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" |
5b316468 | 34 | #include "zoned.h" |
0b86a832 | 35 | |
af902047 ZL |
36 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
37 | [BTRFS_RAID_RAID10] = { | |
38 | .sub_stripes = 2, | |
39 | .dev_stripes = 1, | |
40 | .devs_max = 0, /* 0 == as many as possible */ | |
41 | .devs_min = 4, | |
8789f4fe | 42 | .tolerated_failures = 1, |
af902047 ZL |
43 | .devs_increment = 2, |
44 | .ncopies = 2, | |
b50836ed | 45 | .nparity = 0, |
ed23467b | 46 | .raid_name = "raid10", |
41a6e891 | 47 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 48 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
49 | }, |
50 | [BTRFS_RAID_RAID1] = { | |
51 | .sub_stripes = 1, | |
52 | .dev_stripes = 1, | |
53 | .devs_max = 2, | |
54 | .devs_min = 2, | |
8789f4fe | 55 | .tolerated_failures = 1, |
af902047 ZL |
56 | .devs_increment = 2, |
57 | .ncopies = 2, | |
b50836ed | 58 | .nparity = 0, |
ed23467b | 59 | .raid_name = "raid1", |
41a6e891 | 60 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 61 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 62 | }, |
47e6f742 DS |
63 | [BTRFS_RAID_RAID1C3] = { |
64 | .sub_stripes = 1, | |
65 | .dev_stripes = 1, | |
cf93e15e | 66 | .devs_max = 3, |
47e6f742 DS |
67 | .devs_min = 3, |
68 | .tolerated_failures = 2, | |
69 | .devs_increment = 3, | |
70 | .ncopies = 3, | |
db26a024 | 71 | .nparity = 0, |
47e6f742 DS |
72 | .raid_name = "raid1c3", |
73 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
74 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
75 | }, | |
8d6fac00 DS |
76 | [BTRFS_RAID_RAID1C4] = { |
77 | .sub_stripes = 1, | |
78 | .dev_stripes = 1, | |
cf93e15e | 79 | .devs_max = 4, |
8d6fac00 DS |
80 | .devs_min = 4, |
81 | .tolerated_failures = 3, | |
82 | .devs_increment = 4, | |
83 | .ncopies = 4, | |
db26a024 | 84 | .nparity = 0, |
8d6fac00 DS |
85 | .raid_name = "raid1c4", |
86 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
87 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
88 | }, | |
af902047 ZL |
89 | [BTRFS_RAID_DUP] = { |
90 | .sub_stripes = 1, | |
91 | .dev_stripes = 2, | |
92 | .devs_max = 1, | |
93 | .devs_min = 1, | |
8789f4fe | 94 | .tolerated_failures = 0, |
af902047 ZL |
95 | .devs_increment = 1, |
96 | .ncopies = 2, | |
b50836ed | 97 | .nparity = 0, |
ed23467b | 98 | .raid_name = "dup", |
41a6e891 | 99 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 100 | .mindev_error = 0, |
af902047 ZL |
101 | }, |
102 | [BTRFS_RAID_RAID0] = { | |
103 | .sub_stripes = 1, | |
104 | .dev_stripes = 1, | |
105 | .devs_max = 0, | |
106 | .devs_min = 2, | |
8789f4fe | 107 | .tolerated_failures = 0, |
af902047 ZL |
108 | .devs_increment = 1, |
109 | .ncopies = 1, | |
b50836ed | 110 | .nparity = 0, |
ed23467b | 111 | .raid_name = "raid0", |
41a6e891 | 112 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 113 | .mindev_error = 0, |
af902047 ZL |
114 | }, |
115 | [BTRFS_RAID_SINGLE] = { | |
116 | .sub_stripes = 1, | |
117 | .dev_stripes = 1, | |
118 | .devs_max = 1, | |
119 | .devs_min = 1, | |
8789f4fe | 120 | .tolerated_failures = 0, |
af902047 ZL |
121 | .devs_increment = 1, |
122 | .ncopies = 1, | |
b50836ed | 123 | .nparity = 0, |
ed23467b | 124 | .raid_name = "single", |
41a6e891 | 125 | .bg_flag = 0, |
f9fbcaa2 | 126 | .mindev_error = 0, |
af902047 ZL |
127 | }, |
128 | [BTRFS_RAID_RAID5] = { | |
129 | .sub_stripes = 1, | |
130 | .dev_stripes = 1, | |
131 | .devs_max = 0, | |
132 | .devs_min = 2, | |
8789f4fe | 133 | .tolerated_failures = 1, |
af902047 | 134 | .devs_increment = 1, |
da612e31 | 135 | .ncopies = 1, |
b50836ed | 136 | .nparity = 1, |
ed23467b | 137 | .raid_name = "raid5", |
41a6e891 | 138 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 139 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
140 | }, |
141 | [BTRFS_RAID_RAID6] = { | |
142 | .sub_stripes = 1, | |
143 | .dev_stripes = 1, | |
144 | .devs_max = 0, | |
145 | .devs_min = 3, | |
8789f4fe | 146 | .tolerated_failures = 2, |
af902047 | 147 | .devs_increment = 1, |
da612e31 | 148 | .ncopies = 1, |
b50836ed | 149 | .nparity = 2, |
ed23467b | 150 | .raid_name = "raid6", |
41a6e891 | 151 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 152 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
153 | }, |
154 | }; | |
155 | ||
158da513 | 156 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 157 | { |
158da513 DS |
158 | const int index = btrfs_bg_flags_to_raid_index(flags); |
159 | ||
160 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
161 | return NULL; |
162 | ||
158da513 | 163 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
164 | } |
165 | ||
f89e09cf AJ |
166 | /* |
167 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
168 | * bytes including terminating null byte. | |
169 | */ | |
170 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
171 | { | |
172 | int i; | |
173 | int ret; | |
174 | char *bp = buf; | |
175 | u64 flags = bg_flags; | |
176 | u32 size_bp = size_buf; | |
177 | ||
178 | if (!flags) { | |
179 | strcpy(bp, "NONE"); | |
180 | return; | |
181 | } | |
182 | ||
183 | #define DESCRIBE_FLAG(flag, desc) \ | |
184 | do { \ | |
185 | if (flags & (flag)) { \ | |
186 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
187 | if (ret < 0 || ret >= size_bp) \ | |
188 | goto out_overflow; \ | |
189 | size_bp -= ret; \ | |
190 | bp += ret; \ | |
191 | flags &= ~(flag); \ | |
192 | } \ | |
193 | } while (0) | |
194 | ||
195 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
196 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
197 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
198 | ||
199 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
200 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
201 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
202 | btrfs_raid_array[i].raid_name); | |
203 | #undef DESCRIBE_FLAG | |
204 | ||
205 | if (flags) { | |
206 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
207 | size_bp -= ret; | |
208 | } | |
209 | ||
210 | if (size_bp < size_buf) | |
211 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
212 | ||
213 | /* | |
214 | * The text is trimmed, it's up to the caller to provide sufficiently | |
215 | * large buffer | |
216 | */ | |
217 | out_overflow:; | |
218 | } | |
219 | ||
6f8e0fc7 | 220 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 221 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 222 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 223 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
224 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
225 | enum btrfs_map_op op, | |
226 | u64 logical, u64 *length, | |
227 | struct btrfs_bio **bbio_ret, | |
228 | int mirror_num, int need_raid_map); | |
2b82032c | 229 | |
9c6b1c4d DS |
230 | /* |
231 | * Device locking | |
232 | * ============== | |
233 | * | |
234 | * There are several mutexes that protect manipulation of devices and low-level | |
235 | * structures like chunks but not block groups, extents or files | |
236 | * | |
237 | * uuid_mutex (global lock) | |
238 | * ------------------------ | |
239 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
240 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
241 | * device) or requested by the device= mount option | |
242 | * | |
243 | * the mutex can be very coarse and can cover long-running operations | |
244 | * | |
245 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 246 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
247 | * |
248 | * global::fs_devs - add, remove, updates to the global list | |
249 | * | |
18c850fd JB |
250 | * does not protect: manipulation of the fs_devices::devices list in general |
251 | * but in mount context it could be used to exclude list modifications by eg. | |
252 | * scan ioctl | |
9c6b1c4d DS |
253 | * |
254 | * btrfs_device::name - renames (write side), read is RCU | |
255 | * | |
256 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
257 | * ------------------------------------------------ | |
258 | * protects updates to fs_devices::devices, ie. adding and deleting | |
259 | * | |
260 | * simple list traversal with read-only actions can be done with RCU protection | |
261 | * | |
262 | * may be used to exclude some operations from running concurrently without any | |
263 | * modifications to the list (see write_all_supers) | |
264 | * | |
18c850fd JB |
265 | * Is not required at mount and close times, because our device list is |
266 | * protected by the uuid_mutex at that point. | |
267 | * | |
9c6b1c4d DS |
268 | * balance_mutex |
269 | * ------------- | |
270 | * protects balance structures (status, state) and context accessed from | |
271 | * several places (internally, ioctl) | |
272 | * | |
273 | * chunk_mutex | |
274 | * ----------- | |
275 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
276 | * device is added/removed. Additionally it also protects post_commit_list of |
277 | * individual devices, since they can be added to the transaction's | |
278 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
279 | * |
280 | * cleaner_mutex | |
281 | * ------------- | |
282 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
283 | * cleaning together with relocation or delayed iputs | |
284 | * | |
285 | * | |
286 | * Lock nesting | |
287 | * ============ | |
288 | * | |
289 | * uuid_mutex | |
ae3e715f AJ |
290 | * device_list_mutex |
291 | * chunk_mutex | |
292 | * balance_mutex | |
89595e80 AJ |
293 | * |
294 | * | |
c3e1f96c GR |
295 | * Exclusive operations |
296 | * ==================== | |
89595e80 AJ |
297 | * |
298 | * Maintains the exclusivity of the following operations that apply to the | |
299 | * whole filesystem and cannot run in parallel. | |
300 | * | |
301 | * - Balance (*) | |
302 | * - Device add | |
303 | * - Device remove | |
304 | * - Device replace (*) | |
305 | * - Resize | |
306 | * | |
307 | * The device operations (as above) can be in one of the following states: | |
308 | * | |
309 | * - Running state | |
310 | * - Paused state | |
311 | * - Completed state | |
312 | * | |
313 | * Only device operations marked with (*) can go into the Paused state for the | |
314 | * following reasons: | |
315 | * | |
316 | * - ioctl (only Balance can be Paused through ioctl) | |
317 | * - filesystem remounted as read-only | |
318 | * - filesystem unmounted and mounted as read-only | |
319 | * - system power-cycle and filesystem mounted as read-only | |
320 | * - filesystem or device errors leading to forced read-only | |
321 | * | |
c3e1f96c GR |
322 | * The status of exclusive operation is set and cleared atomically. |
323 | * During the course of Paused state, fs_info::exclusive_operation remains set. | |
89595e80 AJ |
324 | * A device operation in Paused or Running state can be canceled or resumed |
325 | * either by ioctl (Balance only) or when remounted as read-write. | |
c3e1f96c | 326 | * The exclusive status is cleared when the device operation is canceled or |
89595e80 | 327 | * completed. |
9c6b1c4d DS |
328 | */ |
329 | ||
67a2c45e | 330 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 331 | static LIST_HEAD(fs_uuids); |
4143cb8b | 332 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
333 | { |
334 | return &fs_uuids; | |
335 | } | |
8a4b83cc | 336 | |
2dfeca9b DS |
337 | /* |
338 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
339 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
340 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
341 | * |
342 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
343 | * The returned struct is not linked onto any lists and can be destroyed with | |
344 | * kfree() right away. | |
345 | */ | |
7239ff4b NB |
346 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
347 | const u8 *metadata_fsid) | |
2208a378 ID |
348 | { |
349 | struct btrfs_fs_devices *fs_devs; | |
350 | ||
78f2c9e6 | 351 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
352 | if (!fs_devs) |
353 | return ERR_PTR(-ENOMEM); | |
354 | ||
355 | mutex_init(&fs_devs->device_list_mutex); | |
356 | ||
357 | INIT_LIST_HEAD(&fs_devs->devices); | |
358 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 359 | INIT_LIST_HEAD(&fs_devs->fs_list); |
944d3f9f | 360 | INIT_LIST_HEAD(&fs_devs->seed_list); |
2208a378 ID |
361 | if (fsid) |
362 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 363 | |
7239ff4b NB |
364 | if (metadata_fsid) |
365 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
366 | else if (fsid) | |
367 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
368 | ||
2208a378 ID |
369 | return fs_devs; |
370 | } | |
371 | ||
a425f9d4 | 372 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 373 | { |
bbbf7243 | 374 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 375 | rcu_string_free(device->name); |
1c11b63e | 376 | extent_io_tree_release(&device->alloc_state); |
48dae9cf | 377 | bio_put(device->flush_bio); |
5b316468 | 378 | btrfs_destroy_dev_zone_info(device); |
48dae9cf DS |
379 | kfree(device); |
380 | } | |
381 | ||
e4404d6e YZ |
382 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
383 | { | |
384 | struct btrfs_device *device; | |
385 | WARN_ON(fs_devices->opened); | |
386 | while (!list_empty(&fs_devices->devices)) { | |
387 | device = list_entry(fs_devices->devices.next, | |
388 | struct btrfs_device, dev_list); | |
389 | list_del(&device->dev_list); | |
a425f9d4 | 390 | btrfs_free_device(device); |
e4404d6e YZ |
391 | } |
392 | kfree(fs_devices); | |
393 | } | |
394 | ||
ffc5a379 | 395 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
396 | { |
397 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 398 | |
2b82032c YZ |
399 | while (!list_empty(&fs_uuids)) { |
400 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
401 | struct btrfs_fs_devices, fs_list); |
402 | list_del(&fs_devices->fs_list); | |
e4404d6e | 403 | free_fs_devices(fs_devices); |
8a4b83cc | 404 | } |
8a4b83cc CM |
405 | } |
406 | ||
48dae9cf DS |
407 | /* |
408 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
409 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 410 | * btrfs_free_device. |
48dae9cf | 411 | */ |
154f7cb8 | 412 | static struct btrfs_device *__alloc_device(struct btrfs_fs_info *fs_info) |
12bd2fc0 ID |
413 | { |
414 | struct btrfs_device *dev; | |
415 | ||
78f2c9e6 | 416 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
417 | if (!dev) |
418 | return ERR_PTR(-ENOMEM); | |
419 | ||
e0ae9994 DS |
420 | /* |
421 | * Preallocate a bio that's always going to be used for flushing device | |
422 | * barriers and matches the device lifespan | |
423 | */ | |
424 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
425 | if (!dev->flush_bio) { | |
426 | kfree(dev); | |
427 | return ERR_PTR(-ENOMEM); | |
428 | } | |
e0ae9994 | 429 | |
12bd2fc0 ID |
430 | INIT_LIST_HEAD(&dev->dev_list); |
431 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 432 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 | 433 | |
12bd2fc0 | 434 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 435 | atomic_set(&dev->dev_stats_ccnt, 0); |
c41ec452 | 436 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 437 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 438 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
154f7cb8 QW |
439 | extent_io_tree_init(fs_info, &dev->alloc_state, |
440 | IO_TREE_DEVICE_ALLOC_STATE, NULL); | |
12bd2fc0 ID |
441 | |
442 | return dev; | |
443 | } | |
444 | ||
7239ff4b NB |
445 | static noinline struct btrfs_fs_devices *find_fsid( |
446 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 447 | { |
8a4b83cc CM |
448 | struct btrfs_fs_devices *fs_devices; |
449 | ||
7239ff4b NB |
450 | ASSERT(fsid); |
451 | ||
7a62d0f0 | 452 | /* Handle non-split brain cases */ |
c4babc5e | 453 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
454 | if (metadata_fsid) { |
455 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
456 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
457 | BTRFS_FSID_SIZE) == 0) | |
458 | return fs_devices; | |
459 | } else { | |
460 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
461 | return fs_devices; | |
462 | } | |
8a4b83cc CM |
463 | } |
464 | return NULL; | |
465 | } | |
466 | ||
c6730a0e SY |
467 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
468 | struct btrfs_super_block *disk_super) | |
469 | { | |
470 | ||
471 | struct btrfs_fs_devices *fs_devices; | |
472 | ||
473 | /* | |
474 | * Handle scanned device having completed its fsid change but | |
475 | * belonging to a fs_devices that was created by first scanning | |
476 | * a device which didn't have its fsid/metadata_uuid changed | |
477 | * at all and the CHANGING_FSID_V2 flag set. | |
478 | */ | |
479 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
480 | if (fs_devices->fsid_change && | |
481 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
482 | BTRFS_FSID_SIZE) == 0 && | |
483 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
484 | BTRFS_FSID_SIZE) == 0) { | |
485 | return fs_devices; | |
486 | } | |
487 | } | |
488 | /* | |
489 | * Handle scanned device having completed its fsid change but | |
490 | * belonging to a fs_devices that was created by a device that | |
491 | * has an outdated pair of fsid/metadata_uuid and | |
492 | * CHANGING_FSID_V2 flag set. | |
493 | */ | |
494 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
495 | if (fs_devices->fsid_change && | |
496 | memcmp(fs_devices->metadata_uuid, | |
497 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
498 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
499 | BTRFS_FSID_SIZE) == 0) { | |
500 | return fs_devices; | |
501 | } | |
502 | } | |
503 | ||
504 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
505 | } | |
506 | ||
507 | ||
beaf8ab3 SB |
508 | static int |
509 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
510 | int flush, struct block_device **bdev, | |
8f32380d | 511 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
512 | { |
513 | int ret; | |
514 | ||
515 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
516 | ||
517 | if (IS_ERR(*bdev)) { | |
518 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
519 | goto error; |
520 | } | |
521 | ||
522 | if (flush) | |
523 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 524 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
525 | if (ret) { |
526 | blkdev_put(*bdev, flags); | |
527 | goto error; | |
528 | } | |
529 | invalidate_bdev(*bdev); | |
8f32380d JT |
530 | *disk_super = btrfs_read_dev_super(*bdev); |
531 | if (IS_ERR(*disk_super)) { | |
532 | ret = PTR_ERR(*disk_super); | |
beaf8ab3 SB |
533 | blkdev_put(*bdev, flags); |
534 | goto error; | |
535 | } | |
536 | ||
537 | return 0; | |
538 | ||
539 | error: | |
540 | *bdev = NULL; | |
beaf8ab3 SB |
541 | return ret; |
542 | } | |
543 | ||
70bc7088 AJ |
544 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
545 | { | |
546 | int found; | |
547 | ||
548 | rcu_read_lock(); | |
549 | found = strcmp(rcu_str_deref(device->name), path); | |
550 | rcu_read_unlock(); | |
551 | ||
552 | return found == 0; | |
553 | } | |
554 | ||
d8367db3 AJ |
555 | /* |
556 | * Search and remove all stale (devices which are not mounted) devices. | |
557 | * When both inputs are NULL, it will search and release all stale devices. | |
558 | * path: Optional. When provided will it release all unmounted devices | |
559 | * matching this path only. | |
560 | * skip_dev: Optional. Will skip this device when searching for the stale | |
561 | * devices. | |
70bc7088 AJ |
562 | * Return: 0 for success or if @path is NULL. |
563 | * -EBUSY if @path is a mounted device. | |
564 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 565 | */ |
70bc7088 | 566 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 567 | struct btrfs_device *skip_device) |
4fde46f0 | 568 | { |
fa6d2ae5 AJ |
569 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
570 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
571 | int ret = 0; |
572 | ||
573 | if (path) | |
574 | ret = -ENOENT; | |
4fde46f0 | 575 | |
fa6d2ae5 | 576 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 577 | |
70bc7088 | 578 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
579 | list_for_each_entry_safe(device, tmp_device, |
580 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 581 | if (skip_device && skip_device == device) |
d8367db3 | 582 | continue; |
fa6d2ae5 | 583 | if (path && !device->name) |
4fde46f0 | 584 | continue; |
70bc7088 | 585 | if (path && !device_path_matched(path, device)) |
38cf665d | 586 | continue; |
70bc7088 AJ |
587 | if (fs_devices->opened) { |
588 | /* for an already deleted device return 0 */ | |
589 | if (path && ret != 0) | |
590 | ret = -EBUSY; | |
591 | break; | |
592 | } | |
4fde46f0 | 593 | |
4fde46f0 | 594 | /* delete the stale device */ |
7bcb8164 AJ |
595 | fs_devices->num_devices--; |
596 | list_del(&device->dev_list); | |
597 | btrfs_free_device(device); | |
598 | ||
70bc7088 | 599 | ret = 0; |
7bcb8164 AJ |
600 | } |
601 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 602 | |
7bcb8164 AJ |
603 | if (fs_devices->num_devices == 0) { |
604 | btrfs_sysfs_remove_fsid(fs_devices); | |
605 | list_del(&fs_devices->fs_list); | |
606 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
607 | } |
608 | } | |
70bc7088 AJ |
609 | |
610 | return ret; | |
4fde46f0 AJ |
611 | } |
612 | ||
18c850fd JB |
613 | /* |
614 | * This is only used on mount, and we are protected from competing things | |
615 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
616 | * fs_devices->device_list_mutex here. | |
617 | */ | |
0fb08bcc AJ |
618 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
619 | struct btrfs_device *device, fmode_t flags, | |
620 | void *holder) | |
621 | { | |
622 | struct request_queue *q; | |
623 | struct block_device *bdev; | |
0fb08bcc AJ |
624 | struct btrfs_super_block *disk_super; |
625 | u64 devid; | |
626 | int ret; | |
627 | ||
628 | if (device->bdev) | |
629 | return -EINVAL; | |
630 | if (!device->name) | |
631 | return -EINVAL; | |
632 | ||
633 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 634 | &bdev, &disk_super); |
0fb08bcc AJ |
635 | if (ret) |
636 | return ret; | |
637 | ||
0fb08bcc AJ |
638 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
639 | if (devid != device->devid) | |
8f32380d | 640 | goto error_free_page; |
0fb08bcc AJ |
641 | |
642 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 643 | goto error_free_page; |
0fb08bcc AJ |
644 | |
645 | device->generation = btrfs_super_generation(disk_super); | |
646 | ||
647 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
648 | if (btrfs_super_incompat_flags(disk_super) & |
649 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
650 | pr_err( | |
651 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 652 | goto error_free_page; |
7239ff4b NB |
653 | } |
654 | ||
ebbede42 | 655 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 656 | fs_devices->seeding = true; |
0fb08bcc | 657 | } else { |
ebbede42 AJ |
658 | if (bdev_read_only(bdev)) |
659 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
660 | else | |
661 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
662 | } |
663 | ||
664 | q = bdev_get_queue(bdev); | |
0fb08bcc | 665 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 666 | fs_devices->rotating = true; |
0fb08bcc AJ |
667 | |
668 | device->bdev = bdev; | |
e12c9621 | 669 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
670 | device->mode = flags; |
671 | ||
672 | fs_devices->open_devices++; | |
ebbede42 AJ |
673 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
674 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 675 | fs_devices->rw_devices++; |
b1b8e386 | 676 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 677 | } |
8f32380d | 678 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
679 | |
680 | return 0; | |
681 | ||
8f32380d JT |
682 | error_free_page: |
683 | btrfs_release_disk_super(disk_super); | |
0fb08bcc AJ |
684 | blkdev_put(bdev, flags); |
685 | ||
686 | return -EINVAL; | |
687 | } | |
688 | ||
7a62d0f0 NB |
689 | /* |
690 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
691 | * being created with a disk that has already completed its fsid change. Such |
692 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
693 | * Handle both cases here. | |
7a62d0f0 NB |
694 | */ |
695 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
696 | struct btrfs_super_block *disk_super) | |
697 | { | |
698 | struct btrfs_fs_devices *fs_devices; | |
699 | ||
700 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
701 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
702 | BTRFS_FSID_SIZE) != 0 && | |
703 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
704 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
705 | return fs_devices; | |
706 | } | |
707 | } | |
708 | ||
c0d81c7c | 709 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
710 | } |
711 | ||
cc5de4e7 NB |
712 | |
713 | static struct btrfs_fs_devices *find_fsid_changed( | |
714 | struct btrfs_super_block *disk_super) | |
715 | { | |
716 | struct btrfs_fs_devices *fs_devices; | |
717 | ||
718 | /* | |
719 | * Handles the case where scanned device is part of an fs that had | |
720 | * multiple successful changes of FSID but curently device didn't | |
05840710 NB |
721 | * observe it. Meaning our fsid will be different than theirs. We need |
722 | * to handle two subcases : | |
723 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
724 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
725 | * are equal). | |
cc5de4e7 NB |
726 | */ |
727 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 728 | /* Changed UUIDs */ |
cc5de4e7 NB |
729 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
730 | BTRFS_FSID_SIZE) != 0 && | |
731 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
732 | BTRFS_FSID_SIZE) == 0 && | |
733 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
734 | BTRFS_FSID_SIZE) != 0) |
735 | return fs_devices; | |
736 | ||
737 | /* Unchanged UUIDs */ | |
738 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
739 | BTRFS_FSID_SIZE) == 0 && | |
740 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
741 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 742 | return fs_devices; |
cc5de4e7 NB |
743 | } |
744 | ||
745 | return NULL; | |
746 | } | |
1362089d NB |
747 | |
748 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
749 | struct btrfs_super_block *disk_super) | |
750 | { | |
751 | struct btrfs_fs_devices *fs_devices; | |
752 | ||
753 | /* | |
754 | * Handle the case where the scanned device is part of an fs whose last | |
755 | * metadata UUID change reverted it to the original FSID. At the same | |
756 | * time * fs_devices was first created by another constitutent device | |
757 | * which didn't fully observe the operation. This results in an | |
758 | * btrfs_fs_devices created with metadata/fsid different AND | |
759 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
760 | * fs_devices equal to the FSID of the disk. | |
761 | */ | |
762 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
763 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
764 | BTRFS_FSID_SIZE) != 0 && | |
765 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
766 | BTRFS_FSID_SIZE) == 0 && | |
767 | fs_devices->fsid_change) | |
768 | return fs_devices; | |
769 | } | |
770 | ||
771 | return NULL; | |
772 | } | |
60999ca4 DS |
773 | /* |
774 | * Add new device to list of registered devices | |
775 | * | |
776 | * Returns: | |
e124ece5 AJ |
777 | * device pointer which was just added or updated when successful |
778 | * error pointer when failed | |
60999ca4 | 779 | */ |
e124ece5 | 780 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
781 | struct btrfs_super_block *disk_super, |
782 | bool *new_device_added) | |
8a4b83cc CM |
783 | { |
784 | struct btrfs_device *device; | |
7a62d0f0 | 785 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 786 | struct rcu_string *name; |
8a4b83cc | 787 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 788 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
789 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
790 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
791 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
792 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 793 | |
cc5de4e7 | 794 | if (fsid_change_in_progress) { |
c0d81c7c | 795 | if (!has_metadata_uuid) |
cc5de4e7 | 796 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 797 | else |
cc5de4e7 | 798 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 799 | } else if (has_metadata_uuid) { |
c6730a0e | 800 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 801 | } else { |
1362089d NB |
802 | fs_devices = find_fsid_reverted_metadata(disk_super); |
803 | if (!fs_devices) | |
804 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
805 | } |
806 | ||
8a4b83cc | 807 | |
8a4b83cc | 808 | if (!fs_devices) { |
7239ff4b NB |
809 | if (has_metadata_uuid) |
810 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
811 | disk_super->metadata_uuid); | |
812 | else | |
813 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
814 | ||
2208a378 | 815 | if (IS_ERR(fs_devices)) |
e124ece5 | 816 | return ERR_CAST(fs_devices); |
2208a378 | 817 | |
92900e51 AV |
818 | fs_devices->fsid_change = fsid_change_in_progress; |
819 | ||
9c6d173e | 820 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 821 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 822 | |
8a4b83cc CM |
823 | device = NULL; |
824 | } else { | |
9c6d173e | 825 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 | 826 | device = btrfs_find_device(fs_devices, devid, |
b2598edf | 827 | disk_super->dev_item.uuid, NULL); |
7a62d0f0 NB |
828 | |
829 | /* | |
830 | * If this disk has been pulled into an fs devices created by | |
831 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
832 | * metadata_uuid/fsid values of the fs_devices. | |
833 | */ | |
1362089d | 834 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
835 | found_transid > fs_devices->latest_generation) { |
836 | memcpy(fs_devices->fsid, disk_super->fsid, | |
837 | BTRFS_FSID_SIZE); | |
1362089d NB |
838 | |
839 | if (has_metadata_uuid) | |
840 | memcpy(fs_devices->metadata_uuid, | |
841 | disk_super->metadata_uuid, | |
842 | BTRFS_FSID_SIZE); | |
843 | else | |
844 | memcpy(fs_devices->metadata_uuid, | |
845 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
846 | |
847 | fs_devices->fsid_change = false; | |
848 | } | |
8a4b83cc | 849 | } |
443f24fe | 850 | |
8a4b83cc | 851 | if (!device) { |
9c6d173e AJ |
852 | if (fs_devices->opened) { |
853 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 854 | return ERR_PTR(-EBUSY); |
9c6d173e | 855 | } |
2b82032c | 856 | |
12bd2fc0 ID |
857 | device = btrfs_alloc_device(NULL, &devid, |
858 | disk_super->dev_item.uuid); | |
859 | if (IS_ERR(device)) { | |
9c6d173e | 860 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 861 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 862 | return device; |
8a4b83cc | 863 | } |
606686ee JB |
864 | |
865 | name = rcu_string_strdup(path, GFP_NOFS); | |
866 | if (!name) { | |
a425f9d4 | 867 | btrfs_free_device(device); |
9c6d173e | 868 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 869 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 870 | } |
606686ee | 871 | rcu_assign_pointer(device->name, name); |
90519d66 | 872 | |
1f78160c | 873 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 874 | fs_devices->num_devices++; |
e5e9a520 | 875 | |
2b82032c | 876 | device->fs_devices = fs_devices; |
4306a974 | 877 | *new_device_added = true; |
327f18cc AJ |
878 | |
879 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
880 | pr_info( |
881 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
882 | disk_super->label, devid, found_transid, path, | |
883 | current->comm, task_pid_nr(current)); | |
327f18cc | 884 | else |
aa6c0df7 AJ |
885 | pr_info( |
886 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
887 | disk_super->fsid, devid, found_transid, path, | |
888 | current->comm, task_pid_nr(current)); | |
327f18cc | 889 | |
606686ee | 890 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
891 | /* |
892 | * When FS is already mounted. | |
893 | * 1. If you are here and if the device->name is NULL that | |
894 | * means this device was missing at time of FS mount. | |
895 | * 2. If you are here and if the device->name is different | |
896 | * from 'path' that means either | |
897 | * a. The same device disappeared and reappeared with | |
898 | * different name. or | |
899 | * b. The missing-disk-which-was-replaced, has | |
900 | * reappeared now. | |
901 | * | |
902 | * We must allow 1 and 2a above. But 2b would be a spurious | |
903 | * and unintentional. | |
904 | * | |
905 | * Further in case of 1 and 2a above, the disk at 'path' | |
906 | * would have missed some transaction when it was away and | |
907 | * in case of 2a the stale bdev has to be updated as well. | |
908 | * 2b must not be allowed at all time. | |
909 | */ | |
910 | ||
911 | /* | |
0f23ae74 CM |
912 | * For now, we do allow update to btrfs_fs_device through the |
913 | * btrfs dev scan cli after FS has been mounted. We're still | |
914 | * tracking a problem where systems fail mount by subvolume id | |
915 | * when we reject replacement on a mounted FS. | |
b96de000 | 916 | */ |
0f23ae74 | 917 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
918 | /* |
919 | * That is if the FS is _not_ mounted and if you | |
920 | * are here, that means there is more than one | |
921 | * disk with same uuid and devid.We keep the one | |
922 | * with larger generation number or the last-in if | |
923 | * generation are equal. | |
924 | */ | |
9c6d173e | 925 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 926 | return ERR_PTR(-EEXIST); |
77bdae4d | 927 | } |
b96de000 | 928 | |
a9261d41 AJ |
929 | /* |
930 | * We are going to replace the device path for a given devid, | |
931 | * make sure it's the same device if the device is mounted | |
932 | */ | |
933 | if (device->bdev) { | |
4e7b5671 CH |
934 | int error; |
935 | dev_t path_dev; | |
a9261d41 | 936 | |
4e7b5671 CH |
937 | error = lookup_bdev(path, &path_dev); |
938 | if (error) { | |
a9261d41 | 939 | mutex_unlock(&fs_devices->device_list_mutex); |
4e7b5671 | 940 | return ERR_PTR(error); |
a9261d41 AJ |
941 | } |
942 | ||
4e7b5671 | 943 | if (device->bdev->bd_dev != path_dev) { |
a9261d41 | 944 | mutex_unlock(&fs_devices->device_list_mutex); |
0697d9a6 JT |
945 | /* |
946 | * device->fs_info may not be reliable here, so | |
947 | * pass in a NULL instead. This avoids a | |
948 | * possible use-after-free when the fs_info and | |
949 | * fs_info->sb are already torn down. | |
950 | */ | |
951 | btrfs_warn_in_rcu(NULL, | |
79dae17d AJ |
952 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
953 | path, devid, found_transid, | |
954 | current->comm, | |
955 | task_pid_nr(current)); | |
a9261d41 AJ |
956 | return ERR_PTR(-EEXIST); |
957 | } | |
a9261d41 | 958 | btrfs_info_in_rcu(device->fs_info, |
79dae17d AJ |
959 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
960 | devid, rcu_str_deref(device->name), | |
961 | path, current->comm, | |
962 | task_pid_nr(current)); | |
a9261d41 AJ |
963 | } |
964 | ||
606686ee | 965 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
966 | if (!name) { |
967 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 968 | return ERR_PTR(-ENOMEM); |
9c6d173e | 969 | } |
606686ee JB |
970 | rcu_string_free(device->name); |
971 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 972 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 973 | fs_devices->missing_devices--; |
e6e674bd | 974 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 975 | } |
8a4b83cc CM |
976 | } |
977 | ||
77bdae4d AJ |
978 | /* |
979 | * Unmount does not free the btrfs_device struct but would zero | |
980 | * generation along with most of the other members. So just update | |
981 | * it back. We need it to pick the disk with largest generation | |
982 | * (as above). | |
983 | */ | |
d1a63002 | 984 | if (!fs_devices->opened) { |
77bdae4d | 985 | device->generation = found_transid; |
d1a63002 NB |
986 | fs_devices->latest_generation = max_t(u64, found_transid, |
987 | fs_devices->latest_generation); | |
988 | } | |
77bdae4d | 989 | |
f2788d2f AJ |
990 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
991 | ||
9c6d173e | 992 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 993 | return device; |
8a4b83cc CM |
994 | } |
995 | ||
e4404d6e YZ |
996 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
997 | { | |
998 | struct btrfs_fs_devices *fs_devices; | |
999 | struct btrfs_device *device; | |
1000 | struct btrfs_device *orig_dev; | |
d2979aa2 | 1001 | int ret = 0; |
e4404d6e | 1002 | |
7239ff4b | 1003 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1004 | if (IS_ERR(fs_devices)) |
1005 | return fs_devices; | |
e4404d6e | 1006 | |
adbbb863 | 1007 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1008 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1009 | |
1010 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1011 | struct rcu_string *name; |
1012 | ||
12bd2fc0 ID |
1013 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1014 | orig_dev->uuid); | |
d2979aa2 AJ |
1015 | if (IS_ERR(device)) { |
1016 | ret = PTR_ERR(device); | |
e4404d6e | 1017 | goto error; |
d2979aa2 | 1018 | } |
e4404d6e | 1019 | |
606686ee JB |
1020 | /* |
1021 | * This is ok to do without rcu read locked because we hold the | |
1022 | * uuid mutex so nothing we touch in here is going to disappear. | |
1023 | */ | |
e755f780 | 1024 | if (orig_dev->name) { |
78f2c9e6 DS |
1025 | name = rcu_string_strdup(orig_dev->name->str, |
1026 | GFP_KERNEL); | |
e755f780 | 1027 | if (!name) { |
a425f9d4 | 1028 | btrfs_free_device(device); |
d2979aa2 | 1029 | ret = -ENOMEM; |
e755f780 AJ |
1030 | goto error; |
1031 | } | |
1032 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1033 | } |
e4404d6e | 1034 | |
e4404d6e YZ |
1035 | list_add(&device->dev_list, &fs_devices->devices); |
1036 | device->fs_devices = fs_devices; | |
1037 | fs_devices->num_devices++; | |
1038 | } | |
adbbb863 | 1039 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1040 | return fs_devices; |
1041 | error: | |
adbbb863 | 1042 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1043 | free_fs_devices(fs_devices); |
d2979aa2 | 1044 | return ERR_PTR(ret); |
e4404d6e YZ |
1045 | } |
1046 | ||
3712ccb7 | 1047 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
bacce86a | 1048 | struct btrfs_device **latest_dev) |
dfe25020 | 1049 | { |
c6e30871 | 1050 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1051 | |
46224705 | 1052 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1053 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1054 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1055 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1056 | &device->dev_state) && |
998a0671 AJ |
1057 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1058 | &device->dev_state) && | |
3712ccb7 NB |
1059 | (!*latest_dev || |
1060 | device->generation > (*latest_dev)->generation)) { | |
1061 | *latest_dev = device; | |
a6b0d5c8 | 1062 | } |
2b82032c | 1063 | continue; |
a6b0d5c8 | 1064 | } |
2b82032c | 1065 | |
cf89af14 AJ |
1066 | /* |
1067 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, | |
1068 | * in btrfs_init_dev_replace() so just continue. | |
1069 | */ | |
1070 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) | |
1071 | continue; | |
1072 | ||
2b82032c | 1073 | if (device->bdev) { |
d4d77629 | 1074 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1075 | device->bdev = NULL; |
1076 | fs_devices->open_devices--; | |
1077 | } | |
ebbede42 | 1078 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1079 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1080 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 1081 | } |
e4404d6e YZ |
1082 | list_del_init(&device->dev_list); |
1083 | fs_devices->num_devices--; | |
a425f9d4 | 1084 | btrfs_free_device(device); |
dfe25020 | 1085 | } |
2b82032c | 1086 | |
3712ccb7 NB |
1087 | } |
1088 | ||
1089 | /* | |
1090 | * After we have read the system tree and know devids belonging to this | |
1091 | * filesystem, remove the device which does not belong there. | |
1092 | */ | |
bacce86a | 1093 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
3712ccb7 NB |
1094 | { |
1095 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1096 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1097 | |
1098 | mutex_lock(&uuid_mutex); | |
bacce86a | 1099 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
944d3f9f NB |
1100 | |
1101 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
bacce86a | 1102 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
2b82032c | 1103 | |
443f24fe | 1104 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1105 | |
dfe25020 | 1106 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1107 | } |
a0af469b | 1108 | |
14238819 AJ |
1109 | static void btrfs_close_bdev(struct btrfs_device *device) |
1110 | { | |
08ffcae8 DS |
1111 | if (!device->bdev) |
1112 | return; | |
1113 | ||
ebbede42 | 1114 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1115 | sync_blockdev(device->bdev); |
1116 | invalidate_bdev(device->bdev); | |
1117 | } | |
1118 | ||
08ffcae8 | 1119 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1120 | } |
1121 | ||
959b1c04 | 1122 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1123 | { |
1124 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1125 | |
ebbede42 | 1126 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1127 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1128 | list_del_init(&device->dev_alloc_list); | |
1129 | fs_devices->rw_devices--; | |
1130 | } | |
1131 | ||
e6e674bd | 1132 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1133 | fs_devices->missing_devices--; |
1134 | ||
959b1c04 | 1135 | btrfs_close_bdev(device); |
321f69f8 | 1136 | if (device->bdev) { |
3fff3975 | 1137 | fs_devices->open_devices--; |
321f69f8 | 1138 | device->bdev = NULL; |
f448341a | 1139 | } |
321f69f8 | 1140 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
5b316468 | 1141 | btrfs_destroy_dev_zone_info(device); |
f448341a | 1142 | |
321f69f8 JT |
1143 | device->fs_info = NULL; |
1144 | atomic_set(&device->dev_stats_ccnt, 0); | |
1145 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1146 | |
321f69f8 JT |
1147 | /* Verify the device is back in a pristine state */ |
1148 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1149 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1150 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1151 | ASSERT(list_empty(&device->post_commit_list)); | |
1152 | ASSERT(atomic_read(&device->reada_in_flight) == 0); | |
f448341a AJ |
1153 | } |
1154 | ||
54eed6ae | 1155 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1156 | { |
2037a093 | 1157 | struct btrfs_device *device, *tmp; |
e4404d6e | 1158 | |
425c6ed6 JB |
1159 | lockdep_assert_held(&uuid_mutex); |
1160 | ||
2b82032c | 1161 | if (--fs_devices->opened > 0) |
54eed6ae | 1162 | return; |
8a4b83cc | 1163 | |
425c6ed6 | 1164 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1165 | btrfs_close_one_device(device); |
c9513edb | 1166 | |
e4404d6e YZ |
1167 | WARN_ON(fs_devices->open_devices); |
1168 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1169 | fs_devices->opened = 0; |
0395d84f | 1170 | fs_devices->seeding = false; |
c4989c2f | 1171 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1172 | } |
1173 | ||
54eed6ae | 1174 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1175 | { |
944d3f9f NB |
1176 | LIST_HEAD(list); |
1177 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1178 | |
1179 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1180 | close_fs_devices(fs_devices); |
944d3f9f NB |
1181 | if (!fs_devices->opened) |
1182 | list_splice_init(&fs_devices->seed_list, &list); | |
e4404d6e | 1183 | |
944d3f9f | 1184 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
0226e0eb | 1185 | close_fs_devices(fs_devices); |
944d3f9f | 1186 | list_del(&fs_devices->seed_list); |
e4404d6e YZ |
1187 | free_fs_devices(fs_devices); |
1188 | } | |
425c6ed6 | 1189 | mutex_unlock(&uuid_mutex); |
2b82032c YZ |
1190 | } |
1191 | ||
897fb573 | 1192 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1193 | fmode_t flags, void *holder) |
8a4b83cc | 1194 | { |
8a4b83cc | 1195 | struct btrfs_device *device; |
443f24fe | 1196 | struct btrfs_device *latest_dev = NULL; |
96c2e067 | 1197 | struct btrfs_device *tmp_device; |
8a4b83cc | 1198 | |
d4d77629 TH |
1199 | flags |= FMODE_EXCL; |
1200 | ||
96c2e067 AJ |
1201 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
1202 | dev_list) { | |
1203 | int ret; | |
a0af469b | 1204 | |
96c2e067 AJ |
1205 | ret = btrfs_open_one_device(fs_devices, device, flags, holder); |
1206 | if (ret == 0 && | |
1207 | (!latest_dev || device->generation > latest_dev->generation)) { | |
9f050db4 | 1208 | latest_dev = device; |
96c2e067 AJ |
1209 | } else if (ret == -ENODATA) { |
1210 | fs_devices->num_devices--; | |
1211 | list_del(&device->dev_list); | |
1212 | btrfs_free_device(device); | |
1213 | } | |
8a4b83cc | 1214 | } |
1ed802c9 AJ |
1215 | if (fs_devices->open_devices == 0) |
1216 | return -EINVAL; | |
1217 | ||
2b82032c | 1218 | fs_devices->opened = 1; |
443f24fe | 1219 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1220 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1221 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
33fd2f71 | 1222 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
1ed802c9 AJ |
1223 | |
1224 | return 0; | |
2b82032c YZ |
1225 | } |
1226 | ||
f8e10cd3 AJ |
1227 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1228 | { | |
1229 | struct btrfs_device *dev1, *dev2; | |
1230 | ||
1231 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1232 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1233 | ||
1234 | if (dev1->devid < dev2->devid) | |
1235 | return -1; | |
1236 | else if (dev1->devid > dev2->devid) | |
1237 | return 1; | |
1238 | return 0; | |
1239 | } | |
1240 | ||
2b82032c | 1241 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1242 | fmode_t flags, void *holder) |
2b82032c YZ |
1243 | { |
1244 | int ret; | |
1245 | ||
f5194e34 | 1246 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1247 | /* |
1248 | * The device_list_mutex cannot be taken here in case opening the | |
1249 | * underlying device takes further locks like bd_mutex. | |
1250 | * | |
1251 | * We also don't need the lock here as this is called during mount and | |
1252 | * exclusion is provided by uuid_mutex | |
1253 | */ | |
f5194e34 | 1254 | |
2b82032c | 1255 | if (fs_devices->opened) { |
e4404d6e YZ |
1256 | fs_devices->opened++; |
1257 | ret = 0; | |
2b82032c | 1258 | } else { |
f8e10cd3 | 1259 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1260 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1261 | } |
542c5908 | 1262 | |
8a4b83cc CM |
1263 | return ret; |
1264 | } | |
1265 | ||
8f32380d | 1266 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1267 | { |
8f32380d JT |
1268 | struct page *page = virt_to_page(super); |
1269 | ||
6cf86a00 AJ |
1270 | put_page(page); |
1271 | } | |
1272 | ||
b335eab8 | 1273 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
12659251 | 1274 | u64 bytenr, u64 bytenr_orig) |
6cf86a00 | 1275 | { |
b335eab8 NB |
1276 | struct btrfs_super_block *disk_super; |
1277 | struct page *page; | |
6cf86a00 AJ |
1278 | void *p; |
1279 | pgoff_t index; | |
1280 | ||
1281 | /* make sure our super fits in the device */ | |
1282 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
b335eab8 | 1283 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1284 | |
1285 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1286 | if (sizeof(*disk_super) > PAGE_SIZE) |
1287 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1288 | |
1289 | /* make sure our super doesn't straddle pages on disk */ | |
1290 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1291 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1292 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1293 | |
1294 | /* pull in the page with our super */ | |
b335eab8 | 1295 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1296 | |
b335eab8 NB |
1297 | if (IS_ERR(page)) |
1298 | return ERR_CAST(page); | |
6cf86a00 | 1299 | |
b335eab8 | 1300 | p = page_address(page); |
6cf86a00 AJ |
1301 | |
1302 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1303 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1304 | |
12659251 | 1305 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
b335eab8 | 1306 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
8f32380d | 1307 | btrfs_release_disk_super(p); |
b335eab8 | 1308 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1309 | } |
1310 | ||
b335eab8 NB |
1311 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1312 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1313 | |
b335eab8 | 1314 | return disk_super; |
6cf86a00 AJ |
1315 | } |
1316 | ||
228a73ab AJ |
1317 | int btrfs_forget_devices(const char *path) |
1318 | { | |
1319 | int ret; | |
1320 | ||
1321 | mutex_lock(&uuid_mutex); | |
1322 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1323 | mutex_unlock(&uuid_mutex); | |
1324 | ||
1325 | return ret; | |
1326 | } | |
1327 | ||
6f60cbd3 DS |
1328 | /* |
1329 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1330 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1331 | * is read via pagecache | |
1332 | */ | |
36350e95 GJ |
1333 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1334 | void *holder) | |
8a4b83cc CM |
1335 | { |
1336 | struct btrfs_super_block *disk_super; | |
4306a974 | 1337 | bool new_device_added = false; |
36350e95 | 1338 | struct btrfs_device *device = NULL; |
8a4b83cc | 1339 | struct block_device *bdev; |
12659251 NA |
1340 | u64 bytenr, bytenr_orig; |
1341 | int ret; | |
8a4b83cc | 1342 | |
899f9307 DS |
1343 | lockdep_assert_held(&uuid_mutex); |
1344 | ||
6f60cbd3 DS |
1345 | /* |
1346 | * we would like to check all the supers, but that would make | |
1347 | * a btrfs mount succeed after a mkfs from a different FS. | |
1348 | * So, we need to add a special mount option to scan for | |
1349 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1350 | */ | |
d4d77629 | 1351 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1352 | |
1353 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1354 | if (IS_ERR(bdev)) |
36350e95 | 1355 | return ERR_CAST(bdev); |
6f60cbd3 | 1356 | |
12659251 NA |
1357 | bytenr_orig = btrfs_sb_offset(0); |
1358 | ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); | |
1359 | if (ret) | |
1360 | return ERR_PTR(ret); | |
1361 | ||
1362 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); | |
b335eab8 NB |
1363 | if (IS_ERR(disk_super)) { |
1364 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1365 | goto error_bdev_put; |
05a5c55d | 1366 | } |
6f60cbd3 | 1367 | |
4306a974 | 1368 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1369 | if (!IS_ERR(device)) { |
4306a974 AJ |
1370 | if (new_device_added) |
1371 | btrfs_free_stale_devices(path, device); | |
1372 | } | |
6f60cbd3 | 1373 | |
8f32380d | 1374 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1375 | |
1376 | error_bdev_put: | |
d4d77629 | 1377 | blkdev_put(bdev, flags); |
b6ed73bc | 1378 | |
36350e95 | 1379 | return device; |
8a4b83cc | 1380 | } |
0b86a832 | 1381 | |
1c11b63e JM |
1382 | /* |
1383 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1384 | * such is found, record the end of it in *start | |
1385 | */ | |
1c11b63e JM |
1386 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1387 | u64 len) | |
6df9a95e | 1388 | { |
1c11b63e | 1389 | u64 physical_start, physical_end; |
6df9a95e | 1390 | |
1c11b63e | 1391 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1392 | |
1c11b63e JM |
1393 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1394 | &physical_start, &physical_end, | |
1395 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1396 | |
1c11b63e JM |
1397 | if (in_range(physical_start, *start, len) || |
1398 | in_range(*start, physical_start, | |
1399 | physical_end - physical_start)) { | |
1400 | *start = physical_end + 1; | |
1401 | return true; | |
6df9a95e JB |
1402 | } |
1403 | } | |
1c11b63e | 1404 | return false; |
6df9a95e JB |
1405 | } |
1406 | ||
3b4ffa40 NA |
1407 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1408 | { | |
1409 | switch (device->fs_devices->chunk_alloc_policy) { | |
1410 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1411 | /* | |
1412 | * We don't want to overwrite the superblock on the drive nor | |
1413 | * any area used by the boot loader (grub for example), so we | |
1414 | * make sure to start at an offset of at least 1MB. | |
1415 | */ | |
1416 | return max_t(u64, start, SZ_1M); | |
1cd6121f NA |
1417 | case BTRFS_CHUNK_ALLOC_ZONED: |
1418 | /* | |
1419 | * We don't care about the starting region like regular | |
1420 | * allocator, because we anyway use/reserve the first two zones | |
1421 | * for superblock logging. | |
1422 | */ | |
1423 | return ALIGN(start, device->zone_info->zone_size); | |
3b4ffa40 NA |
1424 | default: |
1425 | BUG(); | |
1426 | } | |
1427 | } | |
1428 | ||
1cd6121f NA |
1429 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
1430 | u64 *hole_start, u64 *hole_size, | |
1431 | u64 num_bytes) | |
1432 | { | |
1433 | u64 zone_size = device->zone_info->zone_size; | |
1434 | u64 pos; | |
1435 | int ret; | |
1436 | bool changed = false; | |
1437 | ||
1438 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); | |
1439 | ||
1440 | while (*hole_size > 0) { | |
1441 | pos = btrfs_find_allocatable_zones(device, *hole_start, | |
1442 | *hole_start + *hole_size, | |
1443 | num_bytes); | |
1444 | if (pos != *hole_start) { | |
1445 | *hole_size = *hole_start + *hole_size - pos; | |
1446 | *hole_start = pos; | |
1447 | changed = true; | |
1448 | if (*hole_size < num_bytes) | |
1449 | break; | |
1450 | } | |
1451 | ||
1452 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); | |
1453 | ||
1454 | /* Range is ensured to be empty */ | |
1455 | if (!ret) | |
1456 | return changed; | |
1457 | ||
1458 | /* Given hole range was invalid (outside of device) */ | |
1459 | if (ret == -ERANGE) { | |
1460 | *hole_start += *hole_size; | |
1461 | *hole_size = 0; | |
1462 | return 1; | |
1463 | } | |
1464 | ||
1465 | *hole_start += zone_size; | |
1466 | *hole_size -= zone_size; | |
1467 | changed = true; | |
1468 | } | |
1469 | ||
1470 | return changed; | |
1471 | } | |
1472 | ||
3b4ffa40 NA |
1473 | /** |
1474 | * dev_extent_hole_check - check if specified hole is suitable for allocation | |
1475 | * @device: the device which we have the hole | |
1476 | * @hole_start: starting position of the hole | |
1477 | * @hole_size: the size of the hole | |
1478 | * @num_bytes: the size of the free space that we need | |
1479 | * | |
1cd6121f | 1480 | * This function may modify @hole_start and @hole_size to reflect the suitable |
3b4ffa40 NA |
1481 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
1482 | */ | |
1483 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1484 | u64 *hole_size, u64 num_bytes) | |
1485 | { | |
1486 | bool changed = false; | |
1487 | u64 hole_end = *hole_start + *hole_size; | |
1488 | ||
1cd6121f NA |
1489 | for (;;) { |
1490 | /* | |
1491 | * Check before we set max_hole_start, otherwise we could end up | |
1492 | * sending back this offset anyway. | |
1493 | */ | |
1494 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1495 | if (hole_end >= *hole_start) | |
1496 | *hole_size = hole_end - *hole_start; | |
1497 | else | |
1498 | *hole_size = 0; | |
1499 | changed = true; | |
1500 | } | |
1501 | ||
1502 | switch (device->fs_devices->chunk_alloc_policy) { | |
1503 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1504 | /* No extra check */ | |
1505 | break; | |
1506 | case BTRFS_CHUNK_ALLOC_ZONED: | |
1507 | if (dev_extent_hole_check_zoned(device, hole_start, | |
1508 | hole_size, num_bytes)) { | |
1509 | changed = true; | |
1510 | /* | |
1511 | * The changed hole can contain pending extent. | |
1512 | * Loop again to check that. | |
1513 | */ | |
1514 | continue; | |
1515 | } | |
1516 | break; | |
1517 | default: | |
1518 | BUG(); | |
1519 | } | |
3b4ffa40 | 1520 | |
3b4ffa40 | 1521 | break; |
3b4ffa40 NA |
1522 | } |
1523 | ||
1524 | return changed; | |
1525 | } | |
6df9a95e | 1526 | |
0b86a832 | 1527 | /* |
499f377f JM |
1528 | * find_free_dev_extent_start - find free space in the specified device |
1529 | * @device: the device which we search the free space in | |
1530 | * @num_bytes: the size of the free space that we need | |
1531 | * @search_start: the position from which to begin the search | |
1532 | * @start: store the start of the free space. | |
1533 | * @len: the size of the free space. that we find, or the size | |
1534 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1535 | * |
0b86a832 CM |
1536 | * this uses a pretty simple search, the expectation is that it is |
1537 | * called very infrequently and that a given device has a small number | |
1538 | * of extents | |
7bfc837d MX |
1539 | * |
1540 | * @start is used to store the start of the free space if we find. But if we | |
1541 | * don't find suitable free space, it will be used to store the start position | |
1542 | * of the max free space. | |
1543 | * | |
1544 | * @len is used to store the size of the free space that we find. | |
1545 | * But if we don't find suitable free space, it is used to store the size of | |
1546 | * the max free space. | |
135da976 QW |
1547 | * |
1548 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1549 | * check to ensure dev extents are not double allocated. | |
1550 | * This makes the function safe to allocate dev extents but may not report | |
1551 | * correct usable device space, as device extent freed in current transaction | |
1552 | * is not reported as avaiable. | |
0b86a832 | 1553 | */ |
9e3246a5 QW |
1554 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1555 | u64 num_bytes, u64 search_start, u64 *start, | |
1556 | u64 *len) | |
0b86a832 | 1557 | { |
0b246afa JM |
1558 | struct btrfs_fs_info *fs_info = device->fs_info; |
1559 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1560 | struct btrfs_key key; |
7bfc837d | 1561 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1562 | struct btrfs_path *path; |
7bfc837d MX |
1563 | u64 hole_size; |
1564 | u64 max_hole_start; | |
1565 | u64 max_hole_size; | |
1566 | u64 extent_end; | |
0b86a832 CM |
1567 | u64 search_end = device->total_bytes; |
1568 | int ret; | |
7bfc837d | 1569 | int slot; |
0b86a832 | 1570 | struct extent_buffer *l; |
8cdc7c5b | 1571 | |
3b4ffa40 | 1572 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1573 | |
1cd6121f NA |
1574 | WARN_ON(device->zone_info && |
1575 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); | |
1576 | ||
6df9a95e JB |
1577 | path = btrfs_alloc_path(); |
1578 | if (!path) | |
1579 | return -ENOMEM; | |
f2ab7618 | 1580 | |
7bfc837d MX |
1581 | max_hole_start = search_start; |
1582 | max_hole_size = 0; | |
1583 | ||
f2ab7618 | 1584 | again: |
401e29c1 AJ |
1585 | if (search_start >= search_end || |
1586 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1587 | ret = -ENOSPC; |
6df9a95e | 1588 | goto out; |
7bfc837d MX |
1589 | } |
1590 | ||
e4058b54 | 1591 | path->reada = READA_FORWARD; |
6df9a95e JB |
1592 | path->search_commit_root = 1; |
1593 | path->skip_locking = 1; | |
7bfc837d | 1594 | |
0b86a832 CM |
1595 | key.objectid = device->devid; |
1596 | key.offset = search_start; | |
1597 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1598 | |
125ccb0a | 1599 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1600 | if (ret < 0) |
7bfc837d | 1601 | goto out; |
1fcbac58 YZ |
1602 | if (ret > 0) { |
1603 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1604 | if (ret < 0) | |
7bfc837d | 1605 | goto out; |
1fcbac58 | 1606 | } |
7bfc837d | 1607 | |
0b86a832 CM |
1608 | while (1) { |
1609 | l = path->nodes[0]; | |
1610 | slot = path->slots[0]; | |
1611 | if (slot >= btrfs_header_nritems(l)) { | |
1612 | ret = btrfs_next_leaf(root, path); | |
1613 | if (ret == 0) | |
1614 | continue; | |
1615 | if (ret < 0) | |
7bfc837d MX |
1616 | goto out; |
1617 | ||
1618 | break; | |
0b86a832 CM |
1619 | } |
1620 | btrfs_item_key_to_cpu(l, &key, slot); | |
1621 | ||
1622 | if (key.objectid < device->devid) | |
1623 | goto next; | |
1624 | ||
1625 | if (key.objectid > device->devid) | |
7bfc837d | 1626 | break; |
0b86a832 | 1627 | |
962a298f | 1628 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1629 | goto next; |
9779b72f | 1630 | |
7bfc837d MX |
1631 | if (key.offset > search_start) { |
1632 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1633 | dev_extent_hole_check(device, &search_start, &hole_size, |
1634 | num_bytes); | |
6df9a95e | 1635 | |
7bfc837d MX |
1636 | if (hole_size > max_hole_size) { |
1637 | max_hole_start = search_start; | |
1638 | max_hole_size = hole_size; | |
1639 | } | |
9779b72f | 1640 | |
7bfc837d MX |
1641 | /* |
1642 | * If this free space is greater than which we need, | |
1643 | * it must be the max free space that we have found | |
1644 | * until now, so max_hole_start must point to the start | |
1645 | * of this free space and the length of this free space | |
1646 | * is stored in max_hole_size. Thus, we return | |
1647 | * max_hole_start and max_hole_size and go back to the | |
1648 | * caller. | |
1649 | */ | |
1650 | if (hole_size >= num_bytes) { | |
1651 | ret = 0; | |
1652 | goto out; | |
0b86a832 CM |
1653 | } |
1654 | } | |
0b86a832 | 1655 | |
0b86a832 | 1656 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1657 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1658 | dev_extent); | |
1659 | if (extent_end > search_start) | |
1660 | search_start = extent_end; | |
0b86a832 CM |
1661 | next: |
1662 | path->slots[0]++; | |
1663 | cond_resched(); | |
1664 | } | |
0b86a832 | 1665 | |
38c01b96 | 1666 | /* |
1667 | * At this point, search_start should be the end of | |
1668 | * allocated dev extents, and when shrinking the device, | |
1669 | * search_end may be smaller than search_start. | |
1670 | */ | |
f2ab7618 | 1671 | if (search_end > search_start) { |
38c01b96 | 1672 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1673 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1674 | num_bytes)) { | |
f2ab7618 ZL |
1675 | btrfs_release_path(path); |
1676 | goto again; | |
1677 | } | |
0b86a832 | 1678 | |
f2ab7618 ZL |
1679 | if (hole_size > max_hole_size) { |
1680 | max_hole_start = search_start; | |
1681 | max_hole_size = hole_size; | |
1682 | } | |
6df9a95e JB |
1683 | } |
1684 | ||
7bfc837d | 1685 | /* See above. */ |
f2ab7618 | 1686 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1687 | ret = -ENOSPC; |
1688 | else | |
1689 | ret = 0; | |
1690 | ||
1691 | out: | |
2b82032c | 1692 | btrfs_free_path(path); |
7bfc837d | 1693 | *start = max_hole_start; |
b2117a39 | 1694 | if (len) |
7bfc837d | 1695 | *len = max_hole_size; |
0b86a832 CM |
1696 | return ret; |
1697 | } | |
1698 | ||
60dfdf25 | 1699 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1700 | u64 *start, u64 *len) |
1701 | { | |
499f377f | 1702 | /* FIXME use last free of some kind */ |
60dfdf25 | 1703 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1704 | } |
1705 | ||
b2950863 | 1706 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1707 | struct btrfs_device *device, |
2196d6e8 | 1708 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1709 | { |
0b246afa JM |
1710 | struct btrfs_fs_info *fs_info = device->fs_info; |
1711 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1712 | int ret; |
1713 | struct btrfs_path *path; | |
8f18cf13 | 1714 | struct btrfs_key key; |
a061fc8d CM |
1715 | struct btrfs_key found_key; |
1716 | struct extent_buffer *leaf = NULL; | |
1717 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1718 | |
1719 | path = btrfs_alloc_path(); | |
1720 | if (!path) | |
1721 | return -ENOMEM; | |
1722 | ||
1723 | key.objectid = device->devid; | |
1724 | key.offset = start; | |
1725 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1726 | again: |
8f18cf13 | 1727 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1728 | if (ret > 0) { |
1729 | ret = btrfs_previous_item(root, path, key.objectid, | |
1730 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1731 | if (ret) |
1732 | goto out; | |
a061fc8d CM |
1733 | leaf = path->nodes[0]; |
1734 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1735 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1736 | struct btrfs_dev_extent); | |
1737 | BUG_ON(found_key.offset > start || found_key.offset + | |
1738 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1739 | key = found_key; |
1740 | btrfs_release_path(path); | |
1741 | goto again; | |
a061fc8d CM |
1742 | } else if (ret == 0) { |
1743 | leaf = path->nodes[0]; | |
1744 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1745 | struct btrfs_dev_extent); | |
79787eaa | 1746 | } else { |
0b246afa | 1747 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1748 | goto out; |
a061fc8d | 1749 | } |
8f18cf13 | 1750 | |
2196d6e8 MX |
1751 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1752 | ||
8f18cf13 | 1753 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1754 | if (ret) { |
0b246afa JM |
1755 | btrfs_handle_fs_error(fs_info, ret, |
1756 | "Failed to remove dev extent item"); | |
13212b54 | 1757 | } else { |
3204d33c | 1758 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1759 | } |
b0b802d7 | 1760 | out: |
8f18cf13 CM |
1761 | btrfs_free_path(path); |
1762 | return ret; | |
1763 | } | |
1764 | ||
48a3b636 ES |
1765 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1766 | struct btrfs_device *device, | |
48a3b636 | 1767 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1768 | { |
1769 | int ret; | |
1770 | struct btrfs_path *path; | |
0b246afa JM |
1771 | struct btrfs_fs_info *fs_info = device->fs_info; |
1772 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1773 | struct btrfs_dev_extent *extent; |
1774 | struct extent_buffer *leaf; | |
1775 | struct btrfs_key key; | |
1776 | ||
e12c9621 | 1777 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1778 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1779 | path = btrfs_alloc_path(); |
1780 | if (!path) | |
1781 | return -ENOMEM; | |
1782 | ||
0b86a832 | 1783 | key.objectid = device->devid; |
2b82032c | 1784 | key.offset = start; |
0b86a832 CM |
1785 | key.type = BTRFS_DEV_EXTENT_KEY; |
1786 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1787 | sizeof(*extent)); | |
2cdcecbc MF |
1788 | if (ret) |
1789 | goto out; | |
0b86a832 CM |
1790 | |
1791 | leaf = path->nodes[0]; | |
1792 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1793 | struct btrfs_dev_extent); | |
b5d9071c NB |
1794 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1795 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1796 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1797 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1798 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1799 | ||
0b86a832 CM |
1800 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1801 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1802 | out: |
0b86a832 CM |
1803 | btrfs_free_path(path); |
1804 | return ret; | |
1805 | } | |
1806 | ||
6df9a95e | 1807 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1808 | { |
6df9a95e JB |
1809 | struct extent_map_tree *em_tree; |
1810 | struct extent_map *em; | |
1811 | struct rb_node *n; | |
1812 | u64 ret = 0; | |
0b86a832 | 1813 | |
c8bf1b67 | 1814 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1815 | read_lock(&em_tree->lock); |
07e1ce09 | 1816 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1817 | if (n) { |
1818 | em = rb_entry(n, struct extent_map, rb_node); | |
1819 | ret = em->start + em->len; | |
0b86a832 | 1820 | } |
6df9a95e JB |
1821 | read_unlock(&em_tree->lock); |
1822 | ||
0b86a832 CM |
1823 | return ret; |
1824 | } | |
1825 | ||
53f10659 ID |
1826 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1827 | u64 *devid_ret) | |
0b86a832 CM |
1828 | { |
1829 | int ret; | |
1830 | struct btrfs_key key; | |
1831 | struct btrfs_key found_key; | |
2b82032c YZ |
1832 | struct btrfs_path *path; |
1833 | ||
2b82032c YZ |
1834 | path = btrfs_alloc_path(); |
1835 | if (!path) | |
1836 | return -ENOMEM; | |
0b86a832 CM |
1837 | |
1838 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1839 | key.type = BTRFS_DEV_ITEM_KEY; | |
1840 | key.offset = (u64)-1; | |
1841 | ||
53f10659 | 1842 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1843 | if (ret < 0) |
1844 | goto error; | |
1845 | ||
a06dee4d AJ |
1846 | if (ret == 0) { |
1847 | /* Corruption */ | |
1848 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1849 | ret = -EUCLEAN; | |
1850 | goto error; | |
1851 | } | |
0b86a832 | 1852 | |
53f10659 ID |
1853 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1854 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1855 | BTRFS_DEV_ITEM_KEY); |
1856 | if (ret) { | |
53f10659 | 1857 | *devid_ret = 1; |
0b86a832 CM |
1858 | } else { |
1859 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1860 | path->slots[0]); | |
53f10659 | 1861 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1862 | } |
1863 | ret = 0; | |
1864 | error: | |
2b82032c | 1865 | btrfs_free_path(path); |
0b86a832 CM |
1866 | return ret; |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * the device information is stored in the chunk root | |
1871 | * the btrfs_device struct should be fully filled in | |
1872 | */ | |
c74a0b02 | 1873 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1874 | struct btrfs_device *device) |
0b86a832 CM |
1875 | { |
1876 | int ret; | |
1877 | struct btrfs_path *path; | |
1878 | struct btrfs_dev_item *dev_item; | |
1879 | struct extent_buffer *leaf; | |
1880 | struct btrfs_key key; | |
1881 | unsigned long ptr; | |
0b86a832 | 1882 | |
0b86a832 CM |
1883 | path = btrfs_alloc_path(); |
1884 | if (!path) | |
1885 | return -ENOMEM; | |
1886 | ||
0b86a832 CM |
1887 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1888 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1889 | key.offset = device->devid; |
0b86a832 | 1890 | |
8e87e856 NB |
1891 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1892 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1893 | if (ret) |
1894 | goto out; | |
1895 | ||
1896 | leaf = path->nodes[0]; | |
1897 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1898 | ||
1899 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1900 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1901 | btrfs_set_device_type(leaf, dev_item, device->type); |
1902 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1903 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1904 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1905 | btrfs_set_device_total_bytes(leaf, dev_item, |
1906 | btrfs_device_get_disk_total_bytes(device)); | |
1907 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1908 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1909 | btrfs_set_device_group(leaf, dev_item, 0); |
1910 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1911 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1912 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1913 | |
410ba3a2 | 1914 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1915 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1916 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1917 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1918 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1919 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1920 | |
2b82032c | 1921 | ret = 0; |
0b86a832 CM |
1922 | out: |
1923 | btrfs_free_path(path); | |
1924 | return ret; | |
1925 | } | |
8f18cf13 | 1926 | |
5a1972bd QW |
1927 | /* |
1928 | * Function to update ctime/mtime for a given device path. | |
1929 | * Mainly used for ctime/mtime based probe like libblkid. | |
1930 | */ | |
da353f6b | 1931 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1932 | { |
1933 | struct file *filp; | |
1934 | ||
1935 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1936 | if (IS_ERR(filp)) |
5a1972bd QW |
1937 | return; |
1938 | file_update_time(filp); | |
1939 | filp_close(filp, NULL); | |
5a1972bd QW |
1940 | } |
1941 | ||
f331a952 | 1942 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1943 | { |
f331a952 | 1944 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1945 | int ret; |
1946 | struct btrfs_path *path; | |
a061fc8d | 1947 | struct btrfs_key key; |
a061fc8d CM |
1948 | struct btrfs_trans_handle *trans; |
1949 | ||
a061fc8d CM |
1950 | path = btrfs_alloc_path(); |
1951 | if (!path) | |
1952 | return -ENOMEM; | |
1953 | ||
a22285a6 | 1954 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1955 | if (IS_ERR(trans)) { |
1956 | btrfs_free_path(path); | |
1957 | return PTR_ERR(trans); | |
1958 | } | |
a061fc8d CM |
1959 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1960 | key.type = BTRFS_DEV_ITEM_KEY; | |
1961 | key.offset = device->devid; | |
1962 | ||
1963 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1964 | if (ret) { |
1965 | if (ret > 0) | |
1966 | ret = -ENOENT; | |
1967 | btrfs_abort_transaction(trans, ret); | |
1968 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1969 | goto out; |
1970 | } | |
1971 | ||
1972 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1973 | if (ret) { |
1974 | btrfs_abort_transaction(trans, ret); | |
1975 | btrfs_end_transaction(trans); | |
1976 | } | |
1977 | ||
a061fc8d CM |
1978 | out: |
1979 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1980 | if (!ret) |
1981 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1982 | return ret; |
1983 | } | |
1984 | ||
3cc31a0d DS |
1985 | /* |
1986 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1987 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1988 | * replace. | |
1989 | */ | |
1990 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1991 | u64 num_devices) | |
a061fc8d | 1992 | { |
a061fc8d | 1993 | u64 all_avail; |
de98ced9 | 1994 | unsigned seq; |
418775a2 | 1995 | int i; |
a061fc8d | 1996 | |
de98ced9 | 1997 | do { |
bd45ffbc | 1998 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1999 | |
bd45ffbc AJ |
2000 | all_avail = fs_info->avail_data_alloc_bits | |
2001 | fs_info->avail_system_alloc_bits | | |
2002 | fs_info->avail_metadata_alloc_bits; | |
2003 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2004 | |
418775a2 | 2005 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2006 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2007 | continue; |
a061fc8d | 2008 | |
418775a2 | 2009 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2010 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2011 | |
418775a2 DS |
2012 | if (ret) |
2013 | return ret; | |
2014 | } | |
53b381b3 DW |
2015 | } |
2016 | ||
bd45ffbc | 2017 | return 0; |
f1fa7f26 AJ |
2018 | } |
2019 | ||
c9162bdf OS |
2020 | static struct btrfs_device * btrfs_find_next_active_device( |
2021 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2022 | { |
2b82032c | 2023 | struct btrfs_device *next_device; |
88acff64 AJ |
2024 | |
2025 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2026 | if (next_device != device && | |
e6e674bd AJ |
2027 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2028 | && next_device->bdev) | |
88acff64 AJ |
2029 | return next_device; |
2030 | } | |
2031 | ||
2032 | return NULL; | |
2033 | } | |
2034 | ||
2035 | /* | |
2036 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2037 | * and replace it with the provided or the next active device, in the context | |
2038 | * where this function called, there should be always be another device (or | |
2039 | * this_dev) which is active. | |
2040 | */ | |
b105e927 | 2041 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
e493e8f9 | 2042 | struct btrfs_device *next_device) |
88acff64 | 2043 | { |
d6507cf1 | 2044 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 | 2045 | |
e493e8f9 | 2046 | if (!next_device) |
88acff64 | 2047 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
e493e8f9 | 2048 | device); |
88acff64 AJ |
2049 | ASSERT(next_device); |
2050 | ||
2051 | if (fs_info->sb->s_bdev && | |
2052 | (fs_info->sb->s_bdev == device->bdev)) | |
2053 | fs_info->sb->s_bdev = next_device->bdev; | |
2054 | ||
2055 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2056 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2057 | } | |
2058 | ||
1da73967 AJ |
2059 | /* |
2060 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2061 | * currently replaced. | |
2062 | */ | |
2063 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2064 | { | |
2065 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2066 | ||
cb5583dd | 2067 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2068 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2069 | ASSERT(num_devices > 1); | |
2070 | num_devices--; | |
2071 | } | |
cb5583dd | 2072 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2073 | |
2074 | return num_devices; | |
2075 | } | |
2076 | ||
313b0858 JB |
2077 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2078 | struct block_device *bdev, | |
2079 | const char *device_path) | |
6fbceb9f | 2080 | { |
6fbceb9f JT |
2081 | struct btrfs_super_block *disk_super; |
2082 | int copy_num; | |
2083 | ||
2084 | if (!bdev) | |
2085 | return; | |
2086 | ||
2087 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
8f32380d JT |
2088 | struct page *page; |
2089 | int ret; | |
6fbceb9f | 2090 | |
8f32380d JT |
2091 | disk_super = btrfs_read_dev_one_super(bdev, copy_num); |
2092 | if (IS_ERR(disk_super)) | |
2093 | continue; | |
6fbceb9f | 2094 | |
12659251 NA |
2095 | if (bdev_is_zoned(bdev)) { |
2096 | btrfs_reset_sb_log_zones(bdev, copy_num); | |
2097 | continue; | |
2098 | } | |
2099 | ||
6fbceb9f | 2100 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); |
8f32380d JT |
2101 | |
2102 | page = virt_to_page(disk_super); | |
2103 | set_page_dirty(page); | |
2104 | lock_page(page); | |
2105 | /* write_on_page() unlocks the page */ | |
2106 | ret = write_one_page(page); | |
2107 | if (ret) | |
2108 | btrfs_warn(fs_info, | |
2109 | "error clearing superblock number %d (%d)", | |
2110 | copy_num, ret); | |
2111 | btrfs_release_disk_super(disk_super); | |
2112 | ||
6fbceb9f JT |
2113 | } |
2114 | ||
2115 | /* Notify udev that device has changed */ | |
2116 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2117 | ||
2118 | /* Update ctime/mtime for device path for libblkid */ | |
2119 | update_dev_time(device_path); | |
2120 | } | |
2121 | ||
da353f6b | 2122 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
53f8a74c | 2123 | u64 devid) |
f1fa7f26 AJ |
2124 | { |
2125 | struct btrfs_device *device; | |
1f78160c | 2126 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2127 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2128 | u64 num_devices; |
a061fc8d CM |
2129 | int ret = 0; |
2130 | ||
a061fc8d CM |
2131 | mutex_lock(&uuid_mutex); |
2132 | ||
1da73967 | 2133 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2134 | |
0b246afa | 2135 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2136 | if (ret) |
a061fc8d | 2137 | goto out; |
a061fc8d | 2138 | |
a27a94c2 NB |
2139 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2140 | ||
2141 | if (IS_ERR(device)) { | |
2142 | if (PTR_ERR(device) == -ENOENT && | |
2143 | strcmp(device_path, "missing") == 0) | |
2144 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2145 | else | |
2146 | ret = PTR_ERR(device); | |
53b381b3 | 2147 | goto out; |
a27a94c2 | 2148 | } |
dfe25020 | 2149 | |
eede2bf3 OS |
2150 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2151 | btrfs_warn_in_rcu(fs_info, | |
2152 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2153 | rcu_str_deref(device->name), device->devid); | |
2154 | ret = -ETXTBSY; | |
2155 | goto out; | |
2156 | } | |
2157 | ||
401e29c1 | 2158 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2159 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2160 | goto out; |
63a212ab SB |
2161 | } |
2162 | ||
ebbede42 AJ |
2163 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2164 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2165 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2166 | goto out; |
2b82032c YZ |
2167 | } |
2168 | ||
ebbede42 | 2169 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2170 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2171 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2172 | device->fs_devices->rw_devices--; |
34441361 | 2173 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2174 | } |
a061fc8d | 2175 | |
d7901554 | 2176 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2177 | ret = btrfs_shrink_device(device, 0); |
66d204a1 FM |
2178 | if (!ret) |
2179 | btrfs_reada_remove_dev(device); | |
d7901554 | 2180 | mutex_lock(&uuid_mutex); |
a061fc8d | 2181 | if (ret) |
9b3517e9 | 2182 | goto error_undo; |
a061fc8d | 2183 | |
63a212ab SB |
2184 | /* |
2185 | * TODO: the superblock still includes this device in its num_devices | |
2186 | * counter although write_all_supers() is not locked out. This | |
2187 | * could give a filesystem state which requires a degraded mount. | |
2188 | */ | |
f331a952 | 2189 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2190 | if (ret) |
9b3517e9 | 2191 | goto error_undo; |
a061fc8d | 2192 | |
e12c9621 | 2193 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2194 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2195 | |
2196 | /* | |
2197 | * the device list mutex makes sure that we don't change | |
2198 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2199 | * the device supers. Whoever is writing all supers, should |
2200 | * lock the device list mutex before getting the number of | |
2201 | * devices in the super block (super_copy). Conversely, | |
2202 | * whoever updates the number of devices in the super block | |
2203 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2204 | */ |
1f78160c | 2205 | |
41a52a0f AJ |
2206 | /* |
2207 | * In normal cases the cur_devices == fs_devices. But in case | |
2208 | * of deleting a seed device, the cur_devices should point to | |
2209 | * its own fs_devices listed under the fs_devices->seed. | |
2210 | */ | |
1f78160c | 2211 | cur_devices = device->fs_devices; |
b5185197 | 2212 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2213 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2214 | |
41a52a0f AJ |
2215 | cur_devices->num_devices--; |
2216 | cur_devices->total_devices--; | |
b4993e64 AJ |
2217 | /* Update total_devices of the parent fs_devices if it's seed */ |
2218 | if (cur_devices != fs_devices) | |
2219 | fs_devices->total_devices--; | |
2b82032c | 2220 | |
e6e674bd | 2221 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2222 | cur_devices->missing_devices--; |
cd02dca5 | 2223 | |
d6507cf1 | 2224 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2225 | |
0bfaa9c5 | 2226 | if (device->bdev) { |
41a52a0f | 2227 | cur_devices->open_devices--; |
0bfaa9c5 | 2228 | /* remove sysfs entry */ |
53f8a74c | 2229 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2230 | } |
99994cde | 2231 | |
0b246afa JM |
2232 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2233 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2234 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2235 | |
cea67ab9 JM |
2236 | /* |
2237 | * at this point, the device is zero sized and detached from | |
2238 | * the devices list. All that's left is to zero out the old | |
2239 | * supers and free the device. | |
2240 | */ | |
ebbede42 | 2241 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
8f32380d JT |
2242 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2243 | device->name->str); | |
cea67ab9 JM |
2244 | |
2245 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2246 | synchronize_rcu(); |
2247 | btrfs_free_device(device); | |
cea67ab9 | 2248 | |
1f78160c | 2249 | if (cur_devices->open_devices == 0) { |
944d3f9f | 2250 | list_del_init(&cur_devices->seed_list); |
0226e0eb | 2251 | close_fs_devices(cur_devices); |
1f78160c | 2252 | free_fs_devices(cur_devices); |
2b82032c YZ |
2253 | } |
2254 | ||
a061fc8d CM |
2255 | out: |
2256 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2257 | return ret; |
24fc572f | 2258 | |
9b3517e9 | 2259 | error_undo: |
66d204a1 | 2260 | btrfs_reada_undo_remove_dev(device); |
ebbede42 | 2261 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2262 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2263 | list_add(&device->dev_alloc_list, |
b5185197 | 2264 | &fs_devices->alloc_list); |
c3929c36 | 2265 | device->fs_devices->rw_devices++; |
34441361 | 2266 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2267 | } |
24fc572f | 2268 | goto out; |
a061fc8d CM |
2269 | } |
2270 | ||
68a9db5f | 2271 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2272 | { |
d51908ce AJ |
2273 | struct btrfs_fs_devices *fs_devices; |
2274 | ||
68a9db5f | 2275 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2276 | |
25e8e911 AJ |
2277 | /* |
2278 | * in case of fs with no seed, srcdev->fs_devices will point | |
2279 | * to fs_devices of fs_info. However when the dev being replaced is | |
2280 | * a seed dev it will point to the seed's local fs_devices. In short | |
2281 | * srcdev will have its correct fs_devices in both the cases. | |
2282 | */ | |
2283 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2284 | |
e93c89c1 | 2285 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2286 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2287 | fs_devices->num_devices--; |
e6e674bd | 2288 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2289 | fs_devices->missing_devices--; |
e93c89c1 | 2290 | |
ebbede42 | 2291 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2292 | fs_devices->rw_devices--; |
1357272f | 2293 | |
82372bc8 | 2294 | if (srcdev->bdev) |
d51908ce | 2295 | fs_devices->open_devices--; |
084b6e7c QW |
2296 | } |
2297 | ||
65237ee3 | 2298 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2299 | { |
2300 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2301 | |
a466c85e JB |
2302 | mutex_lock(&uuid_mutex); |
2303 | ||
14238819 | 2304 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2305 | synchronize_rcu(); |
2306 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2307 | |
94d5f0c2 AJ |
2308 | /* if this is no devs we rather delete the fs_devices */ |
2309 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2310 | /* |
2311 | * On a mounted FS, num_devices can't be zero unless it's a | |
2312 | * seed. In case of a seed device being replaced, the replace | |
2313 | * target added to the sprout FS, so there will be no more | |
2314 | * device left under the seed FS. | |
2315 | */ | |
2316 | ASSERT(fs_devices->seeding); | |
2317 | ||
944d3f9f | 2318 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2319 | close_fs_devices(fs_devices); |
8bef8401 | 2320 | free_fs_devices(fs_devices); |
94d5f0c2 | 2321 | } |
a466c85e | 2322 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2323 | } |
2324 | ||
4f5ad7bd | 2325 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2326 | { |
4f5ad7bd | 2327 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2328 | |
d9a071f0 | 2329 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2330 | |
53f8a74c | 2331 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2332 | |
779bf3fe | 2333 | if (tgtdev->bdev) |
d9a071f0 | 2334 | fs_devices->open_devices--; |
779bf3fe | 2335 | |
d9a071f0 | 2336 | fs_devices->num_devices--; |
e93c89c1 | 2337 | |
d6507cf1 | 2338 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2339 | |
e93c89c1 | 2340 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2341 | |
d9a071f0 | 2342 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2343 | |
2344 | /* | |
2345 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2346 | * may lead to a call to btrfs_show_devname() which will try | |
2347 | * to hold device_list_mutex. And here this device | |
2348 | * is already out of device list, so we don't have to hold | |
2349 | * the device_list_mutex lock. | |
2350 | */ | |
8f32380d JT |
2351 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2352 | tgtdev->name->str); | |
14238819 AJ |
2353 | |
2354 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2355 | synchronize_rcu(); |
2356 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2357 | } |
2358 | ||
b444ad46 NB |
2359 | static struct btrfs_device *btrfs_find_device_by_path( |
2360 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2361 | { |
2362 | int ret = 0; | |
2363 | struct btrfs_super_block *disk_super; | |
2364 | u64 devid; | |
2365 | u8 *dev_uuid; | |
2366 | struct block_device *bdev; | |
b444ad46 | 2367 | struct btrfs_device *device; |
7ba15b7d | 2368 | |
7ba15b7d | 2369 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
8f32380d | 2370 | fs_info->bdev_holder, 0, &bdev, &disk_super); |
7ba15b7d | 2371 | if (ret) |
b444ad46 | 2372 | return ERR_PTR(ret); |
8f32380d | 2373 | |
7ba15b7d SB |
2374 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2375 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2376 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2377 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2378 | disk_super->metadata_uuid); |
7239ff4b | 2379 | else |
e4319cd9 | 2380 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2381 | disk_super->fsid); |
7239ff4b | 2382 | |
8f32380d | 2383 | btrfs_release_disk_super(disk_super); |
b444ad46 NB |
2384 | if (!device) |
2385 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2386 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2387 | return device; |
7ba15b7d SB |
2388 | } |
2389 | ||
5c5c0df0 DS |
2390 | /* |
2391 | * Lookup a device given by device id, or the path if the id is 0. | |
2392 | */ | |
a27a94c2 | 2393 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2394 | struct btrfs_fs_info *fs_info, u64 devid, |
2395 | const char *device_path) | |
24e0474b | 2396 | { |
a27a94c2 | 2397 | struct btrfs_device *device; |
24e0474b | 2398 | |
5c5c0df0 | 2399 | if (devid) { |
e4319cd9 | 2400 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
b2598edf | 2401 | NULL); |
a27a94c2 NB |
2402 | if (!device) |
2403 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2404 | return device; |
2405 | } | |
2406 | ||
2407 | if (!device_path || !device_path[0]) | |
2408 | return ERR_PTR(-EINVAL); | |
2409 | ||
2410 | if (strcmp(device_path, "missing") == 0) { | |
2411 | /* Find first missing device */ | |
2412 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2413 | dev_list) { | |
2414 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2415 | &device->dev_state) && !device->bdev) | |
2416 | return device; | |
d95a830c | 2417 | } |
6e927ceb | 2418 | return ERR_PTR(-ENOENT); |
24e0474b | 2419 | } |
6e927ceb AJ |
2420 | |
2421 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2422 | } |
2423 | ||
2b82032c YZ |
2424 | /* |
2425 | * does all the dirty work required for changing file system's UUID. | |
2426 | */ | |
2ff7e61e | 2427 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2428 | { |
0b246afa | 2429 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2430 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2431 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2432 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2433 | struct btrfs_device *device; |
2434 | u64 super_flags; | |
2435 | ||
a32bf9a3 | 2436 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2437 | if (!fs_devices->seeding) |
2b82032c YZ |
2438 | return -EINVAL; |
2439 | ||
427c8fdd NB |
2440 | /* |
2441 | * Private copy of the seed devices, anchored at | |
2442 | * fs_info->fs_devices->seed_list | |
2443 | */ | |
7239ff4b | 2444 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2445 | if (IS_ERR(seed_devices)) |
2446 | return PTR_ERR(seed_devices); | |
2b82032c | 2447 | |
427c8fdd NB |
2448 | /* |
2449 | * It's necessary to retain a copy of the original seed fs_devices in | |
2450 | * fs_uuids so that filesystems which have been seeded can successfully | |
2451 | * reference the seed device from open_seed_devices. This also supports | |
2452 | * multiple fs seed. | |
2453 | */ | |
e4404d6e YZ |
2454 | old_devices = clone_fs_devices(fs_devices); |
2455 | if (IS_ERR(old_devices)) { | |
2456 | kfree(seed_devices); | |
2457 | return PTR_ERR(old_devices); | |
2b82032c | 2458 | } |
e4404d6e | 2459 | |
c4babc5e | 2460 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2461 | |
e4404d6e YZ |
2462 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2463 | seed_devices->opened = 1; | |
2464 | INIT_LIST_HEAD(&seed_devices->devices); | |
2465 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2466 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2467 | |
321a4bf7 | 2468 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2469 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2470 | synchronize_rcu); | |
2196d6e8 MX |
2471 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2472 | device->fs_devices = seed_devices; | |
c9513edb | 2473 | |
0395d84f | 2474 | fs_devices->seeding = false; |
2b82032c YZ |
2475 | fs_devices->num_devices = 0; |
2476 | fs_devices->open_devices = 0; | |
69611ac8 | 2477 | fs_devices->missing_devices = 0; |
7f0432d0 | 2478 | fs_devices->rotating = false; |
944d3f9f | 2479 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2480 | |
2481 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2482 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2483 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2484 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2485 | |
2b82032c YZ |
2486 | super_flags = btrfs_super_flags(disk_super) & |
2487 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2488 | btrfs_set_super_flags(disk_super, super_flags); | |
2489 | ||
2490 | return 0; | |
2491 | } | |
2492 | ||
2493 | /* | |
01327610 | 2494 | * Store the expected generation for seed devices in device items. |
2b82032c | 2495 | */ |
5c466629 | 2496 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2497 | { |
5c466629 | 2498 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2499 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2500 | struct btrfs_path *path; |
2501 | struct extent_buffer *leaf; | |
2502 | struct btrfs_dev_item *dev_item; | |
2503 | struct btrfs_device *device; | |
2504 | struct btrfs_key key; | |
44880fdc | 2505 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2506 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2507 | u64 devid; | |
2508 | int ret; | |
2509 | ||
2510 | path = btrfs_alloc_path(); | |
2511 | if (!path) | |
2512 | return -ENOMEM; | |
2513 | ||
2b82032c YZ |
2514 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2515 | key.offset = 0; | |
2516 | key.type = BTRFS_DEV_ITEM_KEY; | |
2517 | ||
2518 | while (1) { | |
2519 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2520 | if (ret < 0) | |
2521 | goto error; | |
2522 | ||
2523 | leaf = path->nodes[0]; | |
2524 | next_slot: | |
2525 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2526 | ret = btrfs_next_leaf(root, path); | |
2527 | if (ret > 0) | |
2528 | break; | |
2529 | if (ret < 0) | |
2530 | goto error; | |
2531 | leaf = path->nodes[0]; | |
2532 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2533 | btrfs_release_path(path); |
2b82032c YZ |
2534 | continue; |
2535 | } | |
2536 | ||
2537 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2538 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2539 | key.type != BTRFS_DEV_ITEM_KEY) | |
2540 | break; | |
2541 | ||
2542 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2543 | struct btrfs_dev_item); | |
2544 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2545 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2546 | BTRFS_UUID_SIZE); |
1473b24e | 2547 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2548 | BTRFS_FSID_SIZE); |
e4319cd9 | 2549 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2550 | fs_uuid); |
79787eaa | 2551 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2552 | |
2553 | if (device->fs_devices->seeding) { | |
2554 | btrfs_set_device_generation(leaf, dev_item, | |
2555 | device->generation); | |
2556 | btrfs_mark_buffer_dirty(leaf); | |
2557 | } | |
2558 | ||
2559 | path->slots[0]++; | |
2560 | goto next_slot; | |
2561 | } | |
2562 | ret = 0; | |
2563 | error: | |
2564 | btrfs_free_path(path); | |
2565 | return ret; | |
2566 | } | |
2567 | ||
da353f6b | 2568 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2569 | { |
5112febb | 2570 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2571 | struct request_queue *q; |
788f20eb CM |
2572 | struct btrfs_trans_handle *trans; |
2573 | struct btrfs_device *device; | |
2574 | struct block_device *bdev; | |
0b246afa | 2575 | struct super_block *sb = fs_info->sb; |
606686ee | 2576 | struct rcu_string *name; |
5da54bc1 | 2577 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2578 | u64 orig_super_total_bytes; |
2579 | u64 orig_super_num_devices; | |
2b82032c | 2580 | int seeding_dev = 0; |
788f20eb | 2581 | int ret = 0; |
44cab9ba | 2582 | bool locked = false; |
788f20eb | 2583 | |
5da54bc1 | 2584 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2585 | return -EROFS; |
788f20eb | 2586 | |
a5d16333 | 2587 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2588 | fs_info->bdev_holder); |
7f59203a JB |
2589 | if (IS_ERR(bdev)) |
2590 | return PTR_ERR(bdev); | |
a2135011 | 2591 | |
b70f5097 NA |
2592 | if (!btrfs_check_device_zone_type(fs_info, bdev)) { |
2593 | ret = -EINVAL; | |
2594 | goto error; | |
2595 | } | |
2596 | ||
5da54bc1 | 2597 | if (fs_devices->seeding) { |
2b82032c YZ |
2598 | seeding_dev = 1; |
2599 | down_write(&sb->s_umount); | |
2600 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2601 | locked = true; |
2b82032c YZ |
2602 | } |
2603 | ||
b9ba017f | 2604 | sync_blockdev(bdev); |
a2135011 | 2605 | |
f4cfa9bd NB |
2606 | rcu_read_lock(); |
2607 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
788f20eb CM |
2608 | if (device->bdev == bdev) { |
2609 | ret = -EEXIST; | |
f4cfa9bd | 2610 | rcu_read_unlock(); |
2b82032c | 2611 | goto error; |
788f20eb CM |
2612 | } |
2613 | } | |
f4cfa9bd | 2614 | rcu_read_unlock(); |
788f20eb | 2615 | |
0b246afa | 2616 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2617 | if (IS_ERR(device)) { |
788f20eb | 2618 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2619 | ret = PTR_ERR(device); |
2b82032c | 2620 | goto error; |
788f20eb CM |
2621 | } |
2622 | ||
78f2c9e6 | 2623 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2624 | if (!name) { |
2b82032c | 2625 | ret = -ENOMEM; |
5c4cf6c9 | 2626 | goto error_free_device; |
788f20eb | 2627 | } |
606686ee | 2628 | rcu_assign_pointer(device->name, name); |
2b82032c | 2629 | |
5b316468 NA |
2630 | device->fs_info = fs_info; |
2631 | device->bdev = bdev; | |
2632 | ||
2633 | ret = btrfs_get_dev_zone_info(device); | |
2634 | if (ret) | |
2635 | goto error_free_device; | |
2636 | ||
a22285a6 | 2637 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2638 | if (IS_ERR(trans)) { |
98d5dc13 | 2639 | ret = PTR_ERR(trans); |
5b316468 | 2640 | goto error_free_zone; |
98d5dc13 TI |
2641 | } |
2642 | ||
d5e2003c | 2643 | q = bdev_get_queue(bdev); |
ebbede42 | 2644 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2645 | device->generation = trans->transid; |
0b246afa JM |
2646 | device->io_width = fs_info->sectorsize; |
2647 | device->io_align = fs_info->sectorsize; | |
2648 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2649 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2650 | fs_info->sectorsize); | |
2cc3c559 | 2651 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2652 | device->commit_total_bytes = device->total_bytes; |
e12c9621 | 2653 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2654 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2655 | device->mode = FMODE_EXCL; |
27087f37 | 2656 | device->dev_stats_valid = 1; |
9f6d2510 | 2657 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2658 | |
2b82032c | 2659 | if (seeding_dev) { |
a0a1db70 | 2660 | btrfs_clear_sb_rdonly(sb); |
2ff7e61e | 2661 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2662 | if (ret) { |
2663 | btrfs_abort_transaction(trans, ret); | |
2664 | goto error_trans; | |
2665 | } | |
2b82032c | 2666 | } |
788f20eb | 2667 | |
5da54bc1 | 2668 | device->fs_devices = fs_devices; |
e5e9a520 | 2669 | |
5da54bc1 | 2670 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2671 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2672 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2673 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2674 | fs_devices->num_devices++; | |
2675 | fs_devices->open_devices++; | |
2676 | fs_devices->rw_devices++; | |
2677 | fs_devices->total_devices++; | |
2678 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2679 | |
a5ed45f8 | 2680 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2681 | |
e884f4f0 | 2682 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2683 | fs_devices->rotating = true; |
c289811c | 2684 | |
39379faa | 2685 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2686 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2687 | round_down(orig_super_total_bytes + device->total_bytes, |
2688 | fs_info->sectorsize)); | |
788f20eb | 2689 | |
39379faa NA |
2690 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2691 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2692 | orig_super_num_devices + 1); | |
0d39376a | 2693 | |
2196d6e8 MX |
2694 | /* |
2695 | * we've got more storage, clear any full flags on the space | |
2696 | * infos | |
2697 | */ | |
0b246afa | 2698 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2699 | |
34441361 | 2700 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2701 | |
2702 | /* Add sysfs device entry */ | |
cd36da2e | 2703 | btrfs_sysfs_add_device(device); |
ca10845a | 2704 | |
5da54bc1 | 2705 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2706 | |
2b82032c | 2707 | if (seeding_dev) { |
34441361 | 2708 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2709 | ret = init_first_rw_device(trans); |
34441361 | 2710 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2711 | if (ret) { |
66642832 | 2712 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2713 | goto error_sysfs; |
005d6427 | 2714 | } |
2196d6e8 MX |
2715 | } |
2716 | ||
8e87e856 | 2717 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2718 | if (ret) { |
66642832 | 2719 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2720 | goto error_sysfs; |
2196d6e8 MX |
2721 | } |
2722 | ||
2723 | if (seeding_dev) { | |
5c466629 | 2724 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2725 | if (ret) { |
66642832 | 2726 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2727 | goto error_sysfs; |
005d6427 | 2728 | } |
b2373f25 | 2729 | |
8e560081 NB |
2730 | /* |
2731 | * fs_devices now represents the newly sprouted filesystem and | |
2732 | * its fsid has been changed by btrfs_prepare_sprout | |
2733 | */ | |
2734 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2735 | } |
2736 | ||
3a45bb20 | 2737 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2738 | |
2b82032c YZ |
2739 | if (seeding_dev) { |
2740 | mutex_unlock(&uuid_mutex); | |
2741 | up_write(&sb->s_umount); | |
44cab9ba | 2742 | locked = false; |
788f20eb | 2743 | |
79787eaa JM |
2744 | if (ret) /* transaction commit */ |
2745 | return ret; | |
2746 | ||
2ff7e61e | 2747 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2748 | if (ret < 0) |
0b246afa | 2749 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2750 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2751 | trans = btrfs_attach_transaction(root); |
2752 | if (IS_ERR(trans)) { | |
2753 | if (PTR_ERR(trans) == -ENOENT) | |
2754 | return 0; | |
7132a262 AJ |
2755 | ret = PTR_ERR(trans); |
2756 | trans = NULL; | |
2757 | goto error_sysfs; | |
671415b7 | 2758 | } |
3a45bb20 | 2759 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2760 | } |
c9e9f97b | 2761 | |
7f551d96 AJ |
2762 | /* |
2763 | * Now that we have written a new super block to this device, check all | |
2764 | * other fs_devices list if device_path alienates any other scanned | |
2765 | * device. | |
2766 | * We can ignore the return value as it typically returns -EINVAL and | |
2767 | * only succeeds if the device was an alien. | |
2768 | */ | |
2769 | btrfs_forget_devices(device_path); | |
2770 | ||
2771 | /* Update ctime/mtime for blkid or udev */ | |
5a1972bd | 2772 | update_dev_time(device_path); |
7f551d96 | 2773 | |
2b82032c | 2774 | return ret; |
79787eaa | 2775 | |
d31c32f6 | 2776 | error_sysfs: |
53f8a74c | 2777 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2778 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2779 | mutex_lock(&fs_info->chunk_mutex); | |
2780 | list_del_rcu(&device->dev_list); | |
2781 | list_del(&device->dev_alloc_list); | |
2782 | fs_info->fs_devices->num_devices--; | |
2783 | fs_info->fs_devices->open_devices--; | |
2784 | fs_info->fs_devices->rw_devices--; | |
2785 | fs_info->fs_devices->total_devices--; | |
2786 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2787 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2788 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2789 | orig_super_total_bytes); | |
2790 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2791 | orig_super_num_devices); | |
2792 | mutex_unlock(&fs_info->chunk_mutex); | |
2793 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2794 | error_trans: |
0af2c4bf | 2795 | if (seeding_dev) |
a0a1db70 | 2796 | btrfs_set_sb_rdonly(sb); |
7132a262 AJ |
2797 | if (trans) |
2798 | btrfs_end_transaction(trans); | |
5b316468 NA |
2799 | error_free_zone: |
2800 | btrfs_destroy_dev_zone_info(device); | |
5c4cf6c9 | 2801 | error_free_device: |
a425f9d4 | 2802 | btrfs_free_device(device); |
2b82032c | 2803 | error: |
e525fd89 | 2804 | blkdev_put(bdev, FMODE_EXCL); |
44cab9ba | 2805 | if (locked) { |
2b82032c YZ |
2806 | mutex_unlock(&uuid_mutex); |
2807 | up_write(&sb->s_umount); | |
2808 | } | |
c9e9f97b | 2809 | return ret; |
788f20eb CM |
2810 | } |
2811 | ||
d397712b CM |
2812 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2813 | struct btrfs_device *device) | |
0b86a832 CM |
2814 | { |
2815 | int ret; | |
2816 | struct btrfs_path *path; | |
0b246afa | 2817 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2818 | struct btrfs_dev_item *dev_item; |
2819 | struct extent_buffer *leaf; | |
2820 | struct btrfs_key key; | |
2821 | ||
0b86a832 CM |
2822 | path = btrfs_alloc_path(); |
2823 | if (!path) | |
2824 | return -ENOMEM; | |
2825 | ||
2826 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2827 | key.type = BTRFS_DEV_ITEM_KEY; | |
2828 | key.offset = device->devid; | |
2829 | ||
2830 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2831 | if (ret < 0) | |
2832 | goto out; | |
2833 | ||
2834 | if (ret > 0) { | |
2835 | ret = -ENOENT; | |
2836 | goto out; | |
2837 | } | |
2838 | ||
2839 | leaf = path->nodes[0]; | |
2840 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2841 | ||
2842 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2843 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2844 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2845 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2846 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2847 | btrfs_set_device_total_bytes(leaf, dev_item, |
2848 | btrfs_device_get_disk_total_bytes(device)); | |
2849 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2850 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2851 | btrfs_mark_buffer_dirty(leaf); |
2852 | ||
2853 | out: | |
2854 | btrfs_free_path(path); | |
2855 | return ret; | |
2856 | } | |
2857 | ||
2196d6e8 | 2858 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2859 | struct btrfs_device *device, u64 new_size) |
2860 | { | |
0b246afa JM |
2861 | struct btrfs_fs_info *fs_info = device->fs_info; |
2862 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2863 | u64 old_total; |
2864 | u64 diff; | |
8f18cf13 | 2865 | |
ebbede42 | 2866 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2867 | return -EACCES; |
2196d6e8 | 2868 | |
7dfb8be1 NB |
2869 | new_size = round_down(new_size, fs_info->sectorsize); |
2870 | ||
34441361 | 2871 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2872 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2873 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2874 | |
63a212ab | 2875 | if (new_size <= device->total_bytes || |
401e29c1 | 2876 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2877 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2878 | return -EINVAL; |
2196d6e8 | 2879 | } |
2b82032c | 2880 | |
7dfb8be1 NB |
2881 | btrfs_set_super_total_bytes(super_copy, |
2882 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2883 | device->fs_devices->total_rw_bytes += diff; |
2884 | ||
7cc8e58d MX |
2885 | btrfs_device_set_total_bytes(device, new_size); |
2886 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2887 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2888 | if (list_empty(&device->post_commit_list)) |
2889 | list_add_tail(&device->post_commit_list, | |
2890 | &trans->transaction->dev_update_list); | |
34441361 | 2891 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2892 | |
8f18cf13 CM |
2893 | return btrfs_update_device(trans, device); |
2894 | } | |
2895 | ||
f4208794 | 2896 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2897 | { |
f4208794 | 2898 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2899 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2900 | int ret; |
2901 | struct btrfs_path *path; | |
2902 | struct btrfs_key key; | |
2903 | ||
8f18cf13 CM |
2904 | path = btrfs_alloc_path(); |
2905 | if (!path) | |
2906 | return -ENOMEM; | |
2907 | ||
408fbf19 | 2908 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2909 | key.offset = chunk_offset; |
2910 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2911 | ||
2912 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2913 | if (ret < 0) |
2914 | goto out; | |
2915 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2916 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2917 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2918 | ret = -ENOENT; |
2919 | goto out; | |
2920 | } | |
8f18cf13 CM |
2921 | |
2922 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2923 | if (ret < 0) |
0b246afa JM |
2924 | btrfs_handle_fs_error(fs_info, ret, |
2925 | "Failed to delete chunk item."); | |
79787eaa | 2926 | out: |
8f18cf13 | 2927 | btrfs_free_path(path); |
65a246c5 | 2928 | return ret; |
8f18cf13 CM |
2929 | } |
2930 | ||
408fbf19 | 2931 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2932 | { |
0b246afa | 2933 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2934 | struct btrfs_disk_key *disk_key; |
2935 | struct btrfs_chunk *chunk; | |
2936 | u8 *ptr; | |
2937 | int ret = 0; | |
2938 | u32 num_stripes; | |
2939 | u32 array_size; | |
2940 | u32 len = 0; | |
2941 | u32 cur; | |
2942 | struct btrfs_key key; | |
2943 | ||
34441361 | 2944 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2945 | array_size = btrfs_super_sys_array_size(super_copy); |
2946 | ||
2947 | ptr = super_copy->sys_chunk_array; | |
2948 | cur = 0; | |
2949 | ||
2950 | while (cur < array_size) { | |
2951 | disk_key = (struct btrfs_disk_key *)ptr; | |
2952 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2953 | ||
2954 | len = sizeof(*disk_key); | |
2955 | ||
2956 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2957 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2958 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2959 | len += btrfs_chunk_item_size(num_stripes); | |
2960 | } else { | |
2961 | ret = -EIO; | |
2962 | break; | |
2963 | } | |
408fbf19 | 2964 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2965 | key.offset == chunk_offset) { |
2966 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2967 | array_size -= len; | |
2968 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2969 | } else { | |
2970 | ptr += len; | |
2971 | cur += len; | |
2972 | } | |
2973 | } | |
34441361 | 2974 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2975 | return ret; |
2976 | } | |
2977 | ||
60ca842e OS |
2978 | /* |
2979 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2980 | * @logical: Logical block offset in bytes. | |
2981 | * @length: Length of extent in bytes. | |
2982 | * | |
2983 | * Return: Chunk mapping or ERR_PTR. | |
2984 | */ | |
2985 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2986 | u64 logical, u64 length) | |
592d92ee LB |
2987 | { |
2988 | struct extent_map_tree *em_tree; | |
2989 | struct extent_map *em; | |
2990 | ||
c8bf1b67 | 2991 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2992 | read_lock(&em_tree->lock); |
2993 | em = lookup_extent_mapping(em_tree, logical, length); | |
2994 | read_unlock(&em_tree->lock); | |
2995 | ||
2996 | if (!em) { | |
2997 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2998 | logical, length); | |
2999 | return ERR_PTR(-EINVAL); | |
3000 | } | |
3001 | ||
3002 | if (em->start > logical || em->start + em->len < logical) { | |
3003 | btrfs_crit(fs_info, | |
3004 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
3005 | logical, length, em->start, em->start + em->len); | |
3006 | free_extent_map(em); | |
3007 | return ERR_PTR(-EINVAL); | |
3008 | } | |
3009 | ||
3010 | /* callers are responsible for dropping em's ref. */ | |
3011 | return em; | |
3012 | } | |
3013 | ||
97aff912 | 3014 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3015 | { |
97aff912 | 3016 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3017 | struct extent_map *em; |
3018 | struct map_lookup *map; | |
2196d6e8 | 3019 | u64 dev_extent_len = 0; |
47ab2a6c | 3020 | int i, ret = 0; |
0b246afa | 3021 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3022 | |
60ca842e | 3023 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3024 | if (IS_ERR(em)) { |
47ab2a6c JB |
3025 | /* |
3026 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3027 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3028 | * do anything we still error out. |
3029 | */ | |
3030 | ASSERT(0); | |
592d92ee | 3031 | return PTR_ERR(em); |
47ab2a6c | 3032 | } |
95617d69 | 3033 | map = em->map_lookup; |
34441361 | 3034 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 3035 | check_system_chunk(trans, map->type); |
34441361 | 3036 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 3037 | |
57ba4cb8 FM |
3038 | /* |
3039 | * Take the device list mutex to prevent races with the final phase of | |
3040 | * a device replace operation that replaces the device object associated | |
3041 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
3042 | */ | |
3043 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3044 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3045 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3046 | ret = btrfs_free_dev_extent(trans, device, |
3047 | map->stripes[i].physical, | |
3048 | &dev_extent_len); | |
47ab2a6c | 3049 | if (ret) { |
57ba4cb8 | 3050 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3051 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3052 | goto out; |
3053 | } | |
a061fc8d | 3054 | |
2196d6e8 | 3055 | if (device->bytes_used > 0) { |
34441361 | 3056 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3057 | btrfs_device_set_bytes_used(device, |
3058 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3059 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3060 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3061 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3062 | } |
a061fc8d | 3063 | |
64bc6c2a NB |
3064 | ret = btrfs_update_device(trans, device); |
3065 | if (ret) { | |
3066 | mutex_unlock(&fs_devices->device_list_mutex); | |
3067 | btrfs_abort_transaction(trans, ret); | |
3068 | goto out; | |
dfe25020 | 3069 | } |
8f18cf13 | 3070 | } |
57ba4cb8 FM |
3071 | mutex_unlock(&fs_devices->device_list_mutex); |
3072 | ||
f4208794 | 3073 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3074 | if (ret) { |
66642832 | 3075 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3076 | goto out; |
3077 | } | |
8f18cf13 | 3078 | |
6bccf3ab | 3079 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3080 | |
8f18cf13 | 3081 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3082 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3083 | if (ret) { |
66642832 | 3084 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3085 | goto out; |
3086 | } | |
8f18cf13 CM |
3087 | } |
3088 | ||
5a98ec01 | 3089 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3090 | if (ret) { |
66642832 | 3091 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3092 | goto out; |
3093 | } | |
2b82032c | 3094 | |
47ab2a6c | 3095 | out: |
2b82032c YZ |
3096 | /* once for us */ |
3097 | free_extent_map(em); | |
47ab2a6c JB |
3098 | return ret; |
3099 | } | |
2b82032c | 3100 | |
5b4aacef | 3101 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3102 | { |
5b4aacef | 3103 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3104 | struct btrfs_trans_handle *trans; |
b0643e59 | 3105 | struct btrfs_block_group *block_group; |
47ab2a6c | 3106 | int ret; |
2b82032c | 3107 | |
67c5e7d4 FM |
3108 | /* |
3109 | * Prevent races with automatic removal of unused block groups. | |
3110 | * After we relocate and before we remove the chunk with offset | |
3111 | * chunk_offset, automatic removal of the block group can kick in, | |
3112 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3113 | * | |
3114 | * Make sure to acquire this mutex before doing a tree search (dev | |
3115 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3116 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3117 | * we release the path used to search the chunk/dev tree and before | |
3118 | * the current task acquires this mutex and calls us. | |
3119 | */ | |
a32bf9a3 | 3120 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3121 | |
47ab2a6c | 3122 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3123 | btrfs_scrub_pause(fs_info); |
0b246afa | 3124 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3125 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3126 | if (ret) |
3127 | return ret; | |
3128 | ||
b0643e59 DZ |
3129 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3130 | if (!block_group) | |
3131 | return -ENOENT; | |
3132 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
3133 | btrfs_put_block_group(block_group); | |
3134 | ||
19c4d2f9 CM |
3135 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3136 | chunk_offset); | |
3137 | if (IS_ERR(trans)) { | |
3138 | ret = PTR_ERR(trans); | |
3139 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3140 | return ret; | |
3141 | } | |
3142 | ||
47ab2a6c | 3143 | /* |
19c4d2f9 CM |
3144 | * step two, delete the device extents and the |
3145 | * chunk tree entries | |
47ab2a6c | 3146 | */ |
97aff912 | 3147 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3148 | btrfs_end_transaction(trans); |
19c4d2f9 | 3149 | return ret; |
2b82032c YZ |
3150 | } |
3151 | ||
2ff7e61e | 3152 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3153 | { |
0b246afa | 3154 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3155 | struct btrfs_path *path; |
3156 | struct extent_buffer *leaf; | |
3157 | struct btrfs_chunk *chunk; | |
3158 | struct btrfs_key key; | |
3159 | struct btrfs_key found_key; | |
2b82032c | 3160 | u64 chunk_type; |
ba1bf481 JB |
3161 | bool retried = false; |
3162 | int failed = 0; | |
2b82032c YZ |
3163 | int ret; |
3164 | ||
3165 | path = btrfs_alloc_path(); | |
3166 | if (!path) | |
3167 | return -ENOMEM; | |
3168 | ||
ba1bf481 | 3169 | again: |
2b82032c YZ |
3170 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3171 | key.offset = (u64)-1; | |
3172 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3173 | ||
3174 | while (1) { | |
0b246afa | 3175 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3176 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3177 | if (ret < 0) { |
0b246afa | 3178 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3179 | goto error; |
67c5e7d4 | 3180 | } |
79787eaa | 3181 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3182 | |
3183 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3184 | key.type); | |
67c5e7d4 | 3185 | if (ret) |
0b246afa | 3186 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3187 | if (ret < 0) |
3188 | goto error; | |
3189 | if (ret > 0) | |
3190 | break; | |
1a40e23b | 3191 | |
2b82032c YZ |
3192 | leaf = path->nodes[0]; |
3193 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3194 | |
2b82032c YZ |
3195 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3196 | struct btrfs_chunk); | |
3197 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3198 | btrfs_release_path(path); |
8f18cf13 | 3199 | |
2b82032c | 3200 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3201 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3202 | if (ret == -ENOSPC) |
3203 | failed++; | |
14586651 HS |
3204 | else |
3205 | BUG_ON(ret); | |
2b82032c | 3206 | } |
0b246afa | 3207 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3208 | |
2b82032c YZ |
3209 | if (found_key.offset == 0) |
3210 | break; | |
3211 | key.offset = found_key.offset - 1; | |
3212 | } | |
3213 | ret = 0; | |
ba1bf481 JB |
3214 | if (failed && !retried) { |
3215 | failed = 0; | |
3216 | retried = true; | |
3217 | goto again; | |
fae7f21c | 3218 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3219 | ret = -ENOSPC; |
3220 | } | |
2b82032c YZ |
3221 | error: |
3222 | btrfs_free_path(path); | |
3223 | return ret; | |
8f18cf13 CM |
3224 | } |
3225 | ||
a6f93c71 LB |
3226 | /* |
3227 | * return 1 : allocate a data chunk successfully, | |
3228 | * return <0: errors during allocating a data chunk, | |
3229 | * return 0 : no need to allocate a data chunk. | |
3230 | */ | |
3231 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3232 | u64 chunk_offset) | |
3233 | { | |
32da5386 | 3234 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3235 | u64 bytes_used; |
3236 | u64 chunk_type; | |
3237 | ||
3238 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3239 | ASSERT(cache); | |
3240 | chunk_type = cache->flags; | |
3241 | btrfs_put_block_group(cache); | |
3242 | ||
5ae21692 JT |
3243 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3244 | return 0; | |
3245 | ||
3246 | spin_lock(&fs_info->data_sinfo->lock); | |
3247 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3248 | spin_unlock(&fs_info->data_sinfo->lock); | |
3249 | ||
3250 | if (!bytes_used) { | |
3251 | struct btrfs_trans_handle *trans; | |
3252 | int ret; | |
3253 | ||
3254 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3255 | if (IS_ERR(trans)) | |
3256 | return PTR_ERR(trans); | |
3257 | ||
3258 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3259 | btrfs_end_transaction(trans); | |
3260 | if (ret < 0) | |
3261 | return ret; | |
3262 | return 1; | |
a6f93c71 | 3263 | } |
5ae21692 | 3264 | |
a6f93c71 LB |
3265 | return 0; |
3266 | } | |
3267 | ||
6bccf3ab | 3268 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3269 | struct btrfs_balance_control *bctl) |
3270 | { | |
6bccf3ab | 3271 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3272 | struct btrfs_trans_handle *trans; |
3273 | struct btrfs_balance_item *item; | |
3274 | struct btrfs_disk_balance_args disk_bargs; | |
3275 | struct btrfs_path *path; | |
3276 | struct extent_buffer *leaf; | |
3277 | struct btrfs_key key; | |
3278 | int ret, err; | |
3279 | ||
3280 | path = btrfs_alloc_path(); | |
3281 | if (!path) | |
3282 | return -ENOMEM; | |
3283 | ||
3284 | trans = btrfs_start_transaction(root, 0); | |
3285 | if (IS_ERR(trans)) { | |
3286 | btrfs_free_path(path); | |
3287 | return PTR_ERR(trans); | |
3288 | } | |
3289 | ||
3290 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3291 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3292 | key.offset = 0; |
3293 | ||
3294 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3295 | sizeof(*item)); | |
3296 | if (ret) | |
3297 | goto out; | |
3298 | ||
3299 | leaf = path->nodes[0]; | |
3300 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3301 | ||
b159fa28 | 3302 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3303 | |
3304 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3305 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3306 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3307 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3308 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3309 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3310 | ||
3311 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3312 | ||
3313 | btrfs_mark_buffer_dirty(leaf); | |
3314 | out: | |
3315 | btrfs_free_path(path); | |
3a45bb20 | 3316 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3317 | if (err && !ret) |
3318 | ret = err; | |
3319 | return ret; | |
3320 | } | |
3321 | ||
6bccf3ab | 3322 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3323 | { |
6bccf3ab | 3324 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3325 | struct btrfs_trans_handle *trans; |
3326 | struct btrfs_path *path; | |
3327 | struct btrfs_key key; | |
3328 | int ret, err; | |
3329 | ||
3330 | path = btrfs_alloc_path(); | |
3331 | if (!path) | |
3332 | return -ENOMEM; | |
3333 | ||
3502a8c0 | 3334 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3335 | if (IS_ERR(trans)) { |
3336 | btrfs_free_path(path); | |
3337 | return PTR_ERR(trans); | |
3338 | } | |
3339 | ||
3340 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3341 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3342 | key.offset = 0; |
3343 | ||
3344 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3345 | if (ret < 0) | |
3346 | goto out; | |
3347 | if (ret > 0) { | |
3348 | ret = -ENOENT; | |
3349 | goto out; | |
3350 | } | |
3351 | ||
3352 | ret = btrfs_del_item(trans, root, path); | |
3353 | out: | |
3354 | btrfs_free_path(path); | |
3a45bb20 | 3355 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3356 | if (err && !ret) |
3357 | ret = err; | |
3358 | return ret; | |
3359 | } | |
3360 | ||
59641015 ID |
3361 | /* |
3362 | * This is a heuristic used to reduce the number of chunks balanced on | |
3363 | * resume after balance was interrupted. | |
3364 | */ | |
3365 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3366 | { | |
3367 | /* | |
3368 | * Turn on soft mode for chunk types that were being converted. | |
3369 | */ | |
3370 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3371 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3372 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3373 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3374 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3375 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3376 | ||
3377 | /* | |
3378 | * Turn on usage filter if is not already used. The idea is | |
3379 | * that chunks that we have already balanced should be | |
3380 | * reasonably full. Don't do it for chunks that are being | |
3381 | * converted - that will keep us from relocating unconverted | |
3382 | * (albeit full) chunks. | |
3383 | */ | |
3384 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3385 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3386 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3387 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3388 | bctl->data.usage = 90; | |
3389 | } | |
3390 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3391 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3392 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3393 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3394 | bctl->sys.usage = 90; | |
3395 | } | |
3396 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3397 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3398 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3399 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3400 | bctl->meta.usage = 90; | |
3401 | } | |
3402 | } | |
3403 | ||
149196a2 DS |
3404 | /* |
3405 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3406 | */ | |
3407 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3408 | { |
3409 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3410 | int ret; |
c9e9f97b ID |
3411 | |
3412 | BUG_ON(!fs_info->balance_ctl); | |
3413 | ||
3414 | spin_lock(&fs_info->balance_lock); | |
3415 | fs_info->balance_ctl = NULL; | |
3416 | spin_unlock(&fs_info->balance_lock); | |
3417 | ||
3418 | kfree(bctl); | |
149196a2 DS |
3419 | ret = del_balance_item(fs_info); |
3420 | if (ret) | |
3421 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3422 | } |
3423 | ||
ed25e9b2 ID |
3424 | /* |
3425 | * Balance filters. Return 1 if chunk should be filtered out | |
3426 | * (should not be balanced). | |
3427 | */ | |
899c81ea | 3428 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3429 | struct btrfs_balance_args *bargs) |
3430 | { | |
899c81ea ID |
3431 | chunk_type = chunk_to_extended(chunk_type) & |
3432 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3433 | |
899c81ea | 3434 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3435 | return 0; |
3436 | ||
3437 | return 1; | |
3438 | } | |
3439 | ||
dba72cb3 | 3440 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3441 | struct btrfs_balance_args *bargs) |
bc309467 | 3442 | { |
32da5386 | 3443 | struct btrfs_block_group *cache; |
bc309467 DS |
3444 | u64 chunk_used; |
3445 | u64 user_thresh_min; | |
3446 | u64 user_thresh_max; | |
3447 | int ret = 1; | |
3448 | ||
3449 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3450 | chunk_used = cache->used; |
bc309467 DS |
3451 | |
3452 | if (bargs->usage_min == 0) | |
3453 | user_thresh_min = 0; | |
3454 | else | |
b3470b5d DS |
3455 | user_thresh_min = div_factor_fine(cache->length, |
3456 | bargs->usage_min); | |
bc309467 DS |
3457 | |
3458 | if (bargs->usage_max == 0) | |
3459 | user_thresh_max = 1; | |
3460 | else if (bargs->usage_max > 100) | |
b3470b5d | 3461 | user_thresh_max = cache->length; |
bc309467 | 3462 | else |
b3470b5d DS |
3463 | user_thresh_max = div_factor_fine(cache->length, |
3464 | bargs->usage_max); | |
bc309467 DS |
3465 | |
3466 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3467 | ret = 0; | |
3468 | ||
3469 | btrfs_put_block_group(cache); | |
3470 | return ret; | |
3471 | } | |
3472 | ||
dba72cb3 | 3473 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3474 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3475 | { |
32da5386 | 3476 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3477 | u64 chunk_used, user_thresh; |
3478 | int ret = 1; | |
3479 | ||
3480 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3481 | chunk_used = cache->used; |
5ce5b3c0 | 3482 | |
bc309467 | 3483 | if (bargs->usage_min == 0) |
3e39cea6 | 3484 | user_thresh = 1; |
a105bb88 | 3485 | else if (bargs->usage > 100) |
b3470b5d | 3486 | user_thresh = cache->length; |
a105bb88 | 3487 | else |
b3470b5d | 3488 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3489 | |
5ce5b3c0 ID |
3490 | if (chunk_used < user_thresh) |
3491 | ret = 0; | |
3492 | ||
3493 | btrfs_put_block_group(cache); | |
3494 | return ret; | |
3495 | } | |
3496 | ||
409d404b ID |
3497 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3498 | struct btrfs_chunk *chunk, | |
3499 | struct btrfs_balance_args *bargs) | |
3500 | { | |
3501 | struct btrfs_stripe *stripe; | |
3502 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3503 | int i; | |
3504 | ||
3505 | for (i = 0; i < num_stripes; i++) { | |
3506 | stripe = btrfs_stripe_nr(chunk, i); | |
3507 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3508 | return 0; | |
3509 | } | |
3510 | ||
3511 | return 1; | |
3512 | } | |
3513 | ||
946c9256 DS |
3514 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3515 | { | |
3516 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3517 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3518 | const int nparity = btrfs_raid_array[index].nparity; | |
3519 | ||
3520 | if (nparity) | |
3521 | return num_stripes - nparity; | |
3522 | else | |
3523 | return num_stripes / ncopies; | |
3524 | } | |
3525 | ||
94e60d5a ID |
3526 | /* [pstart, pend) */ |
3527 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3528 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3529 | struct btrfs_balance_args *bargs) |
3530 | { | |
3531 | struct btrfs_stripe *stripe; | |
3532 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3533 | u64 stripe_offset; | |
3534 | u64 stripe_length; | |
946c9256 | 3535 | u64 type; |
94e60d5a ID |
3536 | int factor; |
3537 | int i; | |
3538 | ||
3539 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3540 | return 0; | |
3541 | ||
946c9256 DS |
3542 | type = btrfs_chunk_type(leaf, chunk); |
3543 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3544 | |
3545 | for (i = 0; i < num_stripes; i++) { | |
3546 | stripe = btrfs_stripe_nr(chunk, i); | |
3547 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3548 | continue; | |
3549 | ||
3550 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3551 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3552 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3553 | |
3554 | if (stripe_offset < bargs->pend && | |
3555 | stripe_offset + stripe_length > bargs->pstart) | |
3556 | return 0; | |
3557 | } | |
3558 | ||
3559 | return 1; | |
3560 | } | |
3561 | ||
ea67176a ID |
3562 | /* [vstart, vend) */ |
3563 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3564 | struct btrfs_chunk *chunk, | |
3565 | u64 chunk_offset, | |
3566 | struct btrfs_balance_args *bargs) | |
3567 | { | |
3568 | if (chunk_offset < bargs->vend && | |
3569 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3570 | /* at least part of the chunk is inside this vrange */ | |
3571 | return 0; | |
3572 | ||
3573 | return 1; | |
3574 | } | |
3575 | ||
dee32d0a GAP |
3576 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3577 | struct btrfs_chunk *chunk, | |
3578 | struct btrfs_balance_args *bargs) | |
3579 | { | |
3580 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3581 | ||
3582 | if (bargs->stripes_min <= num_stripes | |
3583 | && num_stripes <= bargs->stripes_max) | |
3584 | return 0; | |
3585 | ||
3586 | return 1; | |
3587 | } | |
3588 | ||
899c81ea | 3589 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3590 | struct btrfs_balance_args *bargs) |
3591 | { | |
3592 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3593 | return 0; | |
3594 | ||
899c81ea ID |
3595 | chunk_type = chunk_to_extended(chunk_type) & |
3596 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3597 | |
899c81ea | 3598 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3599 | return 1; |
3600 | ||
3601 | return 0; | |
3602 | } | |
3603 | ||
6ec0896c | 3604 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3605 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3606 | { | |
6ec0896c | 3607 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3608 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3609 | struct btrfs_balance_args *bargs = NULL; |
3610 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3611 | ||
3612 | /* type filter */ | |
3613 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3614 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3615 | return 0; | |
3616 | } | |
3617 | ||
3618 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3619 | bargs = &bctl->data; | |
3620 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3621 | bargs = &bctl->sys; | |
3622 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3623 | bargs = &bctl->meta; | |
3624 | ||
ed25e9b2 ID |
3625 | /* profiles filter */ |
3626 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3627 | chunk_profiles_filter(chunk_type, bargs)) { | |
3628 | return 0; | |
5ce5b3c0 ID |
3629 | } |
3630 | ||
3631 | /* usage filter */ | |
3632 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3633 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3634 | return 0; |
bc309467 | 3635 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3636 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3637 | return 0; |
409d404b ID |
3638 | } |
3639 | ||
3640 | /* devid filter */ | |
3641 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3642 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3643 | return 0; | |
94e60d5a ID |
3644 | } |
3645 | ||
3646 | /* drange filter, makes sense only with devid filter */ | |
3647 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3648 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3649 | return 0; |
ea67176a ID |
3650 | } |
3651 | ||
3652 | /* vrange filter */ | |
3653 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3654 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3655 | return 0; | |
ed25e9b2 ID |
3656 | } |
3657 | ||
dee32d0a GAP |
3658 | /* stripes filter */ |
3659 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3660 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3661 | return 0; | |
3662 | } | |
3663 | ||
cfa4c961 ID |
3664 | /* soft profile changing mode */ |
3665 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3666 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3667 | return 0; | |
3668 | } | |
3669 | ||
7d824b6f DS |
3670 | /* |
3671 | * limited by count, must be the last filter | |
3672 | */ | |
3673 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3674 | if (bargs->limit == 0) | |
3675 | return 0; | |
3676 | else | |
3677 | bargs->limit--; | |
12907fc7 DS |
3678 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3679 | /* | |
3680 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3681 | * determined here because we do not have the global information |
12907fc7 DS |
3682 | * about the count of all chunks that satisfy the filters. |
3683 | */ | |
3684 | if (bargs->limit_max == 0) | |
3685 | return 0; | |
3686 | else | |
3687 | bargs->limit_max--; | |
7d824b6f DS |
3688 | } |
3689 | ||
f43ffb60 ID |
3690 | return 1; |
3691 | } | |
3692 | ||
c9e9f97b | 3693 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3694 | { |
19a39dce | 3695 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3696 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3697 | u64 chunk_type; |
f43ffb60 | 3698 | struct btrfs_chunk *chunk; |
5a488b9d | 3699 | struct btrfs_path *path = NULL; |
ec44a35c | 3700 | struct btrfs_key key; |
ec44a35c | 3701 | struct btrfs_key found_key; |
f43ffb60 ID |
3702 | struct extent_buffer *leaf; |
3703 | int slot; | |
c9e9f97b ID |
3704 | int ret; |
3705 | int enospc_errors = 0; | |
19a39dce | 3706 | bool counting = true; |
12907fc7 | 3707 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3708 | u64 limit_data = bctl->data.limit; |
3709 | u64 limit_meta = bctl->meta.limit; | |
3710 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3711 | u32 count_data = 0; |
3712 | u32 count_meta = 0; | |
3713 | u32 count_sys = 0; | |
2c9fe835 | 3714 | int chunk_reserved = 0; |
ec44a35c | 3715 | |
ec44a35c | 3716 | path = btrfs_alloc_path(); |
17e9f796 MF |
3717 | if (!path) { |
3718 | ret = -ENOMEM; | |
3719 | goto error; | |
3720 | } | |
19a39dce ID |
3721 | |
3722 | /* zero out stat counters */ | |
3723 | spin_lock(&fs_info->balance_lock); | |
3724 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3725 | spin_unlock(&fs_info->balance_lock); | |
3726 | again: | |
7d824b6f | 3727 | if (!counting) { |
12907fc7 DS |
3728 | /* |
3729 | * The single value limit and min/max limits use the same bytes | |
3730 | * in the | |
3731 | */ | |
7d824b6f DS |
3732 | bctl->data.limit = limit_data; |
3733 | bctl->meta.limit = limit_meta; | |
3734 | bctl->sys.limit = limit_sys; | |
3735 | } | |
ec44a35c CM |
3736 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3737 | key.offset = (u64)-1; | |
3738 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3739 | ||
d397712b | 3740 | while (1) { |
19a39dce | 3741 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3742 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3743 | ret = -ECANCELED; |
3744 | goto error; | |
3745 | } | |
3746 | ||
67c5e7d4 | 3747 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3748 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3749 | if (ret < 0) { |
3750 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3751 | goto error; |
67c5e7d4 | 3752 | } |
ec44a35c CM |
3753 | |
3754 | /* | |
3755 | * this shouldn't happen, it means the last relocate | |
3756 | * failed | |
3757 | */ | |
3758 | if (ret == 0) | |
c9e9f97b | 3759 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3760 | |
3761 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3762 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3763 | if (ret) { |
67c5e7d4 | 3764 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3765 | ret = 0; |
ec44a35c | 3766 | break; |
c9e9f97b | 3767 | } |
7d9eb12c | 3768 | |
f43ffb60 ID |
3769 | leaf = path->nodes[0]; |
3770 | slot = path->slots[0]; | |
3771 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3772 | |
67c5e7d4 FM |
3773 | if (found_key.objectid != key.objectid) { |
3774 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3775 | break; |
67c5e7d4 | 3776 | } |
7d9eb12c | 3777 | |
f43ffb60 | 3778 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3779 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3780 | |
19a39dce ID |
3781 | if (!counting) { |
3782 | spin_lock(&fs_info->balance_lock); | |
3783 | bctl->stat.considered++; | |
3784 | spin_unlock(&fs_info->balance_lock); | |
3785 | } | |
3786 | ||
6ec0896c | 3787 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3788 | |
b3b4aa74 | 3789 | btrfs_release_path(path); |
67c5e7d4 FM |
3790 | if (!ret) { |
3791 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3792 | goto loop; |
67c5e7d4 | 3793 | } |
f43ffb60 | 3794 | |
19a39dce | 3795 | if (counting) { |
67c5e7d4 | 3796 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3797 | spin_lock(&fs_info->balance_lock); |
3798 | bctl->stat.expected++; | |
3799 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3800 | |
3801 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3802 | count_data++; | |
3803 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3804 | count_sys++; | |
3805 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3806 | count_meta++; | |
3807 | ||
3808 | goto loop; | |
3809 | } | |
3810 | ||
3811 | /* | |
3812 | * Apply limit_min filter, no need to check if the LIMITS | |
3813 | * filter is used, limit_min is 0 by default | |
3814 | */ | |
3815 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3816 | count_data < bctl->data.limit_min) | |
3817 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3818 | count_meta < bctl->meta.limit_min) | |
3819 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3820 | count_sys < bctl->sys.limit_min)) { | |
3821 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3822 | goto loop; |
3823 | } | |
3824 | ||
a6f93c71 LB |
3825 | if (!chunk_reserved) { |
3826 | /* | |
3827 | * We may be relocating the only data chunk we have, | |
3828 | * which could potentially end up with losing data's | |
3829 | * raid profile, so lets allocate an empty one in | |
3830 | * advance. | |
3831 | */ | |
3832 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3833 | found_key.offset); | |
2c9fe835 ZL |
3834 | if (ret < 0) { |
3835 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3836 | goto error; | |
a6f93c71 LB |
3837 | } else if (ret == 1) { |
3838 | chunk_reserved = 1; | |
2c9fe835 | 3839 | } |
2c9fe835 ZL |
3840 | } |
3841 | ||
5b4aacef | 3842 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3843 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3844 | if (ret == -ENOSPC) { |
c9e9f97b | 3845 | enospc_errors++; |
eede2bf3 OS |
3846 | } else if (ret == -ETXTBSY) { |
3847 | btrfs_info(fs_info, | |
3848 | "skipping relocation of block group %llu due to active swapfile", | |
3849 | found_key.offset); | |
3850 | ret = 0; | |
3851 | } else if (ret) { | |
3852 | goto error; | |
19a39dce ID |
3853 | } else { |
3854 | spin_lock(&fs_info->balance_lock); | |
3855 | bctl->stat.completed++; | |
3856 | spin_unlock(&fs_info->balance_lock); | |
3857 | } | |
f43ffb60 | 3858 | loop: |
795a3321 ID |
3859 | if (found_key.offset == 0) |
3860 | break; | |
ba1bf481 | 3861 | key.offset = found_key.offset - 1; |
ec44a35c | 3862 | } |
c9e9f97b | 3863 | |
19a39dce ID |
3864 | if (counting) { |
3865 | btrfs_release_path(path); | |
3866 | counting = false; | |
3867 | goto again; | |
3868 | } | |
ec44a35c CM |
3869 | error: |
3870 | btrfs_free_path(path); | |
c9e9f97b | 3871 | if (enospc_errors) { |
efe120a0 | 3872 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3873 | enospc_errors); |
c9e9f97b ID |
3874 | if (!ret) |
3875 | ret = -ENOSPC; | |
3876 | } | |
3877 | ||
ec44a35c CM |
3878 | return ret; |
3879 | } | |
3880 | ||
0c460c0d ID |
3881 | /** |
3882 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3883 | * @flags: profile to validate | |
3884 | * @extended: if true @flags is treated as an extended profile | |
3885 | */ | |
3886 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3887 | { | |
3888 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3889 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3890 | ||
3891 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3892 | ||
3893 | /* 1) check that all other bits are zeroed */ | |
3894 | if (flags & ~mask) | |
3895 | return 0; | |
3896 | ||
3897 | /* 2) see if profile is reduced */ | |
3898 | if (flags == 0) | |
3899 | return !extended; /* "0" is valid for usual profiles */ | |
3900 | ||
c1499166 | 3901 | return has_single_bit_set(flags); |
0c460c0d ID |
3902 | } |
3903 | ||
837d5b6e ID |
3904 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3905 | { | |
a7e99c69 ID |
3906 | /* cancel requested || normal exit path */ |
3907 | return atomic_read(&fs_info->balance_cancel_req) || | |
3908 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3909 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3910 | } |
3911 | ||
5ba366c3 DS |
3912 | /* |
3913 | * Validate target profile against allowed profiles and return true if it's OK. | |
3914 | * Otherwise print the error message and return false. | |
3915 | */ | |
3916 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
3917 | const struct btrfs_balance_args *bargs, | |
3918 | u64 allowed, const char *type) | |
bdcd3c97 | 3919 | { |
5ba366c3 DS |
3920 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
3921 | return true; | |
3922 | ||
3923 | /* Profile is valid and does not have bits outside of the allowed set */ | |
3924 | if (alloc_profile_is_valid(bargs->target, 1) && | |
3925 | (bargs->target & ~allowed) == 0) | |
3926 | return true; | |
3927 | ||
3928 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
3929 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
3930 | return false; | |
bdcd3c97 AM |
3931 | } |
3932 | ||
56fc37d9 AJ |
3933 | /* |
3934 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3935 | * @size_buf including the terminating null. The output may be trimmed if it | |
3936 | * does not fit into the provided buffer. | |
3937 | */ | |
3938 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3939 | u32 size_buf) | |
3940 | { | |
3941 | int ret; | |
3942 | u32 size_bp = size_buf; | |
3943 | char *bp = buf; | |
3944 | u64 flags = bargs->flags; | |
3945 | char tmp_buf[128] = {'\0'}; | |
3946 | ||
3947 | if (!flags) | |
3948 | return; | |
3949 | ||
3950 | #define CHECK_APPEND_NOARG(a) \ | |
3951 | do { \ | |
3952 | ret = snprintf(bp, size_bp, (a)); \ | |
3953 | if (ret < 0 || ret >= size_bp) \ | |
3954 | goto out_overflow; \ | |
3955 | size_bp -= ret; \ | |
3956 | bp += ret; \ | |
3957 | } while (0) | |
3958 | ||
3959 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3960 | do { \ | |
3961 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3962 | if (ret < 0 || ret >= size_bp) \ | |
3963 | goto out_overflow; \ | |
3964 | size_bp -= ret; \ | |
3965 | bp += ret; \ | |
3966 | } while (0) | |
3967 | ||
3968 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3969 | do { \ | |
3970 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3971 | if (ret < 0 || ret >= size_bp) \ | |
3972 | goto out_overflow; \ | |
3973 | size_bp -= ret; \ | |
3974 | bp += ret; \ | |
3975 | } while (0) | |
3976 | ||
158da513 DS |
3977 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3978 | CHECK_APPEND_1ARG("convert=%s,", | |
3979 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3980 | |
3981 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3982 | CHECK_APPEND_NOARG("soft,"); | |
3983 | ||
3984 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3985 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3986 | sizeof(tmp_buf)); | |
3987 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3988 | } | |
3989 | ||
3990 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3991 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3992 | ||
3993 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3994 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3995 | bargs->usage_min, bargs->usage_max); | |
3996 | ||
3997 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3998 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3999 | ||
4000 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4001 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4002 | bargs->pstart, bargs->pend); | |
4003 | ||
4004 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4005 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4006 | bargs->vstart, bargs->vend); | |
4007 | ||
4008 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4009 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4010 | ||
4011 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4012 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4013 | bargs->limit_min, bargs->limit_max); | |
4014 | ||
4015 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4016 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4017 | bargs->stripes_min, bargs->stripes_max); | |
4018 | ||
4019 | #undef CHECK_APPEND_2ARG | |
4020 | #undef CHECK_APPEND_1ARG | |
4021 | #undef CHECK_APPEND_NOARG | |
4022 | ||
4023 | out_overflow: | |
4024 | ||
4025 | if (size_bp < size_buf) | |
4026 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4027 | else | |
4028 | buf[0] = '\0'; | |
4029 | } | |
4030 | ||
4031 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4032 | { | |
4033 | u32 size_buf = 1024; | |
4034 | char tmp_buf[192] = {'\0'}; | |
4035 | char *buf; | |
4036 | char *bp; | |
4037 | u32 size_bp = size_buf; | |
4038 | int ret; | |
4039 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4040 | ||
4041 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4042 | if (!buf) | |
4043 | return; | |
4044 | ||
4045 | bp = buf; | |
4046 | ||
4047 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4048 | do { \ | |
4049 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4050 | if (ret < 0 || ret >= size_bp) \ | |
4051 | goto out_overflow; \ | |
4052 | size_bp -= ret; \ | |
4053 | bp += ret; \ | |
4054 | } while (0) | |
4055 | ||
4056 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4057 | CHECK_APPEND_1ARG("%s", "-f "); | |
4058 | ||
4059 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4060 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4061 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4062 | } | |
4063 | ||
4064 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4065 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4066 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4067 | } | |
4068 | ||
4069 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4070 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4071 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4072 | } | |
4073 | ||
4074 | #undef CHECK_APPEND_1ARG | |
4075 | ||
4076 | out_overflow: | |
4077 | ||
4078 | if (size_bp < size_buf) | |
4079 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4080 | btrfs_info(fs_info, "balance: %s %s", | |
4081 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4082 | "resume" : "start", buf); | |
4083 | ||
4084 | kfree(buf); | |
4085 | } | |
4086 | ||
c9e9f97b | 4087 | /* |
dccdb07b | 4088 | * Should be called with balance mutexe held |
c9e9f97b | 4089 | */ |
6fcf6e2b DS |
4090 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4091 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4092 | struct btrfs_ioctl_balance_args *bargs) |
4093 | { | |
14506127 | 4094 | u64 meta_target, data_target; |
f43ffb60 | 4095 | u64 allowed; |
e4837f8f | 4096 | int mixed = 0; |
c9e9f97b | 4097 | int ret; |
8dabb742 | 4098 | u64 num_devices; |
de98ced9 | 4099 | unsigned seq; |
e62869be | 4100 | bool reducing_redundancy; |
081db89b | 4101 | int i; |
c9e9f97b | 4102 | |
837d5b6e | 4103 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4104 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4105 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4106 | ret = -EINVAL; |
4107 | goto out; | |
4108 | } | |
4109 | ||
e4837f8f ID |
4110 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4111 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4112 | mixed = 1; | |
4113 | ||
f43ffb60 ID |
4114 | /* |
4115 | * In case of mixed groups both data and meta should be picked, | |
4116 | * and identical options should be given for both of them. | |
4117 | */ | |
e4837f8f ID |
4118 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4119 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4120 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4121 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4122 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4123 | btrfs_err(fs_info, |
6dac13f8 | 4124 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4125 | ret = -EINVAL; |
4126 | goto out; | |
4127 | } | |
4128 | } | |
4129 | ||
b35cf1f0 JB |
4130 | /* |
4131 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4132 | * are exclusive |
b35cf1f0 JB |
4133 | */ |
4134 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4135 | |
4136 | /* | |
4137 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4138 | * special bit for it, to make it easier to distinguish. Thus we need | |
4139 | * to set it manually, or balance would refuse the profile. | |
4140 | */ | |
4141 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4142 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4143 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4144 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4145 | |
5ba366c3 DS |
4146 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4147 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4148 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4149 | ret = -EINVAL; |
4150 | goto out; | |
4151 | } | |
4152 | ||
6079e12c DS |
4153 | /* |
4154 | * Allow to reduce metadata or system integrity only if force set for | |
4155 | * profiles with redundancy (copies, parity) | |
4156 | */ | |
4157 | allowed = 0; | |
4158 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4159 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4160 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4161 | allowed |= btrfs_raid_array[i].bg_flag; | |
4162 | } | |
de98ced9 MX |
4163 | do { |
4164 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4165 | ||
4166 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4167 | (fs_info->avail_system_alloc_bits & allowed) && | |
4168 | !(bctl->sys.target & allowed)) || | |
4169 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4170 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4171 | !(bctl->meta.target & allowed))) |
e62869be | 4172 | reducing_redundancy = true; |
5a8067c0 | 4173 | else |
e62869be | 4174 | reducing_redundancy = false; |
5a8067c0 FM |
4175 | |
4176 | /* if we're not converting, the target field is uninitialized */ | |
4177 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4178 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4179 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4180 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4181 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4182 | |
e62869be | 4183 | if (reducing_redundancy) { |
5a8067c0 FM |
4184 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4185 | btrfs_info(fs_info, | |
e62869be | 4186 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4187 | } else { |
4188 | btrfs_err(fs_info, | |
e62869be | 4189 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4190 | ret = -EINVAL; |
4191 | goto out; | |
4192 | } | |
4193 | } | |
4194 | ||
14506127 AB |
4195 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4196 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4197 | btrfs_warn(fs_info, |
6dac13f8 | 4198 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4199 | btrfs_bg_type_to_raid_name(meta_target), |
4200 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4201 | } |
4202 | ||
9e967495 FM |
4203 | if (fs_info->send_in_progress) { |
4204 | btrfs_warn_rl(fs_info, | |
4205 | "cannot run balance while send operations are in progress (%d in progress)", | |
4206 | fs_info->send_in_progress); | |
4207 | ret = -EAGAIN; | |
4208 | goto out; | |
4209 | } | |
4210 | ||
6bccf3ab | 4211 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4212 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4213 | goto out; |
4214 | ||
59641015 ID |
4215 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4216 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4217 | BUG_ON(fs_info->balance_ctl); |
4218 | spin_lock(&fs_info->balance_lock); | |
4219 | fs_info->balance_ctl = bctl; | |
4220 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4221 | } else { |
4222 | BUG_ON(ret != -EEXIST); | |
4223 | spin_lock(&fs_info->balance_lock); | |
4224 | update_balance_args(bctl); | |
4225 | spin_unlock(&fs_info->balance_lock); | |
4226 | } | |
c9e9f97b | 4227 | |
3009a62f DS |
4228 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4229 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4230 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4231 | mutex_unlock(&fs_info->balance_mutex); |
4232 | ||
4233 | ret = __btrfs_balance(fs_info); | |
4234 | ||
4235 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4236 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4237 | btrfs_info(fs_info, "balance: paused"); | |
44d354ab QW |
4238 | /* |
4239 | * Balance can be canceled by: | |
4240 | * | |
4241 | * - Regular cancel request | |
4242 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4243 | * | |
4244 | * - Fatal signal to "btrfs" process | |
4245 | * Either the signal caught by wait_reserve_ticket() and callers | |
4246 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4247 | * got -ECANCELED. | |
4248 | * Either way, in this case balance_cancel_req = 0, and | |
4249 | * ret == -EINTR or ret == -ECANCELED. | |
4250 | * | |
4251 | * So here we only check the return value to catch canceled balance. | |
4252 | */ | |
4253 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4254 | btrfs_info(fs_info, "balance: canceled"); |
4255 | else | |
4256 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4257 | ||
3009a62f | 4258 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4259 | |
4260 | if (bargs) { | |
4261 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4262 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4263 | } |
4264 | ||
3a01aa7a ID |
4265 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4266 | balance_need_close(fs_info)) { | |
149196a2 | 4267 | reset_balance_state(fs_info); |
c3e1f96c | 4268 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4269 | } |
4270 | ||
837d5b6e | 4271 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4272 | |
4273 | return ret; | |
4274 | out: | |
59641015 | 4275 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4276 | reset_balance_state(fs_info); |
a17c95df | 4277 | else |
59641015 | 4278 | kfree(bctl); |
c3e1f96c | 4279 | btrfs_exclop_finish(fs_info); |
a17c95df | 4280 | |
59641015 ID |
4281 | return ret; |
4282 | } | |
4283 | ||
4284 | static int balance_kthread(void *data) | |
4285 | { | |
2b6ba629 | 4286 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4287 | int ret = 0; |
59641015 | 4288 | |
59641015 | 4289 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4290 | if (fs_info->balance_ctl) |
6fcf6e2b | 4291 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4292 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4293 | |
59641015 ID |
4294 | return ret; |
4295 | } | |
4296 | ||
2b6ba629 ID |
4297 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4298 | { | |
4299 | struct task_struct *tsk; | |
4300 | ||
1354e1a1 | 4301 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4302 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4303 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4304 | return 0; |
4305 | } | |
1354e1a1 | 4306 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4307 | |
3cdde224 | 4308 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4309 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4310 | return 0; |
4311 | } | |
4312 | ||
02ee654d AJ |
4313 | /* |
4314 | * A ro->rw remount sequence should continue with the paused balance | |
4315 | * regardless of who pauses it, system or the user as of now, so set | |
4316 | * the resume flag. | |
4317 | */ | |
4318 | spin_lock(&fs_info->balance_lock); | |
4319 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4320 | spin_unlock(&fs_info->balance_lock); | |
4321 | ||
2b6ba629 | 4322 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4323 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4324 | } |
4325 | ||
68310a5e | 4326 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4327 | { |
59641015 ID |
4328 | struct btrfs_balance_control *bctl; |
4329 | struct btrfs_balance_item *item; | |
4330 | struct btrfs_disk_balance_args disk_bargs; | |
4331 | struct btrfs_path *path; | |
4332 | struct extent_buffer *leaf; | |
4333 | struct btrfs_key key; | |
4334 | int ret; | |
4335 | ||
4336 | path = btrfs_alloc_path(); | |
4337 | if (!path) | |
4338 | return -ENOMEM; | |
4339 | ||
59641015 | 4340 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4341 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4342 | key.offset = 0; |
4343 | ||
68310a5e | 4344 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4345 | if (ret < 0) |
68310a5e | 4346 | goto out; |
59641015 ID |
4347 | if (ret > 0) { /* ret = -ENOENT; */ |
4348 | ret = 0; | |
68310a5e ID |
4349 | goto out; |
4350 | } | |
4351 | ||
4352 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4353 | if (!bctl) { | |
4354 | ret = -ENOMEM; | |
4355 | goto out; | |
59641015 ID |
4356 | } |
4357 | ||
4358 | leaf = path->nodes[0]; | |
4359 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4360 | ||
68310a5e ID |
4361 | bctl->flags = btrfs_balance_flags(leaf, item); |
4362 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4363 | |
4364 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4365 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4366 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4367 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4368 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4369 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4370 | ||
eee95e3f DS |
4371 | /* |
4372 | * This should never happen, as the paused balance state is recovered | |
4373 | * during mount without any chance of other exclusive ops to collide. | |
4374 | * | |
4375 | * This gives the exclusive op status to balance and keeps in paused | |
4376 | * state until user intervention (cancel or umount). If the ownership | |
4377 | * cannot be assigned, show a message but do not fail. The balance | |
4378 | * is in a paused state and must have fs_info::balance_ctl properly | |
4379 | * set up. | |
4380 | */ | |
c3e1f96c | 4381 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) |
eee95e3f | 4382 | btrfs_warn(fs_info, |
6dac13f8 | 4383 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4384 | |
fb286100 JB |
4385 | btrfs_release_path(path); |
4386 | ||
68310a5e | 4387 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4388 | BUG_ON(fs_info->balance_ctl); |
4389 | spin_lock(&fs_info->balance_lock); | |
4390 | fs_info->balance_ctl = bctl; | |
4391 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4392 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4393 | out: |
4394 | btrfs_free_path(path); | |
ec44a35c CM |
4395 | return ret; |
4396 | } | |
4397 | ||
837d5b6e ID |
4398 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4399 | { | |
4400 | int ret = 0; | |
4401 | ||
4402 | mutex_lock(&fs_info->balance_mutex); | |
4403 | if (!fs_info->balance_ctl) { | |
4404 | mutex_unlock(&fs_info->balance_mutex); | |
4405 | return -ENOTCONN; | |
4406 | } | |
4407 | ||
3009a62f | 4408 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4409 | atomic_inc(&fs_info->balance_pause_req); |
4410 | mutex_unlock(&fs_info->balance_mutex); | |
4411 | ||
4412 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4413 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4414 | |
4415 | mutex_lock(&fs_info->balance_mutex); | |
4416 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4417 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4418 | atomic_dec(&fs_info->balance_pause_req); |
4419 | } else { | |
4420 | ret = -ENOTCONN; | |
4421 | } | |
4422 | ||
4423 | mutex_unlock(&fs_info->balance_mutex); | |
4424 | return ret; | |
4425 | } | |
4426 | ||
a7e99c69 ID |
4427 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4428 | { | |
4429 | mutex_lock(&fs_info->balance_mutex); | |
4430 | if (!fs_info->balance_ctl) { | |
4431 | mutex_unlock(&fs_info->balance_mutex); | |
4432 | return -ENOTCONN; | |
4433 | } | |
4434 | ||
cf7d20f4 DS |
4435 | /* |
4436 | * A paused balance with the item stored on disk can be resumed at | |
4437 | * mount time if the mount is read-write. Otherwise it's still paused | |
4438 | * and we must not allow cancelling as it deletes the item. | |
4439 | */ | |
4440 | if (sb_rdonly(fs_info->sb)) { | |
4441 | mutex_unlock(&fs_info->balance_mutex); | |
4442 | return -EROFS; | |
4443 | } | |
4444 | ||
a7e99c69 ID |
4445 | atomic_inc(&fs_info->balance_cancel_req); |
4446 | /* | |
4447 | * if we are running just wait and return, balance item is | |
4448 | * deleted in btrfs_balance in this case | |
4449 | */ | |
3009a62f | 4450 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4451 | mutex_unlock(&fs_info->balance_mutex); |
4452 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4453 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4454 | mutex_lock(&fs_info->balance_mutex); |
4455 | } else { | |
a7e99c69 | 4456 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4457 | /* |
4458 | * Lock released to allow other waiters to continue, we'll | |
4459 | * reexamine the status again. | |
4460 | */ | |
a7e99c69 ID |
4461 | mutex_lock(&fs_info->balance_mutex); |
4462 | ||
a17c95df | 4463 | if (fs_info->balance_ctl) { |
149196a2 | 4464 | reset_balance_state(fs_info); |
c3e1f96c | 4465 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4466 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4467 | } |
a7e99c69 ID |
4468 | } |
4469 | ||
3009a62f DS |
4470 | BUG_ON(fs_info->balance_ctl || |
4471 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4472 | atomic_dec(&fs_info->balance_cancel_req); |
4473 | mutex_unlock(&fs_info->balance_mutex); | |
4474 | return 0; | |
4475 | } | |
4476 | ||
97f4dd09 | 4477 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4478 | { |
4479 | struct btrfs_fs_info *fs_info = data; | |
4480 | struct btrfs_root *root = fs_info->tree_root; | |
4481 | struct btrfs_key key; | |
803b2f54 SB |
4482 | struct btrfs_path *path = NULL; |
4483 | int ret = 0; | |
4484 | struct extent_buffer *eb; | |
4485 | int slot; | |
4486 | struct btrfs_root_item root_item; | |
4487 | u32 item_size; | |
f45388f3 | 4488 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4489 | bool closing = false; |
803b2f54 SB |
4490 | |
4491 | path = btrfs_alloc_path(); | |
4492 | if (!path) { | |
4493 | ret = -ENOMEM; | |
4494 | goto out; | |
4495 | } | |
4496 | ||
4497 | key.objectid = 0; | |
4498 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4499 | key.offset = 0; | |
4500 | ||
803b2f54 | 4501 | while (1) { |
c94bec2c JB |
4502 | if (btrfs_fs_closing(fs_info)) { |
4503 | closing = true; | |
4504 | break; | |
4505 | } | |
7c829b72 AJ |
4506 | ret = btrfs_search_forward(root, &key, path, |
4507 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4508 | if (ret) { |
4509 | if (ret > 0) | |
4510 | ret = 0; | |
4511 | break; | |
4512 | } | |
4513 | ||
4514 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4515 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4516 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4517 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4518 | goto skip; | |
4519 | ||
4520 | eb = path->nodes[0]; | |
4521 | slot = path->slots[0]; | |
4522 | item_size = btrfs_item_size_nr(eb, slot); | |
4523 | if (item_size < sizeof(root_item)) | |
4524 | goto skip; | |
4525 | ||
803b2f54 SB |
4526 | read_extent_buffer(eb, &root_item, |
4527 | btrfs_item_ptr_offset(eb, slot), | |
4528 | (int)sizeof(root_item)); | |
4529 | if (btrfs_root_refs(&root_item) == 0) | |
4530 | goto skip; | |
f45388f3 FDBM |
4531 | |
4532 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4533 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4534 | if (trans) | |
4535 | goto update_tree; | |
4536 | ||
4537 | btrfs_release_path(path); | |
803b2f54 SB |
4538 | /* |
4539 | * 1 - subvol uuid item | |
4540 | * 1 - received_subvol uuid item | |
4541 | */ | |
4542 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4543 | if (IS_ERR(trans)) { | |
4544 | ret = PTR_ERR(trans); | |
4545 | break; | |
4546 | } | |
f45388f3 FDBM |
4547 | continue; |
4548 | } else { | |
4549 | goto skip; | |
4550 | } | |
4551 | update_tree: | |
9771a5cf | 4552 | btrfs_release_path(path); |
f45388f3 | 4553 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4554 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4555 | BTRFS_UUID_KEY_SUBVOL, |
4556 | key.objectid); | |
4557 | if (ret < 0) { | |
efe120a0 | 4558 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4559 | ret); |
803b2f54 SB |
4560 | break; |
4561 | } | |
4562 | } | |
4563 | ||
4564 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4565 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4566 | root_item.received_uuid, |
4567 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4568 | key.objectid); | |
4569 | if (ret < 0) { | |
efe120a0 | 4570 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4571 | ret); |
803b2f54 SB |
4572 | break; |
4573 | } | |
4574 | } | |
4575 | ||
f45388f3 | 4576 | skip: |
9771a5cf | 4577 | btrfs_release_path(path); |
803b2f54 | 4578 | if (trans) { |
3a45bb20 | 4579 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4580 | trans = NULL; |
803b2f54 SB |
4581 | if (ret) |
4582 | break; | |
4583 | } | |
4584 | ||
803b2f54 SB |
4585 | if (key.offset < (u64)-1) { |
4586 | key.offset++; | |
4587 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4588 | key.offset = 0; | |
4589 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4590 | } else if (key.objectid < (u64)-1) { | |
4591 | key.offset = 0; | |
4592 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4593 | key.objectid++; | |
4594 | } else { | |
4595 | break; | |
4596 | } | |
4597 | cond_resched(); | |
4598 | } | |
4599 | ||
4600 | out: | |
4601 | btrfs_free_path(path); | |
f45388f3 | 4602 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4603 | btrfs_end_transaction(trans); |
803b2f54 | 4604 | if (ret) |
efe120a0 | 4605 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4606 | else if (!closing) |
afcdd129 | 4607 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4608 | up(&fs_info->uuid_tree_rescan_sem); |
4609 | return 0; | |
4610 | } | |
4611 | ||
f7a81ea4 SB |
4612 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4613 | { | |
4614 | struct btrfs_trans_handle *trans; | |
4615 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4616 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4617 | struct task_struct *task; |
4618 | int ret; | |
f7a81ea4 SB |
4619 | |
4620 | /* | |
4621 | * 1 - root node | |
4622 | * 1 - root item | |
4623 | */ | |
4624 | trans = btrfs_start_transaction(tree_root, 2); | |
4625 | if (IS_ERR(trans)) | |
4626 | return PTR_ERR(trans); | |
4627 | ||
9b7a2440 | 4628 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4629 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4630 | ret = PTR_ERR(uuid_root); |
66642832 | 4631 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4632 | btrfs_end_transaction(trans); |
6d13f549 | 4633 | return ret; |
f7a81ea4 SB |
4634 | } |
4635 | ||
4636 | fs_info->uuid_root = uuid_root; | |
4637 | ||
3a45bb20 | 4638 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4639 | if (ret) |
4640 | return ret; | |
4641 | ||
4642 | down(&fs_info->uuid_tree_rescan_sem); | |
4643 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4644 | if (IS_ERR(task)) { | |
70f80175 | 4645 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4646 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4647 | up(&fs_info->uuid_tree_rescan_sem); |
4648 | return PTR_ERR(task); | |
4649 | } | |
4650 | ||
4651 | return 0; | |
f7a81ea4 | 4652 | } |
803b2f54 | 4653 | |
8f18cf13 CM |
4654 | /* |
4655 | * shrinking a device means finding all of the device extents past | |
4656 | * the new size, and then following the back refs to the chunks. | |
4657 | * The chunk relocation code actually frees the device extent | |
4658 | */ | |
4659 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4660 | { | |
0b246afa JM |
4661 | struct btrfs_fs_info *fs_info = device->fs_info; |
4662 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4663 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4664 | struct btrfs_dev_extent *dev_extent = NULL; |
4665 | struct btrfs_path *path; | |
4666 | u64 length; | |
8f18cf13 CM |
4667 | u64 chunk_offset; |
4668 | int ret; | |
4669 | int slot; | |
ba1bf481 JB |
4670 | int failed = 0; |
4671 | bool retried = false; | |
8f18cf13 CM |
4672 | struct extent_buffer *l; |
4673 | struct btrfs_key key; | |
0b246afa | 4674 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4675 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4676 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4677 | u64 diff; |
61d0d0d2 | 4678 | u64 start; |
7dfb8be1 NB |
4679 | |
4680 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4681 | start = new_size; |
0e4324a4 | 4682 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4683 | |
401e29c1 | 4684 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4685 | return -EINVAL; |
4686 | ||
8f18cf13 CM |
4687 | path = btrfs_alloc_path(); |
4688 | if (!path) | |
4689 | return -ENOMEM; | |
4690 | ||
0338dff6 | 4691 | path->reada = READA_BACK; |
8f18cf13 | 4692 | |
61d0d0d2 NB |
4693 | trans = btrfs_start_transaction(root, 0); |
4694 | if (IS_ERR(trans)) { | |
4695 | btrfs_free_path(path); | |
4696 | return PTR_ERR(trans); | |
4697 | } | |
4698 | ||
34441361 | 4699 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4700 | |
7cc8e58d | 4701 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4702 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4703 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4704 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4705 | } |
61d0d0d2 NB |
4706 | |
4707 | /* | |
4708 | * Once the device's size has been set to the new size, ensure all | |
4709 | * in-memory chunks are synced to disk so that the loop below sees them | |
4710 | * and relocates them accordingly. | |
4711 | */ | |
1c11b63e | 4712 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4713 | mutex_unlock(&fs_info->chunk_mutex); |
4714 | ret = btrfs_commit_transaction(trans); | |
4715 | if (ret) | |
4716 | goto done; | |
4717 | } else { | |
4718 | mutex_unlock(&fs_info->chunk_mutex); | |
4719 | btrfs_end_transaction(trans); | |
4720 | } | |
8f18cf13 | 4721 | |
ba1bf481 | 4722 | again: |
8f18cf13 CM |
4723 | key.objectid = device->devid; |
4724 | key.offset = (u64)-1; | |
4725 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4726 | ||
213e64da | 4727 | do { |
0b246afa | 4728 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4729 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4730 | if (ret < 0) { |
0b246afa | 4731 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4732 | goto done; |
67c5e7d4 | 4733 | } |
8f18cf13 CM |
4734 | |
4735 | ret = btrfs_previous_item(root, path, 0, key.type); | |
8f18cf13 | 4736 | if (ret) { |
7056bf69 NB |
4737 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
4738 | if (ret < 0) | |
4739 | goto done; | |
8f18cf13 | 4740 | ret = 0; |
b3b4aa74 | 4741 | btrfs_release_path(path); |
bf1fb512 | 4742 | break; |
8f18cf13 CM |
4743 | } |
4744 | ||
4745 | l = path->nodes[0]; | |
4746 | slot = path->slots[0]; | |
4747 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4748 | ||
ba1bf481 | 4749 | if (key.objectid != device->devid) { |
0b246afa | 4750 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4751 | btrfs_release_path(path); |
bf1fb512 | 4752 | break; |
ba1bf481 | 4753 | } |
8f18cf13 CM |
4754 | |
4755 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4756 | length = btrfs_dev_extent_length(l, dev_extent); | |
4757 | ||
ba1bf481 | 4758 | if (key.offset + length <= new_size) { |
0b246afa | 4759 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4760 | btrfs_release_path(path); |
d6397bae | 4761 | break; |
ba1bf481 | 4762 | } |
8f18cf13 | 4763 | |
8f18cf13 | 4764 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4765 | btrfs_release_path(path); |
8f18cf13 | 4766 | |
a6f93c71 LB |
4767 | /* |
4768 | * We may be relocating the only data chunk we have, | |
4769 | * which could potentially end up with losing data's | |
4770 | * raid profile, so lets allocate an empty one in | |
4771 | * advance. | |
4772 | */ | |
4773 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4774 | if (ret < 0) { | |
4775 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4776 | goto done; | |
4777 | } | |
4778 | ||
0b246afa JM |
4779 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4780 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4781 | if (ret == -ENOSPC) { |
ba1bf481 | 4782 | failed++; |
eede2bf3 OS |
4783 | } else if (ret) { |
4784 | if (ret == -ETXTBSY) { | |
4785 | btrfs_warn(fs_info, | |
4786 | "could not shrink block group %llu due to active swapfile", | |
4787 | chunk_offset); | |
4788 | } | |
4789 | goto done; | |
4790 | } | |
213e64da | 4791 | } while (key.offset-- > 0); |
ba1bf481 JB |
4792 | |
4793 | if (failed && !retried) { | |
4794 | failed = 0; | |
4795 | retried = true; | |
4796 | goto again; | |
4797 | } else if (failed && retried) { | |
4798 | ret = -ENOSPC; | |
ba1bf481 | 4799 | goto done; |
8f18cf13 CM |
4800 | } |
4801 | ||
d6397bae | 4802 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4803 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4804 | if (IS_ERR(trans)) { |
4805 | ret = PTR_ERR(trans); | |
4806 | goto done; | |
4807 | } | |
4808 | ||
34441361 | 4809 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4810 | /* Clear all state bits beyond the shrunk device size */ |
4811 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4812 | CHUNK_STATE_MASK); | |
4813 | ||
7cc8e58d | 4814 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4815 | if (list_empty(&device->post_commit_list)) |
4816 | list_add_tail(&device->post_commit_list, | |
4817 | &trans->transaction->dev_update_list); | |
d6397bae | 4818 | |
d6397bae | 4819 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4820 | btrfs_set_super_total_bytes(super_copy, |
4821 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4822 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4823 | |
4824 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4825 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4826 | if (ret < 0) { |
4827 | btrfs_abort_transaction(trans, ret); | |
4828 | btrfs_end_transaction(trans); | |
4829 | } else { | |
4830 | ret = btrfs_commit_transaction(trans); | |
4831 | } | |
8f18cf13 CM |
4832 | done: |
4833 | btrfs_free_path(path); | |
53e489bc | 4834 | if (ret) { |
34441361 | 4835 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4836 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4837 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4838 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4839 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4840 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4841 | } |
8f18cf13 CM |
4842 | return ret; |
4843 | } | |
4844 | ||
2ff7e61e | 4845 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4846 | struct btrfs_key *key, |
4847 | struct btrfs_chunk *chunk, int item_size) | |
4848 | { | |
0b246afa | 4849 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4850 | struct btrfs_disk_key disk_key; |
4851 | u32 array_size; | |
4852 | u8 *ptr; | |
4853 | ||
34441361 | 4854 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4855 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4856 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4857 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4858 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4859 | return -EFBIG; |
fe48a5c0 | 4860 | } |
0b86a832 CM |
4861 | |
4862 | ptr = super_copy->sys_chunk_array + array_size; | |
4863 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4864 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4865 | ptr += sizeof(disk_key); | |
4866 | memcpy(ptr, chunk, item_size); | |
4867 | item_size += sizeof(disk_key); | |
4868 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4869 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4870 | |
0b86a832 CM |
4871 | return 0; |
4872 | } | |
4873 | ||
73c5de00 AJ |
4874 | /* |
4875 | * sort the devices in descending order by max_avail, total_avail | |
4876 | */ | |
4877 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4878 | { |
73c5de00 AJ |
4879 | const struct btrfs_device_info *di_a = a; |
4880 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4881 | |
73c5de00 | 4882 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4883 | return -1; |
73c5de00 | 4884 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4885 | return 1; |
73c5de00 AJ |
4886 | if (di_a->total_avail > di_b->total_avail) |
4887 | return -1; | |
4888 | if (di_a->total_avail < di_b->total_avail) | |
4889 | return 1; | |
4890 | return 0; | |
b2117a39 | 4891 | } |
0b86a832 | 4892 | |
53b381b3 DW |
4893 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4894 | { | |
ffe2d203 | 4895 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4896 | return; |
4897 | ||
ceda0864 | 4898 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4899 | } |
4900 | ||
cfbb825c DS |
4901 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4902 | { | |
4903 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4904 | return; | |
4905 | ||
4906 | btrfs_set_fs_incompat(info, RAID1C34); | |
4907 | } | |
4908 | ||
4f2bafe8 NA |
4909 | /* |
4910 | * Structure used internally for __btrfs_alloc_chunk() function. | |
4911 | * Wraps needed parameters. | |
4912 | */ | |
4913 | struct alloc_chunk_ctl { | |
4914 | u64 start; | |
4915 | u64 type; | |
4916 | /* Total number of stripes to allocate */ | |
4917 | int num_stripes; | |
4918 | /* sub_stripes info for map */ | |
4919 | int sub_stripes; | |
4920 | /* Stripes per device */ | |
4921 | int dev_stripes; | |
4922 | /* Maximum number of devices to use */ | |
4923 | int devs_max; | |
4924 | /* Minimum number of devices to use */ | |
4925 | int devs_min; | |
4926 | /* ndevs has to be a multiple of this */ | |
4927 | int devs_increment; | |
4928 | /* Number of copies */ | |
4929 | int ncopies; | |
4930 | /* Number of stripes worth of bytes to store parity information */ | |
4931 | int nparity; | |
4932 | u64 max_stripe_size; | |
4933 | u64 max_chunk_size; | |
6aafb303 | 4934 | u64 dev_extent_min; |
4f2bafe8 NA |
4935 | u64 stripe_size; |
4936 | u64 chunk_size; | |
4937 | int ndevs; | |
4938 | }; | |
4939 | ||
27c314d5 NA |
4940 | static void init_alloc_chunk_ctl_policy_regular( |
4941 | struct btrfs_fs_devices *fs_devices, | |
4942 | struct alloc_chunk_ctl *ctl) | |
4943 | { | |
4944 | u64 type = ctl->type; | |
4945 | ||
4946 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4947 | ctl->max_stripe_size = SZ_1G; | |
4948 | ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; | |
4949 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4950 | /* For larger filesystems, use larger metadata chunks */ | |
4951 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) | |
4952 | ctl->max_stripe_size = SZ_1G; | |
4953 | else | |
4954 | ctl->max_stripe_size = SZ_256M; | |
4955 | ctl->max_chunk_size = ctl->max_stripe_size; | |
4956 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4957 | ctl->max_stripe_size = SZ_32M; | |
4958 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
4959 | ctl->devs_max = min_t(int, ctl->devs_max, | |
4960 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
4961 | } else { | |
4962 | BUG(); | |
4963 | } | |
4964 | ||
4965 | /* We don't want a chunk larger than 10% of writable space */ | |
4966 | ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4967 | ctl->max_chunk_size); | |
6aafb303 | 4968 | ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; |
27c314d5 NA |
4969 | } |
4970 | ||
1cd6121f NA |
4971 | static void init_alloc_chunk_ctl_policy_zoned( |
4972 | struct btrfs_fs_devices *fs_devices, | |
4973 | struct alloc_chunk_ctl *ctl) | |
4974 | { | |
4975 | u64 zone_size = fs_devices->fs_info->zone_size; | |
4976 | u64 limit; | |
4977 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; | |
4978 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; | |
4979 | u64 min_chunk_size = min_data_stripes * zone_size; | |
4980 | u64 type = ctl->type; | |
4981 | ||
4982 | ctl->max_stripe_size = zone_size; | |
4983 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4984 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, | |
4985 | zone_size); | |
4986 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4987 | ctl->max_chunk_size = ctl->max_stripe_size; | |
4988 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4989 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
4990 | ctl->devs_max = min_t(int, ctl->devs_max, | |
4991 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
4992 | } | |
4993 | ||
4994 | /* We don't want a chunk larger than 10% of writable space */ | |
4995 | limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), | |
4996 | zone_size), | |
4997 | min_chunk_size); | |
4998 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); | |
4999 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; | |
5000 | } | |
5001 | ||
27c314d5 NA |
5002 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
5003 | struct alloc_chunk_ctl *ctl) | |
5004 | { | |
5005 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
5006 | ||
5007 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
5008 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5009 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
5010 | if (!ctl->devs_max) | |
5011 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
5012 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
5013 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
5014 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
5015 | ctl->nparity = btrfs_raid_array[index].nparity; | |
5016 | ctl->ndevs = 0; | |
5017 | ||
5018 | switch (fs_devices->chunk_alloc_policy) { | |
5019 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5020 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
5021 | break; | |
1cd6121f NA |
5022 | case BTRFS_CHUNK_ALLOC_ZONED: |
5023 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); | |
5024 | break; | |
27c314d5 NA |
5025 | default: |
5026 | BUG(); | |
5027 | } | |
5028 | } | |
5029 | ||
560156cb NA |
5030 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
5031 | struct alloc_chunk_ctl *ctl, | |
5032 | struct btrfs_device_info *devices_info) | |
b2117a39 | 5033 | { |
560156cb | 5034 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 5035 | struct btrfs_device *device; |
73c5de00 | 5036 | u64 total_avail; |
560156cb | 5037 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 5038 | int ret; |
560156cb NA |
5039 | int ndevs = 0; |
5040 | u64 max_avail; | |
5041 | u64 dev_offset; | |
0cad8a11 | 5042 | |
9f680ce0 | 5043 | /* |
73c5de00 AJ |
5044 | * in the first pass through the devices list, we gather information |
5045 | * about the available holes on each device. | |
9f680ce0 | 5046 | */ |
ebcc9301 | 5047 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 5048 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5049 | WARN(1, KERN_ERR |
efe120a0 | 5050 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5051 | continue; |
5052 | } | |
b2117a39 | 5053 | |
e12c9621 AJ |
5054 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5055 | &device->dev_state) || | |
401e29c1 | 5056 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5057 | continue; |
b2117a39 | 5058 | |
73c5de00 AJ |
5059 | if (device->total_bytes > device->bytes_used) |
5060 | total_avail = device->total_bytes - device->bytes_used; | |
5061 | else | |
5062 | total_avail = 0; | |
38c01b96 | 5063 | |
5064 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 5065 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 5066 | continue; |
b2117a39 | 5067 | |
560156cb NA |
5068 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
5069 | &max_avail); | |
73c5de00 | 5070 | if (ret && ret != -ENOSPC) |
560156cb | 5071 | return ret; |
b2117a39 | 5072 | |
73c5de00 | 5073 | if (ret == 0) |
560156cb | 5074 | max_avail = dev_extent_want; |
b2117a39 | 5075 | |
6aafb303 | 5076 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
5077 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
5078 | btrfs_debug(info, | |
560156cb | 5079 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 5080 | __func__, device->devid, max_avail, |
6aafb303 | 5081 | ctl->dev_extent_min); |
73c5de00 | 5082 | continue; |
4117f207 | 5083 | } |
b2117a39 | 5084 | |
063d006f ES |
5085 | if (ndevs == fs_devices->rw_devices) { |
5086 | WARN(1, "%s: found more than %llu devices\n", | |
5087 | __func__, fs_devices->rw_devices); | |
5088 | break; | |
5089 | } | |
73c5de00 AJ |
5090 | devices_info[ndevs].dev_offset = dev_offset; |
5091 | devices_info[ndevs].max_avail = max_avail; | |
5092 | devices_info[ndevs].total_avail = total_avail; | |
5093 | devices_info[ndevs].dev = device; | |
5094 | ++ndevs; | |
5095 | } | |
560156cb | 5096 | ctl->ndevs = ndevs; |
b2117a39 | 5097 | |
73c5de00 AJ |
5098 | /* |
5099 | * now sort the devices by hole size / available space | |
5100 | */ | |
560156cb | 5101 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 5102 | btrfs_cmp_device_info, NULL); |
b2117a39 | 5103 | |
560156cb NA |
5104 | return 0; |
5105 | } | |
5106 | ||
5badf512 NA |
5107 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
5108 | struct btrfs_device_info *devices_info) | |
5109 | { | |
5110 | /* Number of stripes that count for block group size */ | |
5111 | int data_stripes; | |
5112 | ||
5113 | /* | |
5114 | * The primary goal is to maximize the number of stripes, so use as | |
5115 | * many devices as possible, even if the stripes are not maximum sized. | |
5116 | * | |
5117 | * The DUP profile stores more than one stripe per device, the | |
5118 | * max_avail is the total size so we have to adjust. | |
5119 | */ | |
5120 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5121 | ctl->dev_stripes); | |
5122 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5123 | ||
5124 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5125 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5126 | ||
5127 | /* | |
5128 | * Use the number of data stripes to figure out how big this chunk is | |
5129 | * really going to be in terms of logical address space, and compare | |
5130 | * that answer with the max chunk size. If it's higher, we try to | |
5131 | * reduce stripe_size. | |
5132 | */ | |
5133 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5134 | /* | |
5135 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5136 | * then use it, unless it ends up being even bigger than the | |
5137 | * previous value we had already. | |
5138 | */ | |
5139 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5140 | data_stripes), SZ_16M), | |
5141 | ctl->stripe_size); | |
5142 | } | |
5143 | ||
5144 | /* Align to BTRFS_STRIPE_LEN */ | |
5145 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5146 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5147 | ||
5148 | return 0; | |
5149 | } | |
5150 | ||
1cd6121f NA |
5151 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
5152 | struct btrfs_device_info *devices_info) | |
5153 | { | |
5154 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; | |
5155 | /* Number of stripes that count for block group size */ | |
5156 | int data_stripes; | |
5157 | ||
5158 | /* | |
5159 | * It should hold because: | |
5160 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes | |
5161 | */ | |
5162 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); | |
5163 | ||
5164 | ctl->stripe_size = zone_size; | |
5165 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5166 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5167 | ||
5168 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ | |
5169 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5170 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, | |
5171 | ctl->stripe_size) + ctl->nparity, | |
5172 | ctl->dev_stripes); | |
5173 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5174 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5175 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); | |
5176 | } | |
5177 | ||
5178 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5179 | ||
5180 | return 0; | |
5181 | } | |
5182 | ||
5badf512 NA |
5183 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
5184 | struct alloc_chunk_ctl *ctl, | |
5185 | struct btrfs_device_info *devices_info) | |
5186 | { | |
5187 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5188 | ||
5189 | /* | |
5190 | * Round down to number of usable stripes, devs_increment can be any | |
5191 | * number so we can't use round_down() that requires power of 2, while | |
5192 | * rounddown is safe. | |
5193 | */ | |
5194 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5195 | ||
5196 | if (ctl->ndevs < ctl->devs_min) { | |
5197 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5198 | btrfs_debug(info, | |
5199 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5200 | __func__, ctl->ndevs, ctl->devs_min); | |
5201 | } | |
5202 | return -ENOSPC; | |
5203 | } | |
5204 | ||
5205 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5206 | ||
5207 | switch (fs_devices->chunk_alloc_policy) { | |
5208 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5209 | return decide_stripe_size_regular(ctl, devices_info); | |
1cd6121f NA |
5210 | case BTRFS_CHUNK_ALLOC_ZONED: |
5211 | return decide_stripe_size_zoned(ctl, devices_info); | |
5badf512 NA |
5212 | default: |
5213 | BUG(); | |
5214 | } | |
5215 | } | |
5216 | ||
dce580ca NA |
5217 | static int create_chunk(struct btrfs_trans_handle *trans, |
5218 | struct alloc_chunk_ctl *ctl, | |
5219 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5220 | { |
5221 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5222 | struct map_lookup *map = NULL; |
5223 | struct extent_map_tree *em_tree; | |
5224 | struct extent_map *em; | |
dce580ca NA |
5225 | u64 start = ctl->start; |
5226 | u64 type = ctl->type; | |
560156cb NA |
5227 | int ret; |
5228 | int i; | |
5229 | int j; | |
5230 | ||
dce580ca NA |
5231 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5232 | if (!map) | |
560156cb | 5233 | return -ENOMEM; |
dce580ca | 5234 | map->num_stripes = ctl->num_stripes; |
560156cb | 5235 | |
dce580ca NA |
5236 | for (i = 0; i < ctl->ndevs; ++i) { |
5237 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5238 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5239 | map->stripes[s].dev = devices_info[i].dev; |
5240 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5241 | j * ctl->stripe_size; |
6324fbf3 | 5242 | } |
6324fbf3 | 5243 | } |
500ceed8 NB |
5244 | map->stripe_len = BTRFS_STRIPE_LEN; |
5245 | map->io_align = BTRFS_STRIPE_LEN; | |
5246 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5247 | map->type = type; |
dce580ca | 5248 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5249 | |
dce580ca | 5250 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5251 | |
172ddd60 | 5252 | em = alloc_extent_map(); |
2b82032c | 5253 | if (!em) { |
298a8f9c | 5254 | kfree(map); |
dce580ca | 5255 | return -ENOMEM; |
593060d7 | 5256 | } |
298a8f9c | 5257 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5258 | em->map_lookup = map; |
2b82032c | 5259 | em->start = start; |
dce580ca | 5260 | em->len = ctl->chunk_size; |
2b82032c YZ |
5261 | em->block_start = 0; |
5262 | em->block_len = em->len; | |
dce580ca | 5263 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5264 | |
c8bf1b67 | 5265 | em_tree = &info->mapping_tree; |
890871be | 5266 | write_lock(&em_tree->lock); |
09a2a8f9 | 5267 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5268 | if (ret) { |
1efb72a3 | 5269 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5270 | free_extent_map(em); |
dce580ca | 5271 | return ret; |
0f5d42b2 | 5272 | } |
1efb72a3 NB |
5273 | write_unlock(&em_tree->lock); |
5274 | ||
dce580ca | 5275 | ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); |
6df9a95e JB |
5276 | if (ret) |
5277 | goto error_del_extent; | |
2b82032c | 5278 | |
bbbf7243 NB |
5279 | for (i = 0; i < map->num_stripes; i++) { |
5280 | struct btrfs_device *dev = map->stripes[i].dev; | |
5281 | ||
4f2bafe8 | 5282 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5283 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5284 | if (list_empty(&dev->post_commit_list)) |
5285 | list_add_tail(&dev->post_commit_list, | |
5286 | &trans->transaction->dev_update_list); | |
5287 | } | |
43530c46 | 5288 | |
dce580ca | 5289 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5290 | &info->free_chunk_space); |
1c116187 | 5291 | |
0f5d42b2 | 5292 | free_extent_map(em); |
0b246afa | 5293 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5294 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5295 | |
2b82032c | 5296 | return 0; |
b2117a39 | 5297 | |
6df9a95e | 5298 | error_del_extent: |
0f5d42b2 JB |
5299 | write_lock(&em_tree->lock); |
5300 | remove_extent_mapping(em_tree, em); | |
5301 | write_unlock(&em_tree->lock); | |
5302 | ||
5303 | /* One for our allocation */ | |
5304 | free_extent_map(em); | |
5305 | /* One for the tree reference */ | |
5306 | free_extent_map(em); | |
dce580ca NA |
5307 | |
5308 | return ret; | |
5309 | } | |
5310 | ||
11c67b1a | 5311 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
dce580ca NA |
5312 | { |
5313 | struct btrfs_fs_info *info = trans->fs_info; | |
5314 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5315 | struct btrfs_device_info *devices_info = NULL; | |
5316 | struct alloc_chunk_ctl ctl; | |
5317 | int ret; | |
5318 | ||
11c67b1a NB |
5319 | lockdep_assert_held(&info->chunk_mutex); |
5320 | ||
dce580ca NA |
5321 | if (!alloc_profile_is_valid(type, 0)) { |
5322 | ASSERT(0); | |
5323 | return -EINVAL; | |
5324 | } | |
5325 | ||
5326 | if (list_empty(&fs_devices->alloc_list)) { | |
5327 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5328 | btrfs_debug(info, "%s: no writable device", __func__); | |
5329 | return -ENOSPC; | |
5330 | } | |
5331 | ||
5332 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5333 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5334 | ASSERT(0); | |
5335 | return -EINVAL; | |
5336 | } | |
5337 | ||
11c67b1a | 5338 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5339 | ctl.type = type; |
5340 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5341 | ||
5342 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5343 | GFP_NOFS); | |
5344 | if (!devices_info) | |
5345 | return -ENOMEM; | |
5346 | ||
5347 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
5348 | if (ret < 0) | |
5349 | goto out; | |
5350 | ||
5351 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
5352 | if (ret < 0) | |
5353 | goto out; | |
5354 | ||
5355 | ret = create_chunk(trans, &ctl, devices_info); | |
5356 | ||
5357 | out: | |
b2117a39 MX |
5358 | kfree(devices_info); |
5359 | return ret; | |
2b82032c YZ |
5360 | } |
5361 | ||
11c67b1a NB |
5362 | /* |
5363 | * Chunk allocation falls into two parts. The first part does work | |
5364 | * that makes the new allocated chunk usable, but does not do any operation | |
5365 | * that modifies the chunk tree. The second part does the work that | |
5366 | * requires modifying the chunk tree. This division is important for the | |
5367 | * bootstrap process of adding storage to a seed btrfs. | |
5368 | */ | |
6df9a95e | 5369 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5370 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5371 | { |
97aff912 | 5372 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5373 | struct btrfs_root *extent_root = fs_info->extent_root; |
5374 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5375 | struct btrfs_key key; |
2b82032c YZ |
5376 | struct btrfs_device *device; |
5377 | struct btrfs_chunk *chunk; | |
5378 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5379 | struct extent_map *em; |
5380 | struct map_lookup *map; | |
5381 | size_t item_size; | |
5382 | u64 dev_offset; | |
5383 | u64 stripe_size; | |
5384 | int i = 0; | |
140e639f | 5385 | int ret = 0; |
2b82032c | 5386 | |
60ca842e | 5387 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5388 | if (IS_ERR(em)) |
5389 | return PTR_ERR(em); | |
6df9a95e | 5390 | |
95617d69 | 5391 | map = em->map_lookup; |
6df9a95e JB |
5392 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5393 | stripe_size = em->orig_block_len; | |
5394 | ||
2b82032c | 5395 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5396 | if (!chunk) { |
5397 | ret = -ENOMEM; | |
5398 | goto out; | |
5399 | } | |
5400 | ||
50460e37 FM |
5401 | /* |
5402 | * Take the device list mutex to prevent races with the final phase of | |
5403 | * a device replace operation that replaces the device object associated | |
5404 | * with the map's stripes, because the device object's id can change | |
5405 | * at any time during that final phase of the device replace operation | |
5406 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5407 | */ | |
0b246afa | 5408 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5409 | for (i = 0; i < map->num_stripes; i++) { |
5410 | device = map->stripes[i].dev; | |
5411 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5412 | |
0b86a832 | 5413 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5414 | if (ret) |
50460e37 | 5415 | break; |
b5d9071c NB |
5416 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5417 | dev_offset, stripe_size); | |
6df9a95e | 5418 | if (ret) |
50460e37 FM |
5419 | break; |
5420 | } | |
5421 | if (ret) { | |
0b246afa | 5422 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5423 | goto out; |
2b82032c YZ |
5424 | } |
5425 | ||
2b82032c | 5426 | stripe = &chunk->stripe; |
6df9a95e JB |
5427 | for (i = 0; i < map->num_stripes; i++) { |
5428 | device = map->stripes[i].dev; | |
5429 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5430 | |
e17cade2 CM |
5431 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5432 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5433 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5434 | stripe++; |
0b86a832 | 5435 | } |
0b246afa | 5436 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5437 | |
2b82032c | 5438 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5439 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5440 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5441 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5442 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5443 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5444 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5445 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5446 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5447 | |
2b82032c YZ |
5448 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5449 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5450 | key.offset = chunk_offset; | |
0b86a832 | 5451 | |
2b82032c | 5452 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5453 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5454 | /* | |
5455 | * TODO: Cleanup of inserted chunk root in case of | |
5456 | * failure. | |
5457 | */ | |
2ff7e61e | 5458 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5459 | } |
1abe9b8a | 5460 | |
6df9a95e | 5461 | out: |
0b86a832 | 5462 | kfree(chunk); |
6df9a95e | 5463 | free_extent_map(em); |
4ed1d16e | 5464 | return ret; |
2b82032c | 5465 | } |
0b86a832 | 5466 | |
6f8e0fc7 | 5467 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5468 | { |
6f8e0fc7 | 5469 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5470 | u64 alloc_profile; |
2b82032c YZ |
5471 | int ret; |
5472 | ||
1b86826d | 5473 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
11c67b1a | 5474 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa JM |
5475 | if (ret) |
5476 | return ret; | |
2b82032c | 5477 | |
1b86826d | 5478 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
11c67b1a | 5479 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa | 5480 | return ret; |
2b82032c YZ |
5481 | } |
5482 | ||
d20983b4 MX |
5483 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5484 | { | |
fc9a2ac7 | 5485 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5486 | |
fc9a2ac7 | 5487 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5488 | } |
5489 | ||
2ff7e61e | 5490 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5491 | { |
5492 | struct extent_map *em; | |
5493 | struct map_lookup *map; | |
2b82032c | 5494 | int readonly = 0; |
d20983b4 | 5495 | int miss_ndevs = 0; |
2b82032c YZ |
5496 | int i; |
5497 | ||
60ca842e | 5498 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5499 | if (IS_ERR(em)) |
2b82032c YZ |
5500 | return 1; |
5501 | ||
95617d69 | 5502 | map = em->map_lookup; |
2b82032c | 5503 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5504 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5505 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5506 | miss_ndevs++; |
5507 | continue; | |
5508 | } | |
ebbede42 AJ |
5509 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5510 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5511 | readonly = 1; |
d20983b4 | 5512 | goto end; |
2b82032c YZ |
5513 | } |
5514 | } | |
d20983b4 MX |
5515 | |
5516 | /* | |
5517 | * If the number of missing devices is larger than max errors, | |
5518 | * we can not write the data into that chunk successfully, so | |
5519 | * set it readonly. | |
5520 | */ | |
5521 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5522 | readonly = 1; | |
5523 | end: | |
0b86a832 | 5524 | free_extent_map(em); |
2b82032c | 5525 | return readonly; |
0b86a832 CM |
5526 | } |
5527 | ||
c8bf1b67 | 5528 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5529 | { |
5530 | struct extent_map *em; | |
5531 | ||
d397712b | 5532 | while (1) { |
c8bf1b67 DS |
5533 | write_lock(&tree->lock); |
5534 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5535 | if (em) |
c8bf1b67 DS |
5536 | remove_extent_mapping(tree, em); |
5537 | write_unlock(&tree->lock); | |
0b86a832 CM |
5538 | if (!em) |
5539 | break; | |
0b86a832 CM |
5540 | /* once for us */ |
5541 | free_extent_map(em); | |
5542 | /* once for the tree */ | |
5543 | free_extent_map(em); | |
5544 | } | |
5545 | } | |
5546 | ||
5d964051 | 5547 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5548 | { |
5549 | struct extent_map *em; | |
5550 | struct map_lookup *map; | |
f188591e CM |
5551 | int ret; |
5552 | ||
60ca842e | 5553 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5554 | if (IS_ERR(em)) |
5555 | /* | |
5556 | * We could return errors for these cases, but that could get | |
5557 | * ugly and we'd probably do the same thing which is just not do | |
5558 | * anything else and exit, so return 1 so the callers don't try | |
5559 | * to use other copies. | |
5560 | */ | |
fb7669b5 | 5561 | return 1; |
fb7669b5 | 5562 | |
95617d69 | 5563 | map = em->map_lookup; |
c7369b3f | 5564 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5565 | ret = map->num_stripes; |
321aecc6 CM |
5566 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5567 | ret = map->sub_stripes; | |
53b381b3 DW |
5568 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5569 | ret = 2; | |
5570 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5571 | /* |
5572 | * There could be two corrupted data stripes, we need | |
5573 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5574 | * |
8810f751 LB |
5575 | * Fail a stripe at a time on every retry except the |
5576 | * stripe under reconstruction. | |
5577 | */ | |
5578 | ret = map->num_stripes; | |
f188591e CM |
5579 | else |
5580 | ret = 1; | |
5581 | free_extent_map(em); | |
ad6d620e | 5582 | |
cb5583dd | 5583 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5584 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5585 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5586 | ret++; |
cb5583dd | 5587 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5588 | |
f188591e CM |
5589 | return ret; |
5590 | } | |
5591 | ||
2ff7e61e | 5592 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5593 | u64 logical) |
5594 | { | |
5595 | struct extent_map *em; | |
5596 | struct map_lookup *map; | |
0b246afa | 5597 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5598 | |
60ca842e | 5599 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5600 | |
69f03f13 NB |
5601 | if (!WARN_ON(IS_ERR(em))) { |
5602 | map = em->map_lookup; | |
5603 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5604 | len = map->stripe_len * nr_data_stripes(map); | |
5605 | free_extent_map(em); | |
5606 | } | |
53b381b3 DW |
5607 | return len; |
5608 | } | |
5609 | ||
e4ff5fb5 | 5610 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5611 | { |
5612 | struct extent_map *em; | |
5613 | struct map_lookup *map; | |
53b381b3 DW |
5614 | int ret = 0; |
5615 | ||
60ca842e | 5616 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5617 | |
69f03f13 NB |
5618 | if(!WARN_ON(IS_ERR(em))) { |
5619 | map = em->map_lookup; | |
5620 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5621 | ret = 1; | |
5622 | free_extent_map(em); | |
5623 | } | |
53b381b3 DW |
5624 | return ret; |
5625 | } | |
5626 | ||
30d9861f | 5627 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5628 | struct map_lookup *map, int first, |
8ba0ae78 | 5629 | int dev_replace_is_ongoing) |
dfe25020 CM |
5630 | { |
5631 | int i; | |
99f92a7c | 5632 | int num_stripes; |
8ba0ae78 | 5633 | int preferred_mirror; |
30d9861f SB |
5634 | int tolerance; |
5635 | struct btrfs_device *srcdev; | |
5636 | ||
99f92a7c | 5637 | ASSERT((map->type & |
c7369b3f | 5638 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5639 | |
5640 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5641 | num_stripes = map->sub_stripes; | |
5642 | else | |
5643 | num_stripes = map->num_stripes; | |
5644 | ||
33fd2f71 AJ |
5645 | switch (fs_info->fs_devices->read_policy) { |
5646 | default: | |
5647 | /* Shouldn't happen, just warn and use pid instead of failing */ | |
5648 | btrfs_warn_rl(fs_info, | |
5649 | "unknown read_policy type %u, reset to pid", | |
5650 | fs_info->fs_devices->read_policy); | |
5651 | fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; | |
5652 | fallthrough; | |
5653 | case BTRFS_READ_POLICY_PID: | |
5654 | preferred_mirror = first + (current->pid % num_stripes); | |
5655 | break; | |
5656 | } | |
8ba0ae78 | 5657 | |
30d9861f SB |
5658 | if (dev_replace_is_ongoing && |
5659 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5660 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5661 | srcdev = fs_info->dev_replace.srcdev; | |
5662 | else | |
5663 | srcdev = NULL; | |
5664 | ||
5665 | /* | |
5666 | * try to avoid the drive that is the source drive for a | |
5667 | * dev-replace procedure, only choose it if no other non-missing | |
5668 | * mirror is available | |
5669 | */ | |
5670 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5671 | if (map->stripes[preferred_mirror].dev->bdev && |
5672 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5673 | return preferred_mirror; | |
99f92a7c | 5674 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5675 | if (map->stripes[i].dev->bdev && |
5676 | (tolerance || map->stripes[i].dev != srcdev)) | |
5677 | return i; | |
5678 | } | |
dfe25020 | 5679 | } |
30d9861f | 5680 | |
dfe25020 CM |
5681 | /* we couldn't find one that doesn't fail. Just return something |
5682 | * and the io error handling code will clean up eventually | |
5683 | */ | |
8ba0ae78 | 5684 | return preferred_mirror; |
dfe25020 CM |
5685 | } |
5686 | ||
53b381b3 | 5687 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ |
8e5cfb55 | 5688 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 | 5689 | { |
53b381b3 | 5690 | int i; |
53b381b3 DW |
5691 | int again = 1; |
5692 | ||
5693 | while (again) { | |
5694 | again = 0; | |
cc7539ed | 5695 | for (i = 0; i < num_stripes - 1; i++) { |
eeb6f172 DS |
5696 | /* Swap if parity is on a smaller index */ |
5697 | if (bbio->raid_map[i] > bbio->raid_map[i + 1]) { | |
5698 | swap(bbio->stripes[i], bbio->stripes[i + 1]); | |
5699 | swap(bbio->raid_map[i], bbio->raid_map[i + 1]); | |
53b381b3 DW |
5700 | again = 1; |
5701 | } | |
5702 | } | |
5703 | } | |
5704 | } | |
5705 | ||
6e9606d2 ZL |
5706 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5707 | { | |
5708 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5709 | /* the size of the btrfs_bio */ |
6e9606d2 | 5710 | sizeof(struct btrfs_bio) + |
e57cf21e | 5711 | /* plus the variable array for the stripes */ |
6e9606d2 | 5712 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5713 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5714 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5715 | /* |
5716 | * plus the raid_map, which includes both the tgt dev | |
5717 | * and the stripes | |
5718 | */ | |
5719 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5720 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5721 | |
5722 | atomic_set(&bbio->error, 0); | |
140475ae | 5723 | refcount_set(&bbio->refs, 1); |
6e9606d2 | 5724 | |
608769a4 NB |
5725 | bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes); |
5726 | bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes); | |
5727 | ||
6e9606d2 ZL |
5728 | return bbio; |
5729 | } | |
5730 | ||
5731 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5732 | { | |
140475ae ER |
5733 | WARN_ON(!refcount_read(&bbio->refs)); |
5734 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5735 | } |
5736 | ||
5737 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5738 | { | |
5739 | if (!bbio) | |
5740 | return; | |
140475ae | 5741 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5742 | kfree(bbio); |
5743 | } | |
5744 | ||
0b3d4cd3 LB |
5745 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5746 | /* | |
5747 | * Please note that, discard won't be sent to target device of device | |
5748 | * replace. | |
5749 | */ | |
5750 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5751 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5752 | struct btrfs_bio **bbio_ret) |
5753 | { | |
5754 | struct extent_map *em; | |
5755 | struct map_lookup *map; | |
5756 | struct btrfs_bio *bbio; | |
6b7faadd | 5757 | u64 length = *length_ret; |
0b3d4cd3 LB |
5758 | u64 offset; |
5759 | u64 stripe_nr; | |
5760 | u64 stripe_nr_end; | |
5761 | u64 stripe_end_offset; | |
5762 | u64 stripe_cnt; | |
5763 | u64 stripe_len; | |
5764 | u64 stripe_offset; | |
5765 | u64 num_stripes; | |
5766 | u32 stripe_index; | |
5767 | u32 factor = 0; | |
5768 | u32 sub_stripes = 0; | |
5769 | u64 stripes_per_dev = 0; | |
5770 | u32 remaining_stripes = 0; | |
5771 | u32 last_stripe = 0; | |
5772 | int ret = 0; | |
5773 | int i; | |
5774 | ||
5775 | /* discard always return a bbio */ | |
5776 | ASSERT(bbio_ret); | |
5777 | ||
60ca842e | 5778 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5779 | if (IS_ERR(em)) |
5780 | return PTR_ERR(em); | |
5781 | ||
5782 | map = em->map_lookup; | |
5783 | /* we don't discard raid56 yet */ | |
5784 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5785 | ret = -EOPNOTSUPP; | |
5786 | goto out; | |
5787 | } | |
5788 | ||
5789 | offset = logical - em->start; | |
2d974619 | 5790 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5791 | *length_ret = length; |
0b3d4cd3 LB |
5792 | |
5793 | stripe_len = map->stripe_len; | |
5794 | /* | |
5795 | * stripe_nr counts the total number of stripes we have to stride | |
5796 | * to get to this block | |
5797 | */ | |
5798 | stripe_nr = div64_u64(offset, stripe_len); | |
5799 | ||
5800 | /* stripe_offset is the offset of this block in its stripe */ | |
5801 | stripe_offset = offset - stripe_nr * stripe_len; | |
5802 | ||
5803 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5804 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5805 | stripe_cnt = stripe_nr_end - stripe_nr; |
5806 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5807 | (offset + length); | |
5808 | /* | |
5809 | * after this, stripe_nr is the number of stripes on this | |
5810 | * device we have to walk to find the data, and stripe_index is | |
5811 | * the number of our device in the stripe array | |
5812 | */ | |
5813 | num_stripes = 1; | |
5814 | stripe_index = 0; | |
5815 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5816 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5817 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5818 | sub_stripes = 1; | |
5819 | else | |
5820 | sub_stripes = map->sub_stripes; | |
5821 | ||
5822 | factor = map->num_stripes / sub_stripes; | |
5823 | num_stripes = min_t(u64, map->num_stripes, | |
5824 | sub_stripes * stripe_cnt); | |
5825 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5826 | stripe_index *= sub_stripes; | |
5827 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5828 | &remaining_stripes); | |
5829 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5830 | last_stripe *= sub_stripes; | |
c7369b3f | 5831 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5832 | BTRFS_BLOCK_GROUP_DUP)) { |
5833 | num_stripes = map->num_stripes; | |
5834 | } else { | |
5835 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5836 | &stripe_index); | |
5837 | } | |
5838 | ||
5839 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5840 | if (!bbio) { | |
5841 | ret = -ENOMEM; | |
5842 | goto out; | |
5843 | } | |
5844 | ||
5845 | for (i = 0; i < num_stripes; i++) { | |
5846 | bbio->stripes[i].physical = | |
5847 | map->stripes[stripe_index].physical + | |
5848 | stripe_offset + stripe_nr * map->stripe_len; | |
5849 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5850 | ||
5851 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5852 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5853 | bbio->stripes[i].length = stripes_per_dev * | |
5854 | map->stripe_len; | |
5855 | ||
5856 | if (i / sub_stripes < remaining_stripes) | |
5857 | bbio->stripes[i].length += | |
5858 | map->stripe_len; | |
5859 | ||
5860 | /* | |
5861 | * Special for the first stripe and | |
5862 | * the last stripe: | |
5863 | * | |
5864 | * |-------|...|-------| | |
5865 | * |----------| | |
5866 | * off end_off | |
5867 | */ | |
5868 | if (i < sub_stripes) | |
5869 | bbio->stripes[i].length -= | |
5870 | stripe_offset; | |
5871 | ||
5872 | if (stripe_index >= last_stripe && | |
5873 | stripe_index <= (last_stripe + | |
5874 | sub_stripes - 1)) | |
5875 | bbio->stripes[i].length -= | |
5876 | stripe_end_offset; | |
5877 | ||
5878 | if (i == sub_stripes - 1) | |
5879 | stripe_offset = 0; | |
5880 | } else { | |
5881 | bbio->stripes[i].length = length; | |
5882 | } | |
5883 | ||
5884 | stripe_index++; | |
5885 | if (stripe_index == map->num_stripes) { | |
5886 | stripe_index = 0; | |
5887 | stripe_nr++; | |
5888 | } | |
5889 | } | |
5890 | ||
5891 | *bbio_ret = bbio; | |
5892 | bbio->map_type = map->type; | |
5893 | bbio->num_stripes = num_stripes; | |
5894 | out: | |
5895 | free_extent_map(em); | |
5896 | return ret; | |
5897 | } | |
5898 | ||
5ab56090 LB |
5899 | /* |
5900 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5901 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5902 | * left cursor can also be read from the target drive. | |
5903 | * | |
5904 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5905 | * array of stripes. | |
5906 | * For READ, it also needs to be supported using the same mirror number. | |
5907 | * | |
5908 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5909 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5910 | * case. | |
5911 | */ | |
5912 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5913 | u64 logical, u64 length, | |
5914 | u64 srcdev_devid, int *mirror_num, | |
5915 | u64 *physical) | |
5916 | { | |
5917 | struct btrfs_bio *bbio = NULL; | |
5918 | int num_stripes; | |
5919 | int index_srcdev = 0; | |
5920 | int found = 0; | |
5921 | u64 physical_of_found = 0; | |
5922 | int i; | |
5923 | int ret = 0; | |
5924 | ||
5925 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5926 | logical, &length, &bbio, 0, 0); | |
5927 | if (ret) { | |
5928 | ASSERT(bbio == NULL); | |
5929 | return ret; | |
5930 | } | |
5931 | ||
5932 | num_stripes = bbio->num_stripes; | |
5933 | if (*mirror_num > num_stripes) { | |
5934 | /* | |
5935 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5936 | * that means that the requested area is not left of the left | |
5937 | * cursor | |
5938 | */ | |
5939 | btrfs_put_bbio(bbio); | |
5940 | return -EIO; | |
5941 | } | |
5942 | ||
5943 | /* | |
5944 | * process the rest of the function using the mirror_num of the source | |
5945 | * drive. Therefore look it up first. At the end, patch the device | |
5946 | * pointer to the one of the target drive. | |
5947 | */ | |
5948 | for (i = 0; i < num_stripes; i++) { | |
5949 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5950 | continue; | |
5951 | ||
5952 | /* | |
5953 | * In case of DUP, in order to keep it simple, only add the | |
5954 | * mirror with the lowest physical address | |
5955 | */ | |
5956 | if (found && | |
5957 | physical_of_found <= bbio->stripes[i].physical) | |
5958 | continue; | |
5959 | ||
5960 | index_srcdev = i; | |
5961 | found = 1; | |
5962 | physical_of_found = bbio->stripes[i].physical; | |
5963 | } | |
5964 | ||
5965 | btrfs_put_bbio(bbio); | |
5966 | ||
5967 | ASSERT(found); | |
5968 | if (!found) | |
5969 | return -EIO; | |
5970 | ||
5971 | *mirror_num = index_srcdev + 1; | |
5972 | *physical = physical_of_found; | |
5973 | return ret; | |
5974 | } | |
5975 | ||
73c0f228 LB |
5976 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5977 | struct btrfs_bio **bbio_ret, | |
5978 | struct btrfs_dev_replace *dev_replace, | |
5979 | int *num_stripes_ret, int *max_errors_ret) | |
5980 | { | |
5981 | struct btrfs_bio *bbio = *bbio_ret; | |
5982 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5983 | int tgtdev_indexes = 0; | |
5984 | int num_stripes = *num_stripes_ret; | |
5985 | int max_errors = *max_errors_ret; | |
5986 | int i; | |
5987 | ||
5988 | if (op == BTRFS_MAP_WRITE) { | |
5989 | int index_where_to_add; | |
5990 | ||
5991 | /* | |
5992 | * duplicate the write operations while the dev replace | |
5993 | * procedure is running. Since the copying of the old disk to | |
5994 | * the new disk takes place at run time while the filesystem is | |
5995 | * mounted writable, the regular write operations to the old | |
5996 | * disk have to be duplicated to go to the new disk as well. | |
5997 | * | |
5998 | * Note that device->missing is handled by the caller, and that | |
5999 | * the write to the old disk is already set up in the stripes | |
6000 | * array. | |
6001 | */ | |
6002 | index_where_to_add = num_stripes; | |
6003 | for (i = 0; i < num_stripes; i++) { | |
6004 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
6005 | /* write to new disk, too */ | |
6006 | struct btrfs_bio_stripe *new = | |
6007 | bbio->stripes + index_where_to_add; | |
6008 | struct btrfs_bio_stripe *old = | |
6009 | bbio->stripes + i; | |
6010 | ||
6011 | new->physical = old->physical; | |
6012 | new->length = old->length; | |
6013 | new->dev = dev_replace->tgtdev; | |
6014 | bbio->tgtdev_map[i] = index_where_to_add; | |
6015 | index_where_to_add++; | |
6016 | max_errors++; | |
6017 | tgtdev_indexes++; | |
6018 | } | |
6019 | } | |
6020 | num_stripes = index_where_to_add; | |
6021 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
6022 | int index_srcdev = 0; | |
6023 | int found = 0; | |
6024 | u64 physical_of_found = 0; | |
6025 | ||
6026 | /* | |
6027 | * During the dev-replace procedure, the target drive can also | |
6028 | * be used to read data in case it is needed to repair a corrupt | |
6029 | * block elsewhere. This is possible if the requested area is | |
6030 | * left of the left cursor. In this area, the target drive is a | |
6031 | * full copy of the source drive. | |
6032 | */ | |
6033 | for (i = 0; i < num_stripes; i++) { | |
6034 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
6035 | /* | |
6036 | * In case of DUP, in order to keep it simple, | |
6037 | * only add the mirror with the lowest physical | |
6038 | * address | |
6039 | */ | |
6040 | if (found && | |
6041 | physical_of_found <= | |
6042 | bbio->stripes[i].physical) | |
6043 | continue; | |
6044 | index_srcdev = i; | |
6045 | found = 1; | |
6046 | physical_of_found = bbio->stripes[i].physical; | |
6047 | } | |
6048 | } | |
6049 | if (found) { | |
6050 | struct btrfs_bio_stripe *tgtdev_stripe = | |
6051 | bbio->stripes + num_stripes; | |
6052 | ||
6053 | tgtdev_stripe->physical = physical_of_found; | |
6054 | tgtdev_stripe->length = | |
6055 | bbio->stripes[index_srcdev].length; | |
6056 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
6057 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
6058 | ||
6059 | tgtdev_indexes++; | |
6060 | num_stripes++; | |
6061 | } | |
6062 | } | |
6063 | ||
6064 | *num_stripes_ret = num_stripes; | |
6065 | *max_errors_ret = max_errors; | |
6066 | bbio->num_tgtdevs = tgtdev_indexes; | |
6067 | *bbio_ret = bbio; | |
6068 | } | |
6069 | ||
2b19a1fe LB |
6070 | static bool need_full_stripe(enum btrfs_map_op op) |
6071 | { | |
6072 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
6073 | } | |
6074 | ||
5f141126 | 6075 | /* |
42034313 MR |
6076 | * Calculate the geometry of a particular (address, len) tuple. This |
6077 | * information is used to calculate how big a particular bio can get before it | |
6078 | * straddles a stripe. | |
5f141126 | 6079 | * |
42034313 MR |
6080 | * @fs_info: the filesystem |
6081 | * @em: mapping containing the logical extent | |
6082 | * @op: type of operation - write or read | |
6083 | * @logical: address that we want to figure out the geometry of | |
6084 | * @len: the length of IO we are going to perform, starting at @logical | |
6085 | * @io_geom: pointer used to return values | |
5f141126 NB |
6086 | * |
6087 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
6088 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
6089 | */ | |
42034313 MR |
6090 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, |
6091 | enum btrfs_map_op op, u64 logical, u64 len, | |
6092 | struct btrfs_io_geometry *io_geom) | |
5f141126 | 6093 | { |
5f141126 NB |
6094 | struct map_lookup *map; |
6095 | u64 offset; | |
6096 | u64 stripe_offset; | |
6097 | u64 stripe_nr; | |
6098 | u64 stripe_len; | |
6099 | u64 raid56_full_stripe_start = (u64)-1; | |
6100 | int data_stripes; | |
6101 | ||
6102 | ASSERT(op != BTRFS_MAP_DISCARD); | |
6103 | ||
5f141126 NB |
6104 | map = em->map_lookup; |
6105 | /* Offset of this logical address in the chunk */ | |
6106 | offset = logical - em->start; | |
6107 | /* Len of a stripe in a chunk */ | |
6108 | stripe_len = map->stripe_len; | |
6109 | /* Stripe wher this block falls in */ | |
6110 | stripe_nr = div64_u64(offset, stripe_len); | |
6111 | /* Offset of stripe in the chunk */ | |
6112 | stripe_offset = stripe_nr * stripe_len; | |
6113 | if (offset < stripe_offset) { | |
6114 | btrfs_crit(fs_info, | |
6115 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
6116 | stripe_offset, offset, em->start, logical, stripe_len); | |
42034313 | 6117 | return -EINVAL; |
5f141126 NB |
6118 | } |
6119 | ||
6120 | /* stripe_offset is the offset of this block in its stripe */ | |
6121 | stripe_offset = offset - stripe_offset; | |
6122 | data_stripes = nr_data_stripes(map); | |
6123 | ||
6124 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
6125 | u64 max_len = stripe_len - stripe_offset; | |
6126 | ||
6127 | /* | |
6128 | * In case of raid56, we need to know the stripe aligned start | |
6129 | */ | |
6130 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
6131 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
6132 | raid56_full_stripe_start = offset; | |
6133 | ||
6134 | /* | |
6135 | * Allow a write of a full stripe, but make sure we | |
6136 | * don't allow straddling of stripes | |
6137 | */ | |
6138 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
6139 | full_stripe_len); | |
6140 | raid56_full_stripe_start *= full_stripe_len; | |
6141 | ||
6142 | /* | |
6143 | * For writes to RAID[56], allow a full stripeset across | |
6144 | * all disks. For other RAID types and for RAID[56] | |
6145 | * reads, just allow a single stripe (on a single disk). | |
6146 | */ | |
6147 | if (op == BTRFS_MAP_WRITE) { | |
6148 | max_len = stripe_len * data_stripes - | |
6149 | (offset - raid56_full_stripe_start); | |
6150 | } | |
6151 | } | |
6152 | len = min_t(u64, em->len - offset, max_len); | |
6153 | } else { | |
6154 | len = em->len - offset; | |
6155 | } | |
6156 | ||
6157 | io_geom->len = len; | |
6158 | io_geom->offset = offset; | |
6159 | io_geom->stripe_len = stripe_len; | |
6160 | io_geom->stripe_nr = stripe_nr; | |
6161 | io_geom->stripe_offset = stripe_offset; | |
6162 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
6163 | ||
42034313 | 6164 | return 0; |
5f141126 NB |
6165 | } |
6166 | ||
cf8cddd3 CH |
6167 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
6168 | enum btrfs_map_op op, | |
f2d8d74d | 6169 | u64 logical, u64 *length, |
a1d3c478 | 6170 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6171 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6172 | { |
6173 | struct extent_map *em; | |
6174 | struct map_lookup *map; | |
593060d7 CM |
6175 | u64 stripe_offset; |
6176 | u64 stripe_nr; | |
53b381b3 | 6177 | u64 stripe_len; |
9d644a62 | 6178 | u32 stripe_index; |
cff82672 | 6179 | int data_stripes; |
cea9e445 | 6180 | int i; |
de11cc12 | 6181 | int ret = 0; |
f2d8d74d | 6182 | int num_stripes; |
a236aed1 | 6183 | int max_errors = 0; |
2c8cdd6e | 6184 | int tgtdev_indexes = 0; |
a1d3c478 | 6185 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6186 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6187 | int dev_replace_is_ongoing = 0; | |
6188 | int num_alloc_stripes; | |
ad6d620e SB |
6189 | int patch_the_first_stripe_for_dev_replace = 0; |
6190 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6191 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6192 | struct btrfs_io_geometry geom; |
6193 | ||
6194 | ASSERT(bbio_ret); | |
75fb2e9e | 6195 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6196 | |
42034313 MR |
6197 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
6198 | ASSERT(!IS_ERR(em)); | |
6199 | ||
6200 | ret = btrfs_get_io_geometry(fs_info, em, op, logical, *length, &geom); | |
89b798ad NB |
6201 | if (ret < 0) |
6202 | return ret; | |
0b86a832 | 6203 | |
95617d69 | 6204 | map = em->map_lookup; |
593060d7 | 6205 | |
89b798ad | 6206 | *length = geom.len; |
89b798ad NB |
6207 | stripe_len = geom.stripe_len; |
6208 | stripe_nr = geom.stripe_nr; | |
6209 | stripe_offset = geom.stripe_offset; | |
6210 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6211 | data_stripes = nr_data_stripes(map); |
593060d7 | 6212 | |
cb5583dd | 6213 | down_read(&dev_replace->rwsem); |
472262f3 | 6214 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6215 | /* |
6216 | * Hold the semaphore for read during the whole operation, write is | |
6217 | * requested at commit time but must wait. | |
6218 | */ | |
472262f3 | 6219 | if (!dev_replace_is_ongoing) |
cb5583dd | 6220 | up_read(&dev_replace->rwsem); |
472262f3 | 6221 | |
ad6d620e | 6222 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6223 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6224 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6225 | dev_replace->srcdev->devid, | |
6226 | &mirror_num, | |
6227 | &physical_to_patch_in_first_stripe); | |
6228 | if (ret) | |
ad6d620e | 6229 | goto out; |
5ab56090 LB |
6230 | else |
6231 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6232 | } else if (mirror_num > map->num_stripes) { |
6233 | mirror_num = 0; | |
6234 | } | |
6235 | ||
f2d8d74d | 6236 | num_stripes = 1; |
cea9e445 | 6237 | stripe_index = 0; |
fce3bb9a | 6238 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6239 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6240 | &stripe_index); | |
de483734 | 6241 | if (!need_full_stripe(op)) |
28e1cc7d | 6242 | mirror_num = 1; |
c7369b3f | 6243 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6244 | if (need_full_stripe(op)) |
f2d8d74d | 6245 | num_stripes = map->num_stripes; |
2fff734f | 6246 | else if (mirror_num) |
f188591e | 6247 | stripe_index = mirror_num - 1; |
dfe25020 | 6248 | else { |
30d9861f | 6249 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6250 | dev_replace_is_ongoing); |
a1d3c478 | 6251 | mirror_num = stripe_index + 1; |
dfe25020 | 6252 | } |
2fff734f | 6253 | |
611f0e00 | 6254 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6255 | if (need_full_stripe(op)) { |
f2d8d74d | 6256 | num_stripes = map->num_stripes; |
a1d3c478 | 6257 | } else if (mirror_num) { |
f188591e | 6258 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6259 | } else { |
6260 | mirror_num = 1; | |
6261 | } | |
2fff734f | 6262 | |
321aecc6 | 6263 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6264 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6265 | |
47c5713f | 6266 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6267 | stripe_index *= map->sub_stripes; |
6268 | ||
de483734 | 6269 | if (need_full_stripe(op)) |
f2d8d74d | 6270 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6271 | else if (mirror_num) |
6272 | stripe_index += mirror_num - 1; | |
dfe25020 | 6273 | else { |
3e74317a | 6274 | int old_stripe_index = stripe_index; |
30d9861f SB |
6275 | stripe_index = find_live_mirror(fs_info, map, |
6276 | stripe_index, | |
30d9861f | 6277 | dev_replace_is_ongoing); |
3e74317a | 6278 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6279 | } |
53b381b3 | 6280 | |
ffe2d203 | 6281 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6282 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6283 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6284 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6285 | stripe_len * data_stripes); |
53b381b3 DW |
6286 | |
6287 | /* RAID[56] write or recovery. Return all stripes */ | |
6288 | num_stripes = map->num_stripes; | |
6289 | max_errors = nr_parity_stripes(map); | |
6290 | ||
53b381b3 DW |
6291 | *length = map->stripe_len; |
6292 | stripe_index = 0; | |
6293 | stripe_offset = 0; | |
6294 | } else { | |
6295 | /* | |
6296 | * Mirror #0 or #1 means the original data block. | |
6297 | * Mirror #2 is RAID5 parity block. | |
6298 | * Mirror #3 is RAID6 Q block. | |
6299 | */ | |
47c5713f | 6300 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6301 | data_stripes, &stripe_index); |
53b381b3 | 6302 | if (mirror_num > 1) |
cff82672 | 6303 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6304 | |
6305 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6306 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6307 | &stripe_index); | |
de483734 | 6308 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6309 | mirror_num = 1; |
53b381b3 | 6310 | } |
8790d502 CM |
6311 | } else { |
6312 | /* | |
47c5713f DS |
6313 | * after this, stripe_nr is the number of stripes on this |
6314 | * device we have to walk to find the data, and stripe_index is | |
6315 | * the number of our device in the stripe array | |
8790d502 | 6316 | */ |
47c5713f DS |
6317 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6318 | &stripe_index); | |
a1d3c478 | 6319 | mirror_num = stripe_index + 1; |
8790d502 | 6320 | } |
e042d1ec | 6321 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6322 | btrfs_crit(fs_info, |
6323 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6324 | stripe_index, map->num_stripes); |
6325 | ret = -EINVAL; | |
6326 | goto out; | |
6327 | } | |
cea9e445 | 6328 | |
472262f3 | 6329 | num_alloc_stripes = num_stripes; |
6fad823f | 6330 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6331 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6332 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6333 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6334 | num_alloc_stripes++; |
2c8cdd6e | 6335 | tgtdev_indexes = num_stripes; |
ad6d620e | 6336 | } |
2c8cdd6e | 6337 | |
6e9606d2 | 6338 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6339 | if (!bbio) { |
6340 | ret = -ENOMEM; | |
6341 | goto out; | |
6342 | } | |
608769a4 NB |
6343 | |
6344 | for (i = 0; i < num_stripes; i++) { | |
6345 | bbio->stripes[i].physical = map->stripes[stripe_index].physical + | |
6346 | stripe_offset + stripe_nr * map->stripe_len; | |
6347 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
6348 | stripe_index++; | |
6349 | } | |
de11cc12 | 6350 | |
8e5cfb55 | 6351 | /* build raid_map */ |
2b19a1fe LB |
6352 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6353 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6354 | u64 tmp; |
9d644a62 | 6355 | unsigned rot; |
8e5cfb55 | 6356 | |
8e5cfb55 | 6357 | /* Work out the disk rotation on this stripe-set */ |
47c5713f | 6358 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6359 | |
6360 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6361 | tmp = stripe_nr * data_stripes; |
6362 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6363 | bbio->raid_map[(i+rot) % num_stripes] = |
6364 | em->start + (tmp + i) * map->stripe_len; | |
6365 | ||
6366 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6367 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6368 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6369 | RAID6_Q_STRIPE; | |
8e5cfb55 | 6370 | |
608769a4 | 6371 | sort_parity_stripes(bbio, num_stripes); |
593060d7 | 6372 | } |
de11cc12 | 6373 | |
2b19a1fe | 6374 | if (need_full_stripe(op)) |
d20983b4 | 6375 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6376 | |
73c0f228 | 6377 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6378 | need_full_stripe(op)) { |
73c0f228 LB |
6379 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6380 | &max_errors); | |
472262f3 SB |
6381 | } |
6382 | ||
de11cc12 | 6383 | *bbio_ret = bbio; |
10f11900 | 6384 | bbio->map_type = map->type; |
de11cc12 LZ |
6385 | bbio->num_stripes = num_stripes; |
6386 | bbio->max_errors = max_errors; | |
6387 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6388 | |
6389 | /* | |
6390 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6391 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6392 | * available as a mirror | |
6393 | */ | |
6394 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6395 | WARN_ON(num_stripes > 1); | |
6396 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6397 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6398 | bbio->mirror_num = map->num_stripes + 1; | |
6399 | } | |
cea9e445 | 6400 | out: |
73beece9 | 6401 | if (dev_replace_is_ongoing) { |
53176dde DS |
6402 | lockdep_assert_held(&dev_replace->rwsem); |
6403 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6404 | up_read(&dev_replace->rwsem); |
73beece9 | 6405 | } |
0b86a832 | 6406 | free_extent_map(em); |
de11cc12 | 6407 | return ret; |
0b86a832 CM |
6408 | } |
6409 | ||
cf8cddd3 | 6410 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6411 | u64 logical, u64 *length, |
a1d3c478 | 6412 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6413 | { |
75fb2e9e DS |
6414 | if (op == BTRFS_MAP_DISCARD) |
6415 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6416 | length, bbio_ret); | |
6417 | ||
b3d3fa51 | 6418 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6419 | mirror_num, 0); |
f2d8d74d CM |
6420 | } |
6421 | ||
af8e2d1d | 6422 | /* For Scrub/replace */ |
cf8cddd3 | 6423 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6424 | u64 logical, u64 *length, |
825ad4c9 | 6425 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6426 | { |
825ad4c9 | 6427 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6428 | } |
6429 | ||
4246a0b6 | 6430 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6431 | { |
326e1dbb MS |
6432 | bio->bi_private = bbio->private; |
6433 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6434 | bio_endio(bio); |
326e1dbb | 6435 | |
6e9606d2 | 6436 | btrfs_put_bbio(bbio); |
8408c716 MX |
6437 | } |
6438 | ||
4246a0b6 | 6439 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6440 | { |
9be3395b | 6441 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6442 | int is_orig_bio = 0; |
8790d502 | 6443 | |
4e4cbee9 | 6444 | if (bio->bi_status) { |
a1d3c478 | 6445 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6446 | if (bio->bi_status == BLK_STS_IOERR || |
6447 | bio->bi_status == BLK_STS_TARGET) { | |
c31efbdf | 6448 | struct btrfs_device *dev = btrfs_io_bio(bio)->device; |
442a4f63 | 6449 | |
3eee86c8 | 6450 | ASSERT(dev->bdev); |
cfe94440 | 6451 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) |
3eee86c8 | 6452 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6453 | BTRFS_DEV_STAT_WRITE_ERRS); |
3eee86c8 NB |
6454 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
6455 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6456 | BTRFS_DEV_STAT_READ_ERRS); |
3eee86c8 NB |
6457 | if (bio->bi_opf & REQ_PREFLUSH) |
6458 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6459 | BTRFS_DEV_STAT_FLUSH_ERRS); |
442a4f63 SB |
6460 | } |
6461 | } | |
8790d502 | 6462 | |
a1d3c478 | 6463 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6464 | is_orig_bio = 1; |
6465 | ||
c404e0dc MX |
6466 | btrfs_bio_counter_dec(bbio->fs_info); |
6467 | ||
a1d3c478 | 6468 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6469 | if (!is_orig_bio) { |
6470 | bio_put(bio); | |
a1d3c478 | 6471 | bio = bbio->orig_bio; |
7d2b4daa | 6472 | } |
c7b22bb1 | 6473 | |
9be3395b | 6474 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6475 | /* only send an error to the higher layers if it is |
53b381b3 | 6476 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6477 | */ |
a1d3c478 | 6478 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6479 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6480 | } else { |
1259ab75 CM |
6481 | /* |
6482 | * this bio is actually up to date, we didn't | |
6483 | * go over the max number of errors | |
6484 | */ | |
2dbe0c77 | 6485 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6486 | } |
c55f1396 | 6487 | |
4246a0b6 | 6488 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6489 | } else if (!is_orig_bio) { |
8790d502 CM |
6490 | bio_put(bio); |
6491 | } | |
8790d502 CM |
6492 | } |
6493 | ||
2ff7e61e | 6494 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
c31efbdf | 6495 | u64 physical, struct btrfs_device *dev) |
de1ee92a | 6496 | { |
2ff7e61e | 6497 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6498 | |
6499 | bio->bi_private = bbio; | |
c31efbdf | 6500 | btrfs_io_bio(bio)->device = dev; |
de1ee92a | 6501 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6502 | bio->bi_iter.bi_sector = physical >> 9; |
d8e3fb10 NA |
6503 | /* |
6504 | * For zone append writing, bi_sector must point the beginning of the | |
6505 | * zone | |
6506 | */ | |
6507 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { | |
6508 | if (btrfs_dev_is_sequential(dev, physical)) { | |
6509 | u64 zone_start = round_down(physical, fs_info->zone_size); | |
6510 | ||
6511 | bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; | |
6512 | } else { | |
6513 | bio->bi_opf &= ~REQ_OP_ZONE_APPEND; | |
6514 | bio->bi_opf |= REQ_OP_WRITE; | |
6515 | } | |
6516 | } | |
672d5990 MT |
6517 | btrfs_debug_in_rcu(fs_info, |
6518 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
1201b58b | 6519 | bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, |
1db45a35 DS |
6520 | (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), |
6521 | dev->devid, bio->bi_iter.bi_size); | |
74d46992 | 6522 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6523 | |
2ff7e61e | 6524 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6525 | |
08635bae | 6526 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6527 | } |
6528 | ||
de1ee92a JB |
6529 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6530 | { | |
6531 | atomic_inc(&bbio->error); | |
6532 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6533 | /* Should be the original bio. */ |
8408c716 MX |
6534 | WARN_ON(bio != bbio->orig_bio); |
6535 | ||
9be3395b | 6536 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6537 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6538 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6539 | bio->bi_status = BLK_STS_IOERR; | |
6540 | else | |
6541 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6542 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6543 | } |
6544 | } | |
6545 | ||
58efbc9f | 6546 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6547 | int mirror_num) |
0b86a832 | 6548 | { |
0b86a832 | 6549 | struct btrfs_device *dev; |
8790d502 | 6550 | struct bio *first_bio = bio; |
1201b58b | 6551 | u64 logical = bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6552 | u64 length = 0; |
6553 | u64 map_length; | |
0b86a832 | 6554 | int ret; |
08da757d ZL |
6555 | int dev_nr; |
6556 | int total_devs; | |
a1d3c478 | 6557 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6558 | |
4f024f37 | 6559 | length = bio->bi_iter.bi_size; |
0b86a832 | 6560 | map_length = length; |
cea9e445 | 6561 | |
0b246afa | 6562 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6563 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6564 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6565 | if (ret) { |
0b246afa | 6566 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6567 | return errno_to_blk_status(ret); |
c404e0dc | 6568 | } |
cea9e445 | 6569 | |
a1d3c478 | 6570 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6571 | bbio->orig_bio = first_bio; |
6572 | bbio->private = first_bio->bi_private; | |
6573 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6574 | bbio->fs_info = fs_info; |
53b381b3 DW |
6575 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6576 | ||
ad1ba2a0 | 6577 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cfe94440 | 6578 | ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6579 | /* In this case, map_length has been set to the length of |
6580 | a single stripe; not the whole write */ | |
cfe94440 | 6581 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) { |
2ff7e61e JM |
6582 | ret = raid56_parity_write(fs_info, bio, bbio, |
6583 | map_length); | |
53b381b3 | 6584 | } else { |
2ff7e61e JM |
6585 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6586 | map_length, mirror_num, 1); | |
53b381b3 | 6587 | } |
4245215d | 6588 | |
0b246afa | 6589 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6590 | return errno_to_blk_status(ret); |
53b381b3 DW |
6591 | } |
6592 | ||
cea9e445 | 6593 | if (map_length < length) { |
0b246afa | 6594 | btrfs_crit(fs_info, |
5d163e0e JM |
6595 | "mapping failed logical %llu bio len %llu len %llu", |
6596 | logical, length, map_length); | |
cea9e445 CM |
6597 | BUG(); |
6598 | } | |
a1d3c478 | 6599 | |
08da757d | 6600 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6601 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6602 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6603 | &dev->dev_state) || | |
cfe94440 | 6604 | (btrfs_op(first_bio) == BTRFS_MAP_WRITE && |
ebbede42 | 6605 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { |
de1ee92a | 6606 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6607 | continue; |
6608 | } | |
6609 | ||
3aa8e074 | 6610 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6611 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6612 | else |
a1d3c478 | 6613 | bio = first_bio; |
de1ee92a | 6614 | |
c31efbdf | 6615 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev); |
8790d502 | 6616 | } |
0b246afa | 6617 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6618 | return BLK_STS_OK; |
0b86a832 CM |
6619 | } |
6620 | ||
09ba3bc9 AJ |
6621 | /* |
6622 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6623 | * return NULL. | |
6624 | * | |
6625 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6626 | * only devid is used. | |
6627 | * | |
6628 | * If @seed is true, traverse through the seed devices. | |
6629 | */ | |
e4319cd9 | 6630 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
b2598edf | 6631 | u64 devid, u8 *uuid, u8 *fsid) |
0b86a832 | 6632 | { |
2b82032c | 6633 | struct btrfs_device *device; |
944d3f9f NB |
6634 | struct btrfs_fs_devices *seed_devs; |
6635 | ||
6636 | if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { | |
6637 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6638 | if (device->devid == devid && | |
6639 | (!uuid || memcmp(device->uuid, uuid, | |
6640 | BTRFS_UUID_SIZE) == 0)) | |
6641 | return device; | |
6642 | } | |
6643 | } | |
2b82032c | 6644 | |
944d3f9f | 6645 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
2b82032c | 6646 | if (!fsid || |
944d3f9f NB |
6647 | !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
6648 | list_for_each_entry(device, &seed_devs->devices, | |
09ba3bc9 AJ |
6649 | dev_list) { |
6650 | if (device->devid == devid && | |
6651 | (!uuid || memcmp(device->uuid, uuid, | |
6652 | BTRFS_UUID_SIZE) == 0)) | |
6653 | return device; | |
6654 | } | |
2b82032c | 6655 | } |
2b82032c | 6656 | } |
944d3f9f | 6657 | |
2b82032c | 6658 | return NULL; |
0b86a832 CM |
6659 | } |
6660 | ||
2ff7e61e | 6661 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6662 | u64 devid, u8 *dev_uuid) |
6663 | { | |
6664 | struct btrfs_device *device; | |
fccc0007 | 6665 | unsigned int nofs_flag; |
dfe25020 | 6666 | |
fccc0007 JB |
6667 | /* |
6668 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6669 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6670 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6671 | * places. | |
6672 | */ | |
6673 | nofs_flag = memalloc_nofs_save(); | |
12bd2fc0 | 6674 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
fccc0007 | 6675 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6676 | if (IS_ERR(device)) |
adfb69af | 6677 | return device; |
12bd2fc0 ID |
6678 | |
6679 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6680 | device->fs_devices = fs_devices; |
dfe25020 | 6681 | fs_devices->num_devices++; |
12bd2fc0 | 6682 | |
e6e674bd | 6683 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6684 | fs_devices->missing_devices++; |
12bd2fc0 | 6685 | |
dfe25020 CM |
6686 | return device; |
6687 | } | |
6688 | ||
12bd2fc0 ID |
6689 | /** |
6690 | * btrfs_alloc_device - allocate struct btrfs_device | |
6691 | * @fs_info: used only for generating a new devid, can be NULL if | |
6692 | * devid is provided (i.e. @devid != NULL). | |
6693 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6694 | * is generated. | |
6695 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6696 | * is generated. | |
6697 | * | |
6698 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6699 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6700 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6701 | */ |
6702 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6703 | const u64 *devid, | |
6704 | const u8 *uuid) | |
6705 | { | |
6706 | struct btrfs_device *dev; | |
6707 | u64 tmp; | |
6708 | ||
fae7f21c | 6709 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6710 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6711 | |
154f7cb8 | 6712 | dev = __alloc_device(fs_info); |
12bd2fc0 ID |
6713 | if (IS_ERR(dev)) |
6714 | return dev; | |
6715 | ||
6716 | if (devid) | |
6717 | tmp = *devid; | |
6718 | else { | |
6719 | int ret; | |
6720 | ||
6721 | ret = find_next_devid(fs_info, &tmp); | |
6722 | if (ret) { | |
a425f9d4 | 6723 | btrfs_free_device(dev); |
12bd2fc0 ID |
6724 | return ERR_PTR(ret); |
6725 | } | |
6726 | } | |
6727 | dev->devid = tmp; | |
6728 | ||
6729 | if (uuid) | |
6730 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6731 | else | |
6732 | generate_random_uuid(dev->uuid); | |
6733 | ||
12bd2fc0 ID |
6734 | return dev; |
6735 | } | |
6736 | ||
5a2b8e60 | 6737 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6738 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6739 | { |
2b902dfc AJ |
6740 | if (error) |
6741 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6742 | devid, uuid); | |
6743 | else | |
6744 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6745 | devid, uuid); | |
5a2b8e60 AJ |
6746 | } |
6747 | ||
39e264a4 NB |
6748 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6749 | { | |
6750 | int index = btrfs_bg_flags_to_raid_index(type); | |
6751 | int ncopies = btrfs_raid_array[index].ncopies; | |
e4f6c6be | 6752 | const int nparity = btrfs_raid_array[index].nparity; |
39e264a4 NB |
6753 | int data_stripes; |
6754 | ||
e4f6c6be DS |
6755 | if (nparity) |
6756 | data_stripes = num_stripes - nparity; | |
6757 | else | |
39e264a4 | 6758 | data_stripes = num_stripes / ncopies; |
e4f6c6be | 6759 | |
39e264a4 NB |
6760 | return div_u64(chunk_len, data_stripes); |
6761 | } | |
6762 | ||
9690ac09 | 6763 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6764 | struct btrfs_chunk *chunk) |
6765 | { | |
9690ac09 | 6766 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6767 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6768 | struct map_lookup *map; |
6769 | struct extent_map *em; | |
6770 | u64 logical; | |
6771 | u64 length; | |
e06cd3dd LB |
6772 | u64 devid; |
6773 | u8 uuid[BTRFS_UUID_SIZE]; | |
6774 | int num_stripes; | |
6775 | int ret; | |
6776 | int i; | |
6777 | ||
6778 | logical = key->offset; | |
6779 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6780 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6781 | ||
075cb3c7 QW |
6782 | /* |
6783 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6784 | * as chunk item in tree block is already verified by tree-checker. | |
6785 | */ | |
6786 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6787 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6788 | if (ret) |
6789 | return ret; | |
6790 | } | |
a061fc8d | 6791 | |
c8bf1b67 DS |
6792 | read_lock(&map_tree->lock); |
6793 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6794 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6795 | |
6796 | /* already mapped? */ | |
6797 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6798 | free_extent_map(em); | |
0b86a832 CM |
6799 | return 0; |
6800 | } else if (em) { | |
6801 | free_extent_map(em); | |
6802 | } | |
0b86a832 | 6803 | |
172ddd60 | 6804 | em = alloc_extent_map(); |
0b86a832 CM |
6805 | if (!em) |
6806 | return -ENOMEM; | |
593060d7 | 6807 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6808 | if (!map) { |
6809 | free_extent_map(em); | |
6810 | return -ENOMEM; | |
6811 | } | |
6812 | ||
298a8f9c | 6813 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6814 | em->map_lookup = map; |
0b86a832 CM |
6815 | em->start = logical; |
6816 | em->len = length; | |
70c8a91c | 6817 | em->orig_start = 0; |
0b86a832 | 6818 | em->block_start = 0; |
c8b97818 | 6819 | em->block_len = em->len; |
0b86a832 | 6820 | |
593060d7 CM |
6821 | map->num_stripes = num_stripes; |
6822 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6823 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6824 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6825 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6826 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6827 | map->verified_stripes = 0; |
39e264a4 NB |
6828 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6829 | map->num_stripes); | |
593060d7 CM |
6830 | for (i = 0; i < num_stripes; i++) { |
6831 | map->stripes[i].physical = | |
6832 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6833 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6834 | read_extent_buffer(leaf, uuid, (unsigned long) |
6835 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6836 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6837 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
b2598edf | 6838 | devid, uuid, NULL); |
3cdde224 | 6839 | if (!map->stripes[i].dev && |
0b246afa | 6840 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6841 | free_extent_map(em); |
2b902dfc | 6842 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6843 | return -ENOENT; |
593060d7 | 6844 | } |
dfe25020 CM |
6845 | if (!map->stripes[i].dev) { |
6846 | map->stripes[i].dev = | |
2ff7e61e JM |
6847 | add_missing_dev(fs_info->fs_devices, devid, |
6848 | uuid); | |
adfb69af | 6849 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6850 | free_extent_map(em); |
adfb69af AJ |
6851 | btrfs_err(fs_info, |
6852 | "failed to init missing dev %llu: %ld", | |
6853 | devid, PTR_ERR(map->stripes[i].dev)); | |
6854 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6855 | } |
2b902dfc | 6856 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6857 | } |
e12c9621 AJ |
6858 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6859 | &(map->stripes[i].dev->dev_state)); | |
6860 | ||
0b86a832 CM |
6861 | } |
6862 | ||
c8bf1b67 DS |
6863 | write_lock(&map_tree->lock); |
6864 | ret = add_extent_mapping(map_tree, em, 0); | |
6865 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6866 | if (ret < 0) { |
6867 | btrfs_err(fs_info, | |
6868 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6869 | em->start, em->len, ret); | |
6870 | } | |
0b86a832 CM |
6871 | free_extent_map(em); |
6872 | ||
64f64f43 | 6873 | return ret; |
0b86a832 CM |
6874 | } |
6875 | ||
143bede5 | 6876 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6877 | struct btrfs_dev_item *dev_item, |
6878 | struct btrfs_device *device) | |
6879 | { | |
6880 | unsigned long ptr; | |
0b86a832 CM |
6881 | |
6882 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6883 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6884 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6885 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6886 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6887 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6888 | device->type = btrfs_device_type(leaf, dev_item); |
6889 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6890 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6891 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6892 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6893 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6894 | |
410ba3a2 | 6895 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6896 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6897 | } |
6898 | ||
2ff7e61e | 6899 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6900 | u8 *fsid) |
2b82032c YZ |
6901 | { |
6902 | struct btrfs_fs_devices *fs_devices; | |
6903 | int ret; | |
6904 | ||
a32bf9a3 | 6905 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6906 | ASSERT(fsid); |
2b82032c | 6907 | |
427c8fdd | 6908 | /* This will match only for multi-device seed fs */ |
944d3f9f | 6909 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 6910 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6911 | return fs_devices; |
6912 | ||
2b82032c | 6913 | |
7239ff4b | 6914 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6915 | if (!fs_devices) { |
0b246afa | 6916 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6917 | return ERR_PTR(-ENOENT); |
6918 | ||
7239ff4b | 6919 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6920 | if (IS_ERR(fs_devices)) |
6921 | return fs_devices; | |
6922 | ||
0395d84f | 6923 | fs_devices->seeding = true; |
5f375835 MX |
6924 | fs_devices->opened = 1; |
6925 | return fs_devices; | |
2b82032c | 6926 | } |
e4404d6e | 6927 | |
427c8fdd NB |
6928 | /* |
6929 | * Upon first call for a seed fs fsid, just create a private copy of the | |
6930 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
6931 | */ | |
e4404d6e | 6932 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
6933 | if (IS_ERR(fs_devices)) |
6934 | return fs_devices; | |
2b82032c | 6935 | |
897fb573 | 6936 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6937 | if (ret) { |
6938 | free_fs_devices(fs_devices); | |
c83b60c0 | 6939 | return ERR_PTR(ret); |
48d28232 | 6940 | } |
2b82032c YZ |
6941 | |
6942 | if (!fs_devices->seeding) { | |
0226e0eb | 6943 | close_fs_devices(fs_devices); |
e4404d6e | 6944 | free_fs_devices(fs_devices); |
c83b60c0 | 6945 | return ERR_PTR(-EINVAL); |
2b82032c YZ |
6946 | } |
6947 | ||
944d3f9f | 6948 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
c83b60c0 | 6949 | |
5f375835 | 6950 | return fs_devices; |
2b82032c YZ |
6951 | } |
6952 | ||
17850759 | 6953 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6954 | struct btrfs_dev_item *dev_item) |
6955 | { | |
17850759 | 6956 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6957 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6958 | struct btrfs_device *device; |
6959 | u64 devid; | |
6960 | int ret; | |
44880fdc | 6961 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6962 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6963 | ||
0b86a832 | 6964 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6965 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6966 | BTRFS_UUID_SIZE); |
1473b24e | 6967 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6968 | BTRFS_FSID_SIZE); |
2b82032c | 6969 | |
de37aa51 | 6970 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6971 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6972 | if (IS_ERR(fs_devices)) |
6973 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6974 | } |
6975 | ||
e4319cd9 | 6976 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 6977 | fs_uuid); |
5f375835 | 6978 | if (!device) { |
c5502451 | 6979 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6980 | btrfs_report_missing_device(fs_info, devid, |
6981 | dev_uuid, true); | |
45dbdbc9 | 6982 | return -ENOENT; |
c5502451 | 6983 | } |
2b82032c | 6984 | |
2ff7e61e | 6985 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6986 | if (IS_ERR(device)) { |
6987 | btrfs_err(fs_info, | |
6988 | "failed to add missing dev %llu: %ld", | |
6989 | devid, PTR_ERR(device)); | |
6990 | return PTR_ERR(device); | |
6991 | } | |
2b902dfc | 6992 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6993 | } else { |
c5502451 | 6994 | if (!device->bdev) { |
2b902dfc AJ |
6995 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6996 | btrfs_report_missing_device(fs_info, | |
6997 | devid, dev_uuid, true); | |
45dbdbc9 | 6998 | return -ENOENT; |
2b902dfc AJ |
6999 | } |
7000 | btrfs_report_missing_device(fs_info, devid, | |
7001 | dev_uuid, false); | |
c5502451 | 7002 | } |
5f375835 | 7003 | |
e6e674bd AJ |
7004 | if (!device->bdev && |
7005 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7006 | /* |
7007 | * this happens when a device that was properly setup | |
7008 | * in the device info lists suddenly goes bad. | |
7009 | * device->bdev is NULL, and so we have to set | |
7010 | * device->missing to one here | |
7011 | */ | |
5f375835 | 7012 | device->fs_devices->missing_devices++; |
e6e674bd | 7013 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7014 | } |
5f375835 MX |
7015 | |
7016 | /* Move the device to its own fs_devices */ | |
7017 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7018 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7019 | &device->dev_state)); | |
5f375835 MX |
7020 | |
7021 | list_move(&device->dev_list, &fs_devices->devices); | |
7022 | device->fs_devices->num_devices--; | |
7023 | fs_devices->num_devices++; | |
7024 | ||
7025 | device->fs_devices->missing_devices--; | |
7026 | fs_devices->missing_devices++; | |
7027 | ||
7028 | device->fs_devices = fs_devices; | |
7029 | } | |
2b82032c YZ |
7030 | } |
7031 | ||
0b246afa | 7032 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7033 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7034 | if (device->generation != |
7035 | btrfs_device_generation(leaf, dev_item)) | |
7036 | return -EINVAL; | |
6324fbf3 | 7037 | } |
0b86a832 CM |
7038 | |
7039 | fill_device_from_item(leaf, dev_item, device); | |
3a160a93 AJ |
7040 | if (device->bdev) { |
7041 | u64 max_total_bytes = i_size_read(device->bdev->bd_inode); | |
7042 | ||
7043 | if (device->total_bytes > max_total_bytes) { | |
7044 | btrfs_err(fs_info, | |
7045 | "device total_bytes should be at most %llu but found %llu", | |
7046 | max_total_bytes, device->total_bytes); | |
7047 | return -EINVAL; | |
7048 | } | |
7049 | } | |
e12c9621 | 7050 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7051 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7052 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7053 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7054 | atomic64_add(device->total_bytes - device->bytes_used, |
7055 | &fs_info->free_chunk_space); | |
2bf64758 | 7056 | } |
0b86a832 | 7057 | ret = 0; |
0b86a832 CM |
7058 | return ret; |
7059 | } | |
7060 | ||
6bccf3ab | 7061 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7062 | { |
6bccf3ab | 7063 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 7064 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7065 | struct extent_buffer *sb; |
0b86a832 | 7066 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7067 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7068 | u8 *array_ptr; |
7069 | unsigned long sb_array_offset; | |
84eed90f | 7070 | int ret = 0; |
0b86a832 CM |
7071 | u32 num_stripes; |
7072 | u32 array_size; | |
7073 | u32 len = 0; | |
1ffb22cf | 7074 | u32 cur_offset; |
e06cd3dd | 7075 | u64 type; |
84eed90f | 7076 | struct btrfs_key key; |
0b86a832 | 7077 | |
0b246afa | 7078 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7079 | /* |
7080 | * This will create extent buffer of nodesize, superblock size is | |
7081 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7082 | * overallocate but we can keep it as-is, only the first page is used. | |
7083 | */ | |
3fbaf258 JB |
7084 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, |
7085 | root->root_key.objectid, 0); | |
c871b0f2 LB |
7086 | if (IS_ERR(sb)) |
7087 | return PTR_ERR(sb); | |
4db8c528 | 7088 | set_extent_buffer_uptodate(sb); |
8a334426 | 7089 | /* |
01327610 | 7090 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7091 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7092 | * pages up-to-date when the page is larger: extent does not cover the |
7093 | * whole page and consequently check_page_uptodate does not find all | |
7094 | * the page's extents up-to-date (the hole beyond sb), | |
7095 | * write_extent_buffer then triggers a WARN_ON. | |
7096 | * | |
7097 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7098 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7099 | * to silence the warning eg. on PowerPC 64. | |
7100 | */ | |
09cbfeaf | 7101 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7102 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7103 | |
a061fc8d | 7104 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7105 | array_size = btrfs_super_sys_array_size(super_copy); |
7106 | ||
1ffb22cf DS |
7107 | array_ptr = super_copy->sys_chunk_array; |
7108 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7109 | cur_offset = 0; | |
0b86a832 | 7110 | |
1ffb22cf DS |
7111 | while (cur_offset < array_size) { |
7112 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7113 | len = sizeof(*disk_key); |
7114 | if (cur_offset + len > array_size) | |
7115 | goto out_short_read; | |
7116 | ||
0b86a832 CM |
7117 | btrfs_disk_key_to_cpu(&key, disk_key); |
7118 | ||
1ffb22cf DS |
7119 | array_ptr += len; |
7120 | sb_array_offset += len; | |
7121 | cur_offset += len; | |
0b86a832 | 7122 | |
32ab3d1b JT |
7123 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
7124 | btrfs_err(fs_info, | |
7125 | "unexpected item type %u in sys_array at offset %u", | |
7126 | (u32)key.type, cur_offset); | |
7127 | ret = -EIO; | |
7128 | break; | |
7129 | } | |
f5cdedd7 | 7130 | |
32ab3d1b JT |
7131 | chunk = (struct btrfs_chunk *)sb_array_offset; |
7132 | /* | |
7133 | * At least one btrfs_chunk with one stripe must be present, | |
7134 | * exact stripe count check comes afterwards | |
7135 | */ | |
7136 | len = btrfs_chunk_item_size(1); | |
7137 | if (cur_offset + len > array_size) | |
7138 | goto out_short_read; | |
e06cd3dd | 7139 | |
32ab3d1b JT |
7140 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
7141 | if (!num_stripes) { | |
7142 | btrfs_err(fs_info, | |
7143 | "invalid number of stripes %u in sys_array at offset %u", | |
7144 | num_stripes, cur_offset); | |
7145 | ret = -EIO; | |
7146 | break; | |
7147 | } | |
e3540eab | 7148 | |
32ab3d1b JT |
7149 | type = btrfs_chunk_type(sb, chunk); |
7150 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7151 | btrfs_err(fs_info, |
32ab3d1b JT |
7152 | "invalid chunk type %llu in sys_array at offset %u", |
7153 | type, cur_offset); | |
84eed90f CM |
7154 | ret = -EIO; |
7155 | break; | |
0b86a832 | 7156 | } |
32ab3d1b JT |
7157 | |
7158 | len = btrfs_chunk_item_size(num_stripes); | |
7159 | if (cur_offset + len > array_size) | |
7160 | goto out_short_read; | |
7161 | ||
7162 | ret = read_one_chunk(&key, sb, chunk); | |
7163 | if (ret) | |
7164 | break; | |
7165 | ||
1ffb22cf DS |
7166 | array_ptr += len; |
7167 | sb_array_offset += len; | |
7168 | cur_offset += len; | |
0b86a832 | 7169 | } |
d865177a | 7170 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7171 | free_extent_buffer_stale(sb); |
84eed90f | 7172 | return ret; |
e3540eab DS |
7173 | |
7174 | out_short_read: | |
ab8d0fc4 | 7175 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7176 | len, cur_offset); |
d865177a | 7177 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7178 | free_extent_buffer_stale(sb); |
e3540eab | 7179 | return -EIO; |
0b86a832 CM |
7180 | } |
7181 | ||
21634a19 QW |
7182 | /* |
7183 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7184 | * | |
6528b99d AJ |
7185 | * If the @failing_dev is specified, it's accounted as missing. |
7186 | * | |
21634a19 QW |
7187 | * Return true if all chunks meet the minimal RW mount requirements. |
7188 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7189 | */ | |
6528b99d AJ |
7190 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7191 | struct btrfs_device *failing_dev) | |
21634a19 | 7192 | { |
c8bf1b67 | 7193 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7194 | struct extent_map *em; |
7195 | u64 next_start = 0; | |
7196 | bool ret = true; | |
7197 | ||
c8bf1b67 DS |
7198 | read_lock(&map_tree->lock); |
7199 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7200 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7201 | /* No chunk at all? Return false anyway */ |
7202 | if (!em) { | |
7203 | ret = false; | |
7204 | goto out; | |
7205 | } | |
7206 | while (em) { | |
7207 | struct map_lookup *map; | |
7208 | int missing = 0; | |
7209 | int max_tolerated; | |
7210 | int i; | |
7211 | ||
7212 | map = em->map_lookup; | |
7213 | max_tolerated = | |
7214 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7215 | map->type); | |
7216 | for (i = 0; i < map->num_stripes; i++) { | |
7217 | struct btrfs_device *dev = map->stripes[i].dev; | |
7218 | ||
e6e674bd AJ |
7219 | if (!dev || !dev->bdev || |
7220 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7221 | dev->last_flush_error) |
7222 | missing++; | |
6528b99d AJ |
7223 | else if (failing_dev && failing_dev == dev) |
7224 | missing++; | |
21634a19 QW |
7225 | } |
7226 | if (missing > max_tolerated) { | |
6528b99d AJ |
7227 | if (!failing_dev) |
7228 | btrfs_warn(fs_info, | |
52042d8e | 7229 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7230 | em->start, missing, max_tolerated); |
7231 | free_extent_map(em); | |
7232 | ret = false; | |
7233 | goto out; | |
7234 | } | |
7235 | next_start = extent_map_end(em); | |
7236 | free_extent_map(em); | |
7237 | ||
c8bf1b67 DS |
7238 | read_lock(&map_tree->lock); |
7239 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7240 | (u64)(-1) - next_start); |
c8bf1b67 | 7241 | read_unlock(&map_tree->lock); |
21634a19 QW |
7242 | } |
7243 | out: | |
7244 | return ret; | |
7245 | } | |
7246 | ||
d85327b1 DS |
7247 | static void readahead_tree_node_children(struct extent_buffer *node) |
7248 | { | |
7249 | int i; | |
7250 | const int nr_items = btrfs_header_nritems(node); | |
7251 | ||
bfb484d9 JB |
7252 | for (i = 0; i < nr_items; i++) |
7253 | btrfs_readahead_node_child(node, i); | |
d85327b1 DS |
7254 | } |
7255 | ||
5b4aacef | 7256 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7257 | { |
5b4aacef | 7258 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7259 | struct btrfs_path *path; |
7260 | struct extent_buffer *leaf; | |
7261 | struct btrfs_key key; | |
7262 | struct btrfs_key found_key; | |
7263 | int ret; | |
7264 | int slot; | |
99e3ecfc | 7265 | u64 total_dev = 0; |
d85327b1 | 7266 | u64 last_ra_node = 0; |
0b86a832 | 7267 | |
0b86a832 CM |
7268 | path = btrfs_alloc_path(); |
7269 | if (!path) | |
7270 | return -ENOMEM; | |
7271 | ||
3dd0f7a3 AJ |
7272 | /* |
7273 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7274 | * otherwise we don't need it. | |
7275 | */ | |
b367e47f | 7276 | mutex_lock(&uuid_mutex); |
b367e47f | 7277 | |
48cfa61b BB |
7278 | /* |
7279 | * It is possible for mount and umount to race in such a way that | |
7280 | * we execute this code path, but open_fs_devices failed to clear | |
7281 | * total_rw_bytes. We certainly want it cleared before reading the | |
7282 | * device items, so clear it here. | |
7283 | */ | |
7284 | fs_info->fs_devices->total_rw_bytes = 0; | |
7285 | ||
395927a9 FDBM |
7286 | /* |
7287 | * Read all device items, and then all the chunk items. All | |
7288 | * device items are found before any chunk item (their object id | |
7289 | * is smaller than the lowest possible object id for a chunk | |
7290 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7291 | */ |
7292 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7293 | key.offset = 0; | |
7294 | key.type = 0; | |
0b86a832 | 7295 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7296 | if (ret < 0) |
7297 | goto error; | |
d397712b | 7298 | while (1) { |
d85327b1 DS |
7299 | struct extent_buffer *node; |
7300 | ||
0b86a832 CM |
7301 | leaf = path->nodes[0]; |
7302 | slot = path->slots[0]; | |
7303 | if (slot >= btrfs_header_nritems(leaf)) { | |
7304 | ret = btrfs_next_leaf(root, path); | |
7305 | if (ret == 0) | |
7306 | continue; | |
7307 | if (ret < 0) | |
7308 | goto error; | |
7309 | break; | |
7310 | } | |
d85327b1 DS |
7311 | /* |
7312 | * The nodes on level 1 are not locked but we don't need to do | |
7313 | * that during mount time as nothing else can access the tree | |
7314 | */ | |
7315 | node = path->nodes[1]; | |
7316 | if (node) { | |
7317 | if (last_ra_node != node->start) { | |
7318 | readahead_tree_node_children(node); | |
7319 | last_ra_node = node->start; | |
7320 | } | |
7321 | } | |
0b86a832 | 7322 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
395927a9 FDBM |
7323 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7324 | struct btrfs_dev_item *dev_item; | |
7325 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7326 | struct btrfs_dev_item); |
17850759 | 7327 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7328 | if (ret) |
7329 | goto error; | |
99e3ecfc | 7330 | total_dev++; |
0b86a832 CM |
7331 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7332 | struct btrfs_chunk *chunk; | |
7333 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
01d01caf | 7334 | mutex_lock(&fs_info->chunk_mutex); |
9690ac09 | 7335 | ret = read_one_chunk(&found_key, leaf, chunk); |
01d01caf | 7336 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c YZ |
7337 | if (ret) |
7338 | goto error; | |
0b86a832 CM |
7339 | } |
7340 | path->slots[0]++; | |
7341 | } | |
99e3ecfc LB |
7342 | |
7343 | /* | |
7344 | * After loading chunk tree, we've got all device information, | |
7345 | * do another round of validation checks. | |
7346 | */ | |
0b246afa JM |
7347 | if (total_dev != fs_info->fs_devices->total_devices) { |
7348 | btrfs_err(fs_info, | |
99e3ecfc | 7349 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7350 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7351 | total_dev); |
7352 | ret = -EINVAL; | |
7353 | goto error; | |
7354 | } | |
0b246afa JM |
7355 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7356 | fs_info->fs_devices->total_rw_bytes) { | |
7357 | btrfs_err(fs_info, | |
99e3ecfc | 7358 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7359 | btrfs_super_total_bytes(fs_info->super_copy), |
7360 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7361 | ret = -EINVAL; |
7362 | goto error; | |
7363 | } | |
0b86a832 CM |
7364 | ret = 0; |
7365 | error: | |
b367e47f LZ |
7366 | mutex_unlock(&uuid_mutex); |
7367 | ||
2b82032c | 7368 | btrfs_free_path(path); |
0b86a832 CM |
7369 | return ret; |
7370 | } | |
442a4f63 | 7371 | |
cb517eab MX |
7372 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7373 | { | |
944d3f9f | 7374 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab MX |
7375 | struct btrfs_device *device; |
7376 | ||
944d3f9f NB |
7377 | fs_devices->fs_info = fs_info; |
7378 | ||
7379 | mutex_lock(&fs_devices->device_list_mutex); | |
7380 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7381 | device->fs_info = fs_info; | |
944d3f9f NB |
7382 | |
7383 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
944d3f9f | 7384 | list_for_each_entry(device, &seed_devs->devices, dev_list) |
fb456252 | 7385 | device->fs_info = fs_info; |
29cc83f6 | 7386 | |
944d3f9f | 7387 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7388 | } |
e17125b5 | 7389 | mutex_unlock(&fs_devices->device_list_mutex); |
cb517eab MX |
7390 | } |
7391 | ||
1dc990df DS |
7392 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7393 | const struct btrfs_dev_stats_item *ptr, | |
7394 | int index) | |
7395 | { | |
7396 | u64 val; | |
7397 | ||
7398 | read_extent_buffer(eb, &val, | |
7399 | offsetof(struct btrfs_dev_stats_item, values) + | |
7400 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7401 | sizeof(val)); | |
7402 | return val; | |
7403 | } | |
7404 | ||
7405 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7406 | struct btrfs_dev_stats_item *ptr, | |
7407 | int index, u64 val) | |
7408 | { | |
7409 | write_extent_buffer(eb, &val, | |
7410 | offsetof(struct btrfs_dev_stats_item, values) + | |
7411 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7412 | sizeof(val)); | |
7413 | } | |
7414 | ||
92e26df4 JB |
7415 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
7416 | struct btrfs_path *path) | |
733f4fbb | 7417 | { |
124604eb | 7418 | struct btrfs_dev_stats_item *ptr; |
733f4fbb | 7419 | struct extent_buffer *eb; |
124604eb JB |
7420 | struct btrfs_key key; |
7421 | int item_size; | |
7422 | int i, ret, slot; | |
7423 | ||
7424 | key.objectid = BTRFS_DEV_STATS_OBJECTID; | |
7425 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
7426 | key.offset = device->devid; | |
7427 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); | |
7428 | if (ret) { | |
7429 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7430 | btrfs_dev_stat_set(device, i, 0); | |
7431 | device->dev_stats_valid = 1; | |
7432 | btrfs_release_path(path); | |
92e26df4 | 7433 | return ret < 0 ? ret : 0; |
124604eb JB |
7434 | } |
7435 | slot = path->slots[0]; | |
7436 | eb = path->nodes[0]; | |
7437 | item_size = btrfs_item_size_nr(eb, slot); | |
7438 | ||
7439 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); | |
7440 | ||
7441 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7442 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7443 | btrfs_dev_stat_set(device, i, | |
7444 | btrfs_dev_stats_value(eb, ptr, i)); | |
7445 | else | |
7446 | btrfs_dev_stat_set(device, i, 0); | |
7447 | } | |
7448 | ||
7449 | device->dev_stats_valid = 1; | |
7450 | btrfs_dev_stat_print_on_load(device); | |
7451 | btrfs_release_path(path); | |
92e26df4 JB |
7452 | |
7453 | return 0; | |
124604eb JB |
7454 | } |
7455 | ||
7456 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7457 | { | |
7458 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
733f4fbb SB |
7459 | struct btrfs_device *device; |
7460 | struct btrfs_path *path = NULL; | |
92e26df4 | 7461 | int ret = 0; |
733f4fbb SB |
7462 | |
7463 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7464 | if (!path) |
7465 | return -ENOMEM; | |
733f4fbb SB |
7466 | |
7467 | mutex_lock(&fs_devices->device_list_mutex); | |
92e26df4 JB |
7468 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
7469 | ret = btrfs_device_init_dev_stats(device, path); | |
7470 | if (ret) | |
7471 | goto out; | |
7472 | } | |
124604eb | 7473 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
92e26df4 JB |
7474 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
7475 | ret = btrfs_device_init_dev_stats(device, path); | |
7476 | if (ret) | |
7477 | goto out; | |
7478 | } | |
733f4fbb | 7479 | } |
92e26df4 | 7480 | out: |
733f4fbb SB |
7481 | mutex_unlock(&fs_devices->device_list_mutex); |
7482 | ||
733f4fbb | 7483 | btrfs_free_path(path); |
92e26df4 | 7484 | return ret; |
733f4fbb SB |
7485 | } |
7486 | ||
7487 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7488 | struct btrfs_device *device) |
7489 | { | |
5495f195 | 7490 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7491 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7492 | struct btrfs_path *path; |
7493 | struct btrfs_key key; | |
7494 | struct extent_buffer *eb; | |
7495 | struct btrfs_dev_stats_item *ptr; | |
7496 | int ret; | |
7497 | int i; | |
7498 | ||
242e2956 DS |
7499 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7500 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7501 | key.offset = device->devid; |
7502 | ||
7503 | path = btrfs_alloc_path(); | |
fa252992 DS |
7504 | if (!path) |
7505 | return -ENOMEM; | |
733f4fbb SB |
7506 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7507 | if (ret < 0) { | |
0b246afa | 7508 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7509 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7510 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7511 | goto out; |
7512 | } | |
7513 | ||
7514 | if (ret == 0 && | |
7515 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7516 | /* need to delete old one and insert a new one */ | |
7517 | ret = btrfs_del_item(trans, dev_root, path); | |
7518 | if (ret != 0) { | |
0b246afa | 7519 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7520 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7521 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7522 | goto out; |
7523 | } | |
7524 | ret = 1; | |
7525 | } | |
7526 | ||
7527 | if (ret == 1) { | |
7528 | /* need to insert a new item */ | |
7529 | btrfs_release_path(path); | |
7530 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7531 | &key, sizeof(*ptr)); | |
7532 | if (ret < 0) { | |
0b246afa | 7533 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7534 | "insert dev_stats item for device %s failed %d", |
7535 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7536 | goto out; |
7537 | } | |
7538 | } | |
7539 | ||
7540 | eb = path->nodes[0]; | |
7541 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7542 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7543 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7544 | btrfs_dev_stat_read(device, i)); | |
7545 | btrfs_mark_buffer_dirty(eb); | |
7546 | ||
7547 | out: | |
7548 | btrfs_free_path(path); | |
7549 | return ret; | |
7550 | } | |
7551 | ||
7552 | /* | |
7553 | * called from commit_transaction. Writes all changed device stats to disk. | |
7554 | */ | |
196c9d8d | 7555 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7556 | { |
196c9d8d | 7557 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7558 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7559 | struct btrfs_device *device; | |
addc3fa7 | 7560 | int stats_cnt; |
733f4fbb SB |
7561 | int ret = 0; |
7562 | ||
7563 | mutex_lock(&fs_devices->device_list_mutex); | |
7564 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7565 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7566 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7567 | continue; |
7568 | ||
9deae968 NB |
7569 | |
7570 | /* | |
7571 | * There is a LOAD-LOAD control dependency between the value of | |
7572 | * dev_stats_ccnt and updating the on-disk values which requires | |
7573 | * reading the in-memory counters. Such control dependencies | |
7574 | * require explicit read memory barriers. | |
7575 | * | |
7576 | * This memory barriers pairs with smp_mb__before_atomic in | |
7577 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7578 | * barrier implied by atomic_xchg in | |
7579 | * btrfs_dev_stats_read_and_reset | |
7580 | */ | |
7581 | smp_rmb(); | |
7582 | ||
5495f195 | 7583 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7584 | if (!ret) |
addc3fa7 | 7585 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7586 | } |
7587 | mutex_unlock(&fs_devices->device_list_mutex); | |
7588 | ||
7589 | return ret; | |
7590 | } | |
7591 | ||
442a4f63 SB |
7592 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7593 | { | |
7594 | btrfs_dev_stat_inc(dev, index); | |
7595 | btrfs_dev_stat_print_on_error(dev); | |
7596 | } | |
7597 | ||
48a3b636 | 7598 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7599 | { |
733f4fbb SB |
7600 | if (!dev->dev_stats_valid) |
7601 | return; | |
fb456252 | 7602 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7603 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7604 | rcu_str_deref(dev->name), |
442a4f63 SB |
7605 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7606 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7607 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7608 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7609 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7610 | } |
c11d2c23 | 7611 | |
733f4fbb SB |
7612 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7613 | { | |
a98cdb85 SB |
7614 | int i; |
7615 | ||
7616 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7617 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7618 | break; | |
7619 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7620 | return; /* all values == 0, suppress message */ | |
7621 | ||
fb456252 | 7622 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7623 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7624 | rcu_str_deref(dev->name), |
733f4fbb SB |
7625 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7626 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7627 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7628 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7629 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7630 | } | |
7631 | ||
2ff7e61e | 7632 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7633 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7634 | { |
7635 | struct btrfs_device *dev; | |
0b246afa | 7636 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7637 | int i; |
7638 | ||
7639 | mutex_lock(&fs_devices->device_list_mutex); | |
b2598edf | 7640 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL); |
c11d2c23 SB |
7641 | mutex_unlock(&fs_devices->device_list_mutex); |
7642 | ||
7643 | if (!dev) { | |
0b246afa | 7644 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7645 | return -ENODEV; |
733f4fbb | 7646 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7647 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7648 | return -ENODEV; |
b27f7c0c | 7649 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7650 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7651 | if (stats->nr_items > i) | |
7652 | stats->values[i] = | |
7653 | btrfs_dev_stat_read_and_reset(dev, i); | |
7654 | else | |
4e411a7d | 7655 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7656 | } |
a69976bc AJ |
7657 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7658 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7659 | } else { |
7660 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7661 | if (stats->nr_items > i) | |
7662 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7663 | } | |
7664 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7665 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7666 | return 0; | |
7667 | } | |
a8a6dab7 | 7668 | |
935e5cc9 | 7669 | /* |
bbbf7243 NB |
7670 | * Update the size and bytes used for each device where it changed. This is |
7671 | * delayed since we would otherwise get errors while writing out the | |
7672 | * superblocks. | |
7673 | * | |
7674 | * Must be invoked during transaction commit. | |
935e5cc9 | 7675 | */ |
bbbf7243 | 7676 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7677 | { |
935e5cc9 MX |
7678 | struct btrfs_device *curr, *next; |
7679 | ||
bbbf7243 | 7680 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7681 | |
bbbf7243 | 7682 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7683 | return; |
7684 | ||
bbbf7243 NB |
7685 | /* |
7686 | * We don't need the device_list_mutex here. This list is owned by the | |
7687 | * transaction and the transaction must complete before the device is | |
7688 | * released. | |
7689 | */ | |
7690 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7691 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7692 | post_commit_list) { | |
7693 | list_del_init(&curr->post_commit_list); | |
7694 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7695 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7696 | } |
bbbf7243 | 7697 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7698 | } |
5a13f430 | 7699 | |
46df06b8 DS |
7700 | /* |
7701 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7702 | */ | |
7703 | int btrfs_bg_type_to_factor(u64 flags) | |
7704 | { | |
44b28ada DS |
7705 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7706 | ||
7707 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7708 | } |
cf90d884 QW |
7709 | |
7710 | ||
cf90d884 QW |
7711 | |
7712 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7713 | u64 chunk_offset, u64 devid, | |
7714 | u64 physical_offset, u64 physical_len) | |
7715 | { | |
c8bf1b67 | 7716 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7717 | struct extent_map *em; |
7718 | struct map_lookup *map; | |
05a37c48 | 7719 | struct btrfs_device *dev; |
cf90d884 QW |
7720 | u64 stripe_len; |
7721 | bool found = false; | |
7722 | int ret = 0; | |
7723 | int i; | |
7724 | ||
7725 | read_lock(&em_tree->lock); | |
7726 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7727 | read_unlock(&em_tree->lock); | |
7728 | ||
7729 | if (!em) { | |
7730 | btrfs_err(fs_info, | |
7731 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7732 | physical_offset, devid); | |
7733 | ret = -EUCLEAN; | |
7734 | goto out; | |
7735 | } | |
7736 | ||
7737 | map = em->map_lookup; | |
7738 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7739 | if (physical_len != stripe_len) { | |
7740 | btrfs_err(fs_info, | |
7741 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7742 | physical_offset, devid, em->start, physical_len, | |
7743 | stripe_len); | |
7744 | ret = -EUCLEAN; | |
7745 | goto out; | |
7746 | } | |
7747 | ||
7748 | for (i = 0; i < map->num_stripes; i++) { | |
7749 | if (map->stripes[i].dev->devid == devid && | |
7750 | map->stripes[i].physical == physical_offset) { | |
7751 | found = true; | |
7752 | if (map->verified_stripes >= map->num_stripes) { | |
7753 | btrfs_err(fs_info, | |
7754 | "too many dev extents for chunk %llu found", | |
7755 | em->start); | |
7756 | ret = -EUCLEAN; | |
7757 | goto out; | |
7758 | } | |
7759 | map->verified_stripes++; | |
7760 | break; | |
7761 | } | |
7762 | } | |
7763 | if (!found) { | |
7764 | btrfs_err(fs_info, | |
7765 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7766 | physical_offset, devid); | |
7767 | ret = -EUCLEAN; | |
7768 | } | |
05a37c48 QW |
7769 | |
7770 | /* Make sure no dev extent is beyond device bondary */ | |
b2598edf | 7771 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL); |
05a37c48 QW |
7772 | if (!dev) { |
7773 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7774 | ret = -EUCLEAN; | |
7775 | goto out; | |
7776 | } | |
1b3922a8 | 7777 | |
05a37c48 QW |
7778 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7779 | btrfs_err(fs_info, | |
7780 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7781 | devid, physical_offset, physical_len, | |
7782 | dev->disk_total_bytes); | |
7783 | ret = -EUCLEAN; | |
7784 | goto out; | |
7785 | } | |
381a696e NA |
7786 | |
7787 | if (dev->zone_info) { | |
7788 | u64 zone_size = dev->zone_info->zone_size; | |
7789 | ||
7790 | if (!IS_ALIGNED(physical_offset, zone_size) || | |
7791 | !IS_ALIGNED(physical_len, zone_size)) { | |
7792 | btrfs_err(fs_info, | |
7793 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", | |
7794 | devid, physical_offset, physical_len); | |
7795 | ret = -EUCLEAN; | |
7796 | goto out; | |
7797 | } | |
7798 | } | |
7799 | ||
cf90d884 QW |
7800 | out: |
7801 | free_extent_map(em); | |
7802 | return ret; | |
7803 | } | |
7804 | ||
7805 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7806 | { | |
c8bf1b67 | 7807 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7808 | struct extent_map *em; |
7809 | struct rb_node *node; | |
7810 | int ret = 0; | |
7811 | ||
7812 | read_lock(&em_tree->lock); | |
07e1ce09 | 7813 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7814 | em = rb_entry(node, struct extent_map, rb_node); |
7815 | if (em->map_lookup->num_stripes != | |
7816 | em->map_lookup->verified_stripes) { | |
7817 | btrfs_err(fs_info, | |
7818 | "chunk %llu has missing dev extent, have %d expect %d", | |
7819 | em->start, em->map_lookup->verified_stripes, | |
7820 | em->map_lookup->num_stripes); | |
7821 | ret = -EUCLEAN; | |
7822 | goto out; | |
7823 | } | |
7824 | } | |
7825 | out: | |
7826 | read_unlock(&em_tree->lock); | |
7827 | return ret; | |
7828 | } | |
7829 | ||
7830 | /* | |
7831 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7832 | * later chunk allocation/free would cause unexpected behavior. | |
7833 | * | |
7834 | * NOTE: This will iterate through the whole device tree, which should be of | |
7835 | * the same size level as the chunk tree. This slightly increases mount time. | |
7836 | */ | |
7837 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7838 | { | |
7839 | struct btrfs_path *path; | |
7840 | struct btrfs_root *root = fs_info->dev_root; | |
7841 | struct btrfs_key key; | |
5eb19381 QW |
7842 | u64 prev_devid = 0; |
7843 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7844 | int ret = 0; |
7845 | ||
42437a63 JB |
7846 | /* |
7847 | * We don't have a dev_root because we mounted with ignorebadroots and | |
7848 | * failed to load the root, so we want to skip the verification in this | |
7849 | * case for sure. | |
7850 | * | |
7851 | * However if the dev root is fine, but the tree itself is corrupted | |
7852 | * we'd still fail to mount. This verification is only to make sure | |
7853 | * writes can happen safely, so instead just bypass this check | |
7854 | * completely in the case of IGNOREBADROOTS. | |
7855 | */ | |
7856 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) | |
7857 | return 0; | |
7858 | ||
cf90d884 QW |
7859 | key.objectid = 1; |
7860 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7861 | key.offset = 0; | |
7862 | ||
7863 | path = btrfs_alloc_path(); | |
7864 | if (!path) | |
7865 | return -ENOMEM; | |
7866 | ||
7867 | path->reada = READA_FORWARD; | |
7868 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7869 | if (ret < 0) | |
7870 | goto out; | |
7871 | ||
7872 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7873 | ret = btrfs_next_item(root, path); | |
7874 | if (ret < 0) | |
7875 | goto out; | |
7876 | /* No dev extents at all? Not good */ | |
7877 | if (ret > 0) { | |
7878 | ret = -EUCLEAN; | |
7879 | goto out; | |
7880 | } | |
7881 | } | |
7882 | while (1) { | |
7883 | struct extent_buffer *leaf = path->nodes[0]; | |
7884 | struct btrfs_dev_extent *dext; | |
7885 | int slot = path->slots[0]; | |
7886 | u64 chunk_offset; | |
7887 | u64 physical_offset; | |
7888 | u64 physical_len; | |
7889 | u64 devid; | |
7890 | ||
7891 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7892 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7893 | break; | |
7894 | devid = key.objectid; | |
7895 | physical_offset = key.offset; | |
7896 | ||
7897 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7898 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7899 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7900 | ||
5eb19381 QW |
7901 | /* Check if this dev extent overlaps with the previous one */ |
7902 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7903 | btrfs_err(fs_info, | |
7904 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7905 | devid, physical_offset, prev_dev_ext_end); | |
7906 | ret = -EUCLEAN; | |
7907 | goto out; | |
7908 | } | |
7909 | ||
cf90d884 QW |
7910 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7911 | physical_offset, physical_len); | |
7912 | if (ret < 0) | |
7913 | goto out; | |
5eb19381 QW |
7914 | prev_devid = devid; |
7915 | prev_dev_ext_end = physical_offset + physical_len; | |
7916 | ||
cf90d884 QW |
7917 | ret = btrfs_next_item(root, path); |
7918 | if (ret < 0) | |
7919 | goto out; | |
7920 | if (ret > 0) { | |
7921 | ret = 0; | |
7922 | break; | |
7923 | } | |
7924 | } | |
7925 | ||
7926 | /* Ensure all chunks have corresponding dev extents */ | |
7927 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7928 | out: | |
7929 | btrfs_free_path(path); | |
7930 | return ret; | |
7931 | } | |
eede2bf3 OS |
7932 | |
7933 | /* | |
7934 | * Check whether the given block group or device is pinned by any inode being | |
7935 | * used as a swapfile. | |
7936 | */ | |
7937 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7938 | { | |
7939 | struct btrfs_swapfile_pin *sp; | |
7940 | struct rb_node *node; | |
7941 | ||
7942 | spin_lock(&fs_info->swapfile_pins_lock); | |
7943 | node = fs_info->swapfile_pins.rb_node; | |
7944 | while (node) { | |
7945 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7946 | if (ptr < sp->ptr) | |
7947 | node = node->rb_left; | |
7948 | else if (ptr > sp->ptr) | |
7949 | node = node->rb_right; | |
7950 | else | |
7951 | break; | |
7952 | } | |
7953 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7954 | return node != NULL; | |
7955 | } |