]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 CM |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
8a4b83cc | 9 | #include <linux/buffer_head.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
784352fe | 17 | #include "misc.h" |
0b86a832 CM |
18 | #include "ctree.h" |
19 | #include "extent_map.h" | |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "print-tree.h" | |
23 | #include "volumes.h" | |
53b381b3 | 24 | #include "raid56.h" |
8b712842 | 25 | #include "async-thread.h" |
21adbd5c | 26 | #include "check-integrity.h" |
606686ee | 27 | #include "rcu-string.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
b0643e59 | 33 | #include "discard.h" |
0b86a832 | 34 | |
af902047 ZL |
35 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
36 | [BTRFS_RAID_RAID10] = { | |
37 | .sub_stripes = 2, | |
38 | .dev_stripes = 1, | |
39 | .devs_max = 0, /* 0 == as many as possible */ | |
40 | .devs_min = 4, | |
8789f4fe | 41 | .tolerated_failures = 1, |
af902047 ZL |
42 | .devs_increment = 2, |
43 | .ncopies = 2, | |
b50836ed | 44 | .nparity = 0, |
ed23467b | 45 | .raid_name = "raid10", |
41a6e891 | 46 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 47 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
48 | }, |
49 | [BTRFS_RAID_RAID1] = { | |
50 | .sub_stripes = 1, | |
51 | .dev_stripes = 1, | |
52 | .devs_max = 2, | |
53 | .devs_min = 2, | |
8789f4fe | 54 | .tolerated_failures = 1, |
af902047 ZL |
55 | .devs_increment = 2, |
56 | .ncopies = 2, | |
b50836ed | 57 | .nparity = 0, |
ed23467b | 58 | .raid_name = "raid1", |
41a6e891 | 59 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 60 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 61 | }, |
47e6f742 DS |
62 | [BTRFS_RAID_RAID1C3] = { |
63 | .sub_stripes = 1, | |
64 | .dev_stripes = 1, | |
cf93e15e | 65 | .devs_max = 3, |
47e6f742 DS |
66 | .devs_min = 3, |
67 | .tolerated_failures = 2, | |
68 | .devs_increment = 3, | |
69 | .ncopies = 3, | |
db26a024 | 70 | .nparity = 0, |
47e6f742 DS |
71 | .raid_name = "raid1c3", |
72 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
73 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
74 | }, | |
8d6fac00 DS |
75 | [BTRFS_RAID_RAID1C4] = { |
76 | .sub_stripes = 1, | |
77 | .dev_stripes = 1, | |
cf93e15e | 78 | .devs_max = 4, |
8d6fac00 DS |
79 | .devs_min = 4, |
80 | .tolerated_failures = 3, | |
81 | .devs_increment = 4, | |
82 | .ncopies = 4, | |
db26a024 | 83 | .nparity = 0, |
8d6fac00 DS |
84 | .raid_name = "raid1c4", |
85 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
86 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
87 | }, | |
af902047 ZL |
88 | [BTRFS_RAID_DUP] = { |
89 | .sub_stripes = 1, | |
90 | .dev_stripes = 2, | |
91 | .devs_max = 1, | |
92 | .devs_min = 1, | |
8789f4fe | 93 | .tolerated_failures = 0, |
af902047 ZL |
94 | .devs_increment = 1, |
95 | .ncopies = 2, | |
b50836ed | 96 | .nparity = 0, |
ed23467b | 97 | .raid_name = "dup", |
41a6e891 | 98 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 99 | .mindev_error = 0, |
af902047 ZL |
100 | }, |
101 | [BTRFS_RAID_RAID0] = { | |
102 | .sub_stripes = 1, | |
103 | .dev_stripes = 1, | |
104 | .devs_max = 0, | |
105 | .devs_min = 2, | |
8789f4fe | 106 | .tolerated_failures = 0, |
af902047 ZL |
107 | .devs_increment = 1, |
108 | .ncopies = 1, | |
b50836ed | 109 | .nparity = 0, |
ed23467b | 110 | .raid_name = "raid0", |
41a6e891 | 111 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 112 | .mindev_error = 0, |
af902047 ZL |
113 | }, |
114 | [BTRFS_RAID_SINGLE] = { | |
115 | .sub_stripes = 1, | |
116 | .dev_stripes = 1, | |
117 | .devs_max = 1, | |
118 | .devs_min = 1, | |
8789f4fe | 119 | .tolerated_failures = 0, |
af902047 ZL |
120 | .devs_increment = 1, |
121 | .ncopies = 1, | |
b50836ed | 122 | .nparity = 0, |
ed23467b | 123 | .raid_name = "single", |
41a6e891 | 124 | .bg_flag = 0, |
f9fbcaa2 | 125 | .mindev_error = 0, |
af902047 ZL |
126 | }, |
127 | [BTRFS_RAID_RAID5] = { | |
128 | .sub_stripes = 1, | |
129 | .dev_stripes = 1, | |
130 | .devs_max = 0, | |
131 | .devs_min = 2, | |
8789f4fe | 132 | .tolerated_failures = 1, |
af902047 | 133 | .devs_increment = 1, |
da612e31 | 134 | .ncopies = 1, |
b50836ed | 135 | .nparity = 1, |
ed23467b | 136 | .raid_name = "raid5", |
41a6e891 | 137 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 138 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
139 | }, |
140 | [BTRFS_RAID_RAID6] = { | |
141 | .sub_stripes = 1, | |
142 | .dev_stripes = 1, | |
143 | .devs_max = 0, | |
144 | .devs_min = 3, | |
8789f4fe | 145 | .tolerated_failures = 2, |
af902047 | 146 | .devs_increment = 1, |
da612e31 | 147 | .ncopies = 1, |
b50836ed | 148 | .nparity = 2, |
ed23467b | 149 | .raid_name = "raid6", |
41a6e891 | 150 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 151 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
152 | }, |
153 | }; | |
154 | ||
158da513 | 155 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 156 | { |
158da513 DS |
157 | const int index = btrfs_bg_flags_to_raid_index(flags); |
158 | ||
159 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
160 | return NULL; |
161 | ||
158da513 | 162 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
163 | } |
164 | ||
f89e09cf AJ |
165 | /* |
166 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
167 | * bytes including terminating null byte. | |
168 | */ | |
169 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
170 | { | |
171 | int i; | |
172 | int ret; | |
173 | char *bp = buf; | |
174 | u64 flags = bg_flags; | |
175 | u32 size_bp = size_buf; | |
176 | ||
177 | if (!flags) { | |
178 | strcpy(bp, "NONE"); | |
179 | return; | |
180 | } | |
181 | ||
182 | #define DESCRIBE_FLAG(flag, desc) \ | |
183 | do { \ | |
184 | if (flags & (flag)) { \ | |
185 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
186 | if (ret < 0 || ret >= size_bp) \ | |
187 | goto out_overflow; \ | |
188 | size_bp -= ret; \ | |
189 | bp += ret; \ | |
190 | flags &= ~(flag); \ | |
191 | } \ | |
192 | } while (0) | |
193 | ||
194 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
195 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
196 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
197 | ||
198 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
199 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
200 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
201 | btrfs_raid_array[i].raid_name); | |
202 | #undef DESCRIBE_FLAG | |
203 | ||
204 | if (flags) { | |
205 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
206 | size_bp -= ret; | |
207 | } | |
208 | ||
209 | if (size_bp < size_buf) | |
210 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
211 | ||
212 | /* | |
213 | * The text is trimmed, it's up to the caller to provide sufficiently | |
214 | * large buffer | |
215 | */ | |
216 | out_overflow:; | |
217 | } | |
218 | ||
6f8e0fc7 | 219 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 220 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 221 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 222 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
223 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
224 | enum btrfs_map_op op, | |
225 | u64 logical, u64 *length, | |
226 | struct btrfs_bio **bbio_ret, | |
227 | int mirror_num, int need_raid_map); | |
2b82032c | 228 | |
9c6b1c4d DS |
229 | /* |
230 | * Device locking | |
231 | * ============== | |
232 | * | |
233 | * There are several mutexes that protect manipulation of devices and low-level | |
234 | * structures like chunks but not block groups, extents or files | |
235 | * | |
236 | * uuid_mutex (global lock) | |
237 | * ------------------------ | |
238 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
239 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
240 | * device) or requested by the device= mount option | |
241 | * | |
242 | * the mutex can be very coarse and can cover long-running operations | |
243 | * | |
244 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 245 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
246 | * |
247 | * global::fs_devs - add, remove, updates to the global list | |
248 | * | |
249 | * does not protect: manipulation of the fs_devices::devices list! | |
250 | * | |
251 | * btrfs_device::name - renames (write side), read is RCU | |
252 | * | |
253 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
254 | * ------------------------------------------------ | |
255 | * protects updates to fs_devices::devices, ie. adding and deleting | |
256 | * | |
257 | * simple list traversal with read-only actions can be done with RCU protection | |
258 | * | |
259 | * may be used to exclude some operations from running concurrently without any | |
260 | * modifications to the list (see write_all_supers) | |
261 | * | |
9c6b1c4d DS |
262 | * balance_mutex |
263 | * ------------- | |
264 | * protects balance structures (status, state) and context accessed from | |
265 | * several places (internally, ioctl) | |
266 | * | |
267 | * chunk_mutex | |
268 | * ----------- | |
269 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
270 | * device is added/removed. Additionally it also protects post_commit_list of |
271 | * individual devices, since they can be added to the transaction's | |
272 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
273 | * |
274 | * cleaner_mutex | |
275 | * ------------- | |
276 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
277 | * cleaning together with relocation or delayed iputs | |
278 | * | |
279 | * | |
280 | * Lock nesting | |
281 | * ============ | |
282 | * | |
283 | * uuid_mutex | |
284 | * volume_mutex | |
285 | * device_list_mutex | |
286 | * chunk_mutex | |
287 | * balance_mutex | |
89595e80 AJ |
288 | * |
289 | * | |
290 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
291 | * ====================================== | |
292 | * | |
293 | * Maintains the exclusivity of the following operations that apply to the | |
294 | * whole filesystem and cannot run in parallel. | |
295 | * | |
296 | * - Balance (*) | |
297 | * - Device add | |
298 | * - Device remove | |
299 | * - Device replace (*) | |
300 | * - Resize | |
301 | * | |
302 | * The device operations (as above) can be in one of the following states: | |
303 | * | |
304 | * - Running state | |
305 | * - Paused state | |
306 | * - Completed state | |
307 | * | |
308 | * Only device operations marked with (*) can go into the Paused state for the | |
309 | * following reasons: | |
310 | * | |
311 | * - ioctl (only Balance can be Paused through ioctl) | |
312 | * - filesystem remounted as read-only | |
313 | * - filesystem unmounted and mounted as read-only | |
314 | * - system power-cycle and filesystem mounted as read-only | |
315 | * - filesystem or device errors leading to forced read-only | |
316 | * | |
317 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
318 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
319 | * A device operation in Paused or Running state can be canceled or resumed | |
320 | * either by ioctl (Balance only) or when remounted as read-write. | |
321 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
322 | * completed. | |
9c6b1c4d DS |
323 | */ |
324 | ||
67a2c45e | 325 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 326 | static LIST_HEAD(fs_uuids); |
4143cb8b | 327 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
328 | { |
329 | return &fs_uuids; | |
330 | } | |
8a4b83cc | 331 | |
2dfeca9b DS |
332 | /* |
333 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
334 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
335 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
336 | * |
337 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
338 | * The returned struct is not linked onto any lists and can be destroyed with | |
339 | * kfree() right away. | |
340 | */ | |
7239ff4b NB |
341 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
342 | const u8 *metadata_fsid) | |
2208a378 ID |
343 | { |
344 | struct btrfs_fs_devices *fs_devs; | |
345 | ||
78f2c9e6 | 346 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
347 | if (!fs_devs) |
348 | return ERR_PTR(-ENOMEM); | |
349 | ||
350 | mutex_init(&fs_devs->device_list_mutex); | |
351 | ||
352 | INIT_LIST_HEAD(&fs_devs->devices); | |
353 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 354 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
355 | if (fsid) |
356 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 357 | |
7239ff4b NB |
358 | if (metadata_fsid) |
359 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
360 | else if (fsid) | |
361 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
362 | ||
2208a378 ID |
363 | return fs_devs; |
364 | } | |
365 | ||
a425f9d4 | 366 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 367 | { |
bbbf7243 | 368 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 369 | rcu_string_free(device->name); |
1c11b63e | 370 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
371 | bio_put(device->flush_bio); |
372 | kfree(device); | |
373 | } | |
374 | ||
e4404d6e YZ |
375 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
376 | { | |
377 | struct btrfs_device *device; | |
378 | WARN_ON(fs_devices->opened); | |
379 | while (!list_empty(&fs_devices->devices)) { | |
380 | device = list_entry(fs_devices->devices.next, | |
381 | struct btrfs_device, dev_list); | |
382 | list_del(&device->dev_list); | |
a425f9d4 | 383 | btrfs_free_device(device); |
e4404d6e YZ |
384 | } |
385 | kfree(fs_devices); | |
386 | } | |
387 | ||
ffc5a379 | 388 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
389 | { |
390 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 391 | |
2b82032c YZ |
392 | while (!list_empty(&fs_uuids)) { |
393 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
394 | struct btrfs_fs_devices, fs_list); |
395 | list_del(&fs_devices->fs_list); | |
e4404d6e | 396 | free_fs_devices(fs_devices); |
8a4b83cc | 397 | } |
8a4b83cc CM |
398 | } |
399 | ||
48dae9cf DS |
400 | /* |
401 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
402 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 403 | * btrfs_free_device. |
48dae9cf | 404 | */ |
12bd2fc0 ID |
405 | static struct btrfs_device *__alloc_device(void) |
406 | { | |
407 | struct btrfs_device *dev; | |
408 | ||
78f2c9e6 | 409 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
410 | if (!dev) |
411 | return ERR_PTR(-ENOMEM); | |
412 | ||
e0ae9994 DS |
413 | /* |
414 | * Preallocate a bio that's always going to be used for flushing device | |
415 | * barriers and matches the device lifespan | |
416 | */ | |
417 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
418 | if (!dev->flush_bio) { | |
419 | kfree(dev); | |
420 | return ERR_PTR(-ENOMEM); | |
421 | } | |
e0ae9994 | 422 | |
12bd2fc0 ID |
423 | INIT_LIST_HEAD(&dev->dev_list); |
424 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 425 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 | 426 | |
12bd2fc0 | 427 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 428 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 429 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 430 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 431 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
1c11b63e | 432 | extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL); |
12bd2fc0 ID |
433 | |
434 | return dev; | |
435 | } | |
436 | ||
7239ff4b NB |
437 | static noinline struct btrfs_fs_devices *find_fsid( |
438 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 439 | { |
8a4b83cc CM |
440 | struct btrfs_fs_devices *fs_devices; |
441 | ||
7239ff4b NB |
442 | ASSERT(fsid); |
443 | ||
7a62d0f0 | 444 | /* Handle non-split brain cases */ |
c4babc5e | 445 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
446 | if (metadata_fsid) { |
447 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
448 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
449 | BTRFS_FSID_SIZE) == 0) | |
450 | return fs_devices; | |
451 | } else { | |
452 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
453 | return fs_devices; | |
454 | } | |
8a4b83cc CM |
455 | } |
456 | return NULL; | |
457 | } | |
458 | ||
c6730a0e SY |
459 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
460 | struct btrfs_super_block *disk_super) | |
461 | { | |
462 | ||
463 | struct btrfs_fs_devices *fs_devices; | |
464 | ||
465 | /* | |
466 | * Handle scanned device having completed its fsid change but | |
467 | * belonging to a fs_devices that was created by first scanning | |
468 | * a device which didn't have its fsid/metadata_uuid changed | |
469 | * at all and the CHANGING_FSID_V2 flag set. | |
470 | */ | |
471 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
472 | if (fs_devices->fsid_change && | |
473 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
474 | BTRFS_FSID_SIZE) == 0 && | |
475 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
476 | BTRFS_FSID_SIZE) == 0) { | |
477 | return fs_devices; | |
478 | } | |
479 | } | |
480 | /* | |
481 | * Handle scanned device having completed its fsid change but | |
482 | * belonging to a fs_devices that was created by a device that | |
483 | * has an outdated pair of fsid/metadata_uuid and | |
484 | * CHANGING_FSID_V2 flag set. | |
485 | */ | |
486 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
487 | if (fs_devices->fsid_change && | |
488 | memcmp(fs_devices->metadata_uuid, | |
489 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
490 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
491 | BTRFS_FSID_SIZE) == 0) { | |
492 | return fs_devices; | |
493 | } | |
494 | } | |
495 | ||
496 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
497 | } | |
498 | ||
499 | ||
beaf8ab3 SB |
500 | static int |
501 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
502 | int flush, struct block_device **bdev, | |
503 | struct buffer_head **bh) | |
504 | { | |
505 | int ret; | |
506 | ||
507 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
508 | ||
509 | if (IS_ERR(*bdev)) { | |
510 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
511 | goto error; |
512 | } | |
513 | ||
514 | if (flush) | |
515 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 516 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
517 | if (ret) { |
518 | blkdev_put(*bdev, flags); | |
519 | goto error; | |
520 | } | |
521 | invalidate_bdev(*bdev); | |
522 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
523 | if (IS_ERR(*bh)) { |
524 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
525 | blkdev_put(*bdev, flags); |
526 | goto error; | |
527 | } | |
528 | ||
529 | return 0; | |
530 | ||
531 | error: | |
532 | *bdev = NULL; | |
533 | *bh = NULL; | |
534 | return ret; | |
535 | } | |
536 | ||
70bc7088 AJ |
537 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
538 | { | |
539 | int found; | |
540 | ||
541 | rcu_read_lock(); | |
542 | found = strcmp(rcu_str_deref(device->name), path); | |
543 | rcu_read_unlock(); | |
544 | ||
545 | return found == 0; | |
546 | } | |
547 | ||
d8367db3 AJ |
548 | /* |
549 | * Search and remove all stale (devices which are not mounted) devices. | |
550 | * When both inputs are NULL, it will search and release all stale devices. | |
551 | * path: Optional. When provided will it release all unmounted devices | |
552 | * matching this path only. | |
553 | * skip_dev: Optional. Will skip this device when searching for the stale | |
554 | * devices. | |
70bc7088 AJ |
555 | * Return: 0 for success or if @path is NULL. |
556 | * -EBUSY if @path is a mounted device. | |
557 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 558 | */ |
70bc7088 | 559 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 560 | struct btrfs_device *skip_device) |
4fde46f0 | 561 | { |
fa6d2ae5 AJ |
562 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
563 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
564 | int ret = 0; |
565 | ||
566 | if (path) | |
567 | ret = -ENOENT; | |
4fde46f0 | 568 | |
fa6d2ae5 | 569 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 570 | |
70bc7088 | 571 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
572 | list_for_each_entry_safe(device, tmp_device, |
573 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 574 | if (skip_device && skip_device == device) |
d8367db3 | 575 | continue; |
fa6d2ae5 | 576 | if (path && !device->name) |
4fde46f0 | 577 | continue; |
70bc7088 | 578 | if (path && !device_path_matched(path, device)) |
38cf665d | 579 | continue; |
70bc7088 AJ |
580 | if (fs_devices->opened) { |
581 | /* for an already deleted device return 0 */ | |
582 | if (path && ret != 0) | |
583 | ret = -EBUSY; | |
584 | break; | |
585 | } | |
4fde46f0 | 586 | |
4fde46f0 | 587 | /* delete the stale device */ |
7bcb8164 AJ |
588 | fs_devices->num_devices--; |
589 | list_del(&device->dev_list); | |
590 | btrfs_free_device(device); | |
591 | ||
70bc7088 | 592 | ret = 0; |
7bcb8164 | 593 | if (fs_devices->num_devices == 0) |
fd649f10 | 594 | break; |
7bcb8164 AJ |
595 | } |
596 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 597 | |
7bcb8164 AJ |
598 | if (fs_devices->num_devices == 0) { |
599 | btrfs_sysfs_remove_fsid(fs_devices); | |
600 | list_del(&fs_devices->fs_list); | |
601 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
602 | } |
603 | } | |
70bc7088 AJ |
604 | |
605 | return ret; | |
4fde46f0 AJ |
606 | } |
607 | ||
0fb08bcc AJ |
608 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
609 | struct btrfs_device *device, fmode_t flags, | |
610 | void *holder) | |
611 | { | |
612 | struct request_queue *q; | |
613 | struct block_device *bdev; | |
614 | struct buffer_head *bh; | |
615 | struct btrfs_super_block *disk_super; | |
616 | u64 devid; | |
617 | int ret; | |
618 | ||
619 | if (device->bdev) | |
620 | return -EINVAL; | |
621 | if (!device->name) | |
622 | return -EINVAL; | |
623 | ||
624 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
625 | &bdev, &bh); | |
626 | if (ret) | |
627 | return ret; | |
628 | ||
629 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
630 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
631 | if (devid != device->devid) | |
632 | goto error_brelse; | |
633 | ||
634 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
635 | goto error_brelse; | |
636 | ||
637 | device->generation = btrfs_super_generation(disk_super); | |
638 | ||
639 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
640 | if (btrfs_super_incompat_flags(disk_super) & |
641 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
642 | pr_err( | |
643 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
644 | goto error_brelse; | |
645 | } | |
646 | ||
ebbede42 | 647 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 648 | fs_devices->seeding = true; |
0fb08bcc | 649 | } else { |
ebbede42 AJ |
650 | if (bdev_read_only(bdev)) |
651 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
652 | else | |
653 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
654 | } |
655 | ||
656 | q = bdev_get_queue(bdev); | |
0fb08bcc | 657 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 658 | fs_devices->rotating = true; |
0fb08bcc AJ |
659 | |
660 | device->bdev = bdev; | |
e12c9621 | 661 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
662 | device->mode = flags; |
663 | ||
664 | fs_devices->open_devices++; | |
ebbede42 AJ |
665 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
666 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 667 | fs_devices->rw_devices++; |
b1b8e386 | 668 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
669 | } |
670 | brelse(bh); | |
671 | ||
672 | return 0; | |
673 | ||
674 | error_brelse: | |
675 | brelse(bh); | |
676 | blkdev_put(bdev, flags); | |
677 | ||
678 | return -EINVAL; | |
679 | } | |
680 | ||
7a62d0f0 NB |
681 | /* |
682 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
683 | * being created with a disk that has already completed its fsid change. Such |
684 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
685 | * Handle both cases here. | |
7a62d0f0 NB |
686 | */ |
687 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
688 | struct btrfs_super_block *disk_super) | |
689 | { | |
690 | struct btrfs_fs_devices *fs_devices; | |
691 | ||
692 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
693 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
694 | BTRFS_FSID_SIZE) != 0 && | |
695 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
696 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
697 | return fs_devices; | |
698 | } | |
699 | } | |
700 | ||
c0d81c7c | 701 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
702 | } |
703 | ||
cc5de4e7 NB |
704 | |
705 | static struct btrfs_fs_devices *find_fsid_changed( | |
706 | struct btrfs_super_block *disk_super) | |
707 | { | |
708 | struct btrfs_fs_devices *fs_devices; | |
709 | ||
710 | /* | |
711 | * Handles the case where scanned device is part of an fs that had | |
712 | * multiple successful changes of FSID but curently device didn't | |
05840710 NB |
713 | * observe it. Meaning our fsid will be different than theirs. We need |
714 | * to handle two subcases : | |
715 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
716 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
717 | * are equal). | |
cc5de4e7 NB |
718 | */ |
719 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 720 | /* Changed UUIDs */ |
cc5de4e7 NB |
721 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
722 | BTRFS_FSID_SIZE) != 0 && | |
723 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
724 | BTRFS_FSID_SIZE) == 0 && | |
725 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
726 | BTRFS_FSID_SIZE) != 0) |
727 | return fs_devices; | |
728 | ||
729 | /* Unchanged UUIDs */ | |
730 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
731 | BTRFS_FSID_SIZE) == 0 && | |
732 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
733 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 734 | return fs_devices; |
cc5de4e7 NB |
735 | } |
736 | ||
737 | return NULL; | |
738 | } | |
1362089d NB |
739 | |
740 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
741 | struct btrfs_super_block *disk_super) | |
742 | { | |
743 | struct btrfs_fs_devices *fs_devices; | |
744 | ||
745 | /* | |
746 | * Handle the case where the scanned device is part of an fs whose last | |
747 | * metadata UUID change reverted it to the original FSID. At the same | |
748 | * time * fs_devices was first created by another constitutent device | |
749 | * which didn't fully observe the operation. This results in an | |
750 | * btrfs_fs_devices created with metadata/fsid different AND | |
751 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
752 | * fs_devices equal to the FSID of the disk. | |
753 | */ | |
754 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
755 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
756 | BTRFS_FSID_SIZE) != 0 && | |
757 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
758 | BTRFS_FSID_SIZE) == 0 && | |
759 | fs_devices->fsid_change) | |
760 | return fs_devices; | |
761 | } | |
762 | ||
763 | return NULL; | |
764 | } | |
60999ca4 DS |
765 | /* |
766 | * Add new device to list of registered devices | |
767 | * | |
768 | * Returns: | |
e124ece5 AJ |
769 | * device pointer which was just added or updated when successful |
770 | * error pointer when failed | |
60999ca4 | 771 | */ |
e124ece5 | 772 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
773 | struct btrfs_super_block *disk_super, |
774 | bool *new_device_added) | |
8a4b83cc CM |
775 | { |
776 | struct btrfs_device *device; | |
7a62d0f0 | 777 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 778 | struct rcu_string *name; |
8a4b83cc | 779 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 780 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
781 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
782 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
783 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
784 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 785 | |
cc5de4e7 | 786 | if (fsid_change_in_progress) { |
c0d81c7c | 787 | if (!has_metadata_uuid) |
cc5de4e7 | 788 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 789 | else |
cc5de4e7 | 790 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 791 | } else if (has_metadata_uuid) { |
c6730a0e | 792 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 793 | } else { |
1362089d NB |
794 | fs_devices = find_fsid_reverted_metadata(disk_super); |
795 | if (!fs_devices) | |
796 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
797 | } |
798 | ||
8a4b83cc | 799 | |
8a4b83cc | 800 | if (!fs_devices) { |
7239ff4b NB |
801 | if (has_metadata_uuid) |
802 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
803 | disk_super->metadata_uuid); | |
804 | else | |
805 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
806 | ||
2208a378 | 807 | if (IS_ERR(fs_devices)) |
e124ece5 | 808 | return ERR_CAST(fs_devices); |
2208a378 | 809 | |
92900e51 AV |
810 | fs_devices->fsid_change = fsid_change_in_progress; |
811 | ||
9c6d173e | 812 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 813 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 814 | |
8a4b83cc CM |
815 | device = NULL; |
816 | } else { | |
9c6d173e | 817 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
818 | device = btrfs_find_device(fs_devices, devid, |
819 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
820 | |
821 | /* | |
822 | * If this disk has been pulled into an fs devices created by | |
823 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
824 | * metadata_uuid/fsid values of the fs_devices. | |
825 | */ | |
1362089d | 826 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
827 | found_transid > fs_devices->latest_generation) { |
828 | memcpy(fs_devices->fsid, disk_super->fsid, | |
829 | BTRFS_FSID_SIZE); | |
1362089d NB |
830 | |
831 | if (has_metadata_uuid) | |
832 | memcpy(fs_devices->metadata_uuid, | |
833 | disk_super->metadata_uuid, | |
834 | BTRFS_FSID_SIZE); | |
835 | else | |
836 | memcpy(fs_devices->metadata_uuid, | |
837 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
838 | |
839 | fs_devices->fsid_change = false; | |
840 | } | |
8a4b83cc | 841 | } |
443f24fe | 842 | |
8a4b83cc | 843 | if (!device) { |
9c6d173e AJ |
844 | if (fs_devices->opened) { |
845 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 846 | return ERR_PTR(-EBUSY); |
9c6d173e | 847 | } |
2b82032c | 848 | |
12bd2fc0 ID |
849 | device = btrfs_alloc_device(NULL, &devid, |
850 | disk_super->dev_item.uuid); | |
851 | if (IS_ERR(device)) { | |
9c6d173e | 852 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 853 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 854 | return device; |
8a4b83cc | 855 | } |
606686ee JB |
856 | |
857 | name = rcu_string_strdup(path, GFP_NOFS); | |
858 | if (!name) { | |
a425f9d4 | 859 | btrfs_free_device(device); |
9c6d173e | 860 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 861 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 862 | } |
606686ee | 863 | rcu_assign_pointer(device->name, name); |
90519d66 | 864 | |
1f78160c | 865 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 866 | fs_devices->num_devices++; |
e5e9a520 | 867 | |
2b82032c | 868 | device->fs_devices = fs_devices; |
4306a974 | 869 | *new_device_added = true; |
327f18cc AJ |
870 | |
871 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
872 | pr_info( |
873 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
874 | disk_super->label, devid, found_transid, path, | |
875 | current->comm, task_pid_nr(current)); | |
327f18cc | 876 | else |
aa6c0df7 AJ |
877 | pr_info( |
878 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
879 | disk_super->fsid, devid, found_transid, path, | |
880 | current->comm, task_pid_nr(current)); | |
327f18cc | 881 | |
606686ee | 882 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
883 | /* |
884 | * When FS is already mounted. | |
885 | * 1. If you are here and if the device->name is NULL that | |
886 | * means this device was missing at time of FS mount. | |
887 | * 2. If you are here and if the device->name is different | |
888 | * from 'path' that means either | |
889 | * a. The same device disappeared and reappeared with | |
890 | * different name. or | |
891 | * b. The missing-disk-which-was-replaced, has | |
892 | * reappeared now. | |
893 | * | |
894 | * We must allow 1 and 2a above. But 2b would be a spurious | |
895 | * and unintentional. | |
896 | * | |
897 | * Further in case of 1 and 2a above, the disk at 'path' | |
898 | * would have missed some transaction when it was away and | |
899 | * in case of 2a the stale bdev has to be updated as well. | |
900 | * 2b must not be allowed at all time. | |
901 | */ | |
902 | ||
903 | /* | |
0f23ae74 CM |
904 | * For now, we do allow update to btrfs_fs_device through the |
905 | * btrfs dev scan cli after FS has been mounted. We're still | |
906 | * tracking a problem where systems fail mount by subvolume id | |
907 | * when we reject replacement on a mounted FS. | |
b96de000 | 908 | */ |
0f23ae74 | 909 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
910 | /* |
911 | * That is if the FS is _not_ mounted and if you | |
912 | * are here, that means there is more than one | |
913 | * disk with same uuid and devid.We keep the one | |
914 | * with larger generation number or the last-in if | |
915 | * generation are equal. | |
916 | */ | |
9c6d173e | 917 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 918 | return ERR_PTR(-EEXIST); |
77bdae4d | 919 | } |
b96de000 | 920 | |
a9261d41 AJ |
921 | /* |
922 | * We are going to replace the device path for a given devid, | |
923 | * make sure it's the same device if the device is mounted | |
924 | */ | |
925 | if (device->bdev) { | |
926 | struct block_device *path_bdev; | |
927 | ||
928 | path_bdev = lookup_bdev(path); | |
929 | if (IS_ERR(path_bdev)) { | |
930 | mutex_unlock(&fs_devices->device_list_mutex); | |
931 | return ERR_CAST(path_bdev); | |
932 | } | |
933 | ||
934 | if (device->bdev != path_bdev) { | |
935 | bdput(path_bdev); | |
936 | mutex_unlock(&fs_devices->device_list_mutex); | |
937 | btrfs_warn_in_rcu(device->fs_info, | |
938 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
939 | disk_super->fsid, devid, | |
940 | rcu_str_deref(device->name), path); | |
941 | return ERR_PTR(-EEXIST); | |
942 | } | |
943 | bdput(path_bdev); | |
944 | btrfs_info_in_rcu(device->fs_info, | |
945 | "device fsid %pU devid %llu moved old:%s new:%s", | |
946 | disk_super->fsid, devid, | |
947 | rcu_str_deref(device->name), path); | |
948 | } | |
949 | ||
606686ee | 950 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
951 | if (!name) { |
952 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 953 | return ERR_PTR(-ENOMEM); |
9c6d173e | 954 | } |
606686ee JB |
955 | rcu_string_free(device->name); |
956 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 957 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 958 | fs_devices->missing_devices--; |
e6e674bd | 959 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 960 | } |
8a4b83cc CM |
961 | } |
962 | ||
77bdae4d AJ |
963 | /* |
964 | * Unmount does not free the btrfs_device struct but would zero | |
965 | * generation along with most of the other members. So just update | |
966 | * it back. We need it to pick the disk with largest generation | |
967 | * (as above). | |
968 | */ | |
d1a63002 | 969 | if (!fs_devices->opened) { |
77bdae4d | 970 | device->generation = found_transid; |
d1a63002 NB |
971 | fs_devices->latest_generation = max_t(u64, found_transid, |
972 | fs_devices->latest_generation); | |
973 | } | |
77bdae4d | 974 | |
f2788d2f AJ |
975 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
976 | ||
9c6d173e | 977 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 978 | return device; |
8a4b83cc CM |
979 | } |
980 | ||
e4404d6e YZ |
981 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
982 | { | |
983 | struct btrfs_fs_devices *fs_devices; | |
984 | struct btrfs_device *device; | |
985 | struct btrfs_device *orig_dev; | |
d2979aa2 | 986 | int ret = 0; |
e4404d6e | 987 | |
7239ff4b | 988 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
989 | if (IS_ERR(fs_devices)) |
990 | return fs_devices; | |
e4404d6e | 991 | |
adbbb863 | 992 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 993 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
994 | |
995 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
996 | struct rcu_string *name; |
997 | ||
12bd2fc0 ID |
998 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
999 | orig_dev->uuid); | |
d2979aa2 AJ |
1000 | if (IS_ERR(device)) { |
1001 | ret = PTR_ERR(device); | |
e4404d6e | 1002 | goto error; |
d2979aa2 | 1003 | } |
e4404d6e | 1004 | |
606686ee JB |
1005 | /* |
1006 | * This is ok to do without rcu read locked because we hold the | |
1007 | * uuid mutex so nothing we touch in here is going to disappear. | |
1008 | */ | |
e755f780 | 1009 | if (orig_dev->name) { |
78f2c9e6 DS |
1010 | name = rcu_string_strdup(orig_dev->name->str, |
1011 | GFP_KERNEL); | |
e755f780 | 1012 | if (!name) { |
a425f9d4 | 1013 | btrfs_free_device(device); |
d2979aa2 | 1014 | ret = -ENOMEM; |
e755f780 AJ |
1015 | goto error; |
1016 | } | |
1017 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1018 | } |
e4404d6e | 1019 | |
e4404d6e YZ |
1020 | list_add(&device->dev_list, &fs_devices->devices); |
1021 | device->fs_devices = fs_devices; | |
1022 | fs_devices->num_devices++; | |
1023 | } | |
adbbb863 | 1024 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1025 | return fs_devices; |
1026 | error: | |
adbbb863 | 1027 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1028 | free_fs_devices(fs_devices); |
d2979aa2 | 1029 | return ERR_PTR(ret); |
e4404d6e YZ |
1030 | } |
1031 | ||
9b99b115 AJ |
1032 | /* |
1033 | * After we have read the system tree and know devids belonging to | |
1034 | * this filesystem, remove the device which does not belong there. | |
1035 | */ | |
1036 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1037 | { |
c6e30871 | 1038 | struct btrfs_device *device, *next; |
443f24fe | 1039 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1040 | |
dfe25020 CM |
1041 | mutex_lock(&uuid_mutex); |
1042 | again: | |
46224705 | 1043 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1044 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1045 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1046 | &device->dev_state)) { | |
401e29c1 AJ |
1047 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1048 | &device->dev_state) && | |
1049 | (!latest_dev || | |
1050 | device->generation > latest_dev->generation)) { | |
443f24fe | 1051 | latest_dev = device; |
a6b0d5c8 | 1052 | } |
2b82032c | 1053 | continue; |
a6b0d5c8 | 1054 | } |
2b82032c | 1055 | |
8dabb742 SB |
1056 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1057 | /* | |
1058 | * In the first step, keep the device which has | |
1059 | * the correct fsid and the devid that is used | |
1060 | * for the dev_replace procedure. | |
1061 | * In the second step, the dev_replace state is | |
1062 | * read from the device tree and it is known | |
1063 | * whether the procedure is really active or | |
1064 | * not, which means whether this device is | |
1065 | * used or whether it should be removed. | |
1066 | */ | |
401e29c1 AJ |
1067 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1068 | &device->dev_state)) { | |
8dabb742 SB |
1069 | continue; |
1070 | } | |
1071 | } | |
2b82032c | 1072 | if (device->bdev) { |
d4d77629 | 1073 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1074 | device->bdev = NULL; |
1075 | fs_devices->open_devices--; | |
1076 | } | |
ebbede42 | 1077 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1078 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1079 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1080 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1081 | &device->dev_state)) | |
8dabb742 | 1082 | fs_devices->rw_devices--; |
2b82032c | 1083 | } |
e4404d6e YZ |
1084 | list_del_init(&device->dev_list); |
1085 | fs_devices->num_devices--; | |
a425f9d4 | 1086 | btrfs_free_device(device); |
dfe25020 | 1087 | } |
2b82032c YZ |
1088 | |
1089 | if (fs_devices->seed) { | |
1090 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1091 | goto again; |
1092 | } | |
1093 | ||
443f24fe | 1094 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1095 | |
dfe25020 | 1096 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1097 | } |
a0af469b | 1098 | |
14238819 AJ |
1099 | static void btrfs_close_bdev(struct btrfs_device *device) |
1100 | { | |
08ffcae8 DS |
1101 | if (!device->bdev) |
1102 | return; | |
1103 | ||
ebbede42 | 1104 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1105 | sync_blockdev(device->bdev); |
1106 | invalidate_bdev(device->bdev); | |
1107 | } | |
1108 | ||
08ffcae8 | 1109 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1110 | } |
1111 | ||
959b1c04 | 1112 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1113 | { |
1114 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1115 | |
ebbede42 | 1116 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1117 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1118 | list_del_init(&device->dev_alloc_list); | |
1119 | fs_devices->rw_devices--; | |
1120 | } | |
1121 | ||
e6e674bd | 1122 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1123 | fs_devices->missing_devices--; |
1124 | ||
959b1c04 | 1125 | btrfs_close_bdev(device); |
321f69f8 | 1126 | if (device->bdev) { |
3fff3975 | 1127 | fs_devices->open_devices--; |
321f69f8 | 1128 | device->bdev = NULL; |
f448341a | 1129 | } |
321f69f8 | 1130 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
f448341a | 1131 | |
321f69f8 JT |
1132 | device->fs_info = NULL; |
1133 | atomic_set(&device->dev_stats_ccnt, 0); | |
1134 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1135 | |
321f69f8 JT |
1136 | /* Verify the device is back in a pristine state */ |
1137 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1138 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1139 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1140 | ASSERT(list_empty(&device->post_commit_list)); | |
1141 | ASSERT(atomic_read(&device->reada_in_flight) == 0); | |
f448341a AJ |
1142 | } |
1143 | ||
0226e0eb | 1144 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1145 | { |
2037a093 | 1146 | struct btrfs_device *device, *tmp; |
e4404d6e | 1147 | |
2b82032c YZ |
1148 | if (--fs_devices->opened > 0) |
1149 | return 0; | |
8a4b83cc | 1150 | |
c9513edb | 1151 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1152 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1153 | btrfs_close_one_device(device); |
8a4b83cc | 1154 | } |
c9513edb XG |
1155 | mutex_unlock(&fs_devices->device_list_mutex); |
1156 | ||
e4404d6e YZ |
1157 | WARN_ON(fs_devices->open_devices); |
1158 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1159 | fs_devices->opened = 0; |
0395d84f | 1160 | fs_devices->seeding = false; |
2b82032c | 1161 | |
8a4b83cc CM |
1162 | return 0; |
1163 | } | |
1164 | ||
2b82032c YZ |
1165 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1166 | { | |
e4404d6e | 1167 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1168 | int ret; |
1169 | ||
1170 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1171 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1172 | if (!fs_devices->opened) { |
1173 | seed_devices = fs_devices->seed; | |
1174 | fs_devices->seed = NULL; | |
1175 | } | |
2b82032c | 1176 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1177 | |
1178 | while (seed_devices) { | |
1179 | fs_devices = seed_devices; | |
1180 | seed_devices = fs_devices->seed; | |
0226e0eb | 1181 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1182 | free_fs_devices(fs_devices); |
1183 | } | |
2b82032c YZ |
1184 | return ret; |
1185 | } | |
1186 | ||
897fb573 | 1187 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1188 | fmode_t flags, void *holder) |
8a4b83cc | 1189 | { |
8a4b83cc | 1190 | struct btrfs_device *device; |
443f24fe | 1191 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1192 | int ret = 0; |
8a4b83cc | 1193 | |
d4d77629 TH |
1194 | flags |= FMODE_EXCL; |
1195 | ||
f117e290 | 1196 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1197 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1198 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1199 | continue; |
a0af469b | 1200 | |
9f050db4 AJ |
1201 | if (!latest_dev || |
1202 | device->generation > latest_dev->generation) | |
1203 | latest_dev = device; | |
8a4b83cc | 1204 | } |
a0af469b | 1205 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1206 | ret = -EINVAL; |
a0af469b CM |
1207 | goto out; |
1208 | } | |
2b82032c | 1209 | fs_devices->opened = 1; |
443f24fe | 1210 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1211 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1212 | out: |
2b82032c YZ |
1213 | return ret; |
1214 | } | |
1215 | ||
f8e10cd3 AJ |
1216 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1217 | { | |
1218 | struct btrfs_device *dev1, *dev2; | |
1219 | ||
1220 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1221 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1222 | ||
1223 | if (dev1->devid < dev2->devid) | |
1224 | return -1; | |
1225 | else if (dev1->devid > dev2->devid) | |
1226 | return 1; | |
1227 | return 0; | |
1228 | } | |
1229 | ||
2b82032c | 1230 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1231 | fmode_t flags, void *holder) |
2b82032c YZ |
1232 | { |
1233 | int ret; | |
1234 | ||
f5194e34 DS |
1235 | lockdep_assert_held(&uuid_mutex); |
1236 | ||
542c5908 | 1237 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1238 | if (fs_devices->opened) { |
e4404d6e YZ |
1239 | fs_devices->opened++; |
1240 | ret = 0; | |
2b82032c | 1241 | } else { |
f8e10cd3 | 1242 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1243 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1244 | } |
542c5908 AJ |
1245 | mutex_unlock(&fs_devices->device_list_mutex); |
1246 | ||
8a4b83cc CM |
1247 | return ret; |
1248 | } | |
1249 | ||
c9162bdf | 1250 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1251 | { |
1252 | kunmap(page); | |
1253 | put_page(page); | |
1254 | } | |
1255 | ||
c9162bdf OS |
1256 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1257 | struct page **page, | |
1258 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1259 | { |
1260 | void *p; | |
1261 | pgoff_t index; | |
1262 | ||
1263 | /* make sure our super fits in the device */ | |
1264 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1265 | return 1; | |
1266 | ||
1267 | /* make sure our super fits in the page */ | |
1268 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1269 | return 1; | |
1270 | ||
1271 | /* make sure our super doesn't straddle pages on disk */ | |
1272 | index = bytenr >> PAGE_SHIFT; | |
1273 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1274 | return 1; | |
1275 | ||
1276 | /* pull in the page with our super */ | |
1277 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1278 | index, GFP_KERNEL); | |
1279 | ||
1280 | if (IS_ERR_OR_NULL(*page)) | |
1281 | return 1; | |
1282 | ||
1283 | p = kmap(*page); | |
1284 | ||
1285 | /* align our pointer to the offset of the super block */ | |
7073017a | 1286 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1287 | |
1288 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1289 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1290 | btrfs_release_disk_super(*page); | |
1291 | return 1; | |
1292 | } | |
1293 | ||
1294 | if ((*disk_super)->label[0] && | |
1295 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1296 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
228a73ab AJ |
1301 | int btrfs_forget_devices(const char *path) |
1302 | { | |
1303 | int ret; | |
1304 | ||
1305 | mutex_lock(&uuid_mutex); | |
1306 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1307 | mutex_unlock(&uuid_mutex); | |
1308 | ||
1309 | return ret; | |
1310 | } | |
1311 | ||
6f60cbd3 DS |
1312 | /* |
1313 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1314 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1315 | * is read via pagecache | |
1316 | */ | |
36350e95 GJ |
1317 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1318 | void *holder) | |
8a4b83cc CM |
1319 | { |
1320 | struct btrfs_super_block *disk_super; | |
4306a974 | 1321 | bool new_device_added = false; |
36350e95 | 1322 | struct btrfs_device *device = NULL; |
8a4b83cc | 1323 | struct block_device *bdev; |
6f60cbd3 | 1324 | struct page *page; |
6f60cbd3 | 1325 | u64 bytenr; |
8a4b83cc | 1326 | |
899f9307 DS |
1327 | lockdep_assert_held(&uuid_mutex); |
1328 | ||
6f60cbd3 DS |
1329 | /* |
1330 | * we would like to check all the supers, but that would make | |
1331 | * a btrfs mount succeed after a mkfs from a different FS. | |
1332 | * So, we need to add a special mount option to scan for | |
1333 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1334 | */ | |
1335 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1336 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1337 | |
1338 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1339 | if (IS_ERR(bdev)) |
36350e95 | 1340 | return ERR_CAST(bdev); |
6f60cbd3 | 1341 | |
05a5c55d | 1342 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1343 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1344 | goto error_bdev_put; |
05a5c55d | 1345 | } |
6f60cbd3 | 1346 | |
4306a974 | 1347 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1348 | if (!IS_ERR(device)) { |
4306a974 AJ |
1349 | if (new_device_added) |
1350 | btrfs_free_stale_devices(path, device); | |
1351 | } | |
6f60cbd3 | 1352 | |
6cf86a00 | 1353 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1354 | |
1355 | error_bdev_put: | |
d4d77629 | 1356 | blkdev_put(bdev, flags); |
b6ed73bc | 1357 | |
36350e95 | 1358 | return device; |
8a4b83cc | 1359 | } |
0b86a832 | 1360 | |
1c11b63e JM |
1361 | /* |
1362 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1363 | * such is found, record the end of it in *start | |
1364 | */ | |
1c11b63e JM |
1365 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1366 | u64 len) | |
6df9a95e | 1367 | { |
1c11b63e | 1368 | u64 physical_start, physical_end; |
6df9a95e | 1369 | |
1c11b63e | 1370 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1371 | |
1c11b63e JM |
1372 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1373 | &physical_start, &physical_end, | |
1374 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1375 | |
1c11b63e JM |
1376 | if (in_range(physical_start, *start, len) || |
1377 | in_range(*start, physical_start, | |
1378 | physical_end - physical_start)) { | |
1379 | *start = physical_end + 1; | |
1380 | return true; | |
6df9a95e JB |
1381 | } |
1382 | } | |
1c11b63e | 1383 | return false; |
6df9a95e JB |
1384 | } |
1385 | ||
1386 | ||
0b86a832 | 1387 | /* |
499f377f JM |
1388 | * find_free_dev_extent_start - find free space in the specified device |
1389 | * @device: the device which we search the free space in | |
1390 | * @num_bytes: the size of the free space that we need | |
1391 | * @search_start: the position from which to begin the search | |
1392 | * @start: store the start of the free space. | |
1393 | * @len: the size of the free space. that we find, or the size | |
1394 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1395 | * |
0b86a832 CM |
1396 | * this uses a pretty simple search, the expectation is that it is |
1397 | * called very infrequently and that a given device has a small number | |
1398 | * of extents | |
7bfc837d MX |
1399 | * |
1400 | * @start is used to store the start of the free space if we find. But if we | |
1401 | * don't find suitable free space, it will be used to store the start position | |
1402 | * of the max free space. | |
1403 | * | |
1404 | * @len is used to store the size of the free space that we find. | |
1405 | * But if we don't find suitable free space, it is used to store the size of | |
1406 | * the max free space. | |
135da976 QW |
1407 | * |
1408 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1409 | * check to ensure dev extents are not double allocated. | |
1410 | * This makes the function safe to allocate dev extents but may not report | |
1411 | * correct usable device space, as device extent freed in current transaction | |
1412 | * is not reported as avaiable. | |
0b86a832 | 1413 | */ |
9e3246a5 QW |
1414 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1415 | u64 num_bytes, u64 search_start, u64 *start, | |
1416 | u64 *len) | |
0b86a832 | 1417 | { |
0b246afa JM |
1418 | struct btrfs_fs_info *fs_info = device->fs_info; |
1419 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1420 | struct btrfs_key key; |
7bfc837d | 1421 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1422 | struct btrfs_path *path; |
7bfc837d MX |
1423 | u64 hole_size; |
1424 | u64 max_hole_start; | |
1425 | u64 max_hole_size; | |
1426 | u64 extent_end; | |
0b86a832 CM |
1427 | u64 search_end = device->total_bytes; |
1428 | int ret; | |
7bfc837d | 1429 | int slot; |
0b86a832 | 1430 | struct extent_buffer *l; |
8cdc7c5b FM |
1431 | |
1432 | /* | |
1433 | * We don't want to overwrite the superblock on the drive nor any area | |
1434 | * used by the boot loader (grub for example), so we make sure to start | |
1435 | * at an offset of at least 1MB. | |
1436 | */ | |
0d0c71b3 | 1437 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1438 | |
6df9a95e JB |
1439 | path = btrfs_alloc_path(); |
1440 | if (!path) | |
1441 | return -ENOMEM; | |
f2ab7618 | 1442 | |
7bfc837d MX |
1443 | max_hole_start = search_start; |
1444 | max_hole_size = 0; | |
1445 | ||
f2ab7618 | 1446 | again: |
401e29c1 AJ |
1447 | if (search_start >= search_end || |
1448 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1449 | ret = -ENOSPC; |
6df9a95e | 1450 | goto out; |
7bfc837d MX |
1451 | } |
1452 | ||
e4058b54 | 1453 | path->reada = READA_FORWARD; |
6df9a95e JB |
1454 | path->search_commit_root = 1; |
1455 | path->skip_locking = 1; | |
7bfc837d | 1456 | |
0b86a832 CM |
1457 | key.objectid = device->devid; |
1458 | key.offset = search_start; | |
1459 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1460 | |
125ccb0a | 1461 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1462 | if (ret < 0) |
7bfc837d | 1463 | goto out; |
1fcbac58 YZ |
1464 | if (ret > 0) { |
1465 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1466 | if (ret < 0) | |
7bfc837d | 1467 | goto out; |
1fcbac58 | 1468 | } |
7bfc837d | 1469 | |
0b86a832 CM |
1470 | while (1) { |
1471 | l = path->nodes[0]; | |
1472 | slot = path->slots[0]; | |
1473 | if (slot >= btrfs_header_nritems(l)) { | |
1474 | ret = btrfs_next_leaf(root, path); | |
1475 | if (ret == 0) | |
1476 | continue; | |
1477 | if (ret < 0) | |
7bfc837d MX |
1478 | goto out; |
1479 | ||
1480 | break; | |
0b86a832 CM |
1481 | } |
1482 | btrfs_item_key_to_cpu(l, &key, slot); | |
1483 | ||
1484 | if (key.objectid < device->devid) | |
1485 | goto next; | |
1486 | ||
1487 | if (key.objectid > device->devid) | |
7bfc837d | 1488 | break; |
0b86a832 | 1489 | |
962a298f | 1490 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1491 | goto next; |
9779b72f | 1492 | |
7bfc837d MX |
1493 | if (key.offset > search_start) { |
1494 | hole_size = key.offset - search_start; | |
9779b72f | 1495 | |
6df9a95e JB |
1496 | /* |
1497 | * Have to check before we set max_hole_start, otherwise | |
1498 | * we could end up sending back this offset anyway. | |
1499 | */ | |
1c11b63e | 1500 | if (contains_pending_extent(device, &search_start, |
1b984508 | 1501 | hole_size)) { |
1c11b63e | 1502 | if (key.offset >= search_start) |
1b984508 | 1503 | hole_size = key.offset - search_start; |
1c11b63e | 1504 | else |
1b984508 | 1505 | hole_size = 0; |
1b984508 | 1506 | } |
6df9a95e | 1507 | |
7bfc837d MX |
1508 | if (hole_size > max_hole_size) { |
1509 | max_hole_start = search_start; | |
1510 | max_hole_size = hole_size; | |
1511 | } | |
9779b72f | 1512 | |
7bfc837d MX |
1513 | /* |
1514 | * If this free space is greater than which we need, | |
1515 | * it must be the max free space that we have found | |
1516 | * until now, so max_hole_start must point to the start | |
1517 | * of this free space and the length of this free space | |
1518 | * is stored in max_hole_size. Thus, we return | |
1519 | * max_hole_start and max_hole_size and go back to the | |
1520 | * caller. | |
1521 | */ | |
1522 | if (hole_size >= num_bytes) { | |
1523 | ret = 0; | |
1524 | goto out; | |
0b86a832 CM |
1525 | } |
1526 | } | |
0b86a832 | 1527 | |
0b86a832 | 1528 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1529 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1530 | dev_extent); | |
1531 | if (extent_end > search_start) | |
1532 | search_start = extent_end; | |
0b86a832 CM |
1533 | next: |
1534 | path->slots[0]++; | |
1535 | cond_resched(); | |
1536 | } | |
0b86a832 | 1537 | |
38c01b96 | 1538 | /* |
1539 | * At this point, search_start should be the end of | |
1540 | * allocated dev extents, and when shrinking the device, | |
1541 | * search_end may be smaller than search_start. | |
1542 | */ | |
f2ab7618 | 1543 | if (search_end > search_start) { |
38c01b96 | 1544 | hole_size = search_end - search_start; |
1545 | ||
1c11b63e | 1546 | if (contains_pending_extent(device, &search_start, hole_size)) { |
f2ab7618 ZL |
1547 | btrfs_release_path(path); |
1548 | goto again; | |
1549 | } | |
0b86a832 | 1550 | |
f2ab7618 ZL |
1551 | if (hole_size > max_hole_size) { |
1552 | max_hole_start = search_start; | |
1553 | max_hole_size = hole_size; | |
1554 | } | |
6df9a95e JB |
1555 | } |
1556 | ||
7bfc837d | 1557 | /* See above. */ |
f2ab7618 | 1558 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1559 | ret = -ENOSPC; |
1560 | else | |
1561 | ret = 0; | |
1562 | ||
1563 | out: | |
2b82032c | 1564 | btrfs_free_path(path); |
7bfc837d | 1565 | *start = max_hole_start; |
b2117a39 | 1566 | if (len) |
7bfc837d | 1567 | *len = max_hole_size; |
0b86a832 CM |
1568 | return ret; |
1569 | } | |
1570 | ||
60dfdf25 | 1571 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1572 | u64 *start, u64 *len) |
1573 | { | |
499f377f | 1574 | /* FIXME use last free of some kind */ |
60dfdf25 | 1575 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1576 | } |
1577 | ||
b2950863 | 1578 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1579 | struct btrfs_device *device, |
2196d6e8 | 1580 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1581 | { |
0b246afa JM |
1582 | struct btrfs_fs_info *fs_info = device->fs_info; |
1583 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1584 | int ret; |
1585 | struct btrfs_path *path; | |
8f18cf13 | 1586 | struct btrfs_key key; |
a061fc8d CM |
1587 | struct btrfs_key found_key; |
1588 | struct extent_buffer *leaf = NULL; | |
1589 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1590 | |
1591 | path = btrfs_alloc_path(); | |
1592 | if (!path) | |
1593 | return -ENOMEM; | |
1594 | ||
1595 | key.objectid = device->devid; | |
1596 | key.offset = start; | |
1597 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1598 | again: |
8f18cf13 | 1599 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1600 | if (ret > 0) { |
1601 | ret = btrfs_previous_item(root, path, key.objectid, | |
1602 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1603 | if (ret) |
1604 | goto out; | |
a061fc8d CM |
1605 | leaf = path->nodes[0]; |
1606 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1607 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1608 | struct btrfs_dev_extent); | |
1609 | BUG_ON(found_key.offset > start || found_key.offset + | |
1610 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1611 | key = found_key; |
1612 | btrfs_release_path(path); | |
1613 | goto again; | |
a061fc8d CM |
1614 | } else if (ret == 0) { |
1615 | leaf = path->nodes[0]; | |
1616 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1617 | struct btrfs_dev_extent); | |
79787eaa | 1618 | } else { |
0b246afa | 1619 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1620 | goto out; |
a061fc8d | 1621 | } |
8f18cf13 | 1622 | |
2196d6e8 MX |
1623 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1624 | ||
8f18cf13 | 1625 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1626 | if (ret) { |
0b246afa JM |
1627 | btrfs_handle_fs_error(fs_info, ret, |
1628 | "Failed to remove dev extent item"); | |
13212b54 | 1629 | } else { |
3204d33c | 1630 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1631 | } |
b0b802d7 | 1632 | out: |
8f18cf13 CM |
1633 | btrfs_free_path(path); |
1634 | return ret; | |
1635 | } | |
1636 | ||
48a3b636 ES |
1637 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1638 | struct btrfs_device *device, | |
48a3b636 | 1639 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1640 | { |
1641 | int ret; | |
1642 | struct btrfs_path *path; | |
0b246afa JM |
1643 | struct btrfs_fs_info *fs_info = device->fs_info; |
1644 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1645 | struct btrfs_dev_extent *extent; |
1646 | struct extent_buffer *leaf; | |
1647 | struct btrfs_key key; | |
1648 | ||
e12c9621 | 1649 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1650 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1651 | path = btrfs_alloc_path(); |
1652 | if (!path) | |
1653 | return -ENOMEM; | |
1654 | ||
0b86a832 | 1655 | key.objectid = device->devid; |
2b82032c | 1656 | key.offset = start; |
0b86a832 CM |
1657 | key.type = BTRFS_DEV_EXTENT_KEY; |
1658 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1659 | sizeof(*extent)); | |
2cdcecbc MF |
1660 | if (ret) |
1661 | goto out; | |
0b86a832 CM |
1662 | |
1663 | leaf = path->nodes[0]; | |
1664 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1665 | struct btrfs_dev_extent); | |
b5d9071c NB |
1666 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1667 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1668 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1669 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1670 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1671 | ||
0b86a832 CM |
1672 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1673 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1674 | out: |
0b86a832 CM |
1675 | btrfs_free_path(path); |
1676 | return ret; | |
1677 | } | |
1678 | ||
6df9a95e | 1679 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1680 | { |
6df9a95e JB |
1681 | struct extent_map_tree *em_tree; |
1682 | struct extent_map *em; | |
1683 | struct rb_node *n; | |
1684 | u64 ret = 0; | |
0b86a832 | 1685 | |
c8bf1b67 | 1686 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1687 | read_lock(&em_tree->lock); |
07e1ce09 | 1688 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1689 | if (n) { |
1690 | em = rb_entry(n, struct extent_map, rb_node); | |
1691 | ret = em->start + em->len; | |
0b86a832 | 1692 | } |
6df9a95e JB |
1693 | read_unlock(&em_tree->lock); |
1694 | ||
0b86a832 CM |
1695 | return ret; |
1696 | } | |
1697 | ||
53f10659 ID |
1698 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1699 | u64 *devid_ret) | |
0b86a832 CM |
1700 | { |
1701 | int ret; | |
1702 | struct btrfs_key key; | |
1703 | struct btrfs_key found_key; | |
2b82032c YZ |
1704 | struct btrfs_path *path; |
1705 | ||
2b82032c YZ |
1706 | path = btrfs_alloc_path(); |
1707 | if (!path) | |
1708 | return -ENOMEM; | |
0b86a832 CM |
1709 | |
1710 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1711 | key.type = BTRFS_DEV_ITEM_KEY; | |
1712 | key.offset = (u64)-1; | |
1713 | ||
53f10659 | 1714 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1715 | if (ret < 0) |
1716 | goto error; | |
1717 | ||
a06dee4d AJ |
1718 | if (ret == 0) { |
1719 | /* Corruption */ | |
1720 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1721 | ret = -EUCLEAN; | |
1722 | goto error; | |
1723 | } | |
0b86a832 | 1724 | |
53f10659 ID |
1725 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1726 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1727 | BTRFS_DEV_ITEM_KEY); |
1728 | if (ret) { | |
53f10659 | 1729 | *devid_ret = 1; |
0b86a832 CM |
1730 | } else { |
1731 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1732 | path->slots[0]); | |
53f10659 | 1733 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1734 | } |
1735 | ret = 0; | |
1736 | error: | |
2b82032c | 1737 | btrfs_free_path(path); |
0b86a832 CM |
1738 | return ret; |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * the device information is stored in the chunk root | |
1743 | * the btrfs_device struct should be fully filled in | |
1744 | */ | |
c74a0b02 | 1745 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1746 | struct btrfs_device *device) |
0b86a832 CM |
1747 | { |
1748 | int ret; | |
1749 | struct btrfs_path *path; | |
1750 | struct btrfs_dev_item *dev_item; | |
1751 | struct extent_buffer *leaf; | |
1752 | struct btrfs_key key; | |
1753 | unsigned long ptr; | |
0b86a832 | 1754 | |
0b86a832 CM |
1755 | path = btrfs_alloc_path(); |
1756 | if (!path) | |
1757 | return -ENOMEM; | |
1758 | ||
0b86a832 CM |
1759 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1760 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1761 | key.offset = device->devid; |
0b86a832 | 1762 | |
8e87e856 NB |
1763 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1764 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1765 | if (ret) |
1766 | goto out; | |
1767 | ||
1768 | leaf = path->nodes[0]; | |
1769 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1770 | ||
1771 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1772 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1773 | btrfs_set_device_type(leaf, dev_item, device->type); |
1774 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1775 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1776 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1777 | btrfs_set_device_total_bytes(leaf, dev_item, |
1778 | btrfs_device_get_disk_total_bytes(device)); | |
1779 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1780 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1781 | btrfs_set_device_group(leaf, dev_item, 0); |
1782 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1783 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1784 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1785 | |
410ba3a2 | 1786 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1787 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1788 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1789 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1790 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1791 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1792 | |
2b82032c | 1793 | ret = 0; |
0b86a832 CM |
1794 | out: |
1795 | btrfs_free_path(path); | |
1796 | return ret; | |
1797 | } | |
8f18cf13 | 1798 | |
5a1972bd QW |
1799 | /* |
1800 | * Function to update ctime/mtime for a given device path. | |
1801 | * Mainly used for ctime/mtime based probe like libblkid. | |
1802 | */ | |
da353f6b | 1803 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1804 | { |
1805 | struct file *filp; | |
1806 | ||
1807 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1808 | if (IS_ERR(filp)) |
5a1972bd QW |
1809 | return; |
1810 | file_update_time(filp); | |
1811 | filp_close(filp, NULL); | |
5a1972bd QW |
1812 | } |
1813 | ||
f331a952 | 1814 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1815 | { |
f331a952 | 1816 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1817 | int ret; |
1818 | struct btrfs_path *path; | |
a061fc8d | 1819 | struct btrfs_key key; |
a061fc8d CM |
1820 | struct btrfs_trans_handle *trans; |
1821 | ||
a061fc8d CM |
1822 | path = btrfs_alloc_path(); |
1823 | if (!path) | |
1824 | return -ENOMEM; | |
1825 | ||
a22285a6 | 1826 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1827 | if (IS_ERR(trans)) { |
1828 | btrfs_free_path(path); | |
1829 | return PTR_ERR(trans); | |
1830 | } | |
a061fc8d CM |
1831 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1832 | key.type = BTRFS_DEV_ITEM_KEY; | |
1833 | key.offset = device->devid; | |
1834 | ||
1835 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1836 | if (ret) { |
1837 | if (ret > 0) | |
1838 | ret = -ENOENT; | |
1839 | btrfs_abort_transaction(trans, ret); | |
1840 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1841 | goto out; |
1842 | } | |
1843 | ||
1844 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1845 | if (ret) { |
1846 | btrfs_abort_transaction(trans, ret); | |
1847 | btrfs_end_transaction(trans); | |
1848 | } | |
1849 | ||
a061fc8d CM |
1850 | out: |
1851 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1852 | if (!ret) |
1853 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1854 | return ret; |
1855 | } | |
1856 | ||
3cc31a0d DS |
1857 | /* |
1858 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1859 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1860 | * replace. | |
1861 | */ | |
1862 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1863 | u64 num_devices) | |
a061fc8d | 1864 | { |
a061fc8d | 1865 | u64 all_avail; |
de98ced9 | 1866 | unsigned seq; |
418775a2 | 1867 | int i; |
a061fc8d | 1868 | |
de98ced9 | 1869 | do { |
bd45ffbc | 1870 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1871 | |
bd45ffbc AJ |
1872 | all_avail = fs_info->avail_data_alloc_bits | |
1873 | fs_info->avail_system_alloc_bits | | |
1874 | fs_info->avail_metadata_alloc_bits; | |
1875 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1876 | |
418775a2 | 1877 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1878 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1879 | continue; |
a061fc8d | 1880 | |
418775a2 | 1881 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1882 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1883 | |
418775a2 DS |
1884 | if (ret) |
1885 | return ret; | |
1886 | } | |
53b381b3 DW |
1887 | } |
1888 | ||
bd45ffbc | 1889 | return 0; |
f1fa7f26 AJ |
1890 | } |
1891 | ||
c9162bdf OS |
1892 | static struct btrfs_device * btrfs_find_next_active_device( |
1893 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1894 | { |
2b82032c | 1895 | struct btrfs_device *next_device; |
88acff64 AJ |
1896 | |
1897 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1898 | if (next_device != device && | |
e6e674bd AJ |
1899 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1900 | && next_device->bdev) | |
88acff64 AJ |
1901 | return next_device; |
1902 | } | |
1903 | ||
1904 | return NULL; | |
1905 | } | |
1906 | ||
1907 | /* | |
1908 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1909 | * and replace it with the provided or the next active device, in the context | |
1910 | * where this function called, there should be always be another device (or | |
1911 | * this_dev) which is active. | |
1912 | */ | |
b105e927 | 1913 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
d6507cf1 | 1914 | struct btrfs_device *this_dev) |
88acff64 | 1915 | { |
d6507cf1 | 1916 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1917 | struct btrfs_device *next_device; |
1918 | ||
1919 | if (this_dev) | |
1920 | next_device = this_dev; | |
1921 | else | |
1922 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1923 | device); | |
1924 | ASSERT(next_device); | |
1925 | ||
1926 | if (fs_info->sb->s_bdev && | |
1927 | (fs_info->sb->s_bdev == device->bdev)) | |
1928 | fs_info->sb->s_bdev = next_device->bdev; | |
1929 | ||
1930 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1931 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1932 | } | |
1933 | ||
1da73967 AJ |
1934 | /* |
1935 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1936 | * currently replaced. | |
1937 | */ | |
1938 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1939 | { | |
1940 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1941 | ||
cb5583dd | 1942 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1943 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
1944 | ASSERT(num_devices > 1); | |
1945 | num_devices--; | |
1946 | } | |
cb5583dd | 1947 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1948 | |
1949 | return num_devices; | |
1950 | } | |
1951 | ||
da353f6b DS |
1952 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
1953 | u64 devid) | |
f1fa7f26 AJ |
1954 | { |
1955 | struct btrfs_device *device; | |
1f78160c | 1956 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 1957 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 1958 | u64 num_devices; |
a061fc8d CM |
1959 | int ret = 0; |
1960 | ||
a061fc8d CM |
1961 | mutex_lock(&uuid_mutex); |
1962 | ||
1da73967 | 1963 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 1964 | |
0b246afa | 1965 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1966 | if (ret) |
a061fc8d | 1967 | goto out; |
a061fc8d | 1968 | |
a27a94c2 NB |
1969 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
1970 | ||
1971 | if (IS_ERR(device)) { | |
1972 | if (PTR_ERR(device) == -ENOENT && | |
1973 | strcmp(device_path, "missing") == 0) | |
1974 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
1975 | else | |
1976 | ret = PTR_ERR(device); | |
53b381b3 | 1977 | goto out; |
a27a94c2 | 1978 | } |
dfe25020 | 1979 | |
eede2bf3 OS |
1980 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
1981 | btrfs_warn_in_rcu(fs_info, | |
1982 | "cannot remove device %s (devid %llu) due to active swapfile", | |
1983 | rcu_str_deref(device->name), device->devid); | |
1984 | ret = -ETXTBSY; | |
1985 | goto out; | |
1986 | } | |
1987 | ||
401e29c1 | 1988 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 1989 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1990 | goto out; |
63a212ab SB |
1991 | } |
1992 | ||
ebbede42 AJ |
1993 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
1994 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 1995 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1996 | goto out; |
2b82032c YZ |
1997 | } |
1998 | ||
ebbede42 | 1999 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2000 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2001 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2002 | device->fs_devices->rw_devices--; |
34441361 | 2003 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2004 | } |
a061fc8d | 2005 | |
d7901554 | 2006 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2007 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2008 | mutex_lock(&uuid_mutex); |
a061fc8d | 2009 | if (ret) |
9b3517e9 | 2010 | goto error_undo; |
a061fc8d | 2011 | |
63a212ab SB |
2012 | /* |
2013 | * TODO: the superblock still includes this device in its num_devices | |
2014 | * counter although write_all_supers() is not locked out. This | |
2015 | * could give a filesystem state which requires a degraded mount. | |
2016 | */ | |
f331a952 | 2017 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2018 | if (ret) |
9b3517e9 | 2019 | goto error_undo; |
a061fc8d | 2020 | |
e12c9621 | 2021 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2022 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2023 | |
2024 | /* | |
2025 | * the device list mutex makes sure that we don't change | |
2026 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2027 | * the device supers. Whoever is writing all supers, should |
2028 | * lock the device list mutex before getting the number of | |
2029 | * devices in the super block (super_copy). Conversely, | |
2030 | * whoever updates the number of devices in the super block | |
2031 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2032 | */ |
1f78160c | 2033 | |
41a52a0f AJ |
2034 | /* |
2035 | * In normal cases the cur_devices == fs_devices. But in case | |
2036 | * of deleting a seed device, the cur_devices should point to | |
2037 | * its own fs_devices listed under the fs_devices->seed. | |
2038 | */ | |
1f78160c | 2039 | cur_devices = device->fs_devices; |
b5185197 | 2040 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2041 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2042 | |
41a52a0f AJ |
2043 | cur_devices->num_devices--; |
2044 | cur_devices->total_devices--; | |
b4993e64 AJ |
2045 | /* Update total_devices of the parent fs_devices if it's seed */ |
2046 | if (cur_devices != fs_devices) | |
2047 | fs_devices->total_devices--; | |
2b82032c | 2048 | |
e6e674bd | 2049 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2050 | cur_devices->missing_devices--; |
cd02dca5 | 2051 | |
d6507cf1 | 2052 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2053 | |
0bfaa9c5 | 2054 | if (device->bdev) { |
41a52a0f | 2055 | cur_devices->open_devices--; |
0bfaa9c5 | 2056 | /* remove sysfs entry */ |
b5185197 | 2057 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2058 | } |
99994cde | 2059 | |
0b246afa JM |
2060 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2061 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2062 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2063 | |
cea67ab9 JM |
2064 | /* |
2065 | * at this point, the device is zero sized and detached from | |
2066 | * the devices list. All that's left is to zero out the old | |
2067 | * supers and free the device. | |
2068 | */ | |
ebbede42 | 2069 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2070 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2071 | ||
2072 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2073 | synchronize_rcu(); |
2074 | btrfs_free_device(device); | |
cea67ab9 | 2075 | |
1f78160c | 2076 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2077 | while (fs_devices) { |
8321cf25 RS |
2078 | if (fs_devices->seed == cur_devices) { |
2079 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2080 | break; |
8321cf25 | 2081 | } |
e4404d6e | 2082 | fs_devices = fs_devices->seed; |
2b82032c | 2083 | } |
1f78160c | 2084 | cur_devices->seed = NULL; |
0226e0eb | 2085 | close_fs_devices(cur_devices); |
1f78160c | 2086 | free_fs_devices(cur_devices); |
2b82032c YZ |
2087 | } |
2088 | ||
a061fc8d CM |
2089 | out: |
2090 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2091 | return ret; |
24fc572f | 2092 | |
9b3517e9 | 2093 | error_undo: |
ebbede42 | 2094 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2095 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2096 | list_add(&device->dev_alloc_list, |
b5185197 | 2097 | &fs_devices->alloc_list); |
c3929c36 | 2098 | device->fs_devices->rw_devices++; |
34441361 | 2099 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2100 | } |
24fc572f | 2101 | goto out; |
a061fc8d CM |
2102 | } |
2103 | ||
68a9db5f | 2104 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2105 | { |
d51908ce AJ |
2106 | struct btrfs_fs_devices *fs_devices; |
2107 | ||
68a9db5f | 2108 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2109 | |
25e8e911 AJ |
2110 | /* |
2111 | * in case of fs with no seed, srcdev->fs_devices will point | |
2112 | * to fs_devices of fs_info. However when the dev being replaced is | |
2113 | * a seed dev it will point to the seed's local fs_devices. In short | |
2114 | * srcdev will have its correct fs_devices in both the cases. | |
2115 | */ | |
2116 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2117 | |
e93c89c1 | 2118 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2119 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2120 | fs_devices->num_devices--; |
e6e674bd | 2121 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2122 | fs_devices->missing_devices--; |
e93c89c1 | 2123 | |
ebbede42 | 2124 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2125 | fs_devices->rw_devices--; |
1357272f | 2126 | |
82372bc8 | 2127 | if (srcdev->bdev) |
d51908ce | 2128 | fs_devices->open_devices--; |
084b6e7c QW |
2129 | } |
2130 | ||
65237ee3 | 2131 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c | 2132 | { |
65237ee3 | 2133 | struct btrfs_fs_info *fs_info = srcdev->fs_info; |
084b6e7c | 2134 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
e93c89c1 | 2135 | |
ebbede42 | 2136 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2137 | /* zero out the old super if it is writable */ |
2138 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2139 | } | |
14238819 AJ |
2140 | |
2141 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2142 | synchronize_rcu(); |
2143 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2144 | |
94d5f0c2 AJ |
2145 | /* if this is no devs we rather delete the fs_devices */ |
2146 | if (!fs_devices->num_devices) { | |
2147 | struct btrfs_fs_devices *tmp_fs_devices; | |
2148 | ||
6dd38f81 AJ |
2149 | /* |
2150 | * On a mounted FS, num_devices can't be zero unless it's a | |
2151 | * seed. In case of a seed device being replaced, the replace | |
2152 | * target added to the sprout FS, so there will be no more | |
2153 | * device left under the seed FS. | |
2154 | */ | |
2155 | ASSERT(fs_devices->seeding); | |
2156 | ||
94d5f0c2 AJ |
2157 | tmp_fs_devices = fs_info->fs_devices; |
2158 | while (tmp_fs_devices) { | |
2159 | if (tmp_fs_devices->seed == fs_devices) { | |
2160 | tmp_fs_devices->seed = fs_devices->seed; | |
2161 | break; | |
2162 | } | |
2163 | tmp_fs_devices = tmp_fs_devices->seed; | |
2164 | } | |
2165 | fs_devices->seed = NULL; | |
0226e0eb | 2166 | close_fs_devices(fs_devices); |
8bef8401 | 2167 | free_fs_devices(fs_devices); |
94d5f0c2 | 2168 | } |
e93c89c1 SB |
2169 | } |
2170 | ||
4f5ad7bd | 2171 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2172 | { |
4f5ad7bd | 2173 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2174 | |
d9a071f0 | 2175 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2176 | |
d9a071f0 | 2177 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2178 | |
779bf3fe | 2179 | if (tgtdev->bdev) |
d9a071f0 | 2180 | fs_devices->open_devices--; |
779bf3fe | 2181 | |
d9a071f0 | 2182 | fs_devices->num_devices--; |
e93c89c1 | 2183 | |
d6507cf1 | 2184 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2185 | |
e93c89c1 | 2186 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2187 | |
d9a071f0 | 2188 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2189 | |
2190 | /* | |
2191 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2192 | * may lead to a call to btrfs_show_devname() which will try | |
2193 | * to hold device_list_mutex. And here this device | |
2194 | * is already out of device list, so we don't have to hold | |
2195 | * the device_list_mutex lock. | |
2196 | */ | |
2197 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2198 | |
2199 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2200 | synchronize_rcu(); |
2201 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2202 | } |
2203 | ||
b444ad46 NB |
2204 | static struct btrfs_device *btrfs_find_device_by_path( |
2205 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2206 | { |
2207 | int ret = 0; | |
2208 | struct btrfs_super_block *disk_super; | |
2209 | u64 devid; | |
2210 | u8 *dev_uuid; | |
2211 | struct block_device *bdev; | |
2212 | struct buffer_head *bh; | |
b444ad46 | 2213 | struct btrfs_device *device; |
7ba15b7d | 2214 | |
7ba15b7d | 2215 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2216 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2217 | if (ret) |
b444ad46 | 2218 | return ERR_PTR(ret); |
7ba15b7d SB |
2219 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2220 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2221 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2222 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2223 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2224 | disk_super->metadata_uuid, true); |
7239ff4b | 2225 | else |
e4319cd9 | 2226 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2227 | disk_super->fsid, true); |
7239ff4b | 2228 | |
7ba15b7d | 2229 | brelse(bh); |
b444ad46 NB |
2230 | if (!device) |
2231 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2232 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2233 | return device; |
7ba15b7d SB |
2234 | } |
2235 | ||
5c5c0df0 DS |
2236 | /* |
2237 | * Lookup a device given by device id, or the path if the id is 0. | |
2238 | */ | |
a27a94c2 | 2239 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2240 | struct btrfs_fs_info *fs_info, u64 devid, |
2241 | const char *device_path) | |
24e0474b | 2242 | { |
a27a94c2 | 2243 | struct btrfs_device *device; |
24e0474b | 2244 | |
5c5c0df0 | 2245 | if (devid) { |
e4319cd9 | 2246 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2247 | NULL, true); |
a27a94c2 NB |
2248 | if (!device) |
2249 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2250 | return device; |
2251 | } | |
2252 | ||
2253 | if (!device_path || !device_path[0]) | |
2254 | return ERR_PTR(-EINVAL); | |
2255 | ||
2256 | if (strcmp(device_path, "missing") == 0) { | |
2257 | /* Find first missing device */ | |
2258 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2259 | dev_list) { | |
2260 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2261 | &device->dev_state) && !device->bdev) | |
2262 | return device; | |
d95a830c | 2263 | } |
6e927ceb | 2264 | return ERR_PTR(-ENOENT); |
24e0474b | 2265 | } |
6e927ceb AJ |
2266 | |
2267 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2268 | } |
2269 | ||
2b82032c YZ |
2270 | /* |
2271 | * does all the dirty work required for changing file system's UUID. | |
2272 | */ | |
2ff7e61e | 2273 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2274 | { |
0b246afa | 2275 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2276 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2277 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2278 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2279 | struct btrfs_device *device; |
2280 | u64 super_flags; | |
2281 | ||
a32bf9a3 | 2282 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2283 | if (!fs_devices->seeding) |
2b82032c YZ |
2284 | return -EINVAL; |
2285 | ||
7239ff4b | 2286 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2287 | if (IS_ERR(seed_devices)) |
2288 | return PTR_ERR(seed_devices); | |
2b82032c | 2289 | |
e4404d6e YZ |
2290 | old_devices = clone_fs_devices(fs_devices); |
2291 | if (IS_ERR(old_devices)) { | |
2292 | kfree(seed_devices); | |
2293 | return PTR_ERR(old_devices); | |
2b82032c | 2294 | } |
e4404d6e | 2295 | |
c4babc5e | 2296 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2297 | |
e4404d6e YZ |
2298 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2299 | seed_devices->opened = 1; | |
2300 | INIT_LIST_HEAD(&seed_devices->devices); | |
2301 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2302 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2303 | |
321a4bf7 | 2304 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2305 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2306 | synchronize_rcu); | |
2196d6e8 MX |
2307 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2308 | device->fs_devices = seed_devices; | |
c9513edb | 2309 | |
34441361 | 2310 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2311 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2312 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2313 | |
0395d84f | 2314 | fs_devices->seeding = false; |
2b82032c YZ |
2315 | fs_devices->num_devices = 0; |
2316 | fs_devices->open_devices = 0; | |
69611ac8 | 2317 | fs_devices->missing_devices = 0; |
7f0432d0 | 2318 | fs_devices->rotating = false; |
e4404d6e | 2319 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2320 | |
2321 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2322 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2323 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2324 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2325 | |
2b82032c YZ |
2326 | super_flags = btrfs_super_flags(disk_super) & |
2327 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2328 | btrfs_set_super_flags(disk_super, super_flags); | |
2329 | ||
2330 | return 0; | |
2331 | } | |
2332 | ||
2333 | /* | |
01327610 | 2334 | * Store the expected generation for seed devices in device items. |
2b82032c | 2335 | */ |
5c466629 | 2336 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2337 | { |
5c466629 | 2338 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2339 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2340 | struct btrfs_path *path; |
2341 | struct extent_buffer *leaf; | |
2342 | struct btrfs_dev_item *dev_item; | |
2343 | struct btrfs_device *device; | |
2344 | struct btrfs_key key; | |
44880fdc | 2345 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2346 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2347 | u64 devid; | |
2348 | int ret; | |
2349 | ||
2350 | path = btrfs_alloc_path(); | |
2351 | if (!path) | |
2352 | return -ENOMEM; | |
2353 | ||
2b82032c YZ |
2354 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2355 | key.offset = 0; | |
2356 | key.type = BTRFS_DEV_ITEM_KEY; | |
2357 | ||
2358 | while (1) { | |
2359 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2360 | if (ret < 0) | |
2361 | goto error; | |
2362 | ||
2363 | leaf = path->nodes[0]; | |
2364 | next_slot: | |
2365 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2366 | ret = btrfs_next_leaf(root, path); | |
2367 | if (ret > 0) | |
2368 | break; | |
2369 | if (ret < 0) | |
2370 | goto error; | |
2371 | leaf = path->nodes[0]; | |
2372 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2373 | btrfs_release_path(path); |
2b82032c YZ |
2374 | continue; |
2375 | } | |
2376 | ||
2377 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2378 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2379 | key.type != BTRFS_DEV_ITEM_KEY) | |
2380 | break; | |
2381 | ||
2382 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2383 | struct btrfs_dev_item); | |
2384 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2385 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2386 | BTRFS_UUID_SIZE); |
1473b24e | 2387 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2388 | BTRFS_FSID_SIZE); |
e4319cd9 | 2389 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2390 | fs_uuid, true); |
79787eaa | 2391 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2392 | |
2393 | if (device->fs_devices->seeding) { | |
2394 | btrfs_set_device_generation(leaf, dev_item, | |
2395 | device->generation); | |
2396 | btrfs_mark_buffer_dirty(leaf); | |
2397 | } | |
2398 | ||
2399 | path->slots[0]++; | |
2400 | goto next_slot; | |
2401 | } | |
2402 | ret = 0; | |
2403 | error: | |
2404 | btrfs_free_path(path); | |
2405 | return ret; | |
2406 | } | |
2407 | ||
da353f6b | 2408 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2409 | { |
5112febb | 2410 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2411 | struct request_queue *q; |
788f20eb CM |
2412 | struct btrfs_trans_handle *trans; |
2413 | struct btrfs_device *device; | |
2414 | struct block_device *bdev; | |
0b246afa | 2415 | struct super_block *sb = fs_info->sb; |
606686ee | 2416 | struct rcu_string *name; |
5da54bc1 | 2417 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2418 | u64 orig_super_total_bytes; |
2419 | u64 orig_super_num_devices; | |
2b82032c | 2420 | int seeding_dev = 0; |
788f20eb | 2421 | int ret = 0; |
7132a262 | 2422 | bool unlocked = false; |
788f20eb | 2423 | |
5da54bc1 | 2424 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2425 | return -EROFS; |
788f20eb | 2426 | |
a5d16333 | 2427 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2428 | fs_info->bdev_holder); |
7f59203a JB |
2429 | if (IS_ERR(bdev)) |
2430 | return PTR_ERR(bdev); | |
a2135011 | 2431 | |
5da54bc1 | 2432 | if (fs_devices->seeding) { |
2b82032c YZ |
2433 | seeding_dev = 1; |
2434 | down_write(&sb->s_umount); | |
2435 | mutex_lock(&uuid_mutex); | |
2436 | } | |
2437 | ||
8c8bee1d | 2438 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2439 | |
5da54bc1 | 2440 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2441 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2442 | if (device->bdev == bdev) { |
2443 | ret = -EEXIST; | |
d25628bd | 2444 | mutex_unlock( |
5da54bc1 | 2445 | &fs_devices->device_list_mutex); |
2b82032c | 2446 | goto error; |
788f20eb CM |
2447 | } |
2448 | } | |
5da54bc1 | 2449 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2450 | |
0b246afa | 2451 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2452 | if (IS_ERR(device)) { |
788f20eb | 2453 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2454 | ret = PTR_ERR(device); |
2b82032c | 2455 | goto error; |
788f20eb CM |
2456 | } |
2457 | ||
78f2c9e6 | 2458 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2459 | if (!name) { |
2b82032c | 2460 | ret = -ENOMEM; |
5c4cf6c9 | 2461 | goto error_free_device; |
788f20eb | 2462 | } |
606686ee | 2463 | rcu_assign_pointer(device->name, name); |
2b82032c | 2464 | |
a22285a6 | 2465 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2466 | if (IS_ERR(trans)) { |
98d5dc13 | 2467 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2468 | goto error_free_device; |
98d5dc13 TI |
2469 | } |
2470 | ||
d5e2003c | 2471 | q = bdev_get_queue(bdev); |
ebbede42 | 2472 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2473 | device->generation = trans->transid; |
0b246afa JM |
2474 | device->io_width = fs_info->sectorsize; |
2475 | device->io_align = fs_info->sectorsize; | |
2476 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2477 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2478 | fs_info->sectorsize); | |
2cc3c559 | 2479 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2480 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2481 | device->fs_info = fs_info; |
788f20eb | 2482 | device->bdev = bdev; |
e12c9621 | 2483 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2484 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2485 | device->mode = FMODE_EXCL; |
27087f37 | 2486 | device->dev_stats_valid = 1; |
9f6d2510 | 2487 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2488 | |
2b82032c | 2489 | if (seeding_dev) { |
1751e8a6 | 2490 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2491 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2492 | if (ret) { |
2493 | btrfs_abort_transaction(trans, ret); | |
2494 | goto error_trans; | |
2495 | } | |
2b82032c | 2496 | } |
788f20eb | 2497 | |
5da54bc1 | 2498 | device->fs_devices = fs_devices; |
e5e9a520 | 2499 | |
5da54bc1 | 2500 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2501 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2502 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2503 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2504 | fs_devices->num_devices++; | |
2505 | fs_devices->open_devices++; | |
2506 | fs_devices->rw_devices++; | |
2507 | fs_devices->total_devices++; | |
2508 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2509 | |
a5ed45f8 | 2510 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2511 | |
e884f4f0 | 2512 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2513 | fs_devices->rotating = true; |
c289811c | 2514 | |
39379faa | 2515 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2516 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2517 | round_down(orig_super_total_bytes + device->total_bytes, |
2518 | fs_info->sectorsize)); | |
788f20eb | 2519 | |
39379faa NA |
2520 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2521 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2522 | orig_super_num_devices + 1); | |
0d39376a AJ |
2523 | |
2524 | /* add sysfs device entry */ | |
5da54bc1 | 2525 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2526 | |
2196d6e8 MX |
2527 | /* |
2528 | * we've got more storage, clear any full flags on the space | |
2529 | * infos | |
2530 | */ | |
0b246afa | 2531 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2532 | |
34441361 | 2533 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2534 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2535 | |
2b82032c | 2536 | if (seeding_dev) { |
34441361 | 2537 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2538 | ret = init_first_rw_device(trans); |
34441361 | 2539 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2540 | if (ret) { |
66642832 | 2541 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2542 | goto error_sysfs; |
005d6427 | 2543 | } |
2196d6e8 MX |
2544 | } |
2545 | ||
8e87e856 | 2546 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2547 | if (ret) { |
66642832 | 2548 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2549 | goto error_sysfs; |
2196d6e8 MX |
2550 | } |
2551 | ||
2552 | if (seeding_dev) { | |
5c466629 | 2553 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2554 | if (ret) { |
66642832 | 2555 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2556 | goto error_sysfs; |
005d6427 | 2557 | } |
b2373f25 | 2558 | |
f93c3997 DS |
2559 | btrfs_sysfs_update_sprout_fsid(fs_devices, |
2560 | fs_info->fs_devices->fsid); | |
2b82032c YZ |
2561 | } |
2562 | ||
3a45bb20 | 2563 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2564 | |
2b82032c YZ |
2565 | if (seeding_dev) { |
2566 | mutex_unlock(&uuid_mutex); | |
2567 | up_write(&sb->s_umount); | |
7132a262 | 2568 | unlocked = true; |
788f20eb | 2569 | |
79787eaa JM |
2570 | if (ret) /* transaction commit */ |
2571 | return ret; | |
2572 | ||
2ff7e61e | 2573 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2574 | if (ret < 0) |
0b246afa | 2575 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2576 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2577 | trans = btrfs_attach_transaction(root); |
2578 | if (IS_ERR(trans)) { | |
2579 | if (PTR_ERR(trans) == -ENOENT) | |
2580 | return 0; | |
7132a262 AJ |
2581 | ret = PTR_ERR(trans); |
2582 | trans = NULL; | |
2583 | goto error_sysfs; | |
671415b7 | 2584 | } |
3a45bb20 | 2585 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2586 | } |
c9e9f97b | 2587 | |
5a1972bd QW |
2588 | /* Update ctime/mtime for libblkid */ |
2589 | update_dev_time(device_path); | |
2b82032c | 2590 | return ret; |
79787eaa | 2591 | |
d31c32f6 | 2592 | error_sysfs: |
5da54bc1 | 2593 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2594 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2595 | mutex_lock(&fs_info->chunk_mutex); | |
2596 | list_del_rcu(&device->dev_list); | |
2597 | list_del(&device->dev_alloc_list); | |
2598 | fs_info->fs_devices->num_devices--; | |
2599 | fs_info->fs_devices->open_devices--; | |
2600 | fs_info->fs_devices->rw_devices--; | |
2601 | fs_info->fs_devices->total_devices--; | |
2602 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2603 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2604 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2605 | orig_super_total_bytes); | |
2606 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2607 | orig_super_num_devices); | |
2608 | mutex_unlock(&fs_info->chunk_mutex); | |
2609 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2610 | error_trans: |
0af2c4bf | 2611 | if (seeding_dev) |
1751e8a6 | 2612 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2613 | if (trans) |
2614 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2615 | error_free_device: |
a425f9d4 | 2616 | btrfs_free_device(device); |
2b82032c | 2617 | error: |
e525fd89 | 2618 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2619 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2620 | mutex_unlock(&uuid_mutex); |
2621 | up_write(&sb->s_umount); | |
2622 | } | |
c9e9f97b | 2623 | return ret; |
788f20eb CM |
2624 | } |
2625 | ||
d397712b CM |
2626 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2627 | struct btrfs_device *device) | |
0b86a832 CM |
2628 | { |
2629 | int ret; | |
2630 | struct btrfs_path *path; | |
0b246afa | 2631 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2632 | struct btrfs_dev_item *dev_item; |
2633 | struct extent_buffer *leaf; | |
2634 | struct btrfs_key key; | |
2635 | ||
0b86a832 CM |
2636 | path = btrfs_alloc_path(); |
2637 | if (!path) | |
2638 | return -ENOMEM; | |
2639 | ||
2640 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2641 | key.type = BTRFS_DEV_ITEM_KEY; | |
2642 | key.offset = device->devid; | |
2643 | ||
2644 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2645 | if (ret < 0) | |
2646 | goto out; | |
2647 | ||
2648 | if (ret > 0) { | |
2649 | ret = -ENOENT; | |
2650 | goto out; | |
2651 | } | |
2652 | ||
2653 | leaf = path->nodes[0]; | |
2654 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2655 | ||
2656 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2657 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2658 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2659 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2660 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2661 | btrfs_set_device_total_bytes(leaf, dev_item, |
2662 | btrfs_device_get_disk_total_bytes(device)); | |
2663 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2664 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2665 | btrfs_mark_buffer_dirty(leaf); |
2666 | ||
2667 | out: | |
2668 | btrfs_free_path(path); | |
2669 | return ret; | |
2670 | } | |
2671 | ||
2196d6e8 | 2672 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2673 | struct btrfs_device *device, u64 new_size) |
2674 | { | |
0b246afa JM |
2675 | struct btrfs_fs_info *fs_info = device->fs_info; |
2676 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2677 | u64 old_total; |
2678 | u64 diff; | |
8f18cf13 | 2679 | |
ebbede42 | 2680 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2681 | return -EACCES; |
2196d6e8 | 2682 | |
7dfb8be1 NB |
2683 | new_size = round_down(new_size, fs_info->sectorsize); |
2684 | ||
34441361 | 2685 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2686 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2687 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2688 | |
63a212ab | 2689 | if (new_size <= device->total_bytes || |
401e29c1 | 2690 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2691 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2692 | return -EINVAL; |
2196d6e8 | 2693 | } |
2b82032c | 2694 | |
7dfb8be1 NB |
2695 | btrfs_set_super_total_bytes(super_copy, |
2696 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2697 | device->fs_devices->total_rw_bytes += diff; |
2698 | ||
7cc8e58d MX |
2699 | btrfs_device_set_total_bytes(device, new_size); |
2700 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2701 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2702 | if (list_empty(&device->post_commit_list)) |
2703 | list_add_tail(&device->post_commit_list, | |
2704 | &trans->transaction->dev_update_list); | |
34441361 | 2705 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2706 | |
8f18cf13 CM |
2707 | return btrfs_update_device(trans, device); |
2708 | } | |
2709 | ||
f4208794 | 2710 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2711 | { |
f4208794 | 2712 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2713 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2714 | int ret; |
2715 | struct btrfs_path *path; | |
2716 | struct btrfs_key key; | |
2717 | ||
8f18cf13 CM |
2718 | path = btrfs_alloc_path(); |
2719 | if (!path) | |
2720 | return -ENOMEM; | |
2721 | ||
408fbf19 | 2722 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2723 | key.offset = chunk_offset; |
2724 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2725 | ||
2726 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2727 | if (ret < 0) |
2728 | goto out; | |
2729 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2730 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2731 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2732 | ret = -ENOENT; |
2733 | goto out; | |
2734 | } | |
8f18cf13 CM |
2735 | |
2736 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2737 | if (ret < 0) |
0b246afa JM |
2738 | btrfs_handle_fs_error(fs_info, ret, |
2739 | "Failed to delete chunk item."); | |
79787eaa | 2740 | out: |
8f18cf13 | 2741 | btrfs_free_path(path); |
65a246c5 | 2742 | return ret; |
8f18cf13 CM |
2743 | } |
2744 | ||
408fbf19 | 2745 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2746 | { |
0b246afa | 2747 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2748 | struct btrfs_disk_key *disk_key; |
2749 | struct btrfs_chunk *chunk; | |
2750 | u8 *ptr; | |
2751 | int ret = 0; | |
2752 | u32 num_stripes; | |
2753 | u32 array_size; | |
2754 | u32 len = 0; | |
2755 | u32 cur; | |
2756 | struct btrfs_key key; | |
2757 | ||
34441361 | 2758 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2759 | array_size = btrfs_super_sys_array_size(super_copy); |
2760 | ||
2761 | ptr = super_copy->sys_chunk_array; | |
2762 | cur = 0; | |
2763 | ||
2764 | while (cur < array_size) { | |
2765 | disk_key = (struct btrfs_disk_key *)ptr; | |
2766 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2767 | ||
2768 | len = sizeof(*disk_key); | |
2769 | ||
2770 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2771 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2772 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2773 | len += btrfs_chunk_item_size(num_stripes); | |
2774 | } else { | |
2775 | ret = -EIO; | |
2776 | break; | |
2777 | } | |
408fbf19 | 2778 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2779 | key.offset == chunk_offset) { |
2780 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2781 | array_size -= len; | |
2782 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2783 | } else { | |
2784 | ptr += len; | |
2785 | cur += len; | |
2786 | } | |
2787 | } | |
34441361 | 2788 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2789 | return ret; |
2790 | } | |
2791 | ||
60ca842e OS |
2792 | /* |
2793 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2794 | * @logical: Logical block offset in bytes. | |
2795 | * @length: Length of extent in bytes. | |
2796 | * | |
2797 | * Return: Chunk mapping or ERR_PTR. | |
2798 | */ | |
2799 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2800 | u64 logical, u64 length) | |
592d92ee LB |
2801 | { |
2802 | struct extent_map_tree *em_tree; | |
2803 | struct extent_map *em; | |
2804 | ||
c8bf1b67 | 2805 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2806 | read_lock(&em_tree->lock); |
2807 | em = lookup_extent_mapping(em_tree, logical, length); | |
2808 | read_unlock(&em_tree->lock); | |
2809 | ||
2810 | if (!em) { | |
2811 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2812 | logical, length); | |
2813 | return ERR_PTR(-EINVAL); | |
2814 | } | |
2815 | ||
2816 | if (em->start > logical || em->start + em->len < logical) { | |
2817 | btrfs_crit(fs_info, | |
2818 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2819 | logical, length, em->start, em->start + em->len); | |
2820 | free_extent_map(em); | |
2821 | return ERR_PTR(-EINVAL); | |
2822 | } | |
2823 | ||
2824 | /* callers are responsible for dropping em's ref. */ | |
2825 | return em; | |
2826 | } | |
2827 | ||
97aff912 | 2828 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2829 | { |
97aff912 | 2830 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2831 | struct extent_map *em; |
2832 | struct map_lookup *map; | |
2196d6e8 | 2833 | u64 dev_extent_len = 0; |
47ab2a6c | 2834 | int i, ret = 0; |
0b246afa | 2835 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2836 | |
60ca842e | 2837 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2838 | if (IS_ERR(em)) { |
47ab2a6c JB |
2839 | /* |
2840 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2841 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2842 | * do anything we still error out. |
2843 | */ | |
2844 | ASSERT(0); | |
592d92ee | 2845 | return PTR_ERR(em); |
47ab2a6c | 2846 | } |
95617d69 | 2847 | map = em->map_lookup; |
34441361 | 2848 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2849 | check_system_chunk(trans, map->type); |
34441361 | 2850 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2851 | |
57ba4cb8 FM |
2852 | /* |
2853 | * Take the device list mutex to prevent races with the final phase of | |
2854 | * a device replace operation that replaces the device object associated | |
2855 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2856 | */ | |
2857 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2858 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2859 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2860 | ret = btrfs_free_dev_extent(trans, device, |
2861 | map->stripes[i].physical, | |
2862 | &dev_extent_len); | |
47ab2a6c | 2863 | if (ret) { |
57ba4cb8 | 2864 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2865 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2866 | goto out; |
2867 | } | |
a061fc8d | 2868 | |
2196d6e8 | 2869 | if (device->bytes_used > 0) { |
34441361 | 2870 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2871 | btrfs_device_set_bytes_used(device, |
2872 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2873 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2874 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2875 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2876 | } |
a061fc8d | 2877 | |
64bc6c2a NB |
2878 | ret = btrfs_update_device(trans, device); |
2879 | if (ret) { | |
2880 | mutex_unlock(&fs_devices->device_list_mutex); | |
2881 | btrfs_abort_transaction(trans, ret); | |
2882 | goto out; | |
dfe25020 | 2883 | } |
8f18cf13 | 2884 | } |
57ba4cb8 FM |
2885 | mutex_unlock(&fs_devices->device_list_mutex); |
2886 | ||
f4208794 | 2887 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2888 | if (ret) { |
66642832 | 2889 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2890 | goto out; |
2891 | } | |
8f18cf13 | 2892 | |
6bccf3ab | 2893 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2894 | |
8f18cf13 | 2895 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2896 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2897 | if (ret) { |
66642832 | 2898 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2899 | goto out; |
2900 | } | |
8f18cf13 CM |
2901 | } |
2902 | ||
5a98ec01 | 2903 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 2904 | if (ret) { |
66642832 | 2905 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2906 | goto out; |
2907 | } | |
2b82032c | 2908 | |
47ab2a6c | 2909 | out: |
2b82032c YZ |
2910 | /* once for us */ |
2911 | free_extent_map(em); | |
47ab2a6c JB |
2912 | return ret; |
2913 | } | |
2b82032c | 2914 | |
5b4aacef | 2915 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2916 | { |
5b4aacef | 2917 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 2918 | struct btrfs_trans_handle *trans; |
b0643e59 | 2919 | struct btrfs_block_group *block_group; |
47ab2a6c | 2920 | int ret; |
2b82032c | 2921 | |
67c5e7d4 FM |
2922 | /* |
2923 | * Prevent races with automatic removal of unused block groups. | |
2924 | * After we relocate and before we remove the chunk with offset | |
2925 | * chunk_offset, automatic removal of the block group can kick in, | |
2926 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2927 | * | |
2928 | * Make sure to acquire this mutex before doing a tree search (dev | |
2929 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2930 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2931 | * we release the path used to search the chunk/dev tree and before | |
2932 | * the current task acquires this mutex and calls us. | |
2933 | */ | |
a32bf9a3 | 2934 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 2935 | |
47ab2a6c | 2936 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 2937 | btrfs_scrub_pause(fs_info); |
0b246afa | 2938 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 2939 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
2940 | if (ret) |
2941 | return ret; | |
2942 | ||
b0643e59 DZ |
2943 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
2944 | if (!block_group) | |
2945 | return -ENOENT; | |
2946 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
2947 | btrfs_put_block_group(block_group); | |
2948 | ||
19c4d2f9 CM |
2949 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2950 | chunk_offset); | |
2951 | if (IS_ERR(trans)) { | |
2952 | ret = PTR_ERR(trans); | |
2953 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2954 | return ret; | |
2955 | } | |
2956 | ||
47ab2a6c | 2957 | /* |
19c4d2f9 CM |
2958 | * step two, delete the device extents and the |
2959 | * chunk tree entries | |
47ab2a6c | 2960 | */ |
97aff912 | 2961 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 2962 | btrfs_end_transaction(trans); |
19c4d2f9 | 2963 | return ret; |
2b82032c YZ |
2964 | } |
2965 | ||
2ff7e61e | 2966 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 2967 | { |
0b246afa | 2968 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
2969 | struct btrfs_path *path; |
2970 | struct extent_buffer *leaf; | |
2971 | struct btrfs_chunk *chunk; | |
2972 | struct btrfs_key key; | |
2973 | struct btrfs_key found_key; | |
2b82032c | 2974 | u64 chunk_type; |
ba1bf481 JB |
2975 | bool retried = false; |
2976 | int failed = 0; | |
2b82032c YZ |
2977 | int ret; |
2978 | ||
2979 | path = btrfs_alloc_path(); | |
2980 | if (!path) | |
2981 | return -ENOMEM; | |
2982 | ||
ba1bf481 | 2983 | again: |
2b82032c YZ |
2984 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2985 | key.offset = (u64)-1; | |
2986 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2987 | ||
2988 | while (1) { | |
0b246afa | 2989 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2990 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2991 | if (ret < 0) { |
0b246afa | 2992 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2993 | goto error; |
67c5e7d4 | 2994 | } |
79787eaa | 2995 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2996 | |
2997 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2998 | key.type); | |
67c5e7d4 | 2999 | if (ret) |
0b246afa | 3000 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3001 | if (ret < 0) |
3002 | goto error; | |
3003 | if (ret > 0) | |
3004 | break; | |
1a40e23b | 3005 | |
2b82032c YZ |
3006 | leaf = path->nodes[0]; |
3007 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3008 | |
2b82032c YZ |
3009 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3010 | struct btrfs_chunk); | |
3011 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3012 | btrfs_release_path(path); |
8f18cf13 | 3013 | |
2b82032c | 3014 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3015 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3016 | if (ret == -ENOSPC) |
3017 | failed++; | |
14586651 HS |
3018 | else |
3019 | BUG_ON(ret); | |
2b82032c | 3020 | } |
0b246afa | 3021 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3022 | |
2b82032c YZ |
3023 | if (found_key.offset == 0) |
3024 | break; | |
3025 | key.offset = found_key.offset - 1; | |
3026 | } | |
3027 | ret = 0; | |
ba1bf481 JB |
3028 | if (failed && !retried) { |
3029 | failed = 0; | |
3030 | retried = true; | |
3031 | goto again; | |
fae7f21c | 3032 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3033 | ret = -ENOSPC; |
3034 | } | |
2b82032c YZ |
3035 | error: |
3036 | btrfs_free_path(path); | |
3037 | return ret; | |
8f18cf13 CM |
3038 | } |
3039 | ||
a6f93c71 LB |
3040 | /* |
3041 | * return 1 : allocate a data chunk successfully, | |
3042 | * return <0: errors during allocating a data chunk, | |
3043 | * return 0 : no need to allocate a data chunk. | |
3044 | */ | |
3045 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3046 | u64 chunk_offset) | |
3047 | { | |
32da5386 | 3048 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3049 | u64 bytes_used; |
3050 | u64 chunk_type; | |
3051 | ||
3052 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3053 | ASSERT(cache); | |
3054 | chunk_type = cache->flags; | |
3055 | btrfs_put_block_group(cache); | |
3056 | ||
5ae21692 JT |
3057 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3058 | return 0; | |
3059 | ||
3060 | spin_lock(&fs_info->data_sinfo->lock); | |
3061 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3062 | spin_unlock(&fs_info->data_sinfo->lock); | |
3063 | ||
3064 | if (!bytes_used) { | |
3065 | struct btrfs_trans_handle *trans; | |
3066 | int ret; | |
3067 | ||
3068 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3069 | if (IS_ERR(trans)) | |
3070 | return PTR_ERR(trans); | |
3071 | ||
3072 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3073 | btrfs_end_transaction(trans); | |
3074 | if (ret < 0) | |
3075 | return ret; | |
3076 | return 1; | |
a6f93c71 | 3077 | } |
5ae21692 | 3078 | |
a6f93c71 LB |
3079 | return 0; |
3080 | } | |
3081 | ||
6bccf3ab | 3082 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3083 | struct btrfs_balance_control *bctl) |
3084 | { | |
6bccf3ab | 3085 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3086 | struct btrfs_trans_handle *trans; |
3087 | struct btrfs_balance_item *item; | |
3088 | struct btrfs_disk_balance_args disk_bargs; | |
3089 | struct btrfs_path *path; | |
3090 | struct extent_buffer *leaf; | |
3091 | struct btrfs_key key; | |
3092 | int ret, err; | |
3093 | ||
3094 | path = btrfs_alloc_path(); | |
3095 | if (!path) | |
3096 | return -ENOMEM; | |
3097 | ||
3098 | trans = btrfs_start_transaction(root, 0); | |
3099 | if (IS_ERR(trans)) { | |
3100 | btrfs_free_path(path); | |
3101 | return PTR_ERR(trans); | |
3102 | } | |
3103 | ||
3104 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3105 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3106 | key.offset = 0; |
3107 | ||
3108 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3109 | sizeof(*item)); | |
3110 | if (ret) | |
3111 | goto out; | |
3112 | ||
3113 | leaf = path->nodes[0]; | |
3114 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3115 | ||
b159fa28 | 3116 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3117 | |
3118 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3119 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3120 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3121 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3122 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3123 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3124 | ||
3125 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3126 | ||
3127 | btrfs_mark_buffer_dirty(leaf); | |
3128 | out: | |
3129 | btrfs_free_path(path); | |
3a45bb20 | 3130 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3131 | if (err && !ret) |
3132 | ret = err; | |
3133 | return ret; | |
3134 | } | |
3135 | ||
6bccf3ab | 3136 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3137 | { |
6bccf3ab | 3138 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3139 | struct btrfs_trans_handle *trans; |
3140 | struct btrfs_path *path; | |
3141 | struct btrfs_key key; | |
3142 | int ret, err; | |
3143 | ||
3144 | path = btrfs_alloc_path(); | |
3145 | if (!path) | |
3146 | return -ENOMEM; | |
3147 | ||
3148 | trans = btrfs_start_transaction(root, 0); | |
3149 | if (IS_ERR(trans)) { | |
3150 | btrfs_free_path(path); | |
3151 | return PTR_ERR(trans); | |
3152 | } | |
3153 | ||
3154 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3155 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3156 | key.offset = 0; |
3157 | ||
3158 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3159 | if (ret < 0) | |
3160 | goto out; | |
3161 | if (ret > 0) { | |
3162 | ret = -ENOENT; | |
3163 | goto out; | |
3164 | } | |
3165 | ||
3166 | ret = btrfs_del_item(trans, root, path); | |
3167 | out: | |
3168 | btrfs_free_path(path); | |
3a45bb20 | 3169 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3170 | if (err && !ret) |
3171 | ret = err; | |
3172 | return ret; | |
3173 | } | |
3174 | ||
59641015 ID |
3175 | /* |
3176 | * This is a heuristic used to reduce the number of chunks balanced on | |
3177 | * resume after balance was interrupted. | |
3178 | */ | |
3179 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3180 | { | |
3181 | /* | |
3182 | * Turn on soft mode for chunk types that were being converted. | |
3183 | */ | |
3184 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3185 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3186 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3187 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3188 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3189 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3190 | ||
3191 | /* | |
3192 | * Turn on usage filter if is not already used. The idea is | |
3193 | * that chunks that we have already balanced should be | |
3194 | * reasonably full. Don't do it for chunks that are being | |
3195 | * converted - that will keep us from relocating unconverted | |
3196 | * (albeit full) chunks. | |
3197 | */ | |
3198 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3199 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3200 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3201 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3202 | bctl->data.usage = 90; | |
3203 | } | |
3204 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3205 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3206 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3207 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3208 | bctl->sys.usage = 90; | |
3209 | } | |
3210 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3211 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3212 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3213 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3214 | bctl->meta.usage = 90; | |
3215 | } | |
3216 | } | |
3217 | ||
149196a2 DS |
3218 | /* |
3219 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3220 | */ | |
3221 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3222 | { |
3223 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3224 | int ret; |
c9e9f97b ID |
3225 | |
3226 | BUG_ON(!fs_info->balance_ctl); | |
3227 | ||
3228 | spin_lock(&fs_info->balance_lock); | |
3229 | fs_info->balance_ctl = NULL; | |
3230 | spin_unlock(&fs_info->balance_lock); | |
3231 | ||
3232 | kfree(bctl); | |
149196a2 DS |
3233 | ret = del_balance_item(fs_info); |
3234 | if (ret) | |
3235 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3236 | } |
3237 | ||
ed25e9b2 ID |
3238 | /* |
3239 | * Balance filters. Return 1 if chunk should be filtered out | |
3240 | * (should not be balanced). | |
3241 | */ | |
899c81ea | 3242 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3243 | struct btrfs_balance_args *bargs) |
3244 | { | |
899c81ea ID |
3245 | chunk_type = chunk_to_extended(chunk_type) & |
3246 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3247 | |
899c81ea | 3248 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3249 | return 0; |
3250 | ||
3251 | return 1; | |
3252 | } | |
3253 | ||
dba72cb3 | 3254 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3255 | struct btrfs_balance_args *bargs) |
bc309467 | 3256 | { |
32da5386 | 3257 | struct btrfs_block_group *cache; |
bc309467 DS |
3258 | u64 chunk_used; |
3259 | u64 user_thresh_min; | |
3260 | u64 user_thresh_max; | |
3261 | int ret = 1; | |
3262 | ||
3263 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3264 | chunk_used = cache->used; |
bc309467 DS |
3265 | |
3266 | if (bargs->usage_min == 0) | |
3267 | user_thresh_min = 0; | |
3268 | else | |
b3470b5d DS |
3269 | user_thresh_min = div_factor_fine(cache->length, |
3270 | bargs->usage_min); | |
bc309467 DS |
3271 | |
3272 | if (bargs->usage_max == 0) | |
3273 | user_thresh_max = 1; | |
3274 | else if (bargs->usage_max > 100) | |
b3470b5d | 3275 | user_thresh_max = cache->length; |
bc309467 | 3276 | else |
b3470b5d DS |
3277 | user_thresh_max = div_factor_fine(cache->length, |
3278 | bargs->usage_max); | |
bc309467 DS |
3279 | |
3280 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3281 | ret = 0; | |
3282 | ||
3283 | btrfs_put_block_group(cache); | |
3284 | return ret; | |
3285 | } | |
3286 | ||
dba72cb3 | 3287 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3288 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3289 | { |
32da5386 | 3290 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3291 | u64 chunk_used, user_thresh; |
3292 | int ret = 1; | |
3293 | ||
3294 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3295 | chunk_used = cache->used; |
5ce5b3c0 | 3296 | |
bc309467 | 3297 | if (bargs->usage_min == 0) |
3e39cea6 | 3298 | user_thresh = 1; |
a105bb88 | 3299 | else if (bargs->usage > 100) |
b3470b5d | 3300 | user_thresh = cache->length; |
a105bb88 | 3301 | else |
b3470b5d | 3302 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3303 | |
5ce5b3c0 ID |
3304 | if (chunk_used < user_thresh) |
3305 | ret = 0; | |
3306 | ||
3307 | btrfs_put_block_group(cache); | |
3308 | return ret; | |
3309 | } | |
3310 | ||
409d404b ID |
3311 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3312 | struct btrfs_chunk *chunk, | |
3313 | struct btrfs_balance_args *bargs) | |
3314 | { | |
3315 | struct btrfs_stripe *stripe; | |
3316 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3317 | int i; | |
3318 | ||
3319 | for (i = 0; i < num_stripes; i++) { | |
3320 | stripe = btrfs_stripe_nr(chunk, i); | |
3321 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3322 | return 0; | |
3323 | } | |
3324 | ||
3325 | return 1; | |
3326 | } | |
3327 | ||
946c9256 DS |
3328 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3329 | { | |
3330 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3331 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3332 | const int nparity = btrfs_raid_array[index].nparity; | |
3333 | ||
3334 | if (nparity) | |
3335 | return num_stripes - nparity; | |
3336 | else | |
3337 | return num_stripes / ncopies; | |
3338 | } | |
3339 | ||
94e60d5a ID |
3340 | /* [pstart, pend) */ |
3341 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3342 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3343 | struct btrfs_balance_args *bargs) |
3344 | { | |
3345 | struct btrfs_stripe *stripe; | |
3346 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3347 | u64 stripe_offset; | |
3348 | u64 stripe_length; | |
946c9256 | 3349 | u64 type; |
94e60d5a ID |
3350 | int factor; |
3351 | int i; | |
3352 | ||
3353 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3354 | return 0; | |
3355 | ||
946c9256 DS |
3356 | type = btrfs_chunk_type(leaf, chunk); |
3357 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3358 | |
3359 | for (i = 0; i < num_stripes; i++) { | |
3360 | stripe = btrfs_stripe_nr(chunk, i); | |
3361 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3362 | continue; | |
3363 | ||
3364 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3365 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3366 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3367 | |
3368 | if (stripe_offset < bargs->pend && | |
3369 | stripe_offset + stripe_length > bargs->pstart) | |
3370 | return 0; | |
3371 | } | |
3372 | ||
3373 | return 1; | |
3374 | } | |
3375 | ||
ea67176a ID |
3376 | /* [vstart, vend) */ |
3377 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3378 | struct btrfs_chunk *chunk, | |
3379 | u64 chunk_offset, | |
3380 | struct btrfs_balance_args *bargs) | |
3381 | { | |
3382 | if (chunk_offset < bargs->vend && | |
3383 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3384 | /* at least part of the chunk is inside this vrange */ | |
3385 | return 0; | |
3386 | ||
3387 | return 1; | |
3388 | } | |
3389 | ||
dee32d0a GAP |
3390 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3391 | struct btrfs_chunk *chunk, | |
3392 | struct btrfs_balance_args *bargs) | |
3393 | { | |
3394 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3395 | ||
3396 | if (bargs->stripes_min <= num_stripes | |
3397 | && num_stripes <= bargs->stripes_max) | |
3398 | return 0; | |
3399 | ||
3400 | return 1; | |
3401 | } | |
3402 | ||
899c81ea | 3403 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3404 | struct btrfs_balance_args *bargs) |
3405 | { | |
3406 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3407 | return 0; | |
3408 | ||
899c81ea ID |
3409 | chunk_type = chunk_to_extended(chunk_type) & |
3410 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3411 | |
899c81ea | 3412 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3413 | return 1; |
3414 | ||
3415 | return 0; | |
3416 | } | |
3417 | ||
6ec0896c | 3418 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3419 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3420 | { | |
6ec0896c | 3421 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3422 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3423 | struct btrfs_balance_args *bargs = NULL; |
3424 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3425 | ||
3426 | /* type filter */ | |
3427 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3428 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3429 | return 0; | |
3430 | } | |
3431 | ||
3432 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3433 | bargs = &bctl->data; | |
3434 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3435 | bargs = &bctl->sys; | |
3436 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3437 | bargs = &bctl->meta; | |
3438 | ||
ed25e9b2 ID |
3439 | /* profiles filter */ |
3440 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3441 | chunk_profiles_filter(chunk_type, bargs)) { | |
3442 | return 0; | |
5ce5b3c0 ID |
3443 | } |
3444 | ||
3445 | /* usage filter */ | |
3446 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3447 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3448 | return 0; |
bc309467 | 3449 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3450 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3451 | return 0; |
409d404b ID |
3452 | } |
3453 | ||
3454 | /* devid filter */ | |
3455 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3456 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3457 | return 0; | |
94e60d5a ID |
3458 | } |
3459 | ||
3460 | /* drange filter, makes sense only with devid filter */ | |
3461 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3462 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3463 | return 0; |
ea67176a ID |
3464 | } |
3465 | ||
3466 | /* vrange filter */ | |
3467 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3468 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3469 | return 0; | |
ed25e9b2 ID |
3470 | } |
3471 | ||
dee32d0a GAP |
3472 | /* stripes filter */ |
3473 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3474 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3475 | return 0; | |
3476 | } | |
3477 | ||
cfa4c961 ID |
3478 | /* soft profile changing mode */ |
3479 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3480 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3481 | return 0; | |
3482 | } | |
3483 | ||
7d824b6f DS |
3484 | /* |
3485 | * limited by count, must be the last filter | |
3486 | */ | |
3487 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3488 | if (bargs->limit == 0) | |
3489 | return 0; | |
3490 | else | |
3491 | bargs->limit--; | |
12907fc7 DS |
3492 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3493 | /* | |
3494 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3495 | * determined here because we do not have the global information |
12907fc7 DS |
3496 | * about the count of all chunks that satisfy the filters. |
3497 | */ | |
3498 | if (bargs->limit_max == 0) | |
3499 | return 0; | |
3500 | else | |
3501 | bargs->limit_max--; | |
7d824b6f DS |
3502 | } |
3503 | ||
f43ffb60 ID |
3504 | return 1; |
3505 | } | |
3506 | ||
c9e9f97b | 3507 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3508 | { |
19a39dce | 3509 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3510 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3511 | u64 chunk_type; |
f43ffb60 | 3512 | struct btrfs_chunk *chunk; |
5a488b9d | 3513 | struct btrfs_path *path = NULL; |
ec44a35c | 3514 | struct btrfs_key key; |
ec44a35c | 3515 | struct btrfs_key found_key; |
f43ffb60 ID |
3516 | struct extent_buffer *leaf; |
3517 | int slot; | |
c9e9f97b ID |
3518 | int ret; |
3519 | int enospc_errors = 0; | |
19a39dce | 3520 | bool counting = true; |
12907fc7 | 3521 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3522 | u64 limit_data = bctl->data.limit; |
3523 | u64 limit_meta = bctl->meta.limit; | |
3524 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3525 | u32 count_data = 0; |
3526 | u32 count_meta = 0; | |
3527 | u32 count_sys = 0; | |
2c9fe835 | 3528 | int chunk_reserved = 0; |
ec44a35c | 3529 | |
ec44a35c | 3530 | path = btrfs_alloc_path(); |
17e9f796 MF |
3531 | if (!path) { |
3532 | ret = -ENOMEM; | |
3533 | goto error; | |
3534 | } | |
19a39dce ID |
3535 | |
3536 | /* zero out stat counters */ | |
3537 | spin_lock(&fs_info->balance_lock); | |
3538 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3539 | spin_unlock(&fs_info->balance_lock); | |
3540 | again: | |
7d824b6f | 3541 | if (!counting) { |
12907fc7 DS |
3542 | /* |
3543 | * The single value limit and min/max limits use the same bytes | |
3544 | * in the | |
3545 | */ | |
7d824b6f DS |
3546 | bctl->data.limit = limit_data; |
3547 | bctl->meta.limit = limit_meta; | |
3548 | bctl->sys.limit = limit_sys; | |
3549 | } | |
ec44a35c CM |
3550 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3551 | key.offset = (u64)-1; | |
3552 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3553 | ||
d397712b | 3554 | while (1) { |
19a39dce | 3555 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3556 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3557 | ret = -ECANCELED; |
3558 | goto error; | |
3559 | } | |
3560 | ||
67c5e7d4 | 3561 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3562 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3563 | if (ret < 0) { |
3564 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3565 | goto error; |
67c5e7d4 | 3566 | } |
ec44a35c CM |
3567 | |
3568 | /* | |
3569 | * this shouldn't happen, it means the last relocate | |
3570 | * failed | |
3571 | */ | |
3572 | if (ret == 0) | |
c9e9f97b | 3573 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3574 | |
3575 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3576 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3577 | if (ret) { |
67c5e7d4 | 3578 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3579 | ret = 0; |
ec44a35c | 3580 | break; |
c9e9f97b | 3581 | } |
7d9eb12c | 3582 | |
f43ffb60 ID |
3583 | leaf = path->nodes[0]; |
3584 | slot = path->slots[0]; | |
3585 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3586 | |
67c5e7d4 FM |
3587 | if (found_key.objectid != key.objectid) { |
3588 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3589 | break; |
67c5e7d4 | 3590 | } |
7d9eb12c | 3591 | |
f43ffb60 | 3592 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3593 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3594 | |
19a39dce ID |
3595 | if (!counting) { |
3596 | spin_lock(&fs_info->balance_lock); | |
3597 | bctl->stat.considered++; | |
3598 | spin_unlock(&fs_info->balance_lock); | |
3599 | } | |
3600 | ||
6ec0896c | 3601 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3602 | |
b3b4aa74 | 3603 | btrfs_release_path(path); |
67c5e7d4 FM |
3604 | if (!ret) { |
3605 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3606 | goto loop; |
67c5e7d4 | 3607 | } |
f43ffb60 | 3608 | |
19a39dce | 3609 | if (counting) { |
67c5e7d4 | 3610 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3611 | spin_lock(&fs_info->balance_lock); |
3612 | bctl->stat.expected++; | |
3613 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3614 | |
3615 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3616 | count_data++; | |
3617 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3618 | count_sys++; | |
3619 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3620 | count_meta++; | |
3621 | ||
3622 | goto loop; | |
3623 | } | |
3624 | ||
3625 | /* | |
3626 | * Apply limit_min filter, no need to check if the LIMITS | |
3627 | * filter is used, limit_min is 0 by default | |
3628 | */ | |
3629 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3630 | count_data < bctl->data.limit_min) | |
3631 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3632 | count_meta < bctl->meta.limit_min) | |
3633 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3634 | count_sys < bctl->sys.limit_min)) { | |
3635 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3636 | goto loop; |
3637 | } | |
3638 | ||
a6f93c71 LB |
3639 | if (!chunk_reserved) { |
3640 | /* | |
3641 | * We may be relocating the only data chunk we have, | |
3642 | * which could potentially end up with losing data's | |
3643 | * raid profile, so lets allocate an empty one in | |
3644 | * advance. | |
3645 | */ | |
3646 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3647 | found_key.offset); | |
2c9fe835 ZL |
3648 | if (ret < 0) { |
3649 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3650 | goto error; | |
a6f93c71 LB |
3651 | } else if (ret == 1) { |
3652 | chunk_reserved = 1; | |
2c9fe835 | 3653 | } |
2c9fe835 ZL |
3654 | } |
3655 | ||
5b4aacef | 3656 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3657 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3658 | if (ret == -ENOSPC) { |
c9e9f97b | 3659 | enospc_errors++; |
eede2bf3 OS |
3660 | } else if (ret == -ETXTBSY) { |
3661 | btrfs_info(fs_info, | |
3662 | "skipping relocation of block group %llu due to active swapfile", | |
3663 | found_key.offset); | |
3664 | ret = 0; | |
3665 | } else if (ret) { | |
3666 | goto error; | |
19a39dce ID |
3667 | } else { |
3668 | spin_lock(&fs_info->balance_lock); | |
3669 | bctl->stat.completed++; | |
3670 | spin_unlock(&fs_info->balance_lock); | |
3671 | } | |
f43ffb60 | 3672 | loop: |
795a3321 ID |
3673 | if (found_key.offset == 0) |
3674 | break; | |
ba1bf481 | 3675 | key.offset = found_key.offset - 1; |
ec44a35c | 3676 | } |
c9e9f97b | 3677 | |
19a39dce ID |
3678 | if (counting) { |
3679 | btrfs_release_path(path); | |
3680 | counting = false; | |
3681 | goto again; | |
3682 | } | |
ec44a35c CM |
3683 | error: |
3684 | btrfs_free_path(path); | |
c9e9f97b | 3685 | if (enospc_errors) { |
efe120a0 | 3686 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3687 | enospc_errors); |
c9e9f97b ID |
3688 | if (!ret) |
3689 | ret = -ENOSPC; | |
3690 | } | |
3691 | ||
ec44a35c CM |
3692 | return ret; |
3693 | } | |
3694 | ||
0c460c0d ID |
3695 | /** |
3696 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3697 | * @flags: profile to validate | |
3698 | * @extended: if true @flags is treated as an extended profile | |
3699 | */ | |
3700 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3701 | { | |
3702 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3703 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3704 | ||
3705 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3706 | ||
3707 | /* 1) check that all other bits are zeroed */ | |
3708 | if (flags & ~mask) | |
3709 | return 0; | |
3710 | ||
3711 | /* 2) see if profile is reduced */ | |
3712 | if (flags == 0) | |
3713 | return !extended; /* "0" is valid for usual profiles */ | |
3714 | ||
c1499166 | 3715 | return has_single_bit_set(flags); |
0c460c0d ID |
3716 | } |
3717 | ||
837d5b6e ID |
3718 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3719 | { | |
a7e99c69 ID |
3720 | /* cancel requested || normal exit path */ |
3721 | return atomic_read(&fs_info->balance_cancel_req) || | |
3722 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3723 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3724 | } |
3725 | ||
bdcd3c97 AM |
3726 | /* Non-zero return value signifies invalidity */ |
3727 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3728 | u64 allowed) | |
3729 | { | |
3730 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3731 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3732 | (bctl_arg->target & ~allowed))); | |
3733 | } | |
3734 | ||
56fc37d9 AJ |
3735 | /* |
3736 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3737 | * @size_buf including the terminating null. The output may be trimmed if it | |
3738 | * does not fit into the provided buffer. | |
3739 | */ | |
3740 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3741 | u32 size_buf) | |
3742 | { | |
3743 | int ret; | |
3744 | u32 size_bp = size_buf; | |
3745 | char *bp = buf; | |
3746 | u64 flags = bargs->flags; | |
3747 | char tmp_buf[128] = {'\0'}; | |
3748 | ||
3749 | if (!flags) | |
3750 | return; | |
3751 | ||
3752 | #define CHECK_APPEND_NOARG(a) \ | |
3753 | do { \ | |
3754 | ret = snprintf(bp, size_bp, (a)); \ | |
3755 | if (ret < 0 || ret >= size_bp) \ | |
3756 | goto out_overflow; \ | |
3757 | size_bp -= ret; \ | |
3758 | bp += ret; \ | |
3759 | } while (0) | |
3760 | ||
3761 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3762 | do { \ | |
3763 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3764 | if (ret < 0 || ret >= size_bp) \ | |
3765 | goto out_overflow; \ | |
3766 | size_bp -= ret; \ | |
3767 | bp += ret; \ | |
3768 | } while (0) | |
3769 | ||
3770 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3771 | do { \ | |
3772 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3773 | if (ret < 0 || ret >= size_bp) \ | |
3774 | goto out_overflow; \ | |
3775 | size_bp -= ret; \ | |
3776 | bp += ret; \ | |
3777 | } while (0) | |
3778 | ||
158da513 DS |
3779 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3780 | CHECK_APPEND_1ARG("convert=%s,", | |
3781 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3782 | |
3783 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3784 | CHECK_APPEND_NOARG("soft,"); | |
3785 | ||
3786 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3787 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3788 | sizeof(tmp_buf)); | |
3789 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3790 | } | |
3791 | ||
3792 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3793 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3794 | ||
3795 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3796 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3797 | bargs->usage_min, bargs->usage_max); | |
3798 | ||
3799 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3800 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3801 | ||
3802 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3803 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3804 | bargs->pstart, bargs->pend); | |
3805 | ||
3806 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3807 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3808 | bargs->vstart, bargs->vend); | |
3809 | ||
3810 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3811 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3812 | ||
3813 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3814 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3815 | bargs->limit_min, bargs->limit_max); | |
3816 | ||
3817 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3818 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3819 | bargs->stripes_min, bargs->stripes_max); | |
3820 | ||
3821 | #undef CHECK_APPEND_2ARG | |
3822 | #undef CHECK_APPEND_1ARG | |
3823 | #undef CHECK_APPEND_NOARG | |
3824 | ||
3825 | out_overflow: | |
3826 | ||
3827 | if (size_bp < size_buf) | |
3828 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3829 | else | |
3830 | buf[0] = '\0'; | |
3831 | } | |
3832 | ||
3833 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3834 | { | |
3835 | u32 size_buf = 1024; | |
3836 | char tmp_buf[192] = {'\0'}; | |
3837 | char *buf; | |
3838 | char *bp; | |
3839 | u32 size_bp = size_buf; | |
3840 | int ret; | |
3841 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3842 | ||
3843 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3844 | if (!buf) | |
3845 | return; | |
3846 | ||
3847 | bp = buf; | |
3848 | ||
3849 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3850 | do { \ | |
3851 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3852 | if (ret < 0 || ret >= size_bp) \ | |
3853 | goto out_overflow; \ | |
3854 | size_bp -= ret; \ | |
3855 | bp += ret; \ | |
3856 | } while (0) | |
3857 | ||
3858 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3859 | CHECK_APPEND_1ARG("%s", "-f "); | |
3860 | ||
3861 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3862 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3863 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3864 | } | |
3865 | ||
3866 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
3867 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
3868 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
3869 | } | |
3870 | ||
3871 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
3872 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
3873 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
3874 | } | |
3875 | ||
3876 | #undef CHECK_APPEND_1ARG | |
3877 | ||
3878 | out_overflow: | |
3879 | ||
3880 | if (size_bp < size_buf) | |
3881 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
3882 | btrfs_info(fs_info, "balance: %s %s", | |
3883 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
3884 | "resume" : "start", buf); | |
3885 | ||
3886 | kfree(buf); | |
3887 | } | |
3888 | ||
c9e9f97b | 3889 | /* |
dccdb07b | 3890 | * Should be called with balance mutexe held |
c9e9f97b | 3891 | */ |
6fcf6e2b DS |
3892 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
3893 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
3894 | struct btrfs_ioctl_balance_args *bargs) |
3895 | { | |
14506127 | 3896 | u64 meta_target, data_target; |
f43ffb60 | 3897 | u64 allowed; |
e4837f8f | 3898 | int mixed = 0; |
c9e9f97b | 3899 | int ret; |
8dabb742 | 3900 | u64 num_devices; |
de98ced9 | 3901 | unsigned seq; |
e62869be | 3902 | bool reducing_redundancy; |
081db89b | 3903 | int i; |
c9e9f97b | 3904 | |
837d5b6e | 3905 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3906 | atomic_read(&fs_info->balance_pause_req) || |
3907 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3908 | ret = -EINVAL; |
3909 | goto out; | |
3910 | } | |
3911 | ||
e4837f8f ID |
3912 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3913 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3914 | mixed = 1; | |
3915 | ||
f43ffb60 ID |
3916 | /* |
3917 | * In case of mixed groups both data and meta should be picked, | |
3918 | * and identical options should be given for both of them. | |
3919 | */ | |
e4837f8f ID |
3920 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3921 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3922 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3923 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3924 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 3925 | btrfs_err(fs_info, |
6dac13f8 | 3926 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
3927 | ret = -EINVAL; |
3928 | goto out; | |
3929 | } | |
3930 | } | |
3931 | ||
b35cf1f0 JB |
3932 | /* |
3933 | * rw_devices will not change at the moment, device add/delete/replace | |
3934 | * are excluded by EXCL_OP | |
3935 | */ | |
3936 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
3937 | |
3938 | /* | |
3939 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
3940 | * special bit for it, to make it easier to distinguish. Thus we need | |
3941 | * to set it manually, or balance would refuse the profile. | |
3942 | */ | |
3943 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
3944 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
3945 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
3946 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 3947 | |
bdcd3c97 | 3948 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e | 3949 | btrfs_err(fs_info, |
6dac13f8 | 3950 | "balance: invalid convert data profile %s", |
158da513 | 3951 | btrfs_bg_type_to_raid_name(bctl->data.target)); |
e4d8ec0f ID |
3952 | ret = -EINVAL; |
3953 | goto out; | |
3954 | } | |
bdcd3c97 | 3955 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3956 | btrfs_err(fs_info, |
6dac13f8 | 3957 | "balance: invalid convert metadata profile %s", |
158da513 | 3958 | btrfs_bg_type_to_raid_name(bctl->meta.target)); |
e4d8ec0f ID |
3959 | ret = -EINVAL; |
3960 | goto out; | |
3961 | } | |
bdcd3c97 | 3962 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3963 | btrfs_err(fs_info, |
6dac13f8 | 3964 | "balance: invalid convert system profile %s", |
158da513 | 3965 | btrfs_bg_type_to_raid_name(bctl->sys.target)); |
e4d8ec0f ID |
3966 | ret = -EINVAL; |
3967 | goto out; | |
3968 | } | |
3969 | ||
6079e12c DS |
3970 | /* |
3971 | * Allow to reduce metadata or system integrity only if force set for | |
3972 | * profiles with redundancy (copies, parity) | |
3973 | */ | |
3974 | allowed = 0; | |
3975 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
3976 | if (btrfs_raid_array[i].ncopies >= 2 || | |
3977 | btrfs_raid_array[i].tolerated_failures >= 1) | |
3978 | allowed |= btrfs_raid_array[i].bg_flag; | |
3979 | } | |
de98ced9 MX |
3980 | do { |
3981 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3982 | ||
3983 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3984 | (fs_info->avail_system_alloc_bits & allowed) && | |
3985 | !(bctl->sys.target & allowed)) || | |
3986 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3987 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 3988 | !(bctl->meta.target & allowed))) |
e62869be | 3989 | reducing_redundancy = true; |
5a8067c0 | 3990 | else |
e62869be | 3991 | reducing_redundancy = false; |
5a8067c0 FM |
3992 | |
3993 | /* if we're not converting, the target field is uninitialized */ | |
3994 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3995 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
3996 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
3997 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 3998 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3999 | |
e62869be | 4000 | if (reducing_redundancy) { |
5a8067c0 FM |
4001 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4002 | btrfs_info(fs_info, | |
e62869be | 4003 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4004 | } else { |
4005 | btrfs_err(fs_info, | |
e62869be | 4006 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4007 | ret = -EINVAL; |
4008 | goto out; | |
4009 | } | |
4010 | } | |
4011 | ||
14506127 AB |
4012 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4013 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4014 | btrfs_warn(fs_info, |
6dac13f8 | 4015 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4016 | btrfs_bg_type_to_raid_name(meta_target), |
4017 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4018 | } |
4019 | ||
9e967495 FM |
4020 | if (fs_info->send_in_progress) { |
4021 | btrfs_warn_rl(fs_info, | |
4022 | "cannot run balance while send operations are in progress (%d in progress)", | |
4023 | fs_info->send_in_progress); | |
4024 | ret = -EAGAIN; | |
4025 | goto out; | |
4026 | } | |
4027 | ||
6bccf3ab | 4028 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4029 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4030 | goto out; |
4031 | ||
59641015 ID |
4032 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4033 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4034 | BUG_ON(fs_info->balance_ctl); |
4035 | spin_lock(&fs_info->balance_lock); | |
4036 | fs_info->balance_ctl = bctl; | |
4037 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4038 | } else { |
4039 | BUG_ON(ret != -EEXIST); | |
4040 | spin_lock(&fs_info->balance_lock); | |
4041 | update_balance_args(bctl); | |
4042 | spin_unlock(&fs_info->balance_lock); | |
4043 | } | |
c9e9f97b | 4044 | |
3009a62f DS |
4045 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4046 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4047 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4048 | mutex_unlock(&fs_info->balance_mutex); |
4049 | ||
4050 | ret = __btrfs_balance(fs_info); | |
4051 | ||
4052 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4053 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4054 | btrfs_info(fs_info, "balance: paused"); | |
4055 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4056 | btrfs_info(fs_info, "balance: canceled"); | |
4057 | else | |
4058 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4059 | ||
3009a62f | 4060 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4061 | |
4062 | if (bargs) { | |
4063 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4064 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4065 | } |
4066 | ||
3a01aa7a ID |
4067 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4068 | balance_need_close(fs_info)) { | |
149196a2 | 4069 | reset_balance_state(fs_info); |
a17c95df | 4070 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4071 | } |
4072 | ||
837d5b6e | 4073 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4074 | |
4075 | return ret; | |
4076 | out: | |
59641015 | 4077 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4078 | reset_balance_state(fs_info); |
a17c95df | 4079 | else |
59641015 | 4080 | kfree(bctl); |
a17c95df DS |
4081 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4082 | ||
59641015 ID |
4083 | return ret; |
4084 | } | |
4085 | ||
4086 | static int balance_kthread(void *data) | |
4087 | { | |
2b6ba629 | 4088 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4089 | int ret = 0; |
59641015 | 4090 | |
59641015 | 4091 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4092 | if (fs_info->balance_ctl) |
6fcf6e2b | 4093 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4094 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4095 | |
59641015 ID |
4096 | return ret; |
4097 | } | |
4098 | ||
2b6ba629 ID |
4099 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4100 | { | |
4101 | struct task_struct *tsk; | |
4102 | ||
1354e1a1 | 4103 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4104 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4105 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4106 | return 0; |
4107 | } | |
1354e1a1 | 4108 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4109 | |
3cdde224 | 4110 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4111 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4112 | return 0; |
4113 | } | |
4114 | ||
02ee654d AJ |
4115 | /* |
4116 | * A ro->rw remount sequence should continue with the paused balance | |
4117 | * regardless of who pauses it, system or the user as of now, so set | |
4118 | * the resume flag. | |
4119 | */ | |
4120 | spin_lock(&fs_info->balance_lock); | |
4121 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4122 | spin_unlock(&fs_info->balance_lock); | |
4123 | ||
2b6ba629 | 4124 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4125 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4126 | } |
4127 | ||
68310a5e | 4128 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4129 | { |
59641015 ID |
4130 | struct btrfs_balance_control *bctl; |
4131 | struct btrfs_balance_item *item; | |
4132 | struct btrfs_disk_balance_args disk_bargs; | |
4133 | struct btrfs_path *path; | |
4134 | struct extent_buffer *leaf; | |
4135 | struct btrfs_key key; | |
4136 | int ret; | |
4137 | ||
4138 | path = btrfs_alloc_path(); | |
4139 | if (!path) | |
4140 | return -ENOMEM; | |
4141 | ||
59641015 | 4142 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4143 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4144 | key.offset = 0; |
4145 | ||
68310a5e | 4146 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4147 | if (ret < 0) |
68310a5e | 4148 | goto out; |
59641015 ID |
4149 | if (ret > 0) { /* ret = -ENOENT; */ |
4150 | ret = 0; | |
68310a5e ID |
4151 | goto out; |
4152 | } | |
4153 | ||
4154 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4155 | if (!bctl) { | |
4156 | ret = -ENOMEM; | |
4157 | goto out; | |
59641015 ID |
4158 | } |
4159 | ||
4160 | leaf = path->nodes[0]; | |
4161 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4162 | ||
68310a5e ID |
4163 | bctl->flags = btrfs_balance_flags(leaf, item); |
4164 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4165 | |
4166 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4167 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4168 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4169 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4170 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4171 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4172 | ||
eee95e3f DS |
4173 | /* |
4174 | * This should never happen, as the paused balance state is recovered | |
4175 | * during mount without any chance of other exclusive ops to collide. | |
4176 | * | |
4177 | * This gives the exclusive op status to balance and keeps in paused | |
4178 | * state until user intervention (cancel or umount). If the ownership | |
4179 | * cannot be assigned, show a message but do not fail. The balance | |
4180 | * is in a paused state and must have fs_info::balance_ctl properly | |
4181 | * set up. | |
4182 | */ | |
4183 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4184 | btrfs_warn(fs_info, | |
6dac13f8 | 4185 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4186 | |
68310a5e | 4187 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4188 | BUG_ON(fs_info->balance_ctl); |
4189 | spin_lock(&fs_info->balance_lock); | |
4190 | fs_info->balance_ctl = bctl; | |
4191 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4192 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4193 | out: |
4194 | btrfs_free_path(path); | |
ec44a35c CM |
4195 | return ret; |
4196 | } | |
4197 | ||
837d5b6e ID |
4198 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4199 | { | |
4200 | int ret = 0; | |
4201 | ||
4202 | mutex_lock(&fs_info->balance_mutex); | |
4203 | if (!fs_info->balance_ctl) { | |
4204 | mutex_unlock(&fs_info->balance_mutex); | |
4205 | return -ENOTCONN; | |
4206 | } | |
4207 | ||
3009a62f | 4208 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4209 | atomic_inc(&fs_info->balance_pause_req); |
4210 | mutex_unlock(&fs_info->balance_mutex); | |
4211 | ||
4212 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4213 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4214 | |
4215 | mutex_lock(&fs_info->balance_mutex); | |
4216 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4217 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4218 | atomic_dec(&fs_info->balance_pause_req); |
4219 | } else { | |
4220 | ret = -ENOTCONN; | |
4221 | } | |
4222 | ||
4223 | mutex_unlock(&fs_info->balance_mutex); | |
4224 | return ret; | |
4225 | } | |
4226 | ||
a7e99c69 ID |
4227 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4228 | { | |
4229 | mutex_lock(&fs_info->balance_mutex); | |
4230 | if (!fs_info->balance_ctl) { | |
4231 | mutex_unlock(&fs_info->balance_mutex); | |
4232 | return -ENOTCONN; | |
4233 | } | |
4234 | ||
cf7d20f4 DS |
4235 | /* |
4236 | * A paused balance with the item stored on disk can be resumed at | |
4237 | * mount time if the mount is read-write. Otherwise it's still paused | |
4238 | * and we must not allow cancelling as it deletes the item. | |
4239 | */ | |
4240 | if (sb_rdonly(fs_info->sb)) { | |
4241 | mutex_unlock(&fs_info->balance_mutex); | |
4242 | return -EROFS; | |
4243 | } | |
4244 | ||
a7e99c69 ID |
4245 | atomic_inc(&fs_info->balance_cancel_req); |
4246 | /* | |
4247 | * if we are running just wait and return, balance item is | |
4248 | * deleted in btrfs_balance in this case | |
4249 | */ | |
3009a62f | 4250 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4251 | mutex_unlock(&fs_info->balance_mutex); |
4252 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4253 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4254 | mutex_lock(&fs_info->balance_mutex); |
4255 | } else { | |
a7e99c69 | 4256 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4257 | /* |
4258 | * Lock released to allow other waiters to continue, we'll | |
4259 | * reexamine the status again. | |
4260 | */ | |
a7e99c69 ID |
4261 | mutex_lock(&fs_info->balance_mutex); |
4262 | ||
a17c95df | 4263 | if (fs_info->balance_ctl) { |
149196a2 | 4264 | reset_balance_state(fs_info); |
a17c95df | 4265 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4266 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4267 | } |
a7e99c69 ID |
4268 | } |
4269 | ||
3009a62f DS |
4270 | BUG_ON(fs_info->balance_ctl || |
4271 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4272 | atomic_dec(&fs_info->balance_cancel_req); |
4273 | mutex_unlock(&fs_info->balance_mutex); | |
4274 | return 0; | |
4275 | } | |
4276 | ||
803b2f54 SB |
4277 | static int btrfs_uuid_scan_kthread(void *data) |
4278 | { | |
4279 | struct btrfs_fs_info *fs_info = data; | |
4280 | struct btrfs_root *root = fs_info->tree_root; | |
4281 | struct btrfs_key key; | |
803b2f54 SB |
4282 | struct btrfs_path *path = NULL; |
4283 | int ret = 0; | |
4284 | struct extent_buffer *eb; | |
4285 | int slot; | |
4286 | struct btrfs_root_item root_item; | |
4287 | u32 item_size; | |
f45388f3 | 4288 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4289 | |
4290 | path = btrfs_alloc_path(); | |
4291 | if (!path) { | |
4292 | ret = -ENOMEM; | |
4293 | goto out; | |
4294 | } | |
4295 | ||
4296 | key.objectid = 0; | |
4297 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4298 | key.offset = 0; | |
4299 | ||
803b2f54 | 4300 | while (1) { |
7c829b72 AJ |
4301 | ret = btrfs_search_forward(root, &key, path, |
4302 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4303 | if (ret) { |
4304 | if (ret > 0) | |
4305 | ret = 0; | |
4306 | break; | |
4307 | } | |
4308 | ||
4309 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4310 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4311 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4312 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4313 | goto skip; | |
4314 | ||
4315 | eb = path->nodes[0]; | |
4316 | slot = path->slots[0]; | |
4317 | item_size = btrfs_item_size_nr(eb, slot); | |
4318 | if (item_size < sizeof(root_item)) | |
4319 | goto skip; | |
4320 | ||
803b2f54 SB |
4321 | read_extent_buffer(eb, &root_item, |
4322 | btrfs_item_ptr_offset(eb, slot), | |
4323 | (int)sizeof(root_item)); | |
4324 | if (btrfs_root_refs(&root_item) == 0) | |
4325 | goto skip; | |
f45388f3 FDBM |
4326 | |
4327 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4328 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4329 | if (trans) | |
4330 | goto update_tree; | |
4331 | ||
4332 | btrfs_release_path(path); | |
803b2f54 SB |
4333 | /* |
4334 | * 1 - subvol uuid item | |
4335 | * 1 - received_subvol uuid item | |
4336 | */ | |
4337 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4338 | if (IS_ERR(trans)) { | |
4339 | ret = PTR_ERR(trans); | |
4340 | break; | |
4341 | } | |
f45388f3 FDBM |
4342 | continue; |
4343 | } else { | |
4344 | goto skip; | |
4345 | } | |
4346 | update_tree: | |
4347 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4348 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4349 | BTRFS_UUID_KEY_SUBVOL, |
4350 | key.objectid); | |
4351 | if (ret < 0) { | |
efe120a0 | 4352 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4353 | ret); |
803b2f54 SB |
4354 | break; |
4355 | } | |
4356 | } | |
4357 | ||
4358 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4359 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4360 | root_item.received_uuid, |
4361 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4362 | key.objectid); | |
4363 | if (ret < 0) { | |
efe120a0 | 4364 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4365 | ret); |
803b2f54 SB |
4366 | break; |
4367 | } | |
4368 | } | |
4369 | ||
f45388f3 | 4370 | skip: |
803b2f54 | 4371 | if (trans) { |
3a45bb20 | 4372 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4373 | trans = NULL; |
803b2f54 SB |
4374 | if (ret) |
4375 | break; | |
4376 | } | |
4377 | ||
803b2f54 SB |
4378 | btrfs_release_path(path); |
4379 | if (key.offset < (u64)-1) { | |
4380 | key.offset++; | |
4381 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4382 | key.offset = 0; | |
4383 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4384 | } else if (key.objectid < (u64)-1) { | |
4385 | key.offset = 0; | |
4386 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4387 | key.objectid++; | |
4388 | } else { | |
4389 | break; | |
4390 | } | |
4391 | cond_resched(); | |
4392 | } | |
4393 | ||
4394 | out: | |
4395 | btrfs_free_path(path); | |
f45388f3 | 4396 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4397 | btrfs_end_transaction(trans); |
803b2f54 | 4398 | if (ret) |
efe120a0 | 4399 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4400 | else |
afcdd129 | 4401 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4402 | up(&fs_info->uuid_tree_rescan_sem); |
4403 | return 0; | |
4404 | } | |
4405 | ||
70f80175 SB |
4406 | /* |
4407 | * Callback for btrfs_uuid_tree_iterate(). | |
4408 | * returns: | |
4409 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4410 | * < 0 if an error occurred. |
70f80175 SB |
4411 | * > 0 if the check failed, which means the caller shall remove the entry. |
4412 | */ | |
4413 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4414 | u8 *uuid, u8 type, u64 subid) | |
4415 | { | |
4416 | struct btrfs_key key; | |
4417 | int ret = 0; | |
4418 | struct btrfs_root *subvol_root; | |
4419 | ||
4420 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4421 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4422 | goto out; | |
4423 | ||
4424 | key.objectid = subid; | |
4425 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4426 | key.offset = (u64)-1; | |
4427 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4428 | if (IS_ERR(subvol_root)) { | |
4429 | ret = PTR_ERR(subvol_root); | |
4430 | if (ret == -ENOENT) | |
4431 | ret = 1; | |
4432 | goto out; | |
4433 | } | |
4434 | ||
4435 | switch (type) { | |
4436 | case BTRFS_UUID_KEY_SUBVOL: | |
4437 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4438 | ret = 1; | |
4439 | break; | |
4440 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4441 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4442 | BTRFS_UUID_SIZE)) | |
4443 | ret = 1; | |
4444 | break; | |
4445 | } | |
4446 | ||
4447 | out: | |
4448 | return ret; | |
4449 | } | |
4450 | ||
4451 | static int btrfs_uuid_rescan_kthread(void *data) | |
4452 | { | |
4453 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4454 | int ret; | |
4455 | ||
4456 | /* | |
4457 | * 1st step is to iterate through the existing UUID tree and | |
4458 | * to delete all entries that contain outdated data. | |
4459 | * 2nd step is to add all missing entries to the UUID tree. | |
4460 | */ | |
4461 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4462 | if (ret < 0) { | |
efe120a0 | 4463 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4464 | up(&fs_info->uuid_tree_rescan_sem); |
4465 | return ret; | |
4466 | } | |
4467 | return btrfs_uuid_scan_kthread(data); | |
4468 | } | |
4469 | ||
f7a81ea4 SB |
4470 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4471 | { | |
4472 | struct btrfs_trans_handle *trans; | |
4473 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4474 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4475 | struct task_struct *task; |
4476 | int ret; | |
f7a81ea4 SB |
4477 | |
4478 | /* | |
4479 | * 1 - root node | |
4480 | * 1 - root item | |
4481 | */ | |
4482 | trans = btrfs_start_transaction(tree_root, 2); | |
4483 | if (IS_ERR(trans)) | |
4484 | return PTR_ERR(trans); | |
4485 | ||
9b7a2440 | 4486 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4487 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4488 | ret = PTR_ERR(uuid_root); |
66642832 | 4489 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4490 | btrfs_end_transaction(trans); |
6d13f549 | 4491 | return ret; |
f7a81ea4 SB |
4492 | } |
4493 | ||
4494 | fs_info->uuid_root = uuid_root; | |
4495 | ||
3a45bb20 | 4496 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4497 | if (ret) |
4498 | return ret; | |
4499 | ||
4500 | down(&fs_info->uuid_tree_rescan_sem); | |
4501 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4502 | if (IS_ERR(task)) { | |
70f80175 | 4503 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4504 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4505 | up(&fs_info->uuid_tree_rescan_sem); |
4506 | return PTR_ERR(task); | |
4507 | } | |
4508 | ||
4509 | return 0; | |
f7a81ea4 | 4510 | } |
803b2f54 | 4511 | |
70f80175 SB |
4512 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4513 | { | |
4514 | struct task_struct *task; | |
4515 | ||
4516 | down(&fs_info->uuid_tree_rescan_sem); | |
4517 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4518 | if (IS_ERR(task)) { | |
4519 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4520 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4521 | up(&fs_info->uuid_tree_rescan_sem); |
4522 | return PTR_ERR(task); | |
4523 | } | |
4524 | ||
4525 | return 0; | |
4526 | } | |
4527 | ||
8f18cf13 CM |
4528 | /* |
4529 | * shrinking a device means finding all of the device extents past | |
4530 | * the new size, and then following the back refs to the chunks. | |
4531 | * The chunk relocation code actually frees the device extent | |
4532 | */ | |
4533 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4534 | { | |
0b246afa JM |
4535 | struct btrfs_fs_info *fs_info = device->fs_info; |
4536 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4537 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4538 | struct btrfs_dev_extent *dev_extent = NULL; |
4539 | struct btrfs_path *path; | |
4540 | u64 length; | |
8f18cf13 CM |
4541 | u64 chunk_offset; |
4542 | int ret; | |
4543 | int slot; | |
ba1bf481 JB |
4544 | int failed = 0; |
4545 | bool retried = false; | |
8f18cf13 CM |
4546 | struct extent_buffer *l; |
4547 | struct btrfs_key key; | |
0b246afa | 4548 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4549 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4550 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4551 | u64 diff; |
61d0d0d2 | 4552 | u64 start; |
7dfb8be1 NB |
4553 | |
4554 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4555 | start = new_size; |
0e4324a4 | 4556 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4557 | |
401e29c1 | 4558 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4559 | return -EINVAL; |
4560 | ||
8f18cf13 CM |
4561 | path = btrfs_alloc_path(); |
4562 | if (!path) | |
4563 | return -ENOMEM; | |
4564 | ||
0338dff6 | 4565 | path->reada = READA_BACK; |
8f18cf13 | 4566 | |
61d0d0d2 NB |
4567 | trans = btrfs_start_transaction(root, 0); |
4568 | if (IS_ERR(trans)) { | |
4569 | btrfs_free_path(path); | |
4570 | return PTR_ERR(trans); | |
4571 | } | |
4572 | ||
34441361 | 4573 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4574 | |
7cc8e58d | 4575 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4576 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4577 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4578 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4579 | } |
61d0d0d2 NB |
4580 | |
4581 | /* | |
4582 | * Once the device's size has been set to the new size, ensure all | |
4583 | * in-memory chunks are synced to disk so that the loop below sees them | |
4584 | * and relocates them accordingly. | |
4585 | */ | |
1c11b63e | 4586 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4587 | mutex_unlock(&fs_info->chunk_mutex); |
4588 | ret = btrfs_commit_transaction(trans); | |
4589 | if (ret) | |
4590 | goto done; | |
4591 | } else { | |
4592 | mutex_unlock(&fs_info->chunk_mutex); | |
4593 | btrfs_end_transaction(trans); | |
4594 | } | |
8f18cf13 | 4595 | |
ba1bf481 | 4596 | again: |
8f18cf13 CM |
4597 | key.objectid = device->devid; |
4598 | key.offset = (u64)-1; | |
4599 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4600 | ||
213e64da | 4601 | do { |
0b246afa | 4602 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4603 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4604 | if (ret < 0) { |
0b246afa | 4605 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4606 | goto done; |
67c5e7d4 | 4607 | } |
8f18cf13 CM |
4608 | |
4609 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4610 | if (ret) |
0b246afa | 4611 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4612 | if (ret < 0) |
4613 | goto done; | |
4614 | if (ret) { | |
4615 | ret = 0; | |
b3b4aa74 | 4616 | btrfs_release_path(path); |
bf1fb512 | 4617 | break; |
8f18cf13 CM |
4618 | } |
4619 | ||
4620 | l = path->nodes[0]; | |
4621 | slot = path->slots[0]; | |
4622 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4623 | ||
ba1bf481 | 4624 | if (key.objectid != device->devid) { |
0b246afa | 4625 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4626 | btrfs_release_path(path); |
bf1fb512 | 4627 | break; |
ba1bf481 | 4628 | } |
8f18cf13 CM |
4629 | |
4630 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4631 | length = btrfs_dev_extent_length(l, dev_extent); | |
4632 | ||
ba1bf481 | 4633 | if (key.offset + length <= new_size) { |
0b246afa | 4634 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4635 | btrfs_release_path(path); |
d6397bae | 4636 | break; |
ba1bf481 | 4637 | } |
8f18cf13 | 4638 | |
8f18cf13 | 4639 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4640 | btrfs_release_path(path); |
8f18cf13 | 4641 | |
a6f93c71 LB |
4642 | /* |
4643 | * We may be relocating the only data chunk we have, | |
4644 | * which could potentially end up with losing data's | |
4645 | * raid profile, so lets allocate an empty one in | |
4646 | * advance. | |
4647 | */ | |
4648 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4649 | if (ret < 0) { | |
4650 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4651 | goto done; | |
4652 | } | |
4653 | ||
0b246afa JM |
4654 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4655 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4656 | if (ret == -ENOSPC) { |
ba1bf481 | 4657 | failed++; |
eede2bf3 OS |
4658 | } else if (ret) { |
4659 | if (ret == -ETXTBSY) { | |
4660 | btrfs_warn(fs_info, | |
4661 | "could not shrink block group %llu due to active swapfile", | |
4662 | chunk_offset); | |
4663 | } | |
4664 | goto done; | |
4665 | } | |
213e64da | 4666 | } while (key.offset-- > 0); |
ba1bf481 JB |
4667 | |
4668 | if (failed && !retried) { | |
4669 | failed = 0; | |
4670 | retried = true; | |
4671 | goto again; | |
4672 | } else if (failed && retried) { | |
4673 | ret = -ENOSPC; | |
ba1bf481 | 4674 | goto done; |
8f18cf13 CM |
4675 | } |
4676 | ||
d6397bae | 4677 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4678 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4679 | if (IS_ERR(trans)) { |
4680 | ret = PTR_ERR(trans); | |
4681 | goto done; | |
4682 | } | |
4683 | ||
34441361 | 4684 | mutex_lock(&fs_info->chunk_mutex); |
7cc8e58d | 4685 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4686 | if (list_empty(&device->post_commit_list)) |
4687 | list_add_tail(&device->post_commit_list, | |
4688 | &trans->transaction->dev_update_list); | |
d6397bae | 4689 | |
d6397bae | 4690 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4691 | btrfs_set_super_total_bytes(super_copy, |
4692 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4693 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4694 | |
4695 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4696 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4697 | if (ret < 0) { |
4698 | btrfs_abort_transaction(trans, ret); | |
4699 | btrfs_end_transaction(trans); | |
4700 | } else { | |
4701 | ret = btrfs_commit_transaction(trans); | |
4702 | } | |
8f18cf13 CM |
4703 | done: |
4704 | btrfs_free_path(path); | |
53e489bc | 4705 | if (ret) { |
34441361 | 4706 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4707 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4708 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4709 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4710 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4711 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4712 | } |
8f18cf13 CM |
4713 | return ret; |
4714 | } | |
4715 | ||
2ff7e61e | 4716 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4717 | struct btrfs_key *key, |
4718 | struct btrfs_chunk *chunk, int item_size) | |
4719 | { | |
0b246afa | 4720 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4721 | struct btrfs_disk_key disk_key; |
4722 | u32 array_size; | |
4723 | u8 *ptr; | |
4724 | ||
34441361 | 4725 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4726 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4727 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4728 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4729 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4730 | return -EFBIG; |
fe48a5c0 | 4731 | } |
0b86a832 CM |
4732 | |
4733 | ptr = super_copy->sys_chunk_array + array_size; | |
4734 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4735 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4736 | ptr += sizeof(disk_key); | |
4737 | memcpy(ptr, chunk, item_size); | |
4738 | item_size += sizeof(disk_key); | |
4739 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4740 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4741 | |
0b86a832 CM |
4742 | return 0; |
4743 | } | |
4744 | ||
73c5de00 AJ |
4745 | /* |
4746 | * sort the devices in descending order by max_avail, total_avail | |
4747 | */ | |
4748 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4749 | { |
73c5de00 AJ |
4750 | const struct btrfs_device_info *di_a = a; |
4751 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4752 | |
73c5de00 | 4753 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4754 | return -1; |
73c5de00 | 4755 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4756 | return 1; |
73c5de00 AJ |
4757 | if (di_a->total_avail > di_b->total_avail) |
4758 | return -1; | |
4759 | if (di_a->total_avail < di_b->total_avail) | |
4760 | return 1; | |
4761 | return 0; | |
b2117a39 | 4762 | } |
0b86a832 | 4763 | |
53b381b3 DW |
4764 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4765 | { | |
ffe2d203 | 4766 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4767 | return; |
4768 | ||
ceda0864 | 4769 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4770 | } |
4771 | ||
cfbb825c DS |
4772 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4773 | { | |
4774 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4775 | return; | |
4776 | ||
4777 | btrfs_set_fs_incompat(info, RAID1C34); | |
4778 | } | |
4779 | ||
73c5de00 | 4780 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4781 | u64 start, u64 type) |
b2117a39 | 4782 | { |
2ff7e61e | 4783 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4784 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4785 | struct btrfs_device *device; |
73c5de00 AJ |
4786 | struct map_lookup *map = NULL; |
4787 | struct extent_map_tree *em_tree; | |
4788 | struct extent_map *em; | |
4789 | struct btrfs_device_info *devices_info = NULL; | |
4790 | u64 total_avail; | |
4791 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4792 | int data_stripes; /* number of stripes that count for |
4793 | block group size */ | |
73c5de00 AJ |
4794 | int sub_stripes; /* sub_stripes info for map */ |
4795 | int dev_stripes; /* stripes per dev */ | |
4796 | int devs_max; /* max devs to use */ | |
4797 | int devs_min; /* min devs needed */ | |
4798 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4799 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4800 | int nparity; /* number of stripes worth of bytes to |
4801 | store parity information */ | |
73c5de00 AJ |
4802 | int ret; |
4803 | u64 max_stripe_size; | |
4804 | u64 max_chunk_size; | |
4805 | u64 stripe_size; | |
23f0ff1e | 4806 | u64 chunk_size; |
73c5de00 AJ |
4807 | int ndevs; |
4808 | int i; | |
4809 | int j; | |
31e50229 | 4810 | int index; |
593060d7 | 4811 | |
0c460c0d | 4812 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4813 | |
4117f207 QW |
4814 | if (list_empty(&fs_devices->alloc_list)) { |
4815 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4816 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4817 | return -ENOSPC; |
4117f207 | 4818 | } |
b2117a39 | 4819 | |
3e72ee88 | 4820 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4821 | |
31e50229 LB |
4822 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4823 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4824 | devs_max = btrfs_raid_array[index].devs_max; | |
e3ecdb3f DS |
4825 | if (!devs_max) |
4826 | devs_max = BTRFS_MAX_DEVS(info); | |
31e50229 LB |
4827 | devs_min = btrfs_raid_array[index].devs_min; |
4828 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4829 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4830 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4831 | |
9b3f68b9 | 4832 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4833 | max_stripe_size = SZ_1G; |
fce466ea | 4834 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
9b3f68b9 | 4835 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4836 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4837 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4838 | max_stripe_size = SZ_1G; | |
1100373f | 4839 | else |
ee22184b | 4840 | max_stripe_size = SZ_256M; |
73c5de00 | 4841 | max_chunk_size = max_stripe_size; |
a40a90a0 | 4842 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4843 | max_stripe_size = SZ_32M; |
73c5de00 | 4844 | max_chunk_size = 2 * max_stripe_size; |
c17add7a | 4845 | devs_max = min_t(int, devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); |
73c5de00 | 4846 | } else { |
351fd353 | 4847 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 4848 | type); |
290342f6 | 4849 | BUG(); |
9b3f68b9 CM |
4850 | } |
4851 | ||
52042d8e | 4852 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
4853 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
4854 | max_chunk_size); | |
9b3f68b9 | 4855 | |
31e818fe | 4856 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4857 | GFP_NOFS); |
4858 | if (!devices_info) | |
4859 | return -ENOMEM; | |
0cad8a11 | 4860 | |
9f680ce0 | 4861 | /* |
73c5de00 AJ |
4862 | * in the first pass through the devices list, we gather information |
4863 | * about the available holes on each device. | |
9f680ce0 | 4864 | */ |
73c5de00 | 4865 | ndevs = 0; |
ebcc9301 | 4866 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4867 | u64 max_avail; |
4868 | u64 dev_offset; | |
b2117a39 | 4869 | |
ebbede42 | 4870 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4871 | WARN(1, KERN_ERR |
efe120a0 | 4872 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4873 | continue; |
4874 | } | |
b2117a39 | 4875 | |
e12c9621 AJ |
4876 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4877 | &device->dev_state) || | |
401e29c1 | 4878 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4879 | continue; |
b2117a39 | 4880 | |
73c5de00 AJ |
4881 | if (device->total_bytes > device->bytes_used) |
4882 | total_avail = device->total_bytes - device->bytes_used; | |
4883 | else | |
4884 | total_avail = 0; | |
38c01b96 | 4885 | |
4886 | /* If there is no space on this device, skip it. */ | |
4887 | if (total_avail == 0) | |
4888 | continue; | |
b2117a39 | 4889 | |
60dfdf25 | 4890 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
4891 | max_stripe_size * dev_stripes, |
4892 | &dev_offset, &max_avail); | |
4893 | if (ret && ret != -ENOSPC) | |
4894 | goto error; | |
b2117a39 | 4895 | |
73c5de00 AJ |
4896 | if (ret == 0) |
4897 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4898 | |
4117f207 QW |
4899 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
4900 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4901 | btrfs_debug(info, | |
4902 | "%s: devid %llu has no free space, have=%llu want=%u", | |
4903 | __func__, device->devid, max_avail, | |
4904 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 4905 | continue; |
4117f207 | 4906 | } |
b2117a39 | 4907 | |
063d006f ES |
4908 | if (ndevs == fs_devices->rw_devices) { |
4909 | WARN(1, "%s: found more than %llu devices\n", | |
4910 | __func__, fs_devices->rw_devices); | |
4911 | break; | |
4912 | } | |
73c5de00 AJ |
4913 | devices_info[ndevs].dev_offset = dev_offset; |
4914 | devices_info[ndevs].max_avail = max_avail; | |
4915 | devices_info[ndevs].total_avail = total_avail; | |
4916 | devices_info[ndevs].dev = device; | |
4917 | ++ndevs; | |
4918 | } | |
b2117a39 | 4919 | |
73c5de00 AJ |
4920 | /* |
4921 | * now sort the devices by hole size / available space | |
4922 | */ | |
4923 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4924 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4925 | |
47e6f742 DS |
4926 | /* |
4927 | * Round down to number of usable stripes, devs_increment can be any | |
4928 | * number so we can't use round_down() | |
4929 | */ | |
4930 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4931 | |
ba89b802 | 4932 | if (ndevs < devs_min) { |
73c5de00 | 4933 | ret = -ENOSPC; |
4117f207 QW |
4934 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
4935 | btrfs_debug(info, | |
4936 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 4937 | __func__, ndevs, devs_min); |
4117f207 | 4938 | } |
73c5de00 | 4939 | goto error; |
b2117a39 | 4940 | } |
9f680ce0 | 4941 | |
f148ef4d NB |
4942 | ndevs = min(ndevs, devs_max); |
4943 | ||
73c5de00 | 4944 | /* |
92e222df HK |
4945 | * The primary goal is to maximize the number of stripes, so use as |
4946 | * many devices as possible, even if the stripes are not maximum sized. | |
4947 | * | |
4948 | * The DUP profile stores more than one stripe per device, the | |
4949 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 4950 | */ |
92e222df | 4951 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 4952 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 4953 | |
53b381b3 DW |
4954 | /* |
4955 | * this will have to be fixed for RAID1 and RAID10 over | |
4956 | * more drives | |
4957 | */ | |
b50836ed | 4958 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
4959 | |
4960 | /* | |
4961 | * Use the number of data stripes to figure out how big this chunk | |
4962 | * is really going to be in terms of logical address space, | |
baf92114 HK |
4963 | * and compare that answer with the max chunk size. If it's higher, |
4964 | * we try to reduce stripe_size. | |
86db2578 CM |
4965 | */ |
4966 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 4967 | /* |
baf92114 HK |
4968 | * Reduce stripe_size, round it up to a 16MB boundary again and |
4969 | * then use it, unless it ends up being even bigger than the | |
4970 | * previous value we had already. | |
86db2578 | 4971 | */ |
baf92114 HK |
4972 | stripe_size = min(round_up(div_u64(max_chunk_size, |
4973 | data_stripes), SZ_16M), | |
793ff2c8 | 4974 | stripe_size); |
86db2578 CM |
4975 | } |
4976 | ||
37db63a4 | 4977 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 4978 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
4979 | |
4980 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4981 | if (!map) { | |
4982 | ret = -ENOMEM; | |
4983 | goto error; | |
4984 | } | |
4985 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4986 | |
73c5de00 AJ |
4987 | for (i = 0; i < ndevs; ++i) { |
4988 | for (j = 0; j < dev_stripes; ++j) { | |
4989 | int s = i * dev_stripes + j; | |
4990 | map->stripes[s].dev = devices_info[i].dev; | |
4991 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4992 | j * stripe_size; | |
6324fbf3 | 4993 | } |
6324fbf3 | 4994 | } |
500ceed8 NB |
4995 | map->stripe_len = BTRFS_STRIPE_LEN; |
4996 | map->io_align = BTRFS_STRIPE_LEN; | |
4997 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 4998 | map->type = type; |
2b82032c | 4999 | map->sub_stripes = sub_stripes; |
0b86a832 | 5000 | |
23f0ff1e | 5001 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5002 | |
23f0ff1e | 5003 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5004 | |
172ddd60 | 5005 | em = alloc_extent_map(); |
2b82032c | 5006 | if (!em) { |
298a8f9c | 5007 | kfree(map); |
b2117a39 MX |
5008 | ret = -ENOMEM; |
5009 | goto error; | |
593060d7 | 5010 | } |
298a8f9c | 5011 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5012 | em->map_lookup = map; |
2b82032c | 5013 | em->start = start; |
23f0ff1e | 5014 | em->len = chunk_size; |
2b82032c YZ |
5015 | em->block_start = 0; |
5016 | em->block_len = em->len; | |
6df9a95e | 5017 | em->orig_block_len = stripe_size; |
593060d7 | 5018 | |
c8bf1b67 | 5019 | em_tree = &info->mapping_tree; |
890871be | 5020 | write_lock(&em_tree->lock); |
09a2a8f9 | 5021 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5022 | if (ret) { |
1efb72a3 | 5023 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5024 | free_extent_map(em); |
1dd4602f | 5025 | goto error; |
0f5d42b2 | 5026 | } |
1efb72a3 NB |
5027 | write_unlock(&em_tree->lock); |
5028 | ||
23f0ff1e | 5029 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5030 | if (ret) |
5031 | goto error_del_extent; | |
2b82032c | 5032 | |
bbbf7243 NB |
5033 | for (i = 0; i < map->num_stripes; i++) { |
5034 | struct btrfs_device *dev = map->stripes[i].dev; | |
5035 | ||
5036 | btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size); | |
5037 | if (list_empty(&dev->post_commit_list)) | |
5038 | list_add_tail(&dev->post_commit_list, | |
5039 | &trans->transaction->dev_update_list); | |
5040 | } | |
43530c46 | 5041 | |
a5ed45f8 | 5042 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5043 | |
0f5d42b2 | 5044 | free_extent_map(em); |
0b246afa | 5045 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5046 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5047 | |
b2117a39 | 5048 | kfree(devices_info); |
2b82032c | 5049 | return 0; |
b2117a39 | 5050 | |
6df9a95e | 5051 | error_del_extent: |
0f5d42b2 JB |
5052 | write_lock(&em_tree->lock); |
5053 | remove_extent_mapping(em_tree, em); | |
5054 | write_unlock(&em_tree->lock); | |
5055 | ||
5056 | /* One for our allocation */ | |
5057 | free_extent_map(em); | |
5058 | /* One for the tree reference */ | |
5059 | free_extent_map(em); | |
b2117a39 | 5060 | error: |
b2117a39 MX |
5061 | kfree(devices_info); |
5062 | return ret; | |
2b82032c YZ |
5063 | } |
5064 | ||
6df9a95e | 5065 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5066 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5067 | { |
97aff912 | 5068 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5069 | struct btrfs_root *extent_root = fs_info->extent_root; |
5070 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5071 | struct btrfs_key key; |
2b82032c YZ |
5072 | struct btrfs_device *device; |
5073 | struct btrfs_chunk *chunk; | |
5074 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5075 | struct extent_map *em; |
5076 | struct map_lookup *map; | |
5077 | size_t item_size; | |
5078 | u64 dev_offset; | |
5079 | u64 stripe_size; | |
5080 | int i = 0; | |
140e639f | 5081 | int ret = 0; |
2b82032c | 5082 | |
60ca842e | 5083 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5084 | if (IS_ERR(em)) |
5085 | return PTR_ERR(em); | |
6df9a95e | 5086 | |
95617d69 | 5087 | map = em->map_lookup; |
6df9a95e JB |
5088 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5089 | stripe_size = em->orig_block_len; | |
5090 | ||
2b82032c | 5091 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5092 | if (!chunk) { |
5093 | ret = -ENOMEM; | |
5094 | goto out; | |
5095 | } | |
5096 | ||
50460e37 FM |
5097 | /* |
5098 | * Take the device list mutex to prevent races with the final phase of | |
5099 | * a device replace operation that replaces the device object associated | |
5100 | * with the map's stripes, because the device object's id can change | |
5101 | * at any time during that final phase of the device replace operation | |
5102 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5103 | */ | |
0b246afa | 5104 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5105 | for (i = 0; i < map->num_stripes; i++) { |
5106 | device = map->stripes[i].dev; | |
5107 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5108 | |
0b86a832 | 5109 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5110 | if (ret) |
50460e37 | 5111 | break; |
b5d9071c NB |
5112 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5113 | dev_offset, stripe_size); | |
6df9a95e | 5114 | if (ret) |
50460e37 FM |
5115 | break; |
5116 | } | |
5117 | if (ret) { | |
0b246afa | 5118 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5119 | goto out; |
2b82032c YZ |
5120 | } |
5121 | ||
2b82032c | 5122 | stripe = &chunk->stripe; |
6df9a95e JB |
5123 | for (i = 0; i < map->num_stripes; i++) { |
5124 | device = map->stripes[i].dev; | |
5125 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5126 | |
e17cade2 CM |
5127 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5128 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5129 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5130 | stripe++; |
0b86a832 | 5131 | } |
0b246afa | 5132 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5133 | |
2b82032c | 5134 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5135 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5136 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5137 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5138 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5139 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5140 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5141 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5142 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5143 | |
2b82032c YZ |
5144 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5145 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5146 | key.offset = chunk_offset; | |
0b86a832 | 5147 | |
2b82032c | 5148 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5149 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5150 | /* | |
5151 | * TODO: Cleanup of inserted chunk root in case of | |
5152 | * failure. | |
5153 | */ | |
2ff7e61e | 5154 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5155 | } |
1abe9b8a | 5156 | |
6df9a95e | 5157 | out: |
0b86a832 | 5158 | kfree(chunk); |
6df9a95e | 5159 | free_extent_map(em); |
4ed1d16e | 5160 | return ret; |
2b82032c | 5161 | } |
0b86a832 | 5162 | |
2b82032c | 5163 | /* |
52042d8e AG |
5164 | * Chunk allocation falls into two parts. The first part does work |
5165 | * that makes the new allocated chunk usable, but does not do any operation | |
5166 | * that modifies the chunk tree. The second part does the work that | |
5167 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5168 | * bootstrap process of adding storage to a seed btrfs. |
5169 | */ | |
c216b203 | 5170 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5171 | { |
5172 | u64 chunk_offset; | |
2b82032c | 5173 | |
c216b203 NB |
5174 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5175 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5176 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5177 | } |
5178 | ||
6f8e0fc7 | 5179 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5180 | { |
6f8e0fc7 | 5181 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c YZ |
5182 | u64 chunk_offset; |
5183 | u64 sys_chunk_offset; | |
2b82032c | 5184 | u64 alloc_profile; |
2b82032c YZ |
5185 | int ret; |
5186 | ||
6df9a95e | 5187 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5188 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5189 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5190 | if (ret) |
5191 | return ret; | |
2b82032c | 5192 | |
0b246afa | 5193 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5194 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5195 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5196 | return ret; |
2b82032c YZ |
5197 | } |
5198 | ||
d20983b4 MX |
5199 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5200 | { | |
fc9a2ac7 | 5201 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5202 | |
fc9a2ac7 | 5203 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5204 | } |
5205 | ||
2ff7e61e | 5206 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5207 | { |
5208 | struct extent_map *em; | |
5209 | struct map_lookup *map; | |
2b82032c | 5210 | int readonly = 0; |
d20983b4 | 5211 | int miss_ndevs = 0; |
2b82032c YZ |
5212 | int i; |
5213 | ||
60ca842e | 5214 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5215 | if (IS_ERR(em)) |
2b82032c YZ |
5216 | return 1; |
5217 | ||
95617d69 | 5218 | map = em->map_lookup; |
2b82032c | 5219 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5220 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5221 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5222 | miss_ndevs++; |
5223 | continue; | |
5224 | } | |
ebbede42 AJ |
5225 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5226 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5227 | readonly = 1; |
d20983b4 | 5228 | goto end; |
2b82032c YZ |
5229 | } |
5230 | } | |
d20983b4 MX |
5231 | |
5232 | /* | |
5233 | * If the number of missing devices is larger than max errors, | |
5234 | * we can not write the data into that chunk successfully, so | |
5235 | * set it readonly. | |
5236 | */ | |
5237 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5238 | readonly = 1; | |
5239 | end: | |
0b86a832 | 5240 | free_extent_map(em); |
2b82032c | 5241 | return readonly; |
0b86a832 CM |
5242 | } |
5243 | ||
c8bf1b67 | 5244 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5245 | { |
5246 | struct extent_map *em; | |
5247 | ||
d397712b | 5248 | while (1) { |
c8bf1b67 DS |
5249 | write_lock(&tree->lock); |
5250 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5251 | if (em) |
c8bf1b67 DS |
5252 | remove_extent_mapping(tree, em); |
5253 | write_unlock(&tree->lock); | |
0b86a832 CM |
5254 | if (!em) |
5255 | break; | |
0b86a832 CM |
5256 | /* once for us */ |
5257 | free_extent_map(em); | |
5258 | /* once for the tree */ | |
5259 | free_extent_map(em); | |
5260 | } | |
5261 | } | |
5262 | ||
5d964051 | 5263 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5264 | { |
5265 | struct extent_map *em; | |
5266 | struct map_lookup *map; | |
f188591e CM |
5267 | int ret; |
5268 | ||
60ca842e | 5269 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5270 | if (IS_ERR(em)) |
5271 | /* | |
5272 | * We could return errors for these cases, but that could get | |
5273 | * ugly and we'd probably do the same thing which is just not do | |
5274 | * anything else and exit, so return 1 so the callers don't try | |
5275 | * to use other copies. | |
5276 | */ | |
fb7669b5 | 5277 | return 1; |
fb7669b5 | 5278 | |
95617d69 | 5279 | map = em->map_lookup; |
c7369b3f | 5280 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5281 | ret = map->num_stripes; |
321aecc6 CM |
5282 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5283 | ret = map->sub_stripes; | |
53b381b3 DW |
5284 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5285 | ret = 2; | |
5286 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5287 | /* |
5288 | * There could be two corrupted data stripes, we need | |
5289 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5290 | * |
8810f751 LB |
5291 | * Fail a stripe at a time on every retry except the |
5292 | * stripe under reconstruction. | |
5293 | */ | |
5294 | ret = map->num_stripes; | |
f188591e CM |
5295 | else |
5296 | ret = 1; | |
5297 | free_extent_map(em); | |
ad6d620e | 5298 | |
cb5583dd | 5299 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5300 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5301 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5302 | ret++; |
cb5583dd | 5303 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5304 | |
f188591e CM |
5305 | return ret; |
5306 | } | |
5307 | ||
2ff7e61e | 5308 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5309 | u64 logical) |
5310 | { | |
5311 | struct extent_map *em; | |
5312 | struct map_lookup *map; | |
0b246afa | 5313 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5314 | |
60ca842e | 5315 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5316 | |
69f03f13 NB |
5317 | if (!WARN_ON(IS_ERR(em))) { |
5318 | map = em->map_lookup; | |
5319 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5320 | len = map->stripe_len * nr_data_stripes(map); | |
5321 | free_extent_map(em); | |
5322 | } | |
53b381b3 DW |
5323 | return len; |
5324 | } | |
5325 | ||
e4ff5fb5 | 5326 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5327 | { |
5328 | struct extent_map *em; | |
5329 | struct map_lookup *map; | |
53b381b3 DW |
5330 | int ret = 0; |
5331 | ||
60ca842e | 5332 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5333 | |
69f03f13 NB |
5334 | if(!WARN_ON(IS_ERR(em))) { |
5335 | map = em->map_lookup; | |
5336 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5337 | ret = 1; | |
5338 | free_extent_map(em); | |
5339 | } | |
53b381b3 DW |
5340 | return ret; |
5341 | } | |
5342 | ||
30d9861f | 5343 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5344 | struct map_lookup *map, int first, |
8ba0ae78 | 5345 | int dev_replace_is_ongoing) |
dfe25020 CM |
5346 | { |
5347 | int i; | |
99f92a7c | 5348 | int num_stripes; |
8ba0ae78 | 5349 | int preferred_mirror; |
30d9861f SB |
5350 | int tolerance; |
5351 | struct btrfs_device *srcdev; | |
5352 | ||
99f92a7c | 5353 | ASSERT((map->type & |
c7369b3f | 5354 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5355 | |
5356 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5357 | num_stripes = map->sub_stripes; | |
5358 | else | |
5359 | num_stripes = map->num_stripes; | |
5360 | ||
8ba0ae78 AJ |
5361 | preferred_mirror = first + current->pid % num_stripes; |
5362 | ||
30d9861f SB |
5363 | if (dev_replace_is_ongoing && |
5364 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5365 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5366 | srcdev = fs_info->dev_replace.srcdev; | |
5367 | else | |
5368 | srcdev = NULL; | |
5369 | ||
5370 | /* | |
5371 | * try to avoid the drive that is the source drive for a | |
5372 | * dev-replace procedure, only choose it if no other non-missing | |
5373 | * mirror is available | |
5374 | */ | |
5375 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5376 | if (map->stripes[preferred_mirror].dev->bdev && |
5377 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5378 | return preferred_mirror; | |
99f92a7c | 5379 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5380 | if (map->stripes[i].dev->bdev && |
5381 | (tolerance || map->stripes[i].dev != srcdev)) | |
5382 | return i; | |
5383 | } | |
dfe25020 | 5384 | } |
30d9861f | 5385 | |
dfe25020 CM |
5386 | /* we couldn't find one that doesn't fail. Just return something |
5387 | * and the io error handling code will clean up eventually | |
5388 | */ | |
8ba0ae78 | 5389 | return preferred_mirror; |
dfe25020 CM |
5390 | } |
5391 | ||
53b381b3 DW |
5392 | static inline int parity_smaller(u64 a, u64 b) |
5393 | { | |
5394 | return a > b; | |
5395 | } | |
5396 | ||
5397 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5398 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5399 | { |
5400 | struct btrfs_bio_stripe s; | |
5401 | int i; | |
5402 | u64 l; | |
5403 | int again = 1; | |
5404 | ||
5405 | while (again) { | |
5406 | again = 0; | |
cc7539ed | 5407 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5408 | if (parity_smaller(bbio->raid_map[i], |
5409 | bbio->raid_map[i+1])) { | |
53b381b3 | 5410 | s = bbio->stripes[i]; |
8e5cfb55 | 5411 | l = bbio->raid_map[i]; |
53b381b3 | 5412 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5413 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5414 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5415 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5416 | |
53b381b3 DW |
5417 | again = 1; |
5418 | } | |
5419 | } | |
5420 | } | |
5421 | } | |
5422 | ||
6e9606d2 ZL |
5423 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5424 | { | |
5425 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5426 | /* the size of the btrfs_bio */ |
6e9606d2 | 5427 | sizeof(struct btrfs_bio) + |
e57cf21e | 5428 | /* plus the variable array for the stripes */ |
6e9606d2 | 5429 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5430 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5431 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5432 | /* |
5433 | * plus the raid_map, which includes both the tgt dev | |
5434 | * and the stripes | |
5435 | */ | |
5436 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5437 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5438 | |
5439 | atomic_set(&bbio->error, 0); | |
140475ae | 5440 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5441 | |
5442 | return bbio; | |
5443 | } | |
5444 | ||
5445 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5446 | { | |
140475ae ER |
5447 | WARN_ON(!refcount_read(&bbio->refs)); |
5448 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5449 | } |
5450 | ||
5451 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5452 | { | |
5453 | if (!bbio) | |
5454 | return; | |
140475ae | 5455 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5456 | kfree(bbio); |
5457 | } | |
5458 | ||
0b3d4cd3 LB |
5459 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5460 | /* | |
5461 | * Please note that, discard won't be sent to target device of device | |
5462 | * replace. | |
5463 | */ | |
5464 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5465 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5466 | struct btrfs_bio **bbio_ret) |
5467 | { | |
5468 | struct extent_map *em; | |
5469 | struct map_lookup *map; | |
5470 | struct btrfs_bio *bbio; | |
6b7faadd | 5471 | u64 length = *length_ret; |
0b3d4cd3 LB |
5472 | u64 offset; |
5473 | u64 stripe_nr; | |
5474 | u64 stripe_nr_end; | |
5475 | u64 stripe_end_offset; | |
5476 | u64 stripe_cnt; | |
5477 | u64 stripe_len; | |
5478 | u64 stripe_offset; | |
5479 | u64 num_stripes; | |
5480 | u32 stripe_index; | |
5481 | u32 factor = 0; | |
5482 | u32 sub_stripes = 0; | |
5483 | u64 stripes_per_dev = 0; | |
5484 | u32 remaining_stripes = 0; | |
5485 | u32 last_stripe = 0; | |
5486 | int ret = 0; | |
5487 | int i; | |
5488 | ||
5489 | /* discard always return a bbio */ | |
5490 | ASSERT(bbio_ret); | |
5491 | ||
60ca842e | 5492 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5493 | if (IS_ERR(em)) |
5494 | return PTR_ERR(em); | |
5495 | ||
5496 | map = em->map_lookup; | |
5497 | /* we don't discard raid56 yet */ | |
5498 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5499 | ret = -EOPNOTSUPP; | |
5500 | goto out; | |
5501 | } | |
5502 | ||
5503 | offset = logical - em->start; | |
2d974619 | 5504 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5505 | *length_ret = length; |
0b3d4cd3 LB |
5506 | |
5507 | stripe_len = map->stripe_len; | |
5508 | /* | |
5509 | * stripe_nr counts the total number of stripes we have to stride | |
5510 | * to get to this block | |
5511 | */ | |
5512 | stripe_nr = div64_u64(offset, stripe_len); | |
5513 | ||
5514 | /* stripe_offset is the offset of this block in its stripe */ | |
5515 | stripe_offset = offset - stripe_nr * stripe_len; | |
5516 | ||
5517 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5518 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5519 | stripe_cnt = stripe_nr_end - stripe_nr; |
5520 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5521 | (offset + length); | |
5522 | /* | |
5523 | * after this, stripe_nr is the number of stripes on this | |
5524 | * device we have to walk to find the data, and stripe_index is | |
5525 | * the number of our device in the stripe array | |
5526 | */ | |
5527 | num_stripes = 1; | |
5528 | stripe_index = 0; | |
5529 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5530 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5531 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5532 | sub_stripes = 1; | |
5533 | else | |
5534 | sub_stripes = map->sub_stripes; | |
5535 | ||
5536 | factor = map->num_stripes / sub_stripes; | |
5537 | num_stripes = min_t(u64, map->num_stripes, | |
5538 | sub_stripes * stripe_cnt); | |
5539 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5540 | stripe_index *= sub_stripes; | |
5541 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5542 | &remaining_stripes); | |
5543 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5544 | last_stripe *= sub_stripes; | |
c7369b3f | 5545 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5546 | BTRFS_BLOCK_GROUP_DUP)) { |
5547 | num_stripes = map->num_stripes; | |
5548 | } else { | |
5549 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5550 | &stripe_index); | |
5551 | } | |
5552 | ||
5553 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5554 | if (!bbio) { | |
5555 | ret = -ENOMEM; | |
5556 | goto out; | |
5557 | } | |
5558 | ||
5559 | for (i = 0; i < num_stripes; i++) { | |
5560 | bbio->stripes[i].physical = | |
5561 | map->stripes[stripe_index].physical + | |
5562 | stripe_offset + stripe_nr * map->stripe_len; | |
5563 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5564 | ||
5565 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5566 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5567 | bbio->stripes[i].length = stripes_per_dev * | |
5568 | map->stripe_len; | |
5569 | ||
5570 | if (i / sub_stripes < remaining_stripes) | |
5571 | bbio->stripes[i].length += | |
5572 | map->stripe_len; | |
5573 | ||
5574 | /* | |
5575 | * Special for the first stripe and | |
5576 | * the last stripe: | |
5577 | * | |
5578 | * |-------|...|-------| | |
5579 | * |----------| | |
5580 | * off end_off | |
5581 | */ | |
5582 | if (i < sub_stripes) | |
5583 | bbio->stripes[i].length -= | |
5584 | stripe_offset; | |
5585 | ||
5586 | if (stripe_index >= last_stripe && | |
5587 | stripe_index <= (last_stripe + | |
5588 | sub_stripes - 1)) | |
5589 | bbio->stripes[i].length -= | |
5590 | stripe_end_offset; | |
5591 | ||
5592 | if (i == sub_stripes - 1) | |
5593 | stripe_offset = 0; | |
5594 | } else { | |
5595 | bbio->stripes[i].length = length; | |
5596 | } | |
5597 | ||
5598 | stripe_index++; | |
5599 | if (stripe_index == map->num_stripes) { | |
5600 | stripe_index = 0; | |
5601 | stripe_nr++; | |
5602 | } | |
5603 | } | |
5604 | ||
5605 | *bbio_ret = bbio; | |
5606 | bbio->map_type = map->type; | |
5607 | bbio->num_stripes = num_stripes; | |
5608 | out: | |
5609 | free_extent_map(em); | |
5610 | return ret; | |
5611 | } | |
5612 | ||
5ab56090 LB |
5613 | /* |
5614 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5615 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5616 | * left cursor can also be read from the target drive. | |
5617 | * | |
5618 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5619 | * array of stripes. | |
5620 | * For READ, it also needs to be supported using the same mirror number. | |
5621 | * | |
5622 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5623 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5624 | * case. | |
5625 | */ | |
5626 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5627 | u64 logical, u64 length, | |
5628 | u64 srcdev_devid, int *mirror_num, | |
5629 | u64 *physical) | |
5630 | { | |
5631 | struct btrfs_bio *bbio = NULL; | |
5632 | int num_stripes; | |
5633 | int index_srcdev = 0; | |
5634 | int found = 0; | |
5635 | u64 physical_of_found = 0; | |
5636 | int i; | |
5637 | int ret = 0; | |
5638 | ||
5639 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5640 | logical, &length, &bbio, 0, 0); | |
5641 | if (ret) { | |
5642 | ASSERT(bbio == NULL); | |
5643 | return ret; | |
5644 | } | |
5645 | ||
5646 | num_stripes = bbio->num_stripes; | |
5647 | if (*mirror_num > num_stripes) { | |
5648 | /* | |
5649 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5650 | * that means that the requested area is not left of the left | |
5651 | * cursor | |
5652 | */ | |
5653 | btrfs_put_bbio(bbio); | |
5654 | return -EIO; | |
5655 | } | |
5656 | ||
5657 | /* | |
5658 | * process the rest of the function using the mirror_num of the source | |
5659 | * drive. Therefore look it up first. At the end, patch the device | |
5660 | * pointer to the one of the target drive. | |
5661 | */ | |
5662 | for (i = 0; i < num_stripes; i++) { | |
5663 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5664 | continue; | |
5665 | ||
5666 | /* | |
5667 | * In case of DUP, in order to keep it simple, only add the | |
5668 | * mirror with the lowest physical address | |
5669 | */ | |
5670 | if (found && | |
5671 | physical_of_found <= bbio->stripes[i].physical) | |
5672 | continue; | |
5673 | ||
5674 | index_srcdev = i; | |
5675 | found = 1; | |
5676 | physical_of_found = bbio->stripes[i].physical; | |
5677 | } | |
5678 | ||
5679 | btrfs_put_bbio(bbio); | |
5680 | ||
5681 | ASSERT(found); | |
5682 | if (!found) | |
5683 | return -EIO; | |
5684 | ||
5685 | *mirror_num = index_srcdev + 1; | |
5686 | *physical = physical_of_found; | |
5687 | return ret; | |
5688 | } | |
5689 | ||
73c0f228 LB |
5690 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5691 | struct btrfs_bio **bbio_ret, | |
5692 | struct btrfs_dev_replace *dev_replace, | |
5693 | int *num_stripes_ret, int *max_errors_ret) | |
5694 | { | |
5695 | struct btrfs_bio *bbio = *bbio_ret; | |
5696 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5697 | int tgtdev_indexes = 0; | |
5698 | int num_stripes = *num_stripes_ret; | |
5699 | int max_errors = *max_errors_ret; | |
5700 | int i; | |
5701 | ||
5702 | if (op == BTRFS_MAP_WRITE) { | |
5703 | int index_where_to_add; | |
5704 | ||
5705 | /* | |
5706 | * duplicate the write operations while the dev replace | |
5707 | * procedure is running. Since the copying of the old disk to | |
5708 | * the new disk takes place at run time while the filesystem is | |
5709 | * mounted writable, the regular write operations to the old | |
5710 | * disk have to be duplicated to go to the new disk as well. | |
5711 | * | |
5712 | * Note that device->missing is handled by the caller, and that | |
5713 | * the write to the old disk is already set up in the stripes | |
5714 | * array. | |
5715 | */ | |
5716 | index_where_to_add = num_stripes; | |
5717 | for (i = 0; i < num_stripes; i++) { | |
5718 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5719 | /* write to new disk, too */ | |
5720 | struct btrfs_bio_stripe *new = | |
5721 | bbio->stripes + index_where_to_add; | |
5722 | struct btrfs_bio_stripe *old = | |
5723 | bbio->stripes + i; | |
5724 | ||
5725 | new->physical = old->physical; | |
5726 | new->length = old->length; | |
5727 | new->dev = dev_replace->tgtdev; | |
5728 | bbio->tgtdev_map[i] = index_where_to_add; | |
5729 | index_where_to_add++; | |
5730 | max_errors++; | |
5731 | tgtdev_indexes++; | |
5732 | } | |
5733 | } | |
5734 | num_stripes = index_where_to_add; | |
5735 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5736 | int index_srcdev = 0; | |
5737 | int found = 0; | |
5738 | u64 physical_of_found = 0; | |
5739 | ||
5740 | /* | |
5741 | * During the dev-replace procedure, the target drive can also | |
5742 | * be used to read data in case it is needed to repair a corrupt | |
5743 | * block elsewhere. This is possible if the requested area is | |
5744 | * left of the left cursor. In this area, the target drive is a | |
5745 | * full copy of the source drive. | |
5746 | */ | |
5747 | for (i = 0; i < num_stripes; i++) { | |
5748 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5749 | /* | |
5750 | * In case of DUP, in order to keep it simple, | |
5751 | * only add the mirror with the lowest physical | |
5752 | * address | |
5753 | */ | |
5754 | if (found && | |
5755 | physical_of_found <= | |
5756 | bbio->stripes[i].physical) | |
5757 | continue; | |
5758 | index_srcdev = i; | |
5759 | found = 1; | |
5760 | physical_of_found = bbio->stripes[i].physical; | |
5761 | } | |
5762 | } | |
5763 | if (found) { | |
5764 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5765 | bbio->stripes + num_stripes; | |
5766 | ||
5767 | tgtdev_stripe->physical = physical_of_found; | |
5768 | tgtdev_stripe->length = | |
5769 | bbio->stripes[index_srcdev].length; | |
5770 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5771 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5772 | ||
5773 | tgtdev_indexes++; | |
5774 | num_stripes++; | |
5775 | } | |
5776 | } | |
5777 | ||
5778 | *num_stripes_ret = num_stripes; | |
5779 | *max_errors_ret = max_errors; | |
5780 | bbio->num_tgtdevs = tgtdev_indexes; | |
5781 | *bbio_ret = bbio; | |
5782 | } | |
5783 | ||
2b19a1fe LB |
5784 | static bool need_full_stripe(enum btrfs_map_op op) |
5785 | { | |
5786 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5787 | } | |
5788 | ||
5f141126 NB |
5789 | /* |
5790 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5791 | * tuple. This information is used to calculate how big a | |
5792 | * particular bio can get before it straddles a stripe. | |
5793 | * | |
5794 | * @fs_info - the filesystem | |
5795 | * @logical - address that we want to figure out the geometry of | |
5796 | * @len - the length of IO we are going to perform, starting at @logical | |
5797 | * @op - type of operation - write or read | |
5798 | * @io_geom - pointer used to return values | |
5799 | * | |
5800 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5801 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5802 | */ | |
5803 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5804 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5805 | { |
5806 | struct extent_map *em; | |
5807 | struct map_lookup *map; | |
5808 | u64 offset; | |
5809 | u64 stripe_offset; | |
5810 | u64 stripe_nr; | |
5811 | u64 stripe_len; | |
5812 | u64 raid56_full_stripe_start = (u64)-1; | |
5813 | int data_stripes; | |
373c3b80 | 5814 | int ret = 0; |
5f141126 NB |
5815 | |
5816 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5817 | ||
5818 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5819 | if (IS_ERR(em)) | |
5820 | return PTR_ERR(em); | |
5821 | ||
5822 | map = em->map_lookup; | |
5823 | /* Offset of this logical address in the chunk */ | |
5824 | offset = logical - em->start; | |
5825 | /* Len of a stripe in a chunk */ | |
5826 | stripe_len = map->stripe_len; | |
5827 | /* Stripe wher this block falls in */ | |
5828 | stripe_nr = div64_u64(offset, stripe_len); | |
5829 | /* Offset of stripe in the chunk */ | |
5830 | stripe_offset = stripe_nr * stripe_len; | |
5831 | if (offset < stripe_offset) { | |
5832 | btrfs_crit(fs_info, | |
5833 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5834 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5835 | ret = -EINVAL; |
5836 | goto out; | |
5f141126 NB |
5837 | } |
5838 | ||
5839 | /* stripe_offset is the offset of this block in its stripe */ | |
5840 | stripe_offset = offset - stripe_offset; | |
5841 | data_stripes = nr_data_stripes(map); | |
5842 | ||
5843 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5844 | u64 max_len = stripe_len - stripe_offset; | |
5845 | ||
5846 | /* | |
5847 | * In case of raid56, we need to know the stripe aligned start | |
5848 | */ | |
5849 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5850 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5851 | raid56_full_stripe_start = offset; | |
5852 | ||
5853 | /* | |
5854 | * Allow a write of a full stripe, but make sure we | |
5855 | * don't allow straddling of stripes | |
5856 | */ | |
5857 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5858 | full_stripe_len); | |
5859 | raid56_full_stripe_start *= full_stripe_len; | |
5860 | ||
5861 | /* | |
5862 | * For writes to RAID[56], allow a full stripeset across | |
5863 | * all disks. For other RAID types and for RAID[56] | |
5864 | * reads, just allow a single stripe (on a single disk). | |
5865 | */ | |
5866 | if (op == BTRFS_MAP_WRITE) { | |
5867 | max_len = stripe_len * data_stripes - | |
5868 | (offset - raid56_full_stripe_start); | |
5869 | } | |
5870 | } | |
5871 | len = min_t(u64, em->len - offset, max_len); | |
5872 | } else { | |
5873 | len = em->len - offset; | |
5874 | } | |
5875 | ||
5876 | io_geom->len = len; | |
5877 | io_geom->offset = offset; | |
5878 | io_geom->stripe_len = stripe_len; | |
5879 | io_geom->stripe_nr = stripe_nr; | |
5880 | io_geom->stripe_offset = stripe_offset; | |
5881 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
5882 | ||
373c3b80 JT |
5883 | out: |
5884 | /* once for us */ | |
5885 | free_extent_map(em); | |
5886 | return ret; | |
5f141126 NB |
5887 | } |
5888 | ||
cf8cddd3 CH |
5889 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5890 | enum btrfs_map_op op, | |
f2d8d74d | 5891 | u64 logical, u64 *length, |
a1d3c478 | 5892 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5893 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5894 | { |
5895 | struct extent_map *em; | |
5896 | struct map_lookup *map; | |
593060d7 CM |
5897 | u64 stripe_offset; |
5898 | u64 stripe_nr; | |
53b381b3 | 5899 | u64 stripe_len; |
9d644a62 | 5900 | u32 stripe_index; |
cff82672 | 5901 | int data_stripes; |
cea9e445 | 5902 | int i; |
de11cc12 | 5903 | int ret = 0; |
f2d8d74d | 5904 | int num_stripes; |
a236aed1 | 5905 | int max_errors = 0; |
2c8cdd6e | 5906 | int tgtdev_indexes = 0; |
a1d3c478 | 5907 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5908 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5909 | int dev_replace_is_ongoing = 0; | |
5910 | int num_alloc_stripes; | |
ad6d620e SB |
5911 | int patch_the_first_stripe_for_dev_replace = 0; |
5912 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5913 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
5914 | struct btrfs_io_geometry geom; |
5915 | ||
5916 | ASSERT(bbio_ret); | |
0b86a832 | 5917 | |
0b3d4cd3 LB |
5918 | if (op == BTRFS_MAP_DISCARD) |
5919 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6b7faadd | 5920 | length, bbio_ret); |
0b3d4cd3 | 5921 | |
89b798ad NB |
5922 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
5923 | if (ret < 0) | |
5924 | return ret; | |
0b86a832 | 5925 | |
89b798ad | 5926 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 5927 | ASSERT(!IS_ERR(em)); |
95617d69 | 5928 | map = em->map_lookup; |
593060d7 | 5929 | |
89b798ad | 5930 | *length = geom.len; |
89b798ad NB |
5931 | stripe_len = geom.stripe_len; |
5932 | stripe_nr = geom.stripe_nr; | |
5933 | stripe_offset = geom.stripe_offset; | |
5934 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 5935 | data_stripes = nr_data_stripes(map); |
593060d7 | 5936 | |
cb5583dd | 5937 | down_read(&dev_replace->rwsem); |
472262f3 | 5938 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
5939 | /* |
5940 | * Hold the semaphore for read during the whole operation, write is | |
5941 | * requested at commit time but must wait. | |
5942 | */ | |
472262f3 | 5943 | if (!dev_replace_is_ongoing) |
cb5583dd | 5944 | up_read(&dev_replace->rwsem); |
472262f3 | 5945 | |
ad6d620e | 5946 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 5947 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
5948 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
5949 | dev_replace->srcdev->devid, | |
5950 | &mirror_num, | |
5951 | &physical_to_patch_in_first_stripe); | |
5952 | if (ret) | |
ad6d620e | 5953 | goto out; |
5ab56090 LB |
5954 | else |
5955 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
5956 | } else if (mirror_num > map->num_stripes) { |
5957 | mirror_num = 0; | |
5958 | } | |
5959 | ||
f2d8d74d | 5960 | num_stripes = 1; |
cea9e445 | 5961 | stripe_index = 0; |
fce3bb9a | 5962 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
5963 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5964 | &stripe_index); | |
de483734 | 5965 | if (!need_full_stripe(op)) |
28e1cc7d | 5966 | mirror_num = 1; |
c7369b3f | 5967 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 5968 | if (need_full_stripe(op)) |
f2d8d74d | 5969 | num_stripes = map->num_stripes; |
2fff734f | 5970 | else if (mirror_num) |
f188591e | 5971 | stripe_index = mirror_num - 1; |
dfe25020 | 5972 | else { |
30d9861f | 5973 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 5974 | dev_replace_is_ongoing); |
a1d3c478 | 5975 | mirror_num = stripe_index + 1; |
dfe25020 | 5976 | } |
2fff734f | 5977 | |
611f0e00 | 5978 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 5979 | if (need_full_stripe(op)) { |
f2d8d74d | 5980 | num_stripes = map->num_stripes; |
a1d3c478 | 5981 | } else if (mirror_num) { |
f188591e | 5982 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5983 | } else { |
5984 | mirror_num = 1; | |
5985 | } | |
2fff734f | 5986 | |
321aecc6 | 5987 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5988 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5989 | |
47c5713f | 5990 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5991 | stripe_index *= map->sub_stripes; |
5992 | ||
de483734 | 5993 | if (need_full_stripe(op)) |
f2d8d74d | 5994 | num_stripes = map->sub_stripes; |
321aecc6 CM |
5995 | else if (mirror_num) |
5996 | stripe_index += mirror_num - 1; | |
dfe25020 | 5997 | else { |
3e74317a | 5998 | int old_stripe_index = stripe_index; |
30d9861f SB |
5999 | stripe_index = find_live_mirror(fs_info, map, |
6000 | stripe_index, | |
30d9861f | 6001 | dev_replace_is_ongoing); |
3e74317a | 6002 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6003 | } |
53b381b3 | 6004 | |
ffe2d203 | 6005 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6006 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6007 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6008 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6009 | stripe_len * data_stripes); |
53b381b3 DW |
6010 | |
6011 | /* RAID[56] write or recovery. Return all stripes */ | |
6012 | num_stripes = map->num_stripes; | |
6013 | max_errors = nr_parity_stripes(map); | |
6014 | ||
53b381b3 DW |
6015 | *length = map->stripe_len; |
6016 | stripe_index = 0; | |
6017 | stripe_offset = 0; | |
6018 | } else { | |
6019 | /* | |
6020 | * Mirror #0 or #1 means the original data block. | |
6021 | * Mirror #2 is RAID5 parity block. | |
6022 | * Mirror #3 is RAID6 Q block. | |
6023 | */ | |
47c5713f | 6024 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6025 | data_stripes, &stripe_index); |
53b381b3 | 6026 | if (mirror_num > 1) |
cff82672 | 6027 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6028 | |
6029 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6030 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6031 | &stripe_index); | |
de483734 | 6032 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6033 | mirror_num = 1; |
53b381b3 | 6034 | } |
8790d502 CM |
6035 | } else { |
6036 | /* | |
47c5713f DS |
6037 | * after this, stripe_nr is the number of stripes on this |
6038 | * device we have to walk to find the data, and stripe_index is | |
6039 | * the number of our device in the stripe array | |
8790d502 | 6040 | */ |
47c5713f DS |
6041 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6042 | &stripe_index); | |
a1d3c478 | 6043 | mirror_num = stripe_index + 1; |
8790d502 | 6044 | } |
e042d1ec | 6045 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6046 | btrfs_crit(fs_info, |
6047 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6048 | stripe_index, map->num_stripes); |
6049 | ret = -EINVAL; | |
6050 | goto out; | |
6051 | } | |
cea9e445 | 6052 | |
472262f3 | 6053 | num_alloc_stripes = num_stripes; |
6fad823f | 6054 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6055 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6056 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6057 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6058 | num_alloc_stripes++; |
2c8cdd6e | 6059 | tgtdev_indexes = num_stripes; |
ad6d620e | 6060 | } |
2c8cdd6e | 6061 | |
6e9606d2 | 6062 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6063 | if (!bbio) { |
6064 | ret = -ENOMEM; | |
6065 | goto out; | |
6066 | } | |
6fad823f | 6067 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6068 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6069 | |
8e5cfb55 | 6070 | /* build raid_map */ |
2b19a1fe LB |
6071 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6072 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6073 | u64 tmp; |
9d644a62 | 6074 | unsigned rot; |
8e5cfb55 ZL |
6075 | |
6076 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6077 | sizeof(struct btrfs_bio_stripe) * | |
6078 | num_alloc_stripes + | |
6079 | sizeof(int) * tgtdev_indexes); | |
6080 | ||
6081 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6082 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6083 | |
6084 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6085 | tmp = stripe_nr * data_stripes; |
6086 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6087 | bbio->raid_map[(i+rot) % num_stripes] = |
6088 | em->start + (tmp + i) * map->stripe_len; | |
6089 | ||
6090 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6091 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6092 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6093 | RAID6_Q_STRIPE; | |
6094 | } | |
6095 | ||
b89203f7 | 6096 | |
0b3d4cd3 LB |
6097 | for (i = 0; i < num_stripes; i++) { |
6098 | bbio->stripes[i].physical = | |
6099 | map->stripes[stripe_index].physical + | |
6100 | stripe_offset + | |
6101 | stripe_nr * map->stripe_len; | |
6102 | bbio->stripes[i].dev = | |
6103 | map->stripes[stripe_index].dev; | |
6104 | stripe_index++; | |
593060d7 | 6105 | } |
de11cc12 | 6106 | |
2b19a1fe | 6107 | if (need_full_stripe(op)) |
d20983b4 | 6108 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6109 | |
8e5cfb55 ZL |
6110 | if (bbio->raid_map) |
6111 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6112 | |
73c0f228 | 6113 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6114 | need_full_stripe(op)) { |
73c0f228 LB |
6115 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6116 | &max_errors); | |
472262f3 SB |
6117 | } |
6118 | ||
de11cc12 | 6119 | *bbio_ret = bbio; |
10f11900 | 6120 | bbio->map_type = map->type; |
de11cc12 LZ |
6121 | bbio->num_stripes = num_stripes; |
6122 | bbio->max_errors = max_errors; | |
6123 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6124 | |
6125 | /* | |
6126 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6127 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6128 | * available as a mirror | |
6129 | */ | |
6130 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6131 | WARN_ON(num_stripes > 1); | |
6132 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6133 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6134 | bbio->mirror_num = map->num_stripes + 1; | |
6135 | } | |
cea9e445 | 6136 | out: |
73beece9 | 6137 | if (dev_replace_is_ongoing) { |
53176dde DS |
6138 | lockdep_assert_held(&dev_replace->rwsem); |
6139 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6140 | up_read(&dev_replace->rwsem); |
73beece9 | 6141 | } |
0b86a832 | 6142 | free_extent_map(em); |
de11cc12 | 6143 | return ret; |
0b86a832 CM |
6144 | } |
6145 | ||
cf8cddd3 | 6146 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6147 | u64 logical, u64 *length, |
a1d3c478 | 6148 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6149 | { |
b3d3fa51 | 6150 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6151 | mirror_num, 0); |
f2d8d74d CM |
6152 | } |
6153 | ||
af8e2d1d | 6154 | /* For Scrub/replace */ |
cf8cddd3 | 6155 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6156 | u64 logical, u64 *length, |
825ad4c9 | 6157 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6158 | { |
825ad4c9 | 6159 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6160 | } |
6161 | ||
4246a0b6 | 6162 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6163 | { |
326e1dbb MS |
6164 | bio->bi_private = bbio->private; |
6165 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6166 | bio_endio(bio); |
326e1dbb | 6167 | |
6e9606d2 | 6168 | btrfs_put_bbio(bbio); |
8408c716 MX |
6169 | } |
6170 | ||
4246a0b6 | 6171 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6172 | { |
9be3395b | 6173 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6174 | int is_orig_bio = 0; |
8790d502 | 6175 | |
4e4cbee9 | 6176 | if (bio->bi_status) { |
a1d3c478 | 6177 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6178 | if (bio->bi_status == BLK_STS_IOERR || |
6179 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6180 | unsigned int stripe_index = |
9be3395b | 6181 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6182 | struct btrfs_device *dev; |
442a4f63 SB |
6183 | |
6184 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6185 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6186 | if (dev->bdev) { |
37226b21 | 6187 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6188 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6189 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6190 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6191 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6192 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6193 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6194 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6195 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6196 | } |
442a4f63 SB |
6197 | } |
6198 | } | |
8790d502 | 6199 | |
a1d3c478 | 6200 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6201 | is_orig_bio = 1; |
6202 | ||
c404e0dc MX |
6203 | btrfs_bio_counter_dec(bbio->fs_info); |
6204 | ||
a1d3c478 | 6205 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6206 | if (!is_orig_bio) { |
6207 | bio_put(bio); | |
a1d3c478 | 6208 | bio = bbio->orig_bio; |
7d2b4daa | 6209 | } |
c7b22bb1 | 6210 | |
9be3395b | 6211 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6212 | /* only send an error to the higher layers if it is |
53b381b3 | 6213 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6214 | */ |
a1d3c478 | 6215 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6216 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6217 | } else { |
1259ab75 CM |
6218 | /* |
6219 | * this bio is actually up to date, we didn't | |
6220 | * go over the max number of errors | |
6221 | */ | |
2dbe0c77 | 6222 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6223 | } |
c55f1396 | 6224 | |
4246a0b6 | 6225 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6226 | } else if (!is_orig_bio) { |
8790d502 CM |
6227 | bio_put(bio); |
6228 | } | |
8790d502 CM |
6229 | } |
6230 | ||
2ff7e61e | 6231 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
08635bae | 6232 | u64 physical, int dev_nr) |
de1ee92a JB |
6233 | { |
6234 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6235 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6236 | |
6237 | bio->bi_private = bbio; | |
9be3395b | 6238 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6239 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6240 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6241 | btrfs_debug_in_rcu(fs_info, |
6242 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6243 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6244 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6245 | bio->bi_iter.bi_size); | |
74d46992 | 6246 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6247 | |
2ff7e61e | 6248 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6249 | |
08635bae | 6250 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6251 | } |
6252 | ||
de1ee92a JB |
6253 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6254 | { | |
6255 | atomic_inc(&bbio->error); | |
6256 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6257 | /* Should be the original bio. */ |
8408c716 MX |
6258 | WARN_ON(bio != bbio->orig_bio); |
6259 | ||
9be3395b | 6260 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6261 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6262 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6263 | bio->bi_status = BLK_STS_IOERR; | |
6264 | else | |
6265 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6266 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6267 | } |
6268 | } | |
6269 | ||
58efbc9f | 6270 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6271 | int mirror_num) |
0b86a832 | 6272 | { |
0b86a832 | 6273 | struct btrfs_device *dev; |
8790d502 | 6274 | struct bio *first_bio = bio; |
4f024f37 | 6275 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6276 | u64 length = 0; |
6277 | u64 map_length; | |
0b86a832 | 6278 | int ret; |
08da757d ZL |
6279 | int dev_nr; |
6280 | int total_devs; | |
a1d3c478 | 6281 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6282 | |
4f024f37 | 6283 | length = bio->bi_iter.bi_size; |
0b86a832 | 6284 | map_length = length; |
cea9e445 | 6285 | |
0b246afa | 6286 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6287 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6288 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6289 | if (ret) { |
0b246afa | 6290 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6291 | return errno_to_blk_status(ret); |
c404e0dc | 6292 | } |
cea9e445 | 6293 | |
a1d3c478 | 6294 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6295 | bbio->orig_bio = first_bio; |
6296 | bbio->private = first_bio->bi_private; | |
6297 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6298 | bbio->fs_info = fs_info; |
53b381b3 DW |
6299 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6300 | ||
ad1ba2a0 | 6301 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6302 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6303 | /* In this case, map_length has been set to the length of |
6304 | a single stripe; not the whole write */ | |
37226b21 | 6305 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6306 | ret = raid56_parity_write(fs_info, bio, bbio, |
6307 | map_length); | |
53b381b3 | 6308 | } else { |
2ff7e61e JM |
6309 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6310 | map_length, mirror_num, 1); | |
53b381b3 | 6311 | } |
4245215d | 6312 | |
0b246afa | 6313 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6314 | return errno_to_blk_status(ret); |
53b381b3 DW |
6315 | } |
6316 | ||
cea9e445 | 6317 | if (map_length < length) { |
0b246afa | 6318 | btrfs_crit(fs_info, |
5d163e0e JM |
6319 | "mapping failed logical %llu bio len %llu len %llu", |
6320 | logical, length, map_length); | |
cea9e445 CM |
6321 | BUG(); |
6322 | } | |
a1d3c478 | 6323 | |
08da757d | 6324 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6325 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6326 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6327 | &dev->dev_state) || | |
ebbede42 AJ |
6328 | (bio_op(first_bio) == REQ_OP_WRITE && |
6329 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6330 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6331 | continue; |
6332 | } | |
6333 | ||
3aa8e074 | 6334 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6335 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6336 | else |
a1d3c478 | 6337 | bio = first_bio; |
de1ee92a | 6338 | |
2ff7e61e | 6339 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
08635bae | 6340 | dev_nr); |
8790d502 | 6341 | } |
0b246afa | 6342 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6343 | return BLK_STS_OK; |
0b86a832 CM |
6344 | } |
6345 | ||
09ba3bc9 AJ |
6346 | /* |
6347 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6348 | * return NULL. | |
6349 | * | |
6350 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6351 | * only devid is used. | |
6352 | * | |
6353 | * If @seed is true, traverse through the seed devices. | |
6354 | */ | |
e4319cd9 | 6355 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6356 | u64 devid, u8 *uuid, u8 *fsid, |
6357 | bool seed) | |
0b86a832 | 6358 | { |
2b82032c | 6359 | struct btrfs_device *device; |
2b82032c | 6360 | |
e4319cd9 | 6361 | while (fs_devices) { |
2b82032c | 6362 | if (!fsid || |
e4319cd9 | 6363 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6364 | list_for_each_entry(device, &fs_devices->devices, |
6365 | dev_list) { | |
6366 | if (device->devid == devid && | |
6367 | (!uuid || memcmp(device->uuid, uuid, | |
6368 | BTRFS_UUID_SIZE) == 0)) | |
6369 | return device; | |
6370 | } | |
2b82032c | 6371 | } |
09ba3bc9 AJ |
6372 | if (seed) |
6373 | fs_devices = fs_devices->seed; | |
6374 | else | |
6375 | return NULL; | |
2b82032c YZ |
6376 | } |
6377 | return NULL; | |
0b86a832 CM |
6378 | } |
6379 | ||
2ff7e61e | 6380 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6381 | u64 devid, u8 *dev_uuid) |
6382 | { | |
6383 | struct btrfs_device *device; | |
dfe25020 | 6384 | |
12bd2fc0 ID |
6385 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6386 | if (IS_ERR(device)) | |
adfb69af | 6387 | return device; |
12bd2fc0 ID |
6388 | |
6389 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6390 | device->fs_devices = fs_devices; |
dfe25020 | 6391 | fs_devices->num_devices++; |
12bd2fc0 | 6392 | |
e6e674bd | 6393 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6394 | fs_devices->missing_devices++; |
12bd2fc0 | 6395 | |
dfe25020 CM |
6396 | return device; |
6397 | } | |
6398 | ||
12bd2fc0 ID |
6399 | /** |
6400 | * btrfs_alloc_device - allocate struct btrfs_device | |
6401 | * @fs_info: used only for generating a new devid, can be NULL if | |
6402 | * devid is provided (i.e. @devid != NULL). | |
6403 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6404 | * is generated. | |
6405 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6406 | * is generated. | |
6407 | * | |
6408 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6409 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6410 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6411 | */ |
6412 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6413 | const u64 *devid, | |
6414 | const u8 *uuid) | |
6415 | { | |
6416 | struct btrfs_device *dev; | |
6417 | u64 tmp; | |
6418 | ||
fae7f21c | 6419 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6420 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6421 | |
6422 | dev = __alloc_device(); | |
6423 | if (IS_ERR(dev)) | |
6424 | return dev; | |
6425 | ||
6426 | if (devid) | |
6427 | tmp = *devid; | |
6428 | else { | |
6429 | int ret; | |
6430 | ||
6431 | ret = find_next_devid(fs_info, &tmp); | |
6432 | if (ret) { | |
a425f9d4 | 6433 | btrfs_free_device(dev); |
12bd2fc0 ID |
6434 | return ERR_PTR(ret); |
6435 | } | |
6436 | } | |
6437 | dev->devid = tmp; | |
6438 | ||
6439 | if (uuid) | |
6440 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6441 | else | |
6442 | generate_random_uuid(dev->uuid); | |
6443 | ||
12bd2fc0 ID |
6444 | return dev; |
6445 | } | |
6446 | ||
5a2b8e60 | 6447 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6448 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6449 | { |
2b902dfc AJ |
6450 | if (error) |
6451 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6452 | devid, uuid); | |
6453 | else | |
6454 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6455 | devid, uuid); | |
5a2b8e60 AJ |
6456 | } |
6457 | ||
39e264a4 NB |
6458 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6459 | { | |
6460 | int index = btrfs_bg_flags_to_raid_index(type); | |
6461 | int ncopies = btrfs_raid_array[index].ncopies; | |
e4f6c6be | 6462 | const int nparity = btrfs_raid_array[index].nparity; |
39e264a4 NB |
6463 | int data_stripes; |
6464 | ||
e4f6c6be DS |
6465 | if (nparity) |
6466 | data_stripes = num_stripes - nparity; | |
6467 | else | |
39e264a4 | 6468 | data_stripes = num_stripes / ncopies; |
e4f6c6be | 6469 | |
39e264a4 NB |
6470 | return div_u64(chunk_len, data_stripes); |
6471 | } | |
6472 | ||
9690ac09 | 6473 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6474 | struct btrfs_chunk *chunk) |
6475 | { | |
9690ac09 | 6476 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6477 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6478 | struct map_lookup *map; |
6479 | struct extent_map *em; | |
6480 | u64 logical; | |
6481 | u64 length; | |
e06cd3dd LB |
6482 | u64 devid; |
6483 | u8 uuid[BTRFS_UUID_SIZE]; | |
6484 | int num_stripes; | |
6485 | int ret; | |
6486 | int i; | |
6487 | ||
6488 | logical = key->offset; | |
6489 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6490 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6491 | ||
075cb3c7 QW |
6492 | /* |
6493 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6494 | * as chunk item in tree block is already verified by tree-checker. | |
6495 | */ | |
6496 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6497 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6498 | if (ret) |
6499 | return ret; | |
6500 | } | |
a061fc8d | 6501 | |
c8bf1b67 DS |
6502 | read_lock(&map_tree->lock); |
6503 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6504 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6505 | |
6506 | /* already mapped? */ | |
6507 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6508 | free_extent_map(em); | |
0b86a832 CM |
6509 | return 0; |
6510 | } else if (em) { | |
6511 | free_extent_map(em); | |
6512 | } | |
0b86a832 | 6513 | |
172ddd60 | 6514 | em = alloc_extent_map(); |
0b86a832 CM |
6515 | if (!em) |
6516 | return -ENOMEM; | |
593060d7 | 6517 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6518 | if (!map) { |
6519 | free_extent_map(em); | |
6520 | return -ENOMEM; | |
6521 | } | |
6522 | ||
298a8f9c | 6523 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6524 | em->map_lookup = map; |
0b86a832 CM |
6525 | em->start = logical; |
6526 | em->len = length; | |
70c8a91c | 6527 | em->orig_start = 0; |
0b86a832 | 6528 | em->block_start = 0; |
c8b97818 | 6529 | em->block_len = em->len; |
0b86a832 | 6530 | |
593060d7 CM |
6531 | map->num_stripes = num_stripes; |
6532 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6533 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6534 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6535 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6536 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6537 | map->verified_stripes = 0; |
39e264a4 NB |
6538 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6539 | map->num_stripes); | |
593060d7 CM |
6540 | for (i = 0; i < num_stripes; i++) { |
6541 | map->stripes[i].physical = | |
6542 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6543 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6544 | read_extent_buffer(leaf, uuid, (unsigned long) |
6545 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6546 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6547 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6548 | devid, uuid, NULL, true); |
3cdde224 | 6549 | if (!map->stripes[i].dev && |
0b246afa | 6550 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6551 | free_extent_map(em); |
2b902dfc | 6552 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6553 | return -ENOENT; |
593060d7 | 6554 | } |
dfe25020 CM |
6555 | if (!map->stripes[i].dev) { |
6556 | map->stripes[i].dev = | |
2ff7e61e JM |
6557 | add_missing_dev(fs_info->fs_devices, devid, |
6558 | uuid); | |
adfb69af | 6559 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6560 | free_extent_map(em); |
adfb69af AJ |
6561 | btrfs_err(fs_info, |
6562 | "failed to init missing dev %llu: %ld", | |
6563 | devid, PTR_ERR(map->stripes[i].dev)); | |
6564 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6565 | } |
2b902dfc | 6566 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6567 | } |
e12c9621 AJ |
6568 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6569 | &(map->stripes[i].dev->dev_state)); | |
6570 | ||
0b86a832 CM |
6571 | } |
6572 | ||
c8bf1b67 DS |
6573 | write_lock(&map_tree->lock); |
6574 | ret = add_extent_mapping(map_tree, em, 0); | |
6575 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6576 | if (ret < 0) { |
6577 | btrfs_err(fs_info, | |
6578 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6579 | em->start, em->len, ret); | |
6580 | } | |
0b86a832 CM |
6581 | free_extent_map(em); |
6582 | ||
64f64f43 | 6583 | return ret; |
0b86a832 CM |
6584 | } |
6585 | ||
143bede5 | 6586 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6587 | struct btrfs_dev_item *dev_item, |
6588 | struct btrfs_device *device) | |
6589 | { | |
6590 | unsigned long ptr; | |
0b86a832 CM |
6591 | |
6592 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6593 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6594 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6595 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6596 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6597 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6598 | device->type = btrfs_device_type(leaf, dev_item); |
6599 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6600 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6601 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6602 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6603 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6604 | |
410ba3a2 | 6605 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6606 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6607 | } |
6608 | ||
2ff7e61e | 6609 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6610 | u8 *fsid) |
2b82032c YZ |
6611 | { |
6612 | struct btrfs_fs_devices *fs_devices; | |
6613 | int ret; | |
6614 | ||
a32bf9a3 | 6615 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6616 | ASSERT(fsid); |
2b82032c | 6617 | |
0b246afa | 6618 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6619 | while (fs_devices) { |
44880fdc | 6620 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6621 | return fs_devices; |
6622 | ||
2b82032c YZ |
6623 | fs_devices = fs_devices->seed; |
6624 | } | |
6625 | ||
7239ff4b | 6626 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6627 | if (!fs_devices) { |
0b246afa | 6628 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6629 | return ERR_PTR(-ENOENT); |
6630 | ||
7239ff4b | 6631 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6632 | if (IS_ERR(fs_devices)) |
6633 | return fs_devices; | |
6634 | ||
0395d84f | 6635 | fs_devices->seeding = true; |
5f375835 MX |
6636 | fs_devices->opened = 1; |
6637 | return fs_devices; | |
2b82032c | 6638 | } |
e4404d6e YZ |
6639 | |
6640 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6641 | if (IS_ERR(fs_devices)) |
6642 | return fs_devices; | |
2b82032c | 6643 | |
897fb573 | 6644 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6645 | if (ret) { |
6646 | free_fs_devices(fs_devices); | |
5f375835 | 6647 | fs_devices = ERR_PTR(ret); |
2b82032c | 6648 | goto out; |
48d28232 | 6649 | } |
2b82032c YZ |
6650 | |
6651 | if (!fs_devices->seeding) { | |
0226e0eb | 6652 | close_fs_devices(fs_devices); |
e4404d6e | 6653 | free_fs_devices(fs_devices); |
5f375835 | 6654 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6655 | goto out; |
6656 | } | |
6657 | ||
0b246afa JM |
6658 | fs_devices->seed = fs_info->fs_devices->seed; |
6659 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6660 | out: |
5f375835 | 6661 | return fs_devices; |
2b82032c YZ |
6662 | } |
6663 | ||
17850759 | 6664 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6665 | struct btrfs_dev_item *dev_item) |
6666 | { | |
17850759 | 6667 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6668 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6669 | struct btrfs_device *device; |
6670 | u64 devid; | |
6671 | int ret; | |
44880fdc | 6672 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6673 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6674 | ||
0b86a832 | 6675 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6676 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6677 | BTRFS_UUID_SIZE); |
1473b24e | 6678 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6679 | BTRFS_FSID_SIZE); |
2b82032c | 6680 | |
de37aa51 | 6681 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6682 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6683 | if (IS_ERR(fs_devices)) |
6684 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6685 | } |
6686 | ||
e4319cd9 | 6687 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6688 | fs_uuid, true); |
5f375835 | 6689 | if (!device) { |
c5502451 | 6690 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6691 | btrfs_report_missing_device(fs_info, devid, |
6692 | dev_uuid, true); | |
45dbdbc9 | 6693 | return -ENOENT; |
c5502451 | 6694 | } |
2b82032c | 6695 | |
2ff7e61e | 6696 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6697 | if (IS_ERR(device)) { |
6698 | btrfs_err(fs_info, | |
6699 | "failed to add missing dev %llu: %ld", | |
6700 | devid, PTR_ERR(device)); | |
6701 | return PTR_ERR(device); | |
6702 | } | |
2b902dfc | 6703 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6704 | } else { |
c5502451 | 6705 | if (!device->bdev) { |
2b902dfc AJ |
6706 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6707 | btrfs_report_missing_device(fs_info, | |
6708 | devid, dev_uuid, true); | |
45dbdbc9 | 6709 | return -ENOENT; |
2b902dfc AJ |
6710 | } |
6711 | btrfs_report_missing_device(fs_info, devid, | |
6712 | dev_uuid, false); | |
c5502451 | 6713 | } |
5f375835 | 6714 | |
e6e674bd AJ |
6715 | if (!device->bdev && |
6716 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6717 | /* |
6718 | * this happens when a device that was properly setup | |
6719 | * in the device info lists suddenly goes bad. | |
6720 | * device->bdev is NULL, and so we have to set | |
6721 | * device->missing to one here | |
6722 | */ | |
5f375835 | 6723 | device->fs_devices->missing_devices++; |
e6e674bd | 6724 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6725 | } |
5f375835 MX |
6726 | |
6727 | /* Move the device to its own fs_devices */ | |
6728 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6729 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6730 | &device->dev_state)); | |
5f375835 MX |
6731 | |
6732 | list_move(&device->dev_list, &fs_devices->devices); | |
6733 | device->fs_devices->num_devices--; | |
6734 | fs_devices->num_devices++; | |
6735 | ||
6736 | device->fs_devices->missing_devices--; | |
6737 | fs_devices->missing_devices++; | |
6738 | ||
6739 | device->fs_devices = fs_devices; | |
6740 | } | |
2b82032c YZ |
6741 | } |
6742 | ||
0b246afa | 6743 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6744 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6745 | if (device->generation != |
6746 | btrfs_device_generation(leaf, dev_item)) | |
6747 | return -EINVAL; | |
6324fbf3 | 6748 | } |
0b86a832 CM |
6749 | |
6750 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6751 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6752 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6753 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6754 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6755 | atomic64_add(device->total_bytes - device->bytes_used, |
6756 | &fs_info->free_chunk_space); | |
2bf64758 | 6757 | } |
0b86a832 | 6758 | ret = 0; |
0b86a832 CM |
6759 | return ret; |
6760 | } | |
6761 | ||
6bccf3ab | 6762 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6763 | { |
6bccf3ab | 6764 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6765 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6766 | struct extent_buffer *sb; |
0b86a832 | 6767 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6768 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6769 | u8 *array_ptr; |
6770 | unsigned long sb_array_offset; | |
84eed90f | 6771 | int ret = 0; |
0b86a832 CM |
6772 | u32 num_stripes; |
6773 | u32 array_size; | |
6774 | u32 len = 0; | |
1ffb22cf | 6775 | u32 cur_offset; |
e06cd3dd | 6776 | u64 type; |
84eed90f | 6777 | struct btrfs_key key; |
0b86a832 | 6778 | |
0b246afa | 6779 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6780 | /* |
6781 | * This will create extent buffer of nodesize, superblock size is | |
6782 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6783 | * overallocate but we can keep it as-is, only the first page is used. | |
6784 | */ | |
2ff7e61e | 6785 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6786 | if (IS_ERR(sb)) |
6787 | return PTR_ERR(sb); | |
4db8c528 | 6788 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6789 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6790 | /* |
01327610 | 6791 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6792 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6793 | * pages up-to-date when the page is larger: extent does not cover the |
6794 | * whole page and consequently check_page_uptodate does not find all | |
6795 | * the page's extents up-to-date (the hole beyond sb), | |
6796 | * write_extent_buffer then triggers a WARN_ON. | |
6797 | * | |
6798 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6799 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6800 | * to silence the warning eg. on PowerPC 64. | |
6801 | */ | |
09cbfeaf | 6802 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6803 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6804 | |
a061fc8d | 6805 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6806 | array_size = btrfs_super_sys_array_size(super_copy); |
6807 | ||
1ffb22cf DS |
6808 | array_ptr = super_copy->sys_chunk_array; |
6809 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6810 | cur_offset = 0; | |
0b86a832 | 6811 | |
1ffb22cf DS |
6812 | while (cur_offset < array_size) { |
6813 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6814 | len = sizeof(*disk_key); |
6815 | if (cur_offset + len > array_size) | |
6816 | goto out_short_read; | |
6817 | ||
0b86a832 CM |
6818 | btrfs_disk_key_to_cpu(&key, disk_key); |
6819 | ||
1ffb22cf DS |
6820 | array_ptr += len; |
6821 | sb_array_offset += len; | |
6822 | cur_offset += len; | |
0b86a832 | 6823 | |
32ab3d1b JT |
6824 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
6825 | btrfs_err(fs_info, | |
6826 | "unexpected item type %u in sys_array at offset %u", | |
6827 | (u32)key.type, cur_offset); | |
6828 | ret = -EIO; | |
6829 | break; | |
6830 | } | |
f5cdedd7 | 6831 | |
32ab3d1b JT |
6832 | chunk = (struct btrfs_chunk *)sb_array_offset; |
6833 | /* | |
6834 | * At least one btrfs_chunk with one stripe must be present, | |
6835 | * exact stripe count check comes afterwards | |
6836 | */ | |
6837 | len = btrfs_chunk_item_size(1); | |
6838 | if (cur_offset + len > array_size) | |
6839 | goto out_short_read; | |
e06cd3dd | 6840 | |
32ab3d1b JT |
6841 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
6842 | if (!num_stripes) { | |
6843 | btrfs_err(fs_info, | |
6844 | "invalid number of stripes %u in sys_array at offset %u", | |
6845 | num_stripes, cur_offset); | |
6846 | ret = -EIO; | |
6847 | break; | |
6848 | } | |
e3540eab | 6849 | |
32ab3d1b JT |
6850 | type = btrfs_chunk_type(sb, chunk); |
6851 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6852 | btrfs_err(fs_info, |
32ab3d1b JT |
6853 | "invalid chunk type %llu in sys_array at offset %u", |
6854 | type, cur_offset); | |
84eed90f CM |
6855 | ret = -EIO; |
6856 | break; | |
0b86a832 | 6857 | } |
32ab3d1b JT |
6858 | |
6859 | len = btrfs_chunk_item_size(num_stripes); | |
6860 | if (cur_offset + len > array_size) | |
6861 | goto out_short_read; | |
6862 | ||
6863 | ret = read_one_chunk(&key, sb, chunk); | |
6864 | if (ret) | |
6865 | break; | |
6866 | ||
1ffb22cf DS |
6867 | array_ptr += len; |
6868 | sb_array_offset += len; | |
6869 | cur_offset += len; | |
0b86a832 | 6870 | } |
d865177a | 6871 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6872 | free_extent_buffer_stale(sb); |
84eed90f | 6873 | return ret; |
e3540eab DS |
6874 | |
6875 | out_short_read: | |
ab8d0fc4 | 6876 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6877 | len, cur_offset); |
d865177a | 6878 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6879 | free_extent_buffer_stale(sb); |
e3540eab | 6880 | return -EIO; |
0b86a832 CM |
6881 | } |
6882 | ||
21634a19 QW |
6883 | /* |
6884 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6885 | * | |
6528b99d AJ |
6886 | * If the @failing_dev is specified, it's accounted as missing. |
6887 | * | |
21634a19 QW |
6888 | * Return true if all chunks meet the minimal RW mount requirements. |
6889 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
6890 | */ | |
6528b99d AJ |
6891 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
6892 | struct btrfs_device *failing_dev) | |
21634a19 | 6893 | { |
c8bf1b67 | 6894 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
6895 | struct extent_map *em; |
6896 | u64 next_start = 0; | |
6897 | bool ret = true; | |
6898 | ||
c8bf1b67 DS |
6899 | read_lock(&map_tree->lock); |
6900 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
6901 | read_unlock(&map_tree->lock); | |
21634a19 QW |
6902 | /* No chunk at all? Return false anyway */ |
6903 | if (!em) { | |
6904 | ret = false; | |
6905 | goto out; | |
6906 | } | |
6907 | while (em) { | |
6908 | struct map_lookup *map; | |
6909 | int missing = 0; | |
6910 | int max_tolerated; | |
6911 | int i; | |
6912 | ||
6913 | map = em->map_lookup; | |
6914 | max_tolerated = | |
6915 | btrfs_get_num_tolerated_disk_barrier_failures( | |
6916 | map->type); | |
6917 | for (i = 0; i < map->num_stripes; i++) { | |
6918 | struct btrfs_device *dev = map->stripes[i].dev; | |
6919 | ||
e6e674bd AJ |
6920 | if (!dev || !dev->bdev || |
6921 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
6922 | dev->last_flush_error) |
6923 | missing++; | |
6528b99d AJ |
6924 | else if (failing_dev && failing_dev == dev) |
6925 | missing++; | |
21634a19 QW |
6926 | } |
6927 | if (missing > max_tolerated) { | |
6528b99d AJ |
6928 | if (!failing_dev) |
6929 | btrfs_warn(fs_info, | |
52042d8e | 6930 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
6931 | em->start, missing, max_tolerated); |
6932 | free_extent_map(em); | |
6933 | ret = false; | |
6934 | goto out; | |
6935 | } | |
6936 | next_start = extent_map_end(em); | |
6937 | free_extent_map(em); | |
6938 | ||
c8bf1b67 DS |
6939 | read_lock(&map_tree->lock); |
6940 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 6941 | (u64)(-1) - next_start); |
c8bf1b67 | 6942 | read_unlock(&map_tree->lock); |
21634a19 QW |
6943 | } |
6944 | out: | |
6945 | return ret; | |
6946 | } | |
6947 | ||
5b4aacef | 6948 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6949 | { |
5b4aacef | 6950 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6951 | struct btrfs_path *path; |
6952 | struct extent_buffer *leaf; | |
6953 | struct btrfs_key key; | |
6954 | struct btrfs_key found_key; | |
6955 | int ret; | |
6956 | int slot; | |
99e3ecfc | 6957 | u64 total_dev = 0; |
0b86a832 | 6958 | |
0b86a832 CM |
6959 | path = btrfs_alloc_path(); |
6960 | if (!path) | |
6961 | return -ENOMEM; | |
6962 | ||
3dd0f7a3 AJ |
6963 | /* |
6964 | * uuid_mutex is needed only if we are mounting a sprout FS | |
6965 | * otherwise we don't need it. | |
6966 | */ | |
b367e47f | 6967 | mutex_lock(&uuid_mutex); |
34441361 | 6968 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 6969 | |
395927a9 FDBM |
6970 | /* |
6971 | * Read all device items, and then all the chunk items. All | |
6972 | * device items are found before any chunk item (their object id | |
6973 | * is smaller than the lowest possible object id for a chunk | |
6974 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6975 | */ |
6976 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6977 | key.offset = 0; | |
6978 | key.type = 0; | |
0b86a832 | 6979 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6980 | if (ret < 0) |
6981 | goto error; | |
d397712b | 6982 | while (1) { |
0b86a832 CM |
6983 | leaf = path->nodes[0]; |
6984 | slot = path->slots[0]; | |
6985 | if (slot >= btrfs_header_nritems(leaf)) { | |
6986 | ret = btrfs_next_leaf(root, path); | |
6987 | if (ret == 0) | |
6988 | continue; | |
6989 | if (ret < 0) | |
6990 | goto error; | |
6991 | break; | |
6992 | } | |
6993 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6994 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6995 | struct btrfs_dev_item *dev_item; | |
6996 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6997 | struct btrfs_dev_item); |
17850759 | 6998 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
6999 | if (ret) |
7000 | goto error; | |
99e3ecfc | 7001 | total_dev++; |
0b86a832 CM |
7002 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7003 | struct btrfs_chunk *chunk; | |
7004 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
9690ac09 | 7005 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7006 | if (ret) |
7007 | goto error; | |
0b86a832 CM |
7008 | } |
7009 | path->slots[0]++; | |
7010 | } | |
99e3ecfc LB |
7011 | |
7012 | /* | |
7013 | * After loading chunk tree, we've got all device information, | |
7014 | * do another round of validation checks. | |
7015 | */ | |
0b246afa JM |
7016 | if (total_dev != fs_info->fs_devices->total_devices) { |
7017 | btrfs_err(fs_info, | |
99e3ecfc | 7018 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7019 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7020 | total_dev); |
7021 | ret = -EINVAL; | |
7022 | goto error; | |
7023 | } | |
0b246afa JM |
7024 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7025 | fs_info->fs_devices->total_rw_bytes) { | |
7026 | btrfs_err(fs_info, | |
99e3ecfc | 7027 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7028 | btrfs_super_total_bytes(fs_info->super_copy), |
7029 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7030 | ret = -EINVAL; |
7031 | goto error; | |
7032 | } | |
0b86a832 CM |
7033 | ret = 0; |
7034 | error: | |
34441361 | 7035 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7036 | mutex_unlock(&uuid_mutex); |
7037 | ||
2b82032c | 7038 | btrfs_free_path(path); |
0b86a832 CM |
7039 | return ret; |
7040 | } | |
442a4f63 | 7041 | |
cb517eab MX |
7042 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7043 | { | |
7044 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7045 | struct btrfs_device *device; | |
7046 | ||
29cc83f6 LB |
7047 | while (fs_devices) { |
7048 | mutex_lock(&fs_devices->device_list_mutex); | |
7049 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7050 | device->fs_info = fs_info; |
29cc83f6 LB |
7051 | mutex_unlock(&fs_devices->device_list_mutex); |
7052 | ||
7053 | fs_devices = fs_devices->seed; | |
7054 | } | |
cb517eab MX |
7055 | } |
7056 | ||
1dc990df DS |
7057 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7058 | const struct btrfs_dev_stats_item *ptr, | |
7059 | int index) | |
7060 | { | |
7061 | u64 val; | |
7062 | ||
7063 | read_extent_buffer(eb, &val, | |
7064 | offsetof(struct btrfs_dev_stats_item, values) + | |
7065 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7066 | sizeof(val)); | |
7067 | return val; | |
7068 | } | |
7069 | ||
7070 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7071 | struct btrfs_dev_stats_item *ptr, | |
7072 | int index, u64 val) | |
7073 | { | |
7074 | write_extent_buffer(eb, &val, | |
7075 | offsetof(struct btrfs_dev_stats_item, values) + | |
7076 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7077 | sizeof(val)); | |
7078 | } | |
7079 | ||
733f4fbb SB |
7080 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7081 | { | |
7082 | struct btrfs_key key; | |
733f4fbb SB |
7083 | struct btrfs_root *dev_root = fs_info->dev_root; |
7084 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7085 | struct extent_buffer *eb; | |
7086 | int slot; | |
7087 | int ret = 0; | |
7088 | struct btrfs_device *device; | |
7089 | struct btrfs_path *path = NULL; | |
7090 | int i; | |
7091 | ||
7092 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7093 | if (!path) |
7094 | return -ENOMEM; | |
733f4fbb SB |
7095 | |
7096 | mutex_lock(&fs_devices->device_list_mutex); | |
7097 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7098 | int item_size; | |
7099 | struct btrfs_dev_stats_item *ptr; | |
7100 | ||
242e2956 DS |
7101 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7102 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7103 | key.offset = device->devid; |
7104 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7105 | if (ret) { | |
ae4b9b4c AJ |
7106 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7107 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7108 | device->dev_stats_valid = 1; |
7109 | btrfs_release_path(path); | |
7110 | continue; | |
7111 | } | |
7112 | slot = path->slots[0]; | |
7113 | eb = path->nodes[0]; | |
733f4fbb SB |
7114 | item_size = btrfs_item_size_nr(eb, slot); |
7115 | ||
7116 | ptr = btrfs_item_ptr(eb, slot, | |
7117 | struct btrfs_dev_stats_item); | |
7118 | ||
7119 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7120 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7121 | btrfs_dev_stat_set(device, i, | |
7122 | btrfs_dev_stats_value(eb, ptr, i)); | |
7123 | else | |
4e411a7d | 7124 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7125 | } |
7126 | ||
7127 | device->dev_stats_valid = 1; | |
7128 | btrfs_dev_stat_print_on_load(device); | |
7129 | btrfs_release_path(path); | |
7130 | } | |
7131 | mutex_unlock(&fs_devices->device_list_mutex); | |
7132 | ||
733f4fbb SB |
7133 | btrfs_free_path(path); |
7134 | return ret < 0 ? ret : 0; | |
7135 | } | |
7136 | ||
7137 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7138 | struct btrfs_device *device) |
7139 | { | |
5495f195 | 7140 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7141 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7142 | struct btrfs_path *path; |
7143 | struct btrfs_key key; | |
7144 | struct extent_buffer *eb; | |
7145 | struct btrfs_dev_stats_item *ptr; | |
7146 | int ret; | |
7147 | int i; | |
7148 | ||
242e2956 DS |
7149 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7150 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7151 | key.offset = device->devid; |
7152 | ||
7153 | path = btrfs_alloc_path(); | |
fa252992 DS |
7154 | if (!path) |
7155 | return -ENOMEM; | |
733f4fbb SB |
7156 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7157 | if (ret < 0) { | |
0b246afa | 7158 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7159 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7160 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7161 | goto out; |
7162 | } | |
7163 | ||
7164 | if (ret == 0 && | |
7165 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7166 | /* need to delete old one and insert a new one */ | |
7167 | ret = btrfs_del_item(trans, dev_root, path); | |
7168 | if (ret != 0) { | |
0b246afa | 7169 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7170 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7171 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7172 | goto out; |
7173 | } | |
7174 | ret = 1; | |
7175 | } | |
7176 | ||
7177 | if (ret == 1) { | |
7178 | /* need to insert a new item */ | |
7179 | btrfs_release_path(path); | |
7180 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7181 | &key, sizeof(*ptr)); | |
7182 | if (ret < 0) { | |
0b246afa | 7183 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7184 | "insert dev_stats item for device %s failed %d", |
7185 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7186 | goto out; |
7187 | } | |
7188 | } | |
7189 | ||
7190 | eb = path->nodes[0]; | |
7191 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7192 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7193 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7194 | btrfs_dev_stat_read(device, i)); | |
7195 | btrfs_mark_buffer_dirty(eb); | |
7196 | ||
7197 | out: | |
7198 | btrfs_free_path(path); | |
7199 | return ret; | |
7200 | } | |
7201 | ||
7202 | /* | |
7203 | * called from commit_transaction. Writes all changed device stats to disk. | |
7204 | */ | |
196c9d8d | 7205 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7206 | { |
196c9d8d | 7207 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7208 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7209 | struct btrfs_device *device; | |
addc3fa7 | 7210 | int stats_cnt; |
733f4fbb SB |
7211 | int ret = 0; |
7212 | ||
7213 | mutex_lock(&fs_devices->device_list_mutex); | |
7214 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7215 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7216 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7217 | continue; |
7218 | ||
9deae968 NB |
7219 | |
7220 | /* | |
7221 | * There is a LOAD-LOAD control dependency between the value of | |
7222 | * dev_stats_ccnt and updating the on-disk values which requires | |
7223 | * reading the in-memory counters. Such control dependencies | |
7224 | * require explicit read memory barriers. | |
7225 | * | |
7226 | * This memory barriers pairs with smp_mb__before_atomic in | |
7227 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7228 | * barrier implied by atomic_xchg in | |
7229 | * btrfs_dev_stats_read_and_reset | |
7230 | */ | |
7231 | smp_rmb(); | |
7232 | ||
5495f195 | 7233 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7234 | if (!ret) |
addc3fa7 | 7235 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7236 | } |
7237 | mutex_unlock(&fs_devices->device_list_mutex); | |
7238 | ||
7239 | return ret; | |
7240 | } | |
7241 | ||
442a4f63 SB |
7242 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7243 | { | |
7244 | btrfs_dev_stat_inc(dev, index); | |
7245 | btrfs_dev_stat_print_on_error(dev); | |
7246 | } | |
7247 | ||
48a3b636 | 7248 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7249 | { |
733f4fbb SB |
7250 | if (!dev->dev_stats_valid) |
7251 | return; | |
fb456252 | 7252 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7253 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7254 | rcu_str_deref(dev->name), |
442a4f63 SB |
7255 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7256 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7257 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7258 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7259 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7260 | } |
c11d2c23 | 7261 | |
733f4fbb SB |
7262 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7263 | { | |
a98cdb85 SB |
7264 | int i; |
7265 | ||
7266 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7267 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7268 | break; | |
7269 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7270 | return; /* all values == 0, suppress message */ | |
7271 | ||
fb456252 | 7272 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7273 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7274 | rcu_str_deref(dev->name), |
733f4fbb SB |
7275 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7276 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7277 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7278 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7279 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7280 | } | |
7281 | ||
2ff7e61e | 7282 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7283 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7284 | { |
7285 | struct btrfs_device *dev; | |
0b246afa | 7286 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7287 | int i; |
7288 | ||
7289 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7290 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7291 | true); | |
c11d2c23 SB |
7292 | mutex_unlock(&fs_devices->device_list_mutex); |
7293 | ||
7294 | if (!dev) { | |
0b246afa | 7295 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7296 | return -ENODEV; |
733f4fbb | 7297 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7298 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7299 | return -ENODEV; |
b27f7c0c | 7300 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7301 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7302 | if (stats->nr_items > i) | |
7303 | stats->values[i] = | |
7304 | btrfs_dev_stat_read_and_reset(dev, i); | |
7305 | else | |
4e411a7d | 7306 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7307 | } |
a69976bc AJ |
7308 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7309 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7310 | } else { |
7311 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7312 | if (stats->nr_items > i) | |
7313 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7314 | } | |
7315 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7316 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7317 | return 0; | |
7318 | } | |
a8a6dab7 | 7319 | |
da353f6b | 7320 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7321 | { |
7322 | struct buffer_head *bh; | |
7323 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7324 | int copy_num; |
a8a6dab7 | 7325 | |
12b1c263 AJ |
7326 | if (!bdev) |
7327 | return; | |
a8a6dab7 | 7328 | |
12b1c263 AJ |
7329 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7330 | copy_num++) { | |
a8a6dab7 | 7331 | |
12b1c263 AJ |
7332 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7333 | continue; | |
7334 | ||
7335 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7336 | ||
7337 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7338 | set_buffer_dirty(bh); | |
7339 | sync_dirty_buffer(bh); | |
7340 | brelse(bh); | |
7341 | } | |
7342 | ||
7343 | /* Notify udev that device has changed */ | |
7344 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7345 | ||
7346 | /* Update ctime/mtime for device path for libblkid */ | |
7347 | update_dev_time(device_path); | |
a8a6dab7 | 7348 | } |
935e5cc9 MX |
7349 | |
7350 | /* | |
bbbf7243 NB |
7351 | * Update the size and bytes used for each device where it changed. This is |
7352 | * delayed since we would otherwise get errors while writing out the | |
7353 | * superblocks. | |
7354 | * | |
7355 | * Must be invoked during transaction commit. | |
935e5cc9 | 7356 | */ |
bbbf7243 | 7357 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7358 | { |
935e5cc9 MX |
7359 | struct btrfs_device *curr, *next; |
7360 | ||
bbbf7243 | 7361 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7362 | |
bbbf7243 | 7363 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7364 | return; |
7365 | ||
bbbf7243 NB |
7366 | /* |
7367 | * We don't need the device_list_mutex here. This list is owned by the | |
7368 | * transaction and the transaction must complete before the device is | |
7369 | * released. | |
7370 | */ | |
7371 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7372 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7373 | post_commit_list) { | |
7374 | list_del_init(&curr->post_commit_list); | |
7375 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7376 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7377 | } |
bbbf7243 | 7378 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7379 | } |
5a13f430 AJ |
7380 | |
7381 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7382 | { | |
7383 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7384 | while (fs_devices) { | |
7385 | fs_devices->fs_info = fs_info; | |
7386 | fs_devices = fs_devices->seed; | |
7387 | } | |
7388 | } | |
7389 | ||
7390 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7391 | { | |
7392 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7393 | while (fs_devices) { | |
7394 | fs_devices->fs_info = NULL; | |
7395 | fs_devices = fs_devices->seed; | |
7396 | } | |
7397 | } | |
46df06b8 DS |
7398 | |
7399 | /* | |
7400 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7401 | */ | |
7402 | int btrfs_bg_type_to_factor(u64 flags) | |
7403 | { | |
44b28ada DS |
7404 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7405 | ||
7406 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7407 | } |
cf90d884 QW |
7408 | |
7409 | ||
cf90d884 QW |
7410 | |
7411 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7412 | u64 chunk_offset, u64 devid, | |
7413 | u64 physical_offset, u64 physical_len) | |
7414 | { | |
c8bf1b67 | 7415 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7416 | struct extent_map *em; |
7417 | struct map_lookup *map; | |
05a37c48 | 7418 | struct btrfs_device *dev; |
cf90d884 QW |
7419 | u64 stripe_len; |
7420 | bool found = false; | |
7421 | int ret = 0; | |
7422 | int i; | |
7423 | ||
7424 | read_lock(&em_tree->lock); | |
7425 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7426 | read_unlock(&em_tree->lock); | |
7427 | ||
7428 | if (!em) { | |
7429 | btrfs_err(fs_info, | |
7430 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7431 | physical_offset, devid); | |
7432 | ret = -EUCLEAN; | |
7433 | goto out; | |
7434 | } | |
7435 | ||
7436 | map = em->map_lookup; | |
7437 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7438 | if (physical_len != stripe_len) { | |
7439 | btrfs_err(fs_info, | |
7440 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7441 | physical_offset, devid, em->start, physical_len, | |
7442 | stripe_len); | |
7443 | ret = -EUCLEAN; | |
7444 | goto out; | |
7445 | } | |
7446 | ||
7447 | for (i = 0; i < map->num_stripes; i++) { | |
7448 | if (map->stripes[i].dev->devid == devid && | |
7449 | map->stripes[i].physical == physical_offset) { | |
7450 | found = true; | |
7451 | if (map->verified_stripes >= map->num_stripes) { | |
7452 | btrfs_err(fs_info, | |
7453 | "too many dev extents for chunk %llu found", | |
7454 | em->start); | |
7455 | ret = -EUCLEAN; | |
7456 | goto out; | |
7457 | } | |
7458 | map->verified_stripes++; | |
7459 | break; | |
7460 | } | |
7461 | } | |
7462 | if (!found) { | |
7463 | btrfs_err(fs_info, | |
7464 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7465 | physical_offset, devid); | |
7466 | ret = -EUCLEAN; | |
7467 | } | |
05a37c48 QW |
7468 | |
7469 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7470 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7471 | if (!dev) { |
7472 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7473 | ret = -EUCLEAN; | |
7474 | goto out; | |
7475 | } | |
1b3922a8 QW |
7476 | |
7477 | /* It's possible this device is a dummy for seed device */ | |
7478 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7479 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7480 | NULL, false); | |
1b3922a8 QW |
7481 | if (!dev) { |
7482 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7483 | devid); | |
7484 | ret = -EUCLEAN; | |
7485 | goto out; | |
7486 | } | |
7487 | } | |
7488 | ||
05a37c48 QW |
7489 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7490 | btrfs_err(fs_info, | |
7491 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7492 | devid, physical_offset, physical_len, | |
7493 | dev->disk_total_bytes); | |
7494 | ret = -EUCLEAN; | |
7495 | goto out; | |
7496 | } | |
cf90d884 QW |
7497 | out: |
7498 | free_extent_map(em); | |
7499 | return ret; | |
7500 | } | |
7501 | ||
7502 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7503 | { | |
c8bf1b67 | 7504 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7505 | struct extent_map *em; |
7506 | struct rb_node *node; | |
7507 | int ret = 0; | |
7508 | ||
7509 | read_lock(&em_tree->lock); | |
07e1ce09 | 7510 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7511 | em = rb_entry(node, struct extent_map, rb_node); |
7512 | if (em->map_lookup->num_stripes != | |
7513 | em->map_lookup->verified_stripes) { | |
7514 | btrfs_err(fs_info, | |
7515 | "chunk %llu has missing dev extent, have %d expect %d", | |
7516 | em->start, em->map_lookup->verified_stripes, | |
7517 | em->map_lookup->num_stripes); | |
7518 | ret = -EUCLEAN; | |
7519 | goto out; | |
7520 | } | |
7521 | } | |
7522 | out: | |
7523 | read_unlock(&em_tree->lock); | |
7524 | return ret; | |
7525 | } | |
7526 | ||
7527 | /* | |
7528 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7529 | * later chunk allocation/free would cause unexpected behavior. | |
7530 | * | |
7531 | * NOTE: This will iterate through the whole device tree, which should be of | |
7532 | * the same size level as the chunk tree. This slightly increases mount time. | |
7533 | */ | |
7534 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7535 | { | |
7536 | struct btrfs_path *path; | |
7537 | struct btrfs_root *root = fs_info->dev_root; | |
7538 | struct btrfs_key key; | |
5eb19381 QW |
7539 | u64 prev_devid = 0; |
7540 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7541 | int ret = 0; |
7542 | ||
7543 | key.objectid = 1; | |
7544 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7545 | key.offset = 0; | |
7546 | ||
7547 | path = btrfs_alloc_path(); | |
7548 | if (!path) | |
7549 | return -ENOMEM; | |
7550 | ||
7551 | path->reada = READA_FORWARD; | |
7552 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7553 | if (ret < 0) | |
7554 | goto out; | |
7555 | ||
7556 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7557 | ret = btrfs_next_item(root, path); | |
7558 | if (ret < 0) | |
7559 | goto out; | |
7560 | /* No dev extents at all? Not good */ | |
7561 | if (ret > 0) { | |
7562 | ret = -EUCLEAN; | |
7563 | goto out; | |
7564 | } | |
7565 | } | |
7566 | while (1) { | |
7567 | struct extent_buffer *leaf = path->nodes[0]; | |
7568 | struct btrfs_dev_extent *dext; | |
7569 | int slot = path->slots[0]; | |
7570 | u64 chunk_offset; | |
7571 | u64 physical_offset; | |
7572 | u64 physical_len; | |
7573 | u64 devid; | |
7574 | ||
7575 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7576 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7577 | break; | |
7578 | devid = key.objectid; | |
7579 | physical_offset = key.offset; | |
7580 | ||
7581 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7582 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7583 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7584 | ||
5eb19381 QW |
7585 | /* Check if this dev extent overlaps with the previous one */ |
7586 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7587 | btrfs_err(fs_info, | |
7588 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7589 | devid, physical_offset, prev_dev_ext_end); | |
7590 | ret = -EUCLEAN; | |
7591 | goto out; | |
7592 | } | |
7593 | ||
cf90d884 QW |
7594 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7595 | physical_offset, physical_len); | |
7596 | if (ret < 0) | |
7597 | goto out; | |
5eb19381 QW |
7598 | prev_devid = devid; |
7599 | prev_dev_ext_end = physical_offset + physical_len; | |
7600 | ||
cf90d884 QW |
7601 | ret = btrfs_next_item(root, path); |
7602 | if (ret < 0) | |
7603 | goto out; | |
7604 | if (ret > 0) { | |
7605 | ret = 0; | |
7606 | break; | |
7607 | } | |
7608 | } | |
7609 | ||
7610 | /* Ensure all chunks have corresponding dev extents */ | |
7611 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7612 | out: | |
7613 | btrfs_free_path(path); | |
7614 | return ret; | |
7615 | } | |
eede2bf3 OS |
7616 | |
7617 | /* | |
7618 | * Check whether the given block group or device is pinned by any inode being | |
7619 | * used as a swapfile. | |
7620 | */ | |
7621 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7622 | { | |
7623 | struct btrfs_swapfile_pin *sp; | |
7624 | struct rb_node *node; | |
7625 | ||
7626 | spin_lock(&fs_info->swapfile_pins_lock); | |
7627 | node = fs_info->swapfile_pins.rb_node; | |
7628 | while (node) { | |
7629 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7630 | if (ptr < sp->ptr) | |
7631 | node = node->rb_left; | |
7632 | else if (ptr > sp->ptr) | |
7633 | node = node->rb_right; | |
7634 | else | |
7635 | break; | |
7636 | } | |
7637 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7638 | return node != NULL; | |
7639 | } |