes[i].dev) { |
6666 | map->stripes[i].dev = | |
2ff7e61e JM |
6667 | add_missing_dev(fs_info->fs_devices, devid, |
6668 | uuid); | |
adfb69af | 6669 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6670 | free_extent_map(em); |
adfb69af AJ |
6671 | btrfs_err(fs_info, |
6672 | "failed to init missing dev %llu: %ld", | |
6673 | devid, PTR_ERR(map->stripes[i].dev)); | |
6674 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6675 | } |
2b902dfc | 6676 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6677 | } |
e12c9621 AJ |
6678 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6679 | &(map->stripes[i].dev->dev_state)); | |
6680 | ||
0b86a832 CM |
6681 | } |
6682 | ||
c8bf1b67 DS |
6683 | write_lock(&map_tree->lock); |
6684 | ret = add_extent_mapping(map_tree, em, 0); | |
6685 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6686 | if (ret < 0) { |
6687 | btrfs_err(fs_info, | |
6688 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6689 | em->start, em->len, ret); | |
6690 | } | |
0b86a832 CM |
6691 | free_extent_map(em); |
6692 | ||
64f64f43 | 6693 | return ret; |
0b86a832 CM |
6694 | } |
6695 | ||
143bede5 | 6696 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6697 | struct btrfs_dev_item *dev_item, |
6698 | struct btrfs_device *device) | |
6699 | { | |
6700 | unsigned long ptr; | |
0b86a832 CM |
6701 | |
6702 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6703 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6704 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6705 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6706 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6707 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6708 | device->type = btrfs_device_type(leaf, dev_item); |
6709 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6710 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6711 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6712 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6713 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6714 | |
410ba3a2 | 6715 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6716 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6717 | } |
6718 | ||
2ff7e61e | 6719 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6720 | u8 *fsid) |
2b82032c YZ |
6721 | { |
6722 | struct btrfs_fs_devices *fs_devices; | |
6723 | int ret; | |
6724 | ||
a32bf9a3 | 6725 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6726 | ASSERT(fsid); |
2b82032c | 6727 | |
427c8fdd | 6728 | /* This will match only for multi-device seed fs */ |
944d3f9f | 6729 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 6730 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6731 | return fs_devices; |
6732 | ||
2b82032c | 6733 | |
7239ff4b | 6734 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6735 | if (!fs_devices) { |
0b246afa | 6736 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6737 | return ERR_PTR(-ENOENT); |
6738 | ||
7239ff4b | 6739 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6740 | if (IS_ERR(fs_devices)) |
6741 | return fs_devices; | |
6742 | ||
0395d84f | 6743 | fs_devices->seeding = true; |
5f375835 MX |
6744 | fs_devices->opened = 1; |
6745 | return fs_devices; | |
2b82032c | 6746 | } |
e4404d6e | 6747 | |
427c8fdd NB |
6748 | /* |
6749 | * Upon first call for a seed fs fsid, just create a private copy of the | |
6750 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
6751 | */ | |
e4404d6e | 6752 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
6753 | if (IS_ERR(fs_devices)) |
6754 | return fs_devices; | |
2b82032c | 6755 | |
897fb573 | 6756 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6757 | if (ret) { |
6758 | free_fs_devices(fs_devices); | |
5f375835 | 6759 | fs_devices = ERR_PTR(ret); |
2b82032c | 6760 | goto out; |
48d28232 | 6761 | } |
2b82032c YZ |
6762 | |
6763 | if (!fs_devices->seeding) { | |
0226e0eb | 6764 | close_fs_devices(fs_devices); |
e4404d6e | 6765 | free_fs_devices(fs_devices); |
5f375835 | 6766 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6767 | goto out; |
6768 | } | |
6769 | ||
944d3f9f | 6770 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
2b82032c | 6771 | out: |
5f375835 | 6772 | return fs_devices; |
2b82032c YZ |
6773 | } |
6774 | ||
17850759 | 6775 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6776 | struct btrfs_dev_item *dev_item) |
6777 | { | |
17850759 | 6778 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6779 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6780 | struct btrfs_device *device; |
6781 | u64 devid; | |
6782 | int ret; | |
44880fdc | 6783 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6784 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6785 | ||
0b86a832 | 6786 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6787 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6788 | BTRFS_UUID_SIZE); |
1473b24e | 6789 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6790 | BTRFS_FSID_SIZE); |
2b82032c | 6791 | |
de37aa51 | 6792 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6793 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6794 | if (IS_ERR(fs_devices)) |
6795 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6796 | } |
6797 | ||
e4319cd9 | 6798 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6799 | fs_uuid, true); |
5f375835 | 6800 | if (!device) { |
c5502451 | 6801 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6802 | btrfs_report_missing_device(fs_info, devid, |
6803 | dev_uuid, true); | |
45dbdbc9 | 6804 | return -ENOENT; |
c5502451 | 6805 | } |
2b82032c | 6806 | |
2ff7e61e | 6807 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6808 | if (IS_ERR(device)) { |
6809 | btrfs_err(fs_info, | |
6810 | "failed to add missing dev %llu: %ld", | |
6811 | devid, PTR_ERR(device)); | |
6812 | return PTR_ERR(device); | |
6813 | } | |
2b902dfc | 6814 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6815 | } else { |
c5502451 | 6816 | if (!device->bdev) { |
2b902dfc AJ |
6817 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6818 | btrfs_report_missing_device(fs_info, | |
6819 | devid, dev_uuid, true); | |
45dbdbc9 | 6820 | return -ENOENT; |
2b902dfc AJ |
6821 | } |
6822 | btrfs_report_missing_device(fs_info, devid, | |
6823 | dev_uuid, false); | |
c5502451 | 6824 | } |
5f375835 | 6825 | |
e6e674bd AJ |
6826 | if (!device->bdev && |
6827 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6828 | /* |
6829 | * this happens when a device that was properly setup | |
6830 | * in the device info lists suddenly goes bad. | |
6831 | * device->bdev is NULL, and so we have to set | |
6832 | * device->missing to one here | |
6833 | */ | |
5f375835 | 6834 | device->fs_devices->missing_devices++; |
e6e674bd | 6835 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6836 | } |
5f375835 MX |
6837 | |
6838 | /* Move the device to its own fs_devices */ | |
6839 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6840 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6841 | &device->dev_state)); | |
5f375835 MX |
6842 | |
6843 | list_move(&device->dev_list, &fs_devices->devices); | |
6844 | device->fs_devices->num_devices--; | |
6845 | fs_devices->num_devices++; | |
6846 | ||
6847 | device->fs_devices->missing_devices--; | |
6848 | fs_devices->missing_devices++; | |
6849 | ||
6850 | device->fs_devices = fs_devices; | |
6851 | } | |
2b82032c YZ |
6852 | } |
6853 | ||
0b246afa | 6854 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6855 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6856 | if (device->generation != |
6857 | btrfs_device_generation(leaf, dev_item)) | |
6858 | return -EINVAL; | |
6324fbf3 | 6859 | } |
0b86a832 CM |
6860 | |
6861 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6862 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6863 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6864 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6865 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6866 | atomic64_add(device->total_bytes - device->bytes_used, |
6867 | &fs_info->free_chunk_space); | |
2bf64758 | 6868 | } |
0b86a832 | 6869 | ret = 0; |
0b86a832 CM |
6870 | return ret; |
6871 | } | |
6872 | ||
6bccf3ab | 6873 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6874 | { |
6bccf3ab | 6875 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6876 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6877 | struct extent_buffer *sb; |
0b86a832 | 6878 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6879 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6880 | u8 *array_ptr; |
6881 | unsigned long sb_array_offset; | |
84eed90f | 6882 | int ret = 0; |
0b86a832 CM |
6883 | u32 num_stripes; |
6884 | u32 array_size; | |
6885 | u32 len = 0; | |
1ffb22cf | 6886 | u32 cur_offset; |
e06cd3dd | 6887 | u64 type; |
84eed90f | 6888 | struct btrfs_key key; |
0b86a832 | 6889 | |
0b246afa | 6890 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6891 | /* |
6892 | * This will create extent buffer of nodesize, superblock size is | |
6893 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6894 | * overallocate but we can keep it as-is, only the first page is used. | |
6895 | */ | |
2ff7e61e | 6896 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6897 | if (IS_ERR(sb)) |
6898 | return PTR_ERR(sb); | |
4db8c528 | 6899 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6900 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6901 | /* |
01327610 | 6902 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6903 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6904 | * pages up-to-date when the page is larger: extent does not cover the |
6905 | * whole page and consequently check_page_uptodate does not find all | |
6906 | * the page's extents up-to-date (the hole beyond sb), | |
6907 | * write_extent_buffer then triggers a WARN_ON. | |
6908 | * | |
6909 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6910 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6911 | * to silence the warning eg. on PowerPC 64. | |
6912 | */ | |
09cbfeaf | 6913 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6914 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6915 | |
a061fc8d | 6916 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6917 | array_size = btrfs_super_sys_array_size(super_copy); |
6918 | ||
1ffb22cf DS |
6919 | array_ptr = super_copy->sys_chunk_array; |
6920 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6921 | cur_offset = 0; | |
0b86a832 | 6922 | |
1ffb22cf DS |
6923 | while (cur_offset < array_size) { |
6924 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6925 | len = sizeof(*disk_key); |
6926 | if (cur_offset + len > array_size) | |
6927 | goto out_short_read; | |
6928 | ||
0b86a832 CM |
6929 | btrfs_disk_key_to_cpu(&key, disk_key); |
6930 | ||
1ffb22cf DS |
6931 | array_ptr += len; |
6932 | sb_array_offset += len; | |
6933 | cur_offset += len; | |
0b86a832 | 6934 | |
32ab3d1b JT |
6935 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
6936 | btrfs_err(fs_info, | |
6937 | "unexpected item type %u in sys_array at offset %u", | |
6938 | (u32)key.type, cur_offset); | |
6939 | ret = -EIO; | |
6940 | break; | |
6941 | } | |
f5cdedd7 | 6942 | |
32ab3d1b JT |
6943 | chunk = (struct btrfs_chunk *)sb_array_offset; |
6944 | /* | |
6945 | * At least one btrfs_chunk with one stripe must be present, | |
6946 | * exact stripe count check comes afterwards | |
6947 | */ | |
6948 | len = btrfs_chunk_item_size(1); | |
6949 | if (cur_offset + len > array_size) | |
6950 | goto out_short_read; | |
e06cd3dd | 6951 | |
32ab3d1b JT |
6952 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
6953 | if (!num_stripes) { | |
6954 | btrfs_err(fs_info, | |
6955 | "invalid number of stripes %u in sys_array at offset %u", | |
6956 | num_stripes, cur_offset); | |
6957 | ret = -EIO; | |
6958 | break; | |
6959 | } | |
e3540eab | 6960 | |
32ab3d1b JT |
6961 | type = btrfs_chunk_type(sb, chunk); |
6962 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6963 | btrfs_err(fs_info, |
32ab3d1b JT |
6964 | "invalid chunk type %llu in sys_array at offset %u", |
6965 | type, cur_offset); | |
84eed90f CM |
6966 | ret = -EIO; |
6967 | break; | |
0b86a832 | 6968 | } |
32ab3d1b JT |
6969 | |
6970 | len = btrfs_chunk_item_size(num_stripes); | |
6971 | if (cur_offset + len > array_size) | |
6972 | goto out_short_read; | |
6973 | ||
6974 | ret = read_one_chunk(&key, sb, chunk); | |
6975 | if (ret) | |
6976 | break; | |
6977 | ||
1ffb22cf DS |
6978 | array_ptr += len; |
6979 | sb_array_offset += len; | |
6980 | cur_offset += len; | |
0b86a832 | 6981 | } |
d865177a | 6982 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6983 | free_extent_buffer_stale(sb); |
84eed90f | 6984 | return ret; |
e3540eab DS |
6985 | |
6986 | out_short_read: | |
ab8d0fc4 | 6987 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6988 | len, cur_offset); |
d865177a | 6989 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6990 | free_extent_buffer_stale(sb); |
e3540eab | 6991 | return -EIO; |
0b86a832 CM |
6992 | } |
6993 | ||
21634a19 QW |
6994 | /* |
6995 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6996 | * | |
6528b99d AJ |
6997 | * If the @failing_dev is specified, it's accounted as missing. |
6998 | * | |
21634a19 QW |
6999 | * Return true if all chunks meet the minimal RW mount requirements. |
7000 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7001 | */ | |
6528b99d AJ |
7002 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7003 | struct btrfs_device *failing_dev) | |
21634a19 | 7004 | { |
c8bf1b67 | 7005 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7006 | struct extent_map *em; |
7007 | u64 next_start = 0; | |
7008 | bool ret = true; | |
7009 | ||
c8bf1b67 DS |
7010 | read_lock(&map_tree->lock); |
7011 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7012 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7013 | /* No chunk at all? Return false anyway */ |
7014 | if (!em) { | |
7015 | ret = false; | |
7016 | goto out; | |
7017 | } | |
7018 | while (em) { | |
7019 | struct map_lookup *map; | |
7020 | int missing = 0; | |
7021 | int max_tolerated; | |
7022 | int i; | |
7023 | ||
7024 | map = em->map_lookup; | |
7025 | max_tolerated = | |
7026 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7027 | map->type); | |
7028 | for (i = 0; i < map->num_stripes; i++) { | |
7029 | struct btrfs_device *dev = map->stripes[i].dev; | |
7030 | ||
e6e674bd AJ |
7031 | if (!dev || !dev->bdev || |
7032 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7033 | dev->last_flush_error) |
7034 | missing++; | |
6528b99d AJ |
7035 | else if (failing_dev && failing_dev == dev) |
7036 | missing++; | |
21634a19 QW |
7037 | } |
7038 | if (missing > max_tolerated) { | |
6528b99d AJ |
7039 | if (!failing_dev) |
7040 | btrfs_warn(fs_info, | |
52042d8e | 7041 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7042 | em->start, missing, max_tolerated); |
7043 | free_extent_map(em); | |
7044 | ret = false; | |
7045 | goto out; | |
7046 | } | |
7047 | next_start = extent_map_end(em); | |
7048 | free_extent_map(em); | |
7049 | ||
c8bf1b67 DS |
7050 | read_lock(&map_tree->lock); |
7051 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7052 | (u64)(-1) - next_start); |
c8bf1b67 | 7053 | read_unlock(&map_tree->lock); |
21634a19 QW |
7054 | } |
7055 | out: | |
7056 | return ret; | |
7057 | } | |
7058 | ||
d85327b1 DS |
7059 | static void readahead_tree_node_children(struct extent_buffer *node) |
7060 | { | |
7061 | int i; | |
7062 | const int nr_items = btrfs_header_nritems(node); | |
7063 | ||
7064 | for (i = 0; i < nr_items; i++) { | |
7065 | u64 start; | |
7066 | ||
7067 | start = btrfs_node_blockptr(node, i); | |
7068 | readahead_tree_block(node->fs_info, start); | |
7069 | } | |
7070 | } | |
7071 | ||
5b4aacef | 7072 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7073 | { |
5b4aacef | 7074 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7075 | struct btrfs_path *path; |
7076 | struct extent_buffer *leaf; | |
7077 | struct btrfs_key key; | |
7078 | struct btrfs_key found_key; | |
7079 | int ret; | |
7080 | int slot; | |
99e3ecfc | 7081 | u64 total_dev = 0; |
d85327b1 | 7082 | u64 last_ra_node = 0; |
0b86a832 | 7083 | |
0b86a832 CM |
7084 | path = btrfs_alloc_path(); |
7085 | if (!path) | |
7086 | return -ENOMEM; | |
7087 | ||
3dd0f7a3 AJ |
7088 | /* |
7089 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7090 | * otherwise we don't need it. | |
7091 | */ | |
b367e47f | 7092 | mutex_lock(&uuid_mutex); |
b367e47f | 7093 | |
48cfa61b BB |
7094 | /* |
7095 | * It is possible for mount and umount to race in such a way that | |
7096 | * we execute this code path, but open_fs_devices failed to clear | |
7097 | * total_rw_bytes. We certainly want it cleared before reading the | |
7098 | * device items, so clear it here. | |
7099 | */ | |
7100 | fs_info->fs_devices->total_rw_bytes = 0; | |
7101 | ||
395927a9 FDBM |
7102 | /* |
7103 | * Read all device items, and then all the chunk items. All | |
7104 | * device items are found before any chunk item (their object id | |
7105 | * is smaller than the lowest possible object id for a chunk | |
7106 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7107 | */ |
7108 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7109 | key.offset = 0; | |
7110 | key.type = 0; | |
0b86a832 | 7111 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7112 | if (ret < 0) |
7113 | goto error; | |
d397712b | 7114 | while (1) { |
d85327b1 DS |
7115 | struct extent_buffer *node; |
7116 | ||
0b86a832 CM |
7117 | leaf = path->nodes[0]; |
7118 | slot = path->slots[0]; | |
7119 | if (slot >= btrfs_header_nritems(leaf)) { | |
7120 | ret = btrfs_next_leaf(root, path); | |
7121 | if (ret == 0) | |
7122 | continue; | |
7123 | if (ret < 0) | |
7124 | goto error; | |
7125 | break; | |
7126 | } | |
d85327b1 DS |
7127 | /* |
7128 | * The nodes on level 1 are not locked but we don't need to do | |
7129 | * that during mount time as nothing else can access the tree | |
7130 | */ | |
7131 | node = path->nodes[1]; | |
7132 | if (node) { | |
7133 | if (last_ra_node != node->start) { | |
7134 | readahead_tree_node_children(node); | |
7135 | last_ra_node = node->start; | |
7136 | } | |
7137 | } | |
0b86a832 | 7138 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
395927a9 FDBM |
7139 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7140 | struct btrfs_dev_item *dev_item; | |
7141 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7142 | struct btrfs_dev_item); |
17850759 | 7143 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7144 | if (ret) |
7145 | goto error; | |
99e3ecfc | 7146 | total_dev++; |
0b86a832 CM |
7147 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7148 | struct btrfs_chunk *chunk; | |
7149 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
01d01caf | 7150 | mutex_lock(&fs_info->chunk_mutex); |
9690ac09 | 7151 | ret = read_one_chunk(&found_key, leaf, chunk); |
01d01caf | 7152 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c YZ |
7153 | if (ret) |
7154 | goto error; | |
0b86a832 CM |
7155 | } |
7156 | path->slots[0]++; | |
7157 | } | |
99e3ecfc LB |
7158 | |
7159 | /* | |
7160 | * After loading chunk tree, we've got all device information, | |
7161 | * do another round of validation checks. | |
7162 | */ | |
0b246afa JM |
7163 | if (total_dev != fs_info->fs_devices->total_devices) { |
7164 | btrfs_err(fs_info, | |
99e3ecfc | 7165 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7166 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7167 | total_dev); |
7168 | ret = -EINVAL; | |
7169 | goto error; | |
7170 | } | |
0b246afa JM |
7171 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7172 | fs_info->fs_devices->total_rw_bytes) { | |
7173 | btrfs_err(fs_info, | |
99e3ecfc | 7174 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7175 | btrfs_super_total_bytes(fs_info->super_copy), |
7176 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7177 | ret = -EINVAL; |
7178 | goto error; | |
7179 | } | |
0b86a832 CM |
7180 | ret = 0; |
7181 | error: | |
b367e47f LZ |
7182 | mutex_unlock(&uuid_mutex); |
7183 | ||
2b82032c | 7184 | btrfs_free_path(path); |
0b86a832 CM |
7185 | return ret; |
7186 | } | |
442a4f63 | 7187 | |
cb517eab MX |
7188 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7189 | { | |
944d3f9f | 7190 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab MX |
7191 | struct btrfs_device *device; |
7192 | ||
944d3f9f NB |
7193 | fs_devices->fs_info = fs_info; |
7194 | ||
7195 | mutex_lock(&fs_devices->device_list_mutex); | |
7196 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7197 | device->fs_info = fs_info; | |
7198 | mutex_unlock(&fs_devices->device_list_mutex); | |
7199 | ||
7200 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
7201 | mutex_lock(&seed_devs->device_list_mutex); | |
7202 | list_for_each_entry(device, &seed_devs->devices, dev_list) | |
fb456252 | 7203 | device->fs_info = fs_info; |
944d3f9f | 7204 | mutex_unlock(&seed_devs->device_list_mutex); |
29cc83f6 | 7205 | |
944d3f9f | 7206 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7207 | } |
cb517eab MX |
7208 | } |
7209 | ||
1dc990df DS |
7210 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7211 | const struct btrfs_dev_stats_item *ptr, | |
7212 | int index) | |
7213 | { | |
7214 | u64 val; | |
7215 | ||
7216 | read_extent_buffer(eb, &val, | |
7217 | offsetof(struct btrfs_dev_stats_item, values) + | |
7218 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7219 | sizeof(val)); | |
7220 | return val; | |
7221 | } | |
7222 | ||
7223 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7224 | struct btrfs_dev_stats_item *ptr, | |
7225 | int index, u64 val) | |
7226 | { | |
7227 | write_extent_buffer(eb, &val, | |
7228 | offsetof(struct btrfs_dev_stats_item, values) + | |
7229 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7230 | sizeof(val)); | |
7231 | } | |
7232 | ||
733f4fbb SB |
7233 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7234 | { | |
7235 | struct btrfs_key key; | |
733f4fbb SB |
7236 | struct btrfs_root *dev_root = fs_info->dev_root; |
7237 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7238 | struct extent_buffer *eb; | |
7239 | int slot; | |
7240 | int ret = 0; | |
7241 | struct btrfs_device *device; | |
7242 | struct btrfs_path *path = NULL; | |
7243 | int i; | |
7244 | ||
7245 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7246 | if (!path) |
7247 | return -ENOMEM; | |
733f4fbb SB |
7248 | |
7249 | mutex_lock(&fs_devices->device_list_mutex); | |
7250 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7251 | int item_size; | |
7252 | struct btrfs_dev_stats_item *ptr; | |
7253 | ||
242e2956 DS |
7254 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7255 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7256 | key.offset = device->devid; |
7257 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7258 | if (ret) { | |
ae4b9b4c AJ |
7259 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7260 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7261 | device->dev_stats_valid = 1; |
7262 | btrfs_release_path(path); | |
7263 | continue; | |
7264 | } | |
7265 | slot = path->slots[0]; | |
7266 | eb = path->nodes[0]; | |
733f4fbb SB |
7267 | item_size = btrfs_item_size_nr(eb, slot); |
7268 | ||
7269 | ptr = btrfs_item_ptr(eb, slot, | |
7270 | struct btrfs_dev_stats_item); | |
7271 | ||
7272 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7273 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7274 | btrfs_dev_stat_set(device, i, | |
7275 | btrfs_dev_stats_value(eb, ptr, i)); | |
7276 | else | |
4e411a7d | 7277 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7278 | } |
7279 | ||
7280 | device->dev_stats_valid = 1; | |
7281 | btrfs_dev_stat_print_on_load(device); | |
7282 | btrfs_release_path(path); | |
7283 | } | |
7284 | mutex_unlock(&fs_devices->device_list_mutex); | |
7285 | ||
733f4fbb SB |
7286 | btrfs_free_path(path); |
7287 | return ret < 0 ? ret : 0; | |
7288 | } | |
7289 | ||
7290 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7291 | struct btrfs_device *device) |
7292 | { | |
5495f195 | 7293 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7294 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7295 | struct btrfs_path *path; |
7296 | struct btrfs_key key; | |
7297 | struct extent_buffer *eb; | |
7298 | struct btrfs_dev_stats_item *ptr; | |
7299 | int ret; | |
7300 | int i; | |
7301 | ||
242e2956 DS |
7302 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7303 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7304 | key.offset = device->devid; |
7305 | ||
7306 | path = btrfs_alloc_path(); | |
fa252992 DS |
7307 | if (!path) |
7308 | return -ENOMEM; | |
733f4fbb SB |
7309 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7310 | if (ret < 0) { | |
0b246afa | 7311 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7312 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7313 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7314 | goto out; |
7315 | } | |
7316 | ||
7317 | if (ret == 0 && | |
7318 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7319 | /* need to delete old one and insert a new one */ | |
7320 | ret = btrfs_del_item(trans, dev_root, path); | |
7321 | if (ret != 0) { | |
0b246afa | 7322 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7323 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7324 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7325 | goto out; |
7326 | } | |
7327 | ret = 1; | |
7328 | } | |
7329 | ||
7330 | if (ret == 1) { | |
7331 | /* need to insert a new item */ | |
7332 | btrfs_release_path(path); | |
7333 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7334 | &key, sizeof(*ptr)); | |
7335 | if (ret < 0) { | |
0b246afa | 7336 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7337 | "insert dev_stats item for device %s failed %d", |
7338 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7339 | goto out; |
7340 | } | |
7341 | } | |
7342 | ||
7343 | eb = path->nodes[0]; | |
7344 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7345 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7346 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7347 | btrfs_dev_stat_read(device, i)); | |
7348 | btrfs_mark_buffer_dirty(eb); | |
7349 | ||
7350 | out: | |
7351 | btrfs_free_path(path); | |
7352 | return ret; | |
7353 | } | |
7354 | ||
7355 | /* | |
7356 | * called from commit_transaction. Writes all changed device stats to disk. | |
7357 | */ | |
196c9d8d | 7358 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7359 | { |
196c9d8d | 7360 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7361 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7362 | struct btrfs_device *device; | |
addc3fa7 | 7363 | int stats_cnt; |
733f4fbb SB |
7364 | int ret = 0; |
7365 | ||
7366 | mutex_lock(&fs_devices->device_list_mutex); | |
7367 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7368 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7369 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7370 | continue; |
7371 | ||
9deae968 NB |
7372 | |
7373 | /* | |
7374 | * There is a LOAD-LOAD control dependency between the value of | |
7375 | * dev_stats_ccnt and updating the on-disk values which requires | |
7376 | * reading the in-memory counters. Such control dependencies | |
7377 | * require explicit read memory barriers. | |
7378 | * | |
7379 | * This memory barriers pairs with smp_mb__before_atomic in | |
7380 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7381 | * barrier implied by atomic_xchg in | |
7382 | * btrfs_dev_stats_read_and_reset | |
7383 | */ | |
7384 | smp_rmb(); | |
7385 | ||
5495f195 | 7386 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7387 | if (!ret) |
addc3fa7 | 7388 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7389 | } |
7390 | mutex_unlock(&fs_devices->device_list_mutex); | |
7391 | ||
7392 | return ret; | |
7393 | } | |
7394 | ||
442a4f63 SB |
7395 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7396 | { | |
7397 | btrfs_dev_stat_inc(dev, index); | |
7398 | btrfs_dev_stat_print_on_error(dev); | |
7399 | } | |
7400 | ||
48a3b636 | 7401 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7402 | { |
733f4fbb SB |
7403 | if (!dev->dev_stats_valid) |
7404 | return; | |
fb456252 | 7405 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7406 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7407 | rcu_str_deref(dev->name), |
442a4f63 SB |
7408 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7409 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7410 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7411 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7412 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7413 | } |
c11d2c23 | 7414 | |
733f4fbb SB |
7415 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7416 | { | |
a98cdb85 SB |
7417 | int i; |
7418 | ||
7419 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7420 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7421 | break; | |
7422 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7423 | return; /* all values == 0, suppress message */ | |
7424 | ||
fb456252 | 7425 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7426 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7427 | rcu_str_deref(dev->name), |
733f4fbb SB |
7428 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7429 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7430 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7431 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7432 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7433 | } | |
7434 | ||
2ff7e61e | 7435 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7436 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7437 | { |
7438 | struct btrfs_device *dev; | |
0b246afa | 7439 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7440 | int i; |
7441 | ||
7442 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7443 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7444 | true); | |
c11d2c23 SB |
7445 | mutex_unlock(&fs_devices->device_list_mutex); |
7446 | ||
7447 | if (!dev) { | |
0b246afa | 7448 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7449 | return -ENODEV; |
733f4fbb | 7450 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7451 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7452 | return -ENODEV; |
b27f7c0c | 7453 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7454 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7455 | if (stats->nr_items > i) | |
7456 | stats->values[i] = | |
7457 | btrfs_dev_stat_read_and_reset(dev, i); | |
7458 | else | |
4e411a7d | 7459 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7460 | } |
a69976bc AJ |
7461 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7462 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7463 | } else { |
7464 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7465 | if (stats->nr_items > i) | |
7466 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7467 | } | |
7468 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7469 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7470 | return 0; | |
7471 | } | |
a8a6dab7 | 7472 | |
935e5cc9 | 7473 | /* |
bbbf7243 NB |
7474 | * Update the size and bytes used for each device where it changed. This is |
7475 | * delayed since we would otherwise get errors while writing out the | |
7476 | * superblocks. | |
7477 | * | |
7478 | * Must be invoked during transaction commit. | |
935e5cc9 | 7479 | */ |
bbbf7243 | 7480 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7481 | { |
935e5cc9 MX |
7482 | struct btrfs_device *curr, *next; |
7483 | ||
bbbf7243 | 7484 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7485 | |
bbbf7243 | 7486 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7487 | return; |
7488 | ||
bbbf7243 NB |
7489 | /* |
7490 | * We don't need the device_list_mutex here. This list is owned by the | |
7491 | * transaction and the transaction must complete before the device is | |
7492 | * released. | |
7493 | */ | |
7494 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7495 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7496 | post_commit_list) { | |
7497 | list_del_init(&curr->post_commit_list); | |
7498 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7499 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7500 | } |
bbbf7243 | 7501 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7502 | } |
5a13f430 | 7503 | |
46df06b8 DS |
7504 | /* |
7505 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7506 | */ | |
7507 | int btrfs_bg_type_to_factor(u64 flags) | |
7508 | { | |
44b28ada DS |
7509 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7510 | ||
7511 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7512 | } |
cf90d884 QW |
7513 | |
7514 | ||
cf90d884 QW |
7515 | |
7516 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7517 | u64 chunk_offset, u64 devid, | |
7518 | u64 physical_offset, u64 physical_len) | |
7519 | { | |
c8bf1b67 | 7520 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7521 | struct extent_map *em; |
7522 | struct map_lookup *map; | |
05a37c48 | 7523 | struct btrfs_device *dev; |
cf90d884 QW |
7524 | u64 stripe_len; |
7525 | bool found = false; | |
7526 | int ret = 0; | |
7527 | int i; | |
7528 | ||
7529 | read_lock(&em_tree->lock); | |
7530 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7531 | read_unlock(&em_tree->lock); | |
7532 | ||
7533 | if (!em) { | |
7534 | btrfs_err(fs_info, | |
7535 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7536 | physical_offset, devid); | |
7537 | ret = -EUCLEAN; | |
7538 | goto out; | |
7539 | } | |
7540 | ||
7541 | map = em->map_lookup; | |
7542 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7543 | if (physical_len != stripe_len) { | |
7544 | btrfs_err(fs_info, | |
7545 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7546 | physical_offset, devid, em->start, physical_len, | |
7547 | stripe_len); | |
7548 | ret = -EUCLEAN; | |
7549 | goto out; | |
7550 | } | |
7551 | ||
7552 | for (i = 0; i < map->num_stripes; i++) { | |
7553 | if (map->stripes[i].dev->devid == devid && | |
7554 | map->stripes[i].physical == physical_offset) { | |
7555 | found = true; | |
7556 | if (map->verified_stripes >= map->num_stripes) { | |
7557 | btrfs_err(fs_info, | |
7558 | "too many dev extents for chunk %llu found", | |
7559 | em->start); | |
7560 | ret = -EUCLEAN; | |
7561 | goto out; | |
7562 | } | |
7563 | map->verified_stripes++; | |
7564 | break; | |
7565 | } | |
7566 | } | |
7567 | if (!found) { | |
7568 | btrfs_err(fs_info, | |
7569 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7570 | physical_offset, devid); | |
7571 | ret = -EUCLEAN; | |
7572 | } | |
05a37c48 QW |
7573 | |
7574 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7575 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7576 | if (!dev) { |
7577 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7578 | ret = -EUCLEAN; | |
7579 | goto out; | |
7580 | } | |
1b3922a8 QW |
7581 | |
7582 | /* It's possible this device is a dummy for seed device */ | |
7583 | if (dev->disk_total_bytes == 0) { | |
944d3f9f NB |
7584 | struct btrfs_fs_devices *devs; |
7585 | ||
7586 | devs = list_first_entry(&fs_info->fs_devices->seed_list, | |
7587 | struct btrfs_fs_devices, seed_list); | |
7588 | dev = btrfs_find_device(devs, devid, NULL, NULL, false); | |
1b3922a8 QW |
7589 | if (!dev) { |
7590 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7591 | devid); | |
7592 | ret = -EUCLEAN; | |
7593 | goto out; | |
7594 | } | |
7595 | } | |
7596 | ||
05a37c48 QW |
7597 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7598 | btrfs_err(fs_info, | |
7599 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7600 | devid, physical_offset, physical_len, | |
7601 | dev->disk_total_bytes); | |
7602 | ret = -EUCLEAN; | |
7603 | goto out; | |
7604 | } | |
cf90d884 QW |
7605 | out: |
7606 | free_extent_map(em); | |
7607 | return ret; | |
7608 | } | |
7609 | ||
7610 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7611 | { | |
c8bf1b67 | 7612 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7613 | struct extent_map *em; |
7614 | struct rb_node *node; | |
7615 | int ret = 0; | |
7616 | ||
7617 | read_lock(&em_tree->lock); | |
07e1ce09 | 7618 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7619 | em = rb_entry(node, struct extent_map, rb_node); |
7620 | if (em->map_lookup->num_stripes != | |
7621 | em->map_lookup->verified_stripes) { | |
7622 | btrfs_err(fs_info, | |
7623 | "chunk %llu has missing dev extent, have %d expect %d", | |
7624 | em->start, em->map_lookup->verified_stripes, | |
7625 | em->map_lookup->num_stripes); | |
7626 | ret = -EUCLEAN; | |
7627 | goto out; | |
7628 | } | |
7629 | } | |
7630 | out: | |
7631 | read_unlock(&em_tree->lock); | |
7632 | return ret; | |
7633 | } | |
7634 | ||
7635 | /* | |
7636 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7637 | * later chunk allocation/free would cause unexpected behavior. | |
7638 | * | |
7639 | * NOTE: This will iterate through the whole device tree, which should be of | |
7640 | * the same size level as the chunk tree. This slightly increases mount time. | |
7641 | */ | |
7642 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7643 | { | |
7644 | struct btrfs_path *path; | |
7645 | struct btrfs_root *root = fs_info->dev_root; | |
7646 | struct btrfs_key key; | |
5eb19381 QW |
7647 | u64 prev_devid = 0; |
7648 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7649 | int ret = 0; |
7650 | ||
7651 | key.objectid = 1; | |
7652 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7653 | key.offset = 0; | |
7654 | ||
7655 | path = btrfs_alloc_path(); | |
7656 | if (!path) | |
7657 | return -ENOMEM; | |
7658 | ||
7659 | path->reada = READA_FORWARD; | |
7660 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7661 | if (ret < 0) | |
7662 | goto out; | |
7663 | ||
7664 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7665 | ret = btrfs_next_item(root, path); | |
7666 | if (ret < 0) | |
7667 | goto out; | |
7668 | /* No dev extents at all? Not good */ | |
7669 | if (ret > 0) { | |
7670 | ret = -EUCLEAN; | |
7671 | goto out; | |
7672 | } | |
7673 | } | |
7674 | while (1) { | |
7675 | struct extent_buffer *leaf = path->nodes[0]; | |
7676 | struct btrfs_dev_extent *dext; | |
7677 | int slot = path->slots[0]; | |
7678 | u64 chunk_offset; | |
7679 | u64 physical_offset; | |
7680 | u64 physical_len; | |
7681 | u64 devid; | |
7682 | ||
7683 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7684 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7685 | break; | |
7686 | devid = key.objectid; | |
7687 | physical_offset = key.offset; | |
7688 | ||
7689 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7690 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7691 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7692 | ||
5eb19381 QW |
7693 | /* Check if this dev extent overlaps with the previous one */ |
7694 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7695 | btrfs_err(fs_info, | |
7696 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7697 | devid, physical_offset, prev_dev_ext_end); | |
7698 | ret = -EUCLEAN; | |
7699 | goto out; | |
7700 | } | |
7701 | ||
cf90d884 QW |
7702 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7703 | physical_offset, physical_len); | |
7704 | if (ret < 0) | |
7705 | goto out; | |
5eb19381 QW |
7706 | prev_devid = devid; |
7707 | prev_dev_ext_end = physical_offset + physical_len; | |
7708 | ||
cf90d884 QW |
7709 | ret = btrfs_next_item(root, path); |
7710 | if (ret < 0) | |
7711 | goto out; | |
7712 | if (ret > 0) { | |
7713 | ret = 0; | |
7714 | break; | |
7715 | } | |
7716 | } | |
7717 | ||
7718 | /* Ensure all chunks have corresponding dev extents */ | |
7719 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7720 | out: | |
7721 | btrfs_free_path(path); | |
7722 | return ret; | |
7723 | } | |
eede2bf3 OS |
7724 | |
7725 | /* | |
7726 | * Check whether the given block group or device is pinned by any inode being | |
7727 | * used as a swapfile. | |
7728 | */ | |
7729 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7730 | { | |
7731 | struct btrfs_swapfile_pin *sp; | |
7732 | struct rb_node *node; | |
7733 | ||
7734 | spin_lock(&fs_info->swapfile_pins_lock); | |
7735 | node = fs_info->swapfile_pins.rb_node; | |
7736 | while (node) { | |
7737 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7738 | if (ptr < sp->ptr) | |
7739 | node = node->rb_left; | |
7740 | else if (ptr > sp->ptr) | |
7741 | node = node->rb_right; | |
7742 | else | |
7743 | break; | |
7744 | } | |
7745 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7746 | return node != NULL; | |
7747 | } |