]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 | 6 | #include <linux/sched.h> |
fccc0007 | 7 | #include <linux/sched/mm.h> |
0b86a832 | 8 | #include <linux/bio.h> |
5a0e3ad6 | 9 | #include <linux/slab.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
784352fe | 17 | #include "misc.h" |
0b86a832 CM |
18 | #include "ctree.h" |
19 | #include "extent_map.h" | |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "print-tree.h" | |
23 | #include "volumes.h" | |
53b381b3 | 24 | #include "raid56.h" |
8b712842 | 25 | #include "async-thread.h" |
21adbd5c | 26 | #include "check-integrity.h" |
606686ee | 27 | #include "rcu-string.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
b0643e59 | 33 | #include "discard.h" |
0b86a832 | 34 | |
af902047 ZL |
35 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
36 | [BTRFS_RAID_RAID10] = { | |
37 | .sub_stripes = 2, | |
38 | .dev_stripes = 1, | |
39 | .devs_max = 0, /* 0 == as many as possible */ | |
40 | .devs_min = 4, | |
8789f4fe | 41 | .tolerated_failures = 1, |
af902047 ZL |
42 | .devs_increment = 2, |
43 | .ncopies = 2, | |
b50836ed | 44 | .nparity = 0, |
ed23467b | 45 | .raid_name = "raid10", |
41a6e891 | 46 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 47 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
48 | }, |
49 | [BTRFS_RAID_RAID1] = { | |
50 | .sub_stripes = 1, | |
51 | .dev_stripes = 1, | |
52 | .devs_max = 2, | |
53 | .devs_min = 2, | |
8789f4fe | 54 | .tolerated_failures = 1, |
af902047 ZL |
55 | .devs_increment = 2, |
56 | .ncopies = 2, | |
b50836ed | 57 | .nparity = 0, |
ed23467b | 58 | .raid_name = "raid1", |
41a6e891 | 59 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 60 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 61 | }, |
47e6f742 DS |
62 | [BTRFS_RAID_RAID1C3] = { |
63 | .sub_stripes = 1, | |
64 | .dev_stripes = 1, | |
cf93e15e | 65 | .devs_max = 3, |
47e6f742 DS |
66 | .devs_min = 3, |
67 | .tolerated_failures = 2, | |
68 | .devs_increment = 3, | |
69 | .ncopies = 3, | |
db26a024 | 70 | .nparity = 0, |
47e6f742 DS |
71 | .raid_name = "raid1c3", |
72 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
73 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
74 | }, | |
8d6fac00 DS |
75 | [BTRFS_RAID_RAID1C4] = { |
76 | .sub_stripes = 1, | |
77 | .dev_stripes = 1, | |
cf93e15e | 78 | .devs_max = 4, |
8d6fac00 DS |
79 | .devs_min = 4, |
80 | .tolerated_failures = 3, | |
81 | .devs_increment = 4, | |
82 | .ncopies = 4, | |
db26a024 | 83 | .nparity = 0, |
8d6fac00 DS |
84 | .raid_name = "raid1c4", |
85 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
86 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
87 | }, | |
af902047 ZL |
88 | [BTRFS_RAID_DUP] = { |
89 | .sub_stripes = 1, | |
90 | .dev_stripes = 2, | |
91 | .devs_max = 1, | |
92 | .devs_min = 1, | |
8789f4fe | 93 | .tolerated_failures = 0, |
af902047 ZL |
94 | .devs_increment = 1, |
95 | .ncopies = 2, | |
b50836ed | 96 | .nparity = 0, |
ed23467b | 97 | .raid_name = "dup", |
41a6e891 | 98 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 99 | .mindev_error = 0, |
af902047 ZL |
100 | }, |
101 | [BTRFS_RAID_RAID0] = { | |
102 | .sub_stripes = 1, | |
103 | .dev_stripes = 1, | |
104 | .devs_max = 0, | |
105 | .devs_min = 2, | |
8789f4fe | 106 | .tolerated_failures = 0, |
af902047 ZL |
107 | .devs_increment = 1, |
108 | .ncopies = 1, | |
b50836ed | 109 | .nparity = 0, |
ed23467b | 110 | .raid_name = "raid0", |
41a6e891 | 111 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 112 | .mindev_error = 0, |
af902047 ZL |
113 | }, |
114 | [BTRFS_RAID_SINGLE] = { | |
115 | .sub_stripes = 1, | |
116 | .dev_stripes = 1, | |
117 | .devs_max = 1, | |
118 | .devs_min = 1, | |
8789f4fe | 119 | .tolerated_failures = 0, |
af902047 ZL |
120 | .devs_increment = 1, |
121 | .ncopies = 1, | |
b50836ed | 122 | .nparity = 0, |
ed23467b | 123 | .raid_name = "single", |
41a6e891 | 124 | .bg_flag = 0, |
f9fbcaa2 | 125 | .mindev_error = 0, |
af902047 ZL |
126 | }, |
127 | [BTRFS_RAID_RAID5] = { | |
128 | .sub_stripes = 1, | |
129 | .dev_stripes = 1, | |
130 | .devs_max = 0, | |
131 | .devs_min = 2, | |
8789f4fe | 132 | .tolerated_failures = 1, |
af902047 | 133 | .devs_increment = 1, |
da612e31 | 134 | .ncopies = 1, |
b50836ed | 135 | .nparity = 1, |
ed23467b | 136 | .raid_name = "raid5", |
41a6e891 | 137 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 138 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
139 | }, |
140 | [BTRFS_RAID_RAID6] = { | |
141 | .sub_stripes = 1, | |
142 | .dev_stripes = 1, | |
143 | .devs_max = 0, | |
144 | .devs_min = 3, | |
8789f4fe | 145 | .tolerated_failures = 2, |
af902047 | 146 | .devs_increment = 1, |
da612e31 | 147 | .ncopies = 1, |
b50836ed | 148 | .nparity = 2, |
ed23467b | 149 | .raid_name = "raid6", |
41a6e891 | 150 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 151 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
152 | }, |
153 | }; | |
154 | ||
158da513 | 155 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 156 | { |
158da513 DS |
157 | const int index = btrfs_bg_flags_to_raid_index(flags); |
158 | ||
159 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
160 | return NULL; |
161 | ||
158da513 | 162 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
163 | } |
164 | ||
f89e09cf AJ |
165 | /* |
166 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
167 | * bytes including terminating null byte. | |
168 | */ | |
169 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
170 | { | |
171 | int i; | |
172 | int ret; | |
173 | char *bp = buf; | |
174 | u64 flags = bg_flags; | |
175 | u32 size_bp = size_buf; | |
176 | ||
177 | if (!flags) { | |
178 | strcpy(bp, "NONE"); | |
179 | return; | |
180 | } | |
181 | ||
182 | #define DESCRIBE_FLAG(flag, desc) \ | |
183 | do { \ | |
184 | if (flags & (flag)) { \ | |
185 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
186 | if (ret < 0 || ret >= size_bp) \ | |
187 | goto out_overflow; \ | |
188 | size_bp -= ret; \ | |
189 | bp += ret; \ | |
190 | flags &= ~(flag); \ | |
191 | } \ | |
192 | } while (0) | |
193 | ||
194 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
195 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
196 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
197 | ||
198 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
199 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
200 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
201 | btrfs_raid_array[i].raid_name); | |
202 | #undef DESCRIBE_FLAG | |
203 | ||
204 | if (flags) { | |
205 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
206 | size_bp -= ret; | |
207 | } | |
208 | ||
209 | if (size_bp < size_buf) | |
210 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
211 | ||
212 | /* | |
213 | * The text is trimmed, it's up to the caller to provide sufficiently | |
214 | * large buffer | |
215 | */ | |
216 | out_overflow:; | |
217 | } | |
218 | ||
6f8e0fc7 | 219 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 220 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 221 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 222 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
223 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
224 | enum btrfs_map_op op, | |
225 | u64 logical, u64 *length, | |
226 | struct btrfs_bio **bbio_ret, | |
227 | int mirror_num, int need_raid_map); | |
2b82032c | 228 | |
9c6b1c4d DS |
229 | /* |
230 | * Device locking | |
231 | * ============== | |
232 | * | |
233 | * There are several mutexes that protect manipulation of devices and low-level | |
234 | * structures like chunks but not block groups, extents or files | |
235 | * | |
236 | * uuid_mutex (global lock) | |
237 | * ------------------------ | |
238 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
239 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
240 | * device) or requested by the device= mount option | |
241 | * | |
242 | * the mutex can be very coarse and can cover long-running operations | |
243 | * | |
244 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 245 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
246 | * |
247 | * global::fs_devs - add, remove, updates to the global list | |
248 | * | |
18c850fd JB |
249 | * does not protect: manipulation of the fs_devices::devices list in general |
250 | * but in mount context it could be used to exclude list modifications by eg. | |
251 | * scan ioctl | |
9c6b1c4d DS |
252 | * |
253 | * btrfs_device::name - renames (write side), read is RCU | |
254 | * | |
255 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
256 | * ------------------------------------------------ | |
257 | * protects updates to fs_devices::devices, ie. adding and deleting | |
258 | * | |
259 | * simple list traversal with read-only actions can be done with RCU protection | |
260 | * | |
261 | * may be used to exclude some operations from running concurrently without any | |
262 | * modifications to the list (see write_all_supers) | |
263 | * | |
18c850fd JB |
264 | * Is not required at mount and close times, because our device list is |
265 | * protected by the uuid_mutex at that point. | |
266 | * | |
9c6b1c4d DS |
267 | * balance_mutex |
268 | * ------------- | |
269 | * protects balance structures (status, state) and context accessed from | |
270 | * several places (internally, ioctl) | |
271 | * | |
272 | * chunk_mutex | |
273 | * ----------- | |
274 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
275 | * device is added/removed. Additionally it also protects post_commit_list of |
276 | * individual devices, since they can be added to the transaction's | |
277 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
278 | * |
279 | * cleaner_mutex | |
280 | * ------------- | |
281 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
282 | * cleaning together with relocation or delayed iputs | |
283 | * | |
284 | * | |
285 | * Lock nesting | |
286 | * ============ | |
287 | * | |
288 | * uuid_mutex | |
ae3e715f AJ |
289 | * device_list_mutex |
290 | * chunk_mutex | |
291 | * balance_mutex | |
89595e80 AJ |
292 | * |
293 | * | |
294 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
295 | * ====================================== | |
296 | * | |
297 | * Maintains the exclusivity of the following operations that apply to the | |
298 | * whole filesystem and cannot run in parallel. | |
299 | * | |
300 | * - Balance (*) | |
301 | * - Device add | |
302 | * - Device remove | |
303 | * - Device replace (*) | |
304 | * - Resize | |
305 | * | |
306 | * The device operations (as above) can be in one of the following states: | |
307 | * | |
308 | * - Running state | |
309 | * - Paused state | |
310 | * - Completed state | |
311 | * | |
312 | * Only device operations marked with (*) can go into the Paused state for the | |
313 | * following reasons: | |
314 | * | |
315 | * - ioctl (only Balance can be Paused through ioctl) | |
316 | * - filesystem remounted as read-only | |
317 | * - filesystem unmounted and mounted as read-only | |
318 | * - system power-cycle and filesystem mounted as read-only | |
319 | * - filesystem or device errors leading to forced read-only | |
320 | * | |
321 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
322 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
323 | * A device operation in Paused or Running state can be canceled or resumed | |
324 | * either by ioctl (Balance only) or when remounted as read-write. | |
325 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
326 | * completed. | |
9c6b1c4d DS |
327 | */ |
328 | ||
67a2c45e | 329 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 330 | static LIST_HEAD(fs_uuids); |
4143cb8b | 331 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
332 | { |
333 | return &fs_uuids; | |
334 | } | |
8a4b83cc | 335 | |
2dfeca9b DS |
336 | /* |
337 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
338 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
339 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
340 | * |
341 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
342 | * The returned struct is not linked onto any lists and can be destroyed with | |
343 | * kfree() right away. | |
344 | */ | |
7239ff4b NB |
345 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
346 | const u8 *metadata_fsid) | |
2208a378 ID |
347 | { |
348 | struct btrfs_fs_devices *fs_devs; | |
349 | ||
78f2c9e6 | 350 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
351 | if (!fs_devs) |
352 | return ERR_PTR(-ENOMEM); | |
353 | ||
354 | mutex_init(&fs_devs->device_list_mutex); | |
355 | ||
356 | INIT_LIST_HEAD(&fs_devs->devices); | |
357 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 358 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
359 | if (fsid) |
360 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 361 | |
7239ff4b NB |
362 | if (metadata_fsid) |
363 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
364 | else if (fsid) | |
365 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
366 | ||
2208a378 ID |
367 | return fs_devs; |
368 | } | |
369 | ||
a425f9d4 | 370 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 371 | { |
bbbf7243 | 372 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 373 | rcu_string_free(device->name); |
1c11b63e | 374 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
375 | bio_put(device->flush_bio); |
376 | kfree(device); | |
377 | } | |
378 | ||
e4404d6e YZ |
379 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
380 | { | |
381 | struct btrfs_device *device; | |
382 | WARN_ON(fs_devices->opened); | |
383 | while (!list_empty(&fs_devices->devices)) { | |
384 | device = list_entry(fs_devices->devices.next, | |
385 | struct btrfs_device, dev_list); | |
386 | list_del(&device->dev_list); | |
a425f9d4 | 387 | btrfs_free_device(device); |
e4404d6e YZ |
388 | } |
389 | kfree(fs_devices); | |
390 | } | |
391 | ||
ffc5a379 | 392 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
393 | { |
394 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 395 | |
2b82032c YZ |
396 | while (!list_empty(&fs_uuids)) { |
397 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
398 | struct btrfs_fs_devices, fs_list); |
399 | list_del(&fs_devices->fs_list); | |
e4404d6e | 400 | free_fs_devices(fs_devices); |
8a4b83cc | 401 | } |
8a4b83cc CM |
402 | } |
403 | ||
48dae9cf DS |
404 | /* |
405 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
406 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 407 | * btrfs_free_device. |
48dae9cf | 408 | */ |
12bd2fc0 ID |
409 | static struct btrfs_device *__alloc_device(void) |
410 | { | |
411 | struct btrfs_device *dev; | |
412 | ||
78f2c9e6 | 413 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
414 | if (!dev) |
415 | return ERR_PTR(-ENOMEM); | |
416 | ||
e0ae9994 DS |
417 | /* |
418 | * Preallocate a bio that's always going to be used for flushing device | |
419 | * barriers and matches the device lifespan | |
420 | */ | |
421 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
422 | if (!dev->flush_bio) { | |
423 | kfree(dev); | |
424 | return ERR_PTR(-ENOMEM); | |
425 | } | |
e0ae9994 | 426 | |
12bd2fc0 ID |
427 | INIT_LIST_HEAD(&dev->dev_list); |
428 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 429 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 | 430 | |
12bd2fc0 | 431 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 432 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 433 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 434 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 435 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
1c11b63e | 436 | extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL); |
12bd2fc0 ID |
437 | |
438 | return dev; | |
439 | } | |
440 | ||
7239ff4b NB |
441 | static noinline struct btrfs_fs_devices *find_fsid( |
442 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 443 | { |
8a4b83cc CM |
444 | struct btrfs_fs_devices *fs_devices; |
445 | ||
7239ff4b NB |
446 | ASSERT(fsid); |
447 | ||
7a62d0f0 | 448 | /* Handle non-split brain cases */ |
c4babc5e | 449 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
450 | if (metadata_fsid) { |
451 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
452 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
453 | BTRFS_FSID_SIZE) == 0) | |
454 | return fs_devices; | |
455 | } else { | |
456 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
457 | return fs_devices; | |
458 | } | |
8a4b83cc CM |
459 | } |
460 | return NULL; | |
461 | } | |
462 | ||
c6730a0e SY |
463 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
464 | struct btrfs_super_block *disk_super) | |
465 | { | |
466 | ||
467 | struct btrfs_fs_devices *fs_devices; | |
468 | ||
469 | /* | |
470 | * Handle scanned device having completed its fsid change but | |
471 | * belonging to a fs_devices that was created by first scanning | |
472 | * a device which didn't have its fsid/metadata_uuid changed | |
473 | * at all and the CHANGING_FSID_V2 flag set. | |
474 | */ | |
475 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
476 | if (fs_devices->fsid_change && | |
477 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
478 | BTRFS_FSID_SIZE) == 0 && | |
479 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
480 | BTRFS_FSID_SIZE) == 0) { | |
481 | return fs_devices; | |
482 | } | |
483 | } | |
484 | /* | |
485 | * Handle scanned device having completed its fsid change but | |
486 | * belonging to a fs_devices that was created by a device that | |
487 | * has an outdated pair of fsid/metadata_uuid and | |
488 | * CHANGING_FSID_V2 flag set. | |
489 | */ | |
490 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
491 | if (fs_devices->fsid_change && | |
492 | memcmp(fs_devices->metadata_uuid, | |
493 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
494 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
495 | BTRFS_FSID_SIZE) == 0) { | |
496 | return fs_devices; | |
497 | } | |
498 | } | |
499 | ||
500 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
501 | } | |
502 | ||
503 | ||
beaf8ab3 SB |
504 | static int |
505 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
506 | int flush, struct block_device **bdev, | |
8f32380d | 507 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
508 | { |
509 | int ret; | |
510 | ||
511 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
512 | ||
513 | if (IS_ERR(*bdev)) { | |
514 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
515 | goto error; |
516 | } | |
517 | ||
518 | if (flush) | |
519 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 520 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
521 | if (ret) { |
522 | blkdev_put(*bdev, flags); | |
523 | goto error; | |
524 | } | |
525 | invalidate_bdev(*bdev); | |
8f32380d JT |
526 | *disk_super = btrfs_read_dev_super(*bdev); |
527 | if (IS_ERR(*disk_super)) { | |
528 | ret = PTR_ERR(*disk_super); | |
beaf8ab3 SB |
529 | blkdev_put(*bdev, flags); |
530 | goto error; | |
531 | } | |
532 | ||
533 | return 0; | |
534 | ||
535 | error: | |
536 | *bdev = NULL; | |
beaf8ab3 SB |
537 | return ret; |
538 | } | |
539 | ||
70bc7088 AJ |
540 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
541 | { | |
542 | int found; | |
543 | ||
544 | rcu_read_lock(); | |
545 | found = strcmp(rcu_str_deref(device->name), path); | |
546 | rcu_read_unlock(); | |
547 | ||
548 | return found == 0; | |
549 | } | |
550 | ||
d8367db3 AJ |
551 | /* |
552 | * Search and remove all stale (devices which are not mounted) devices. | |
553 | * When both inputs are NULL, it will search and release all stale devices. | |
554 | * path: Optional. When provided will it release all unmounted devices | |
555 | * matching this path only. | |
556 | * skip_dev: Optional. Will skip this device when searching for the stale | |
557 | * devices. | |
70bc7088 AJ |
558 | * Return: 0 for success or if @path is NULL. |
559 | * -EBUSY if @path is a mounted device. | |
560 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 561 | */ |
70bc7088 | 562 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 563 | struct btrfs_device *skip_device) |
4fde46f0 | 564 | { |
fa6d2ae5 AJ |
565 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
566 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
567 | int ret = 0; |
568 | ||
569 | if (path) | |
570 | ret = -ENOENT; | |
4fde46f0 | 571 | |
fa6d2ae5 | 572 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 573 | |
70bc7088 | 574 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
575 | list_for_each_entry_safe(device, tmp_device, |
576 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 577 | if (skip_device && skip_device == device) |
d8367db3 | 578 | continue; |
fa6d2ae5 | 579 | if (path && !device->name) |
4fde46f0 | 580 | continue; |
70bc7088 | 581 | if (path && !device_path_matched(path, device)) |
38cf665d | 582 | continue; |
70bc7088 AJ |
583 | if (fs_devices->opened) { |
584 | /* for an already deleted device return 0 */ | |
585 | if (path && ret != 0) | |
586 | ret = -EBUSY; | |
587 | break; | |
588 | } | |
4fde46f0 | 589 | |
4fde46f0 | 590 | /* delete the stale device */ |
7bcb8164 AJ |
591 | fs_devices->num_devices--; |
592 | list_del(&device->dev_list); | |
593 | btrfs_free_device(device); | |
594 | ||
70bc7088 | 595 | ret = 0; |
7bcb8164 | 596 | if (fs_devices->num_devices == 0) |
fd649f10 | 597 | break; |
7bcb8164 AJ |
598 | } |
599 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 600 | |
7bcb8164 AJ |
601 | if (fs_devices->num_devices == 0) { |
602 | btrfs_sysfs_remove_fsid(fs_devices); | |
603 | list_del(&fs_devices->fs_list); | |
604 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
605 | } |
606 | } | |
70bc7088 AJ |
607 | |
608 | return ret; | |
4fde46f0 AJ |
609 | } |
610 | ||
18c850fd JB |
611 | /* |
612 | * This is only used on mount, and we are protected from competing things | |
613 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
614 | * fs_devices->device_list_mutex here. | |
615 | */ | |
0fb08bcc AJ |
616 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
617 | struct btrfs_device *device, fmode_t flags, | |
618 | void *holder) | |
619 | { | |
620 | struct request_queue *q; | |
621 | struct block_device *bdev; | |
0fb08bcc AJ |
622 | struct btrfs_super_block *disk_super; |
623 | u64 devid; | |
624 | int ret; | |
625 | ||
626 | if (device->bdev) | |
627 | return -EINVAL; | |
628 | if (!device->name) | |
629 | return -EINVAL; | |
630 | ||
631 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 632 | &bdev, &disk_super); |
0fb08bcc AJ |
633 | if (ret) |
634 | return ret; | |
635 | ||
0fb08bcc AJ |
636 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
637 | if (devid != device->devid) | |
8f32380d | 638 | goto error_free_page; |
0fb08bcc AJ |
639 | |
640 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 641 | goto error_free_page; |
0fb08bcc AJ |
642 | |
643 | device->generation = btrfs_super_generation(disk_super); | |
644 | ||
645 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
646 | if (btrfs_super_incompat_flags(disk_super) & |
647 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
648 | pr_err( | |
649 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 650 | goto error_free_page; |
7239ff4b NB |
651 | } |
652 | ||
ebbede42 | 653 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 654 | fs_devices->seeding = true; |
0fb08bcc | 655 | } else { |
ebbede42 AJ |
656 | if (bdev_read_only(bdev)) |
657 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
658 | else | |
659 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
660 | } |
661 | ||
662 | q = bdev_get_queue(bdev); | |
0fb08bcc | 663 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 664 | fs_devices->rotating = true; |
0fb08bcc AJ |
665 | |
666 | device->bdev = bdev; | |
e12c9621 | 667 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
668 | device->mode = flags; |
669 | ||
670 | fs_devices->open_devices++; | |
ebbede42 AJ |
671 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
672 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 673 | fs_devices->rw_devices++; |
b1b8e386 | 674 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 675 | } |
8f32380d | 676 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
677 | |
678 | return 0; | |
679 | ||
8f32380d JT |
680 | error_free_page: |
681 | btrfs_release_disk_super(disk_super); | |
0fb08bcc AJ |
682 | blkdev_put(bdev, flags); |
683 | ||
684 | return -EINVAL; | |
685 | } | |
686 | ||
7a62d0f0 NB |
687 | /* |
688 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
689 | * being created with a disk that has already completed its fsid change. Such |
690 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
691 | * Handle both cases here. | |
7a62d0f0 NB |
692 | */ |
693 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
694 | struct btrfs_super_block *disk_super) | |
695 | { | |
696 | struct btrfs_fs_devices *fs_devices; | |
697 | ||
698 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
699 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
700 | BTRFS_FSID_SIZE) != 0 && | |
701 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
702 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
703 | return fs_devices; | |
704 | } | |
705 | } | |
706 | ||
c0d81c7c | 707 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
708 | } |
709 | ||
cc5de4e7 NB |
710 | |
711 | static struct btrfs_fs_devices *find_fsid_changed( | |
712 | struct btrfs_super_block *disk_super) | |
713 | { | |
714 | struct btrfs_fs_devices *fs_devices; | |
715 | ||
716 | /* | |
717 | * Handles the case where scanned device is part of an fs that had | |
718 | * multiple successful changes of FSID but curently device didn't | |
05840710 NB |
719 | * observe it. Meaning our fsid will be different than theirs. We need |
720 | * to handle two subcases : | |
721 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
722 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
723 | * are equal). | |
cc5de4e7 NB |
724 | */ |
725 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 726 | /* Changed UUIDs */ |
cc5de4e7 NB |
727 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
728 | BTRFS_FSID_SIZE) != 0 && | |
729 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
730 | BTRFS_FSID_SIZE) == 0 && | |
731 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
732 | BTRFS_FSID_SIZE) != 0) |
733 | return fs_devices; | |
734 | ||
735 | /* Unchanged UUIDs */ | |
736 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
737 | BTRFS_FSID_SIZE) == 0 && | |
738 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
739 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 740 | return fs_devices; |
cc5de4e7 NB |
741 | } |
742 | ||
743 | return NULL; | |
744 | } | |
1362089d NB |
745 | |
746 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
747 | struct btrfs_super_block *disk_super) | |
748 | { | |
749 | struct btrfs_fs_devices *fs_devices; | |
750 | ||
751 | /* | |
752 | * Handle the case where the scanned device is part of an fs whose last | |
753 | * metadata UUID change reverted it to the original FSID. At the same | |
754 | * time * fs_devices was first created by another constitutent device | |
755 | * which didn't fully observe the operation. This results in an | |
756 | * btrfs_fs_devices created with metadata/fsid different AND | |
757 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
758 | * fs_devices equal to the FSID of the disk. | |
759 | */ | |
760 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
761 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
762 | BTRFS_FSID_SIZE) != 0 && | |
763 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
764 | BTRFS_FSID_SIZE) == 0 && | |
765 | fs_devices->fsid_change) | |
766 | return fs_devices; | |
767 | } | |
768 | ||
769 | return NULL; | |
770 | } | |
60999ca4 DS |
771 | /* |
772 | * Add new device to list of registered devices | |
773 | * | |
774 | * Returns: | |
e124ece5 AJ |
775 | * device pointer which was just added or updated when successful |
776 | * error pointer when failed | |
60999ca4 | 777 | */ |
e124ece5 | 778 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
779 | struct btrfs_super_block *disk_super, |
780 | bool *new_device_added) | |
8a4b83cc CM |
781 | { |
782 | struct btrfs_device *device; | |
7a62d0f0 | 783 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 784 | struct rcu_string *name; |
8a4b83cc | 785 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 786 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
787 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
788 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
789 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
790 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 791 | |
cc5de4e7 | 792 | if (fsid_change_in_progress) { |
c0d81c7c | 793 | if (!has_metadata_uuid) |
cc5de4e7 | 794 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 795 | else |
cc5de4e7 | 796 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 797 | } else if (has_metadata_uuid) { |
c6730a0e | 798 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 799 | } else { |
1362089d NB |
800 | fs_devices = find_fsid_reverted_metadata(disk_super); |
801 | if (!fs_devices) | |
802 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
803 | } |
804 | ||
8a4b83cc | 805 | |
8a4b83cc | 806 | if (!fs_devices) { |
7239ff4b NB |
807 | if (has_metadata_uuid) |
808 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
809 | disk_super->metadata_uuid); | |
810 | else | |
811 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
812 | ||
2208a378 | 813 | if (IS_ERR(fs_devices)) |
e124ece5 | 814 | return ERR_CAST(fs_devices); |
2208a378 | 815 | |
92900e51 AV |
816 | fs_devices->fsid_change = fsid_change_in_progress; |
817 | ||
9c6d173e | 818 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 819 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 820 | |
8a4b83cc CM |
821 | device = NULL; |
822 | } else { | |
9c6d173e | 823 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
824 | device = btrfs_find_device(fs_devices, devid, |
825 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
826 | |
827 | /* | |
828 | * If this disk has been pulled into an fs devices created by | |
829 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
830 | * metadata_uuid/fsid values of the fs_devices. | |
831 | */ | |
1362089d | 832 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
833 | found_transid > fs_devices->latest_generation) { |
834 | memcpy(fs_devices->fsid, disk_super->fsid, | |
835 | BTRFS_FSID_SIZE); | |
1362089d NB |
836 | |
837 | if (has_metadata_uuid) | |
838 | memcpy(fs_devices->metadata_uuid, | |
839 | disk_super->metadata_uuid, | |
840 | BTRFS_FSID_SIZE); | |
841 | else | |
842 | memcpy(fs_devices->metadata_uuid, | |
843 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
844 | |
845 | fs_devices->fsid_change = false; | |
846 | } | |
8a4b83cc | 847 | } |
443f24fe | 848 | |
8a4b83cc | 849 | if (!device) { |
9c6d173e AJ |
850 | if (fs_devices->opened) { |
851 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 852 | return ERR_PTR(-EBUSY); |
9c6d173e | 853 | } |
2b82032c | 854 | |
12bd2fc0 ID |
855 | device = btrfs_alloc_device(NULL, &devid, |
856 | disk_super->dev_item.uuid); | |
857 | if (IS_ERR(device)) { | |
9c6d173e | 858 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 859 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 860 | return device; |
8a4b83cc | 861 | } |
606686ee JB |
862 | |
863 | name = rcu_string_strdup(path, GFP_NOFS); | |
864 | if (!name) { | |
a425f9d4 | 865 | btrfs_free_device(device); |
9c6d173e | 866 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 867 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 868 | } |
606686ee | 869 | rcu_assign_pointer(device->name, name); |
90519d66 | 870 | |
1f78160c | 871 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 872 | fs_devices->num_devices++; |
e5e9a520 | 873 | |
2b82032c | 874 | device->fs_devices = fs_devices; |
4306a974 | 875 | *new_device_added = true; |
327f18cc AJ |
876 | |
877 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
878 | pr_info( |
879 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
880 | disk_super->label, devid, found_transid, path, | |
881 | current->comm, task_pid_nr(current)); | |
327f18cc | 882 | else |
aa6c0df7 AJ |
883 | pr_info( |
884 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
885 | disk_super->fsid, devid, found_transid, path, | |
886 | current->comm, task_pid_nr(current)); | |
327f18cc | 887 | |
606686ee | 888 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
889 | /* |
890 | * When FS is already mounted. | |
891 | * 1. If you are here and if the device->name is NULL that | |
892 | * means this device was missing at time of FS mount. | |
893 | * 2. If you are here and if the device->name is different | |
894 | * from 'path' that means either | |
895 | * a. The same device disappeared and reappeared with | |
896 | * different name. or | |
897 | * b. The missing-disk-which-was-replaced, has | |
898 | * reappeared now. | |
899 | * | |
900 | * We must allow 1 and 2a above. But 2b would be a spurious | |
901 | * and unintentional. | |
902 | * | |
903 | * Further in case of 1 and 2a above, the disk at 'path' | |
904 | * would have missed some transaction when it was away and | |
905 | * in case of 2a the stale bdev has to be updated as well. | |
906 | * 2b must not be allowed at all time. | |
907 | */ | |
908 | ||
909 | /* | |
0f23ae74 CM |
910 | * For now, we do allow update to btrfs_fs_device through the |
911 | * btrfs dev scan cli after FS has been mounted. We're still | |
912 | * tracking a problem where systems fail mount by subvolume id | |
913 | * when we reject replacement on a mounted FS. | |
b96de000 | 914 | */ |
0f23ae74 | 915 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
916 | /* |
917 | * That is if the FS is _not_ mounted and if you | |
918 | * are here, that means there is more than one | |
919 | * disk with same uuid and devid.We keep the one | |
920 | * with larger generation number or the last-in if | |
921 | * generation are equal. | |
922 | */ | |
9c6d173e | 923 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 924 | return ERR_PTR(-EEXIST); |
77bdae4d | 925 | } |
b96de000 | 926 | |
a9261d41 AJ |
927 | /* |
928 | * We are going to replace the device path for a given devid, | |
929 | * make sure it's the same device if the device is mounted | |
930 | */ | |
931 | if (device->bdev) { | |
932 | struct block_device *path_bdev; | |
933 | ||
934 | path_bdev = lookup_bdev(path); | |
935 | if (IS_ERR(path_bdev)) { | |
936 | mutex_unlock(&fs_devices->device_list_mutex); | |
937 | return ERR_CAST(path_bdev); | |
938 | } | |
939 | ||
940 | if (device->bdev != path_bdev) { | |
941 | bdput(path_bdev); | |
942 | mutex_unlock(&fs_devices->device_list_mutex); | |
943 | btrfs_warn_in_rcu(device->fs_info, | |
944 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
945 | disk_super->fsid, devid, | |
946 | rcu_str_deref(device->name), path); | |
947 | return ERR_PTR(-EEXIST); | |
948 | } | |
949 | bdput(path_bdev); | |
950 | btrfs_info_in_rcu(device->fs_info, | |
951 | "device fsid %pU devid %llu moved old:%s new:%s", | |
952 | disk_super->fsid, devid, | |
953 | rcu_str_deref(device->name), path); | |
954 | } | |
955 | ||
606686ee | 956 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
957 | if (!name) { |
958 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 959 | return ERR_PTR(-ENOMEM); |
9c6d173e | 960 | } |
606686ee JB |
961 | rcu_string_free(device->name); |
962 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 963 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 964 | fs_devices->missing_devices--; |
e6e674bd | 965 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 966 | } |
8a4b83cc CM |
967 | } |
968 | ||
77bdae4d AJ |
969 | /* |
970 | * Unmount does not free the btrfs_device struct but would zero | |
971 | * generation along with most of the other members. So just update | |
972 | * it back. We need it to pick the disk with largest generation | |
973 | * (as above). | |
974 | */ | |
d1a63002 | 975 | if (!fs_devices->opened) { |
77bdae4d | 976 | device->generation = found_transid; |
d1a63002 NB |
977 | fs_devices->latest_generation = max_t(u64, found_transid, |
978 | fs_devices->latest_generation); | |
979 | } | |
77bdae4d | 980 | |
f2788d2f AJ |
981 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
982 | ||
9c6d173e | 983 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 984 | return device; |
8a4b83cc CM |
985 | } |
986 | ||
e4404d6e YZ |
987 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
988 | { | |
989 | struct btrfs_fs_devices *fs_devices; | |
990 | struct btrfs_device *device; | |
991 | struct btrfs_device *orig_dev; | |
d2979aa2 | 992 | int ret = 0; |
e4404d6e | 993 | |
7239ff4b | 994 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
995 | if (IS_ERR(fs_devices)) |
996 | return fs_devices; | |
e4404d6e | 997 | |
adbbb863 | 998 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 999 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1000 | |
1001 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1002 | struct rcu_string *name; |
1003 | ||
12bd2fc0 ID |
1004 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1005 | orig_dev->uuid); | |
d2979aa2 AJ |
1006 | if (IS_ERR(device)) { |
1007 | ret = PTR_ERR(device); | |
e4404d6e | 1008 | goto error; |
d2979aa2 | 1009 | } |
e4404d6e | 1010 | |
606686ee JB |
1011 | /* |
1012 | * This is ok to do without rcu read locked because we hold the | |
1013 | * uuid mutex so nothing we touch in here is going to disappear. | |
1014 | */ | |
e755f780 | 1015 | if (orig_dev->name) { |
78f2c9e6 DS |
1016 | name = rcu_string_strdup(orig_dev->name->str, |
1017 | GFP_KERNEL); | |
e755f780 | 1018 | if (!name) { |
a425f9d4 | 1019 | btrfs_free_device(device); |
d2979aa2 | 1020 | ret = -ENOMEM; |
e755f780 AJ |
1021 | goto error; |
1022 | } | |
1023 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1024 | } |
e4404d6e | 1025 | |
e4404d6e YZ |
1026 | list_add(&device->dev_list, &fs_devices->devices); |
1027 | device->fs_devices = fs_devices; | |
1028 | fs_devices->num_devices++; | |
1029 | } | |
adbbb863 | 1030 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1031 | return fs_devices; |
1032 | error: | |
adbbb863 | 1033 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1034 | free_fs_devices(fs_devices); |
d2979aa2 | 1035 | return ERR_PTR(ret); |
e4404d6e YZ |
1036 | } |
1037 | ||
9b99b115 AJ |
1038 | /* |
1039 | * After we have read the system tree and know devids belonging to | |
1040 | * this filesystem, remove the device which does not belong there. | |
1041 | */ | |
1042 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1043 | { |
c6e30871 | 1044 | struct btrfs_device *device, *next; |
443f24fe | 1045 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1046 | |
dfe25020 CM |
1047 | mutex_lock(&uuid_mutex); |
1048 | again: | |
46224705 | 1049 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1050 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1051 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1052 | &device->dev_state)) { | |
401e29c1 AJ |
1053 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1054 | &device->dev_state) && | |
998a0671 AJ |
1055 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1056 | &device->dev_state) && | |
401e29c1 AJ |
1057 | (!latest_dev || |
1058 | device->generation > latest_dev->generation)) { | |
443f24fe | 1059 | latest_dev = device; |
a6b0d5c8 | 1060 | } |
2b82032c | 1061 | continue; |
a6b0d5c8 | 1062 | } |
2b82032c | 1063 | |
8dabb742 SB |
1064 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1065 | /* | |
1066 | * In the first step, keep the device which has | |
1067 | * the correct fsid and the devid that is used | |
1068 | * for the dev_replace procedure. | |
1069 | * In the second step, the dev_replace state is | |
1070 | * read from the device tree and it is known | |
1071 | * whether the procedure is really active or | |
1072 | * not, which means whether this device is | |
1073 | * used or whether it should be removed. | |
1074 | */ | |
401e29c1 AJ |
1075 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1076 | &device->dev_state)) { | |
8dabb742 SB |
1077 | continue; |
1078 | } | |
1079 | } | |
2b82032c | 1080 | if (device->bdev) { |
d4d77629 | 1081 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1082 | device->bdev = NULL; |
1083 | fs_devices->open_devices--; | |
1084 | } | |
ebbede42 | 1085 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1086 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1087 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1088 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1089 | &device->dev_state)) | |
8dabb742 | 1090 | fs_devices->rw_devices--; |
2b82032c | 1091 | } |
e4404d6e YZ |
1092 | list_del_init(&device->dev_list); |
1093 | fs_devices->num_devices--; | |
a425f9d4 | 1094 | btrfs_free_device(device); |
dfe25020 | 1095 | } |
2b82032c YZ |
1096 | |
1097 | if (fs_devices->seed) { | |
1098 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1099 | goto again; |
1100 | } | |
1101 | ||
443f24fe | 1102 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1103 | |
dfe25020 | 1104 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1105 | } |
a0af469b | 1106 | |
14238819 AJ |
1107 | static void btrfs_close_bdev(struct btrfs_device *device) |
1108 | { | |
08ffcae8 DS |
1109 | if (!device->bdev) |
1110 | return; | |
1111 | ||
ebbede42 | 1112 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1113 | sync_blockdev(device->bdev); |
1114 | invalidate_bdev(device->bdev); | |
1115 | } | |
1116 | ||
08ffcae8 | 1117 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1118 | } |
1119 | ||
959b1c04 | 1120 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1121 | { |
1122 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1123 | |
ebbede42 | 1124 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1125 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1126 | list_del_init(&device->dev_alloc_list); | |
1127 | fs_devices->rw_devices--; | |
1128 | } | |
1129 | ||
e6e674bd | 1130 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1131 | fs_devices->missing_devices--; |
1132 | ||
959b1c04 | 1133 | btrfs_close_bdev(device); |
321f69f8 | 1134 | if (device->bdev) { |
3fff3975 | 1135 | fs_devices->open_devices--; |
321f69f8 | 1136 | device->bdev = NULL; |
f448341a | 1137 | } |
321f69f8 | 1138 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
f448341a | 1139 | |
321f69f8 JT |
1140 | device->fs_info = NULL; |
1141 | atomic_set(&device->dev_stats_ccnt, 0); | |
1142 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1143 | |
321f69f8 JT |
1144 | /* Verify the device is back in a pristine state */ |
1145 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1146 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1147 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1148 | ASSERT(list_empty(&device->post_commit_list)); | |
1149 | ASSERT(atomic_read(&device->reada_in_flight) == 0); | |
f448341a AJ |
1150 | } |
1151 | ||
0226e0eb | 1152 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1153 | { |
2037a093 | 1154 | struct btrfs_device *device, *tmp; |
e4404d6e | 1155 | |
2b82032c YZ |
1156 | if (--fs_devices->opened > 0) |
1157 | return 0; | |
8a4b83cc | 1158 | |
c9513edb | 1159 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1160 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1161 | btrfs_close_one_device(device); |
8a4b83cc | 1162 | } |
c9513edb XG |
1163 | mutex_unlock(&fs_devices->device_list_mutex); |
1164 | ||
e4404d6e YZ |
1165 | WARN_ON(fs_devices->open_devices); |
1166 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1167 | fs_devices->opened = 0; |
0395d84f | 1168 | fs_devices->seeding = false; |
2b82032c | 1169 | |
8a4b83cc CM |
1170 | return 0; |
1171 | } | |
1172 | ||
2b82032c YZ |
1173 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1174 | { | |
e4404d6e | 1175 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1176 | int ret; |
1177 | ||
1178 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1179 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1180 | if (!fs_devices->opened) { |
1181 | seed_devices = fs_devices->seed; | |
1182 | fs_devices->seed = NULL; | |
1183 | } | |
2b82032c | 1184 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1185 | |
1186 | while (seed_devices) { | |
1187 | fs_devices = seed_devices; | |
1188 | seed_devices = fs_devices->seed; | |
0226e0eb | 1189 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1190 | free_fs_devices(fs_devices); |
1191 | } | |
2b82032c YZ |
1192 | return ret; |
1193 | } | |
1194 | ||
897fb573 | 1195 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1196 | fmode_t flags, void *holder) |
8a4b83cc | 1197 | { |
8a4b83cc | 1198 | struct btrfs_device *device; |
443f24fe | 1199 | struct btrfs_device *latest_dev = NULL; |
8a4b83cc | 1200 | |
d4d77629 TH |
1201 | flags |= FMODE_EXCL; |
1202 | ||
f117e290 | 1203 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1204 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1205 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1206 | continue; |
a0af469b | 1207 | |
9f050db4 AJ |
1208 | if (!latest_dev || |
1209 | device->generation > latest_dev->generation) | |
1210 | latest_dev = device; | |
8a4b83cc | 1211 | } |
1ed802c9 AJ |
1212 | if (fs_devices->open_devices == 0) |
1213 | return -EINVAL; | |
1214 | ||
2b82032c | 1215 | fs_devices->opened = 1; |
443f24fe | 1216 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1217 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1218 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
1ed802c9 AJ |
1219 | |
1220 | return 0; | |
2b82032c YZ |
1221 | } |
1222 | ||
f8e10cd3 AJ |
1223 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1224 | { | |
1225 | struct btrfs_device *dev1, *dev2; | |
1226 | ||
1227 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1228 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1229 | ||
1230 | if (dev1->devid < dev2->devid) | |
1231 | return -1; | |
1232 | else if (dev1->devid > dev2->devid) | |
1233 | return 1; | |
1234 | return 0; | |
1235 | } | |
1236 | ||
2b82032c | 1237 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1238 | fmode_t flags, void *holder) |
2b82032c YZ |
1239 | { |
1240 | int ret; | |
1241 | ||
f5194e34 | 1242 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1243 | /* |
1244 | * The device_list_mutex cannot be taken here in case opening the | |
1245 | * underlying device takes further locks like bd_mutex. | |
1246 | * | |
1247 | * We also don't need the lock here as this is called during mount and | |
1248 | * exclusion is provided by uuid_mutex | |
1249 | */ | |
f5194e34 | 1250 | |
2b82032c | 1251 | if (fs_devices->opened) { |
e4404d6e YZ |
1252 | fs_devices->opened++; |
1253 | ret = 0; | |
2b82032c | 1254 | } else { |
f8e10cd3 | 1255 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1256 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1257 | } |
542c5908 | 1258 | |
8a4b83cc CM |
1259 | return ret; |
1260 | } | |
1261 | ||
8f32380d | 1262 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1263 | { |
8f32380d JT |
1264 | struct page *page = virt_to_page(super); |
1265 | ||
6cf86a00 AJ |
1266 | put_page(page); |
1267 | } | |
1268 | ||
b335eab8 NB |
1269 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
1270 | u64 bytenr) | |
6cf86a00 | 1271 | { |
b335eab8 NB |
1272 | struct btrfs_super_block *disk_super; |
1273 | struct page *page; | |
6cf86a00 AJ |
1274 | void *p; |
1275 | pgoff_t index; | |
1276 | ||
1277 | /* make sure our super fits in the device */ | |
1278 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
b335eab8 | 1279 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1280 | |
1281 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1282 | if (sizeof(*disk_super) > PAGE_SIZE) |
1283 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1284 | |
1285 | /* make sure our super doesn't straddle pages on disk */ | |
1286 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1287 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1288 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1289 | |
1290 | /* pull in the page with our super */ | |
b335eab8 | 1291 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1292 | |
b335eab8 NB |
1293 | if (IS_ERR(page)) |
1294 | return ERR_CAST(page); | |
6cf86a00 | 1295 | |
b335eab8 | 1296 | p = page_address(page); |
6cf86a00 AJ |
1297 | |
1298 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1299 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1300 | |
b335eab8 NB |
1301 | if (btrfs_super_bytenr(disk_super) != bytenr || |
1302 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
8f32380d | 1303 | btrfs_release_disk_super(p); |
b335eab8 | 1304 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1305 | } |
1306 | ||
b335eab8 NB |
1307 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1308 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1309 | |
b335eab8 | 1310 | return disk_super; |
6cf86a00 AJ |
1311 | } |
1312 | ||
228a73ab AJ |
1313 | int btrfs_forget_devices(const char *path) |
1314 | { | |
1315 | int ret; | |
1316 | ||
1317 | mutex_lock(&uuid_mutex); | |
1318 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1319 | mutex_unlock(&uuid_mutex); | |
1320 | ||
1321 | return ret; | |
1322 | } | |
1323 | ||
6f60cbd3 DS |
1324 | /* |
1325 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1326 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1327 | * is read via pagecache | |
1328 | */ | |
36350e95 GJ |
1329 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1330 | void *holder) | |
8a4b83cc CM |
1331 | { |
1332 | struct btrfs_super_block *disk_super; | |
4306a974 | 1333 | bool new_device_added = false; |
36350e95 | 1334 | struct btrfs_device *device = NULL; |
8a4b83cc | 1335 | struct block_device *bdev; |
6f60cbd3 | 1336 | u64 bytenr; |
8a4b83cc | 1337 | |
899f9307 DS |
1338 | lockdep_assert_held(&uuid_mutex); |
1339 | ||
6f60cbd3 DS |
1340 | /* |
1341 | * we would like to check all the supers, but that would make | |
1342 | * a btrfs mount succeed after a mkfs from a different FS. | |
1343 | * So, we need to add a special mount option to scan for | |
1344 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1345 | */ | |
1346 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1347 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1348 | |
1349 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1350 | if (IS_ERR(bdev)) |
36350e95 | 1351 | return ERR_CAST(bdev); |
6f60cbd3 | 1352 | |
b335eab8 NB |
1353 | disk_super = btrfs_read_disk_super(bdev, bytenr); |
1354 | if (IS_ERR(disk_super)) { | |
1355 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1356 | goto error_bdev_put; |
05a5c55d | 1357 | } |
6f60cbd3 | 1358 | |
4306a974 | 1359 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1360 | if (!IS_ERR(device)) { |
4306a974 AJ |
1361 | if (new_device_added) |
1362 | btrfs_free_stale_devices(path, device); | |
1363 | } | |
6f60cbd3 | 1364 | |
8f32380d | 1365 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1366 | |
1367 | error_bdev_put: | |
d4d77629 | 1368 | blkdev_put(bdev, flags); |
b6ed73bc | 1369 | |
36350e95 | 1370 | return device; |
8a4b83cc | 1371 | } |
0b86a832 | 1372 | |
1c11b63e JM |
1373 | /* |
1374 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1375 | * such is found, record the end of it in *start | |
1376 | */ | |
1c11b63e JM |
1377 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1378 | u64 len) | |
6df9a95e | 1379 | { |
1c11b63e | 1380 | u64 physical_start, physical_end; |
6df9a95e | 1381 | |
1c11b63e | 1382 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1383 | |
1c11b63e JM |
1384 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1385 | &physical_start, &physical_end, | |
1386 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1387 | |
1c11b63e JM |
1388 | if (in_range(physical_start, *start, len) || |
1389 | in_range(*start, physical_start, | |
1390 | physical_end - physical_start)) { | |
1391 | *start = physical_end + 1; | |
1392 | return true; | |
6df9a95e JB |
1393 | } |
1394 | } | |
1c11b63e | 1395 | return false; |
6df9a95e JB |
1396 | } |
1397 | ||
3b4ffa40 NA |
1398 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1399 | { | |
1400 | switch (device->fs_devices->chunk_alloc_policy) { | |
1401 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1402 | /* | |
1403 | * We don't want to overwrite the superblock on the drive nor | |
1404 | * any area used by the boot loader (grub for example), so we | |
1405 | * make sure to start at an offset of at least 1MB. | |
1406 | */ | |
1407 | return max_t(u64, start, SZ_1M); | |
1408 | default: | |
1409 | BUG(); | |
1410 | } | |
1411 | } | |
1412 | ||
1413 | /** | |
1414 | * dev_extent_hole_check - check if specified hole is suitable for allocation | |
1415 | * @device: the device which we have the hole | |
1416 | * @hole_start: starting position of the hole | |
1417 | * @hole_size: the size of the hole | |
1418 | * @num_bytes: the size of the free space that we need | |
1419 | * | |
1420 | * This function may modify @hole_start and @hole_end to reflect the suitable | |
1421 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. | |
1422 | */ | |
1423 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1424 | u64 *hole_size, u64 num_bytes) | |
1425 | { | |
1426 | bool changed = false; | |
1427 | u64 hole_end = *hole_start + *hole_size; | |
1428 | ||
1429 | /* | |
1430 | * Check before we set max_hole_start, otherwise we could end up | |
1431 | * sending back this offset anyway. | |
1432 | */ | |
1433 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1434 | if (hole_end >= *hole_start) | |
1435 | *hole_size = hole_end - *hole_start; | |
1436 | else | |
1437 | *hole_size = 0; | |
1438 | changed = true; | |
1439 | } | |
1440 | ||
1441 | switch (device->fs_devices->chunk_alloc_policy) { | |
1442 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1443 | /* No extra check */ | |
1444 | break; | |
1445 | default: | |
1446 | BUG(); | |
1447 | } | |
1448 | ||
1449 | return changed; | |
1450 | } | |
6df9a95e | 1451 | |
0b86a832 | 1452 | /* |
499f377f JM |
1453 | * find_free_dev_extent_start - find free space in the specified device |
1454 | * @device: the device which we search the free space in | |
1455 | * @num_bytes: the size of the free space that we need | |
1456 | * @search_start: the position from which to begin the search | |
1457 | * @start: store the start of the free space. | |
1458 | * @len: the size of the free space. that we find, or the size | |
1459 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1460 | * |
0b86a832 CM |
1461 | * this uses a pretty simple search, the expectation is that it is |
1462 | * called very infrequently and that a given device has a small number | |
1463 | * of extents | |
7bfc837d MX |
1464 | * |
1465 | * @start is used to store the start of the free space if we find. But if we | |
1466 | * don't find suitable free space, it will be used to store the start position | |
1467 | * of the max free space. | |
1468 | * | |
1469 | * @len is used to store the size of the free space that we find. | |
1470 | * But if we don't find suitable free space, it is used to store the size of | |
1471 | * the max free space. | |
135da976 QW |
1472 | * |
1473 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1474 | * check to ensure dev extents are not double allocated. | |
1475 | * This makes the function safe to allocate dev extents but may not report | |
1476 | * correct usable device space, as device extent freed in current transaction | |
1477 | * is not reported as avaiable. | |
0b86a832 | 1478 | */ |
9e3246a5 QW |
1479 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1480 | u64 num_bytes, u64 search_start, u64 *start, | |
1481 | u64 *len) | |
0b86a832 | 1482 | { |
0b246afa JM |
1483 | struct btrfs_fs_info *fs_info = device->fs_info; |
1484 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1485 | struct btrfs_key key; |
7bfc837d | 1486 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1487 | struct btrfs_path *path; |
7bfc837d MX |
1488 | u64 hole_size; |
1489 | u64 max_hole_start; | |
1490 | u64 max_hole_size; | |
1491 | u64 extent_end; | |
0b86a832 CM |
1492 | u64 search_end = device->total_bytes; |
1493 | int ret; | |
7bfc837d | 1494 | int slot; |
0b86a832 | 1495 | struct extent_buffer *l; |
8cdc7c5b | 1496 | |
3b4ffa40 | 1497 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1498 | |
6df9a95e JB |
1499 | path = btrfs_alloc_path(); |
1500 | if (!path) | |
1501 | return -ENOMEM; | |
f2ab7618 | 1502 | |
7bfc837d MX |
1503 | max_hole_start = search_start; |
1504 | max_hole_size = 0; | |
1505 | ||
f2ab7618 | 1506 | again: |
401e29c1 AJ |
1507 | if (search_start >= search_end || |
1508 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1509 | ret = -ENOSPC; |
6df9a95e | 1510 | goto out; |
7bfc837d MX |
1511 | } |
1512 | ||
e4058b54 | 1513 | path->reada = READA_FORWARD; |
6df9a95e JB |
1514 | path->search_commit_root = 1; |
1515 | path->skip_locking = 1; | |
7bfc837d | 1516 | |
0b86a832 CM |
1517 | key.objectid = device->devid; |
1518 | key.offset = search_start; | |
1519 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1520 | |
125ccb0a | 1521 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1522 | if (ret < 0) |
7bfc837d | 1523 | goto out; |
1fcbac58 YZ |
1524 | if (ret > 0) { |
1525 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1526 | if (ret < 0) | |
7bfc837d | 1527 | goto out; |
1fcbac58 | 1528 | } |
7bfc837d | 1529 | |
0b86a832 CM |
1530 | while (1) { |
1531 | l = path->nodes[0]; | |
1532 | slot = path->slots[0]; | |
1533 | if (slot >= btrfs_header_nritems(l)) { | |
1534 | ret = btrfs_next_leaf(root, path); | |
1535 | if (ret == 0) | |
1536 | continue; | |
1537 | if (ret < 0) | |
7bfc837d MX |
1538 | goto out; |
1539 | ||
1540 | break; | |
0b86a832 CM |
1541 | } |
1542 | btrfs_item_key_to_cpu(l, &key, slot); | |
1543 | ||
1544 | if (key.objectid < device->devid) | |
1545 | goto next; | |
1546 | ||
1547 | if (key.objectid > device->devid) | |
7bfc837d | 1548 | break; |
0b86a832 | 1549 | |
962a298f | 1550 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1551 | goto next; |
9779b72f | 1552 | |
7bfc837d MX |
1553 | if (key.offset > search_start) { |
1554 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1555 | dev_extent_hole_check(device, &search_start, &hole_size, |
1556 | num_bytes); | |
6df9a95e | 1557 | |
7bfc837d MX |
1558 | if (hole_size > max_hole_size) { |
1559 | max_hole_start = search_start; | |
1560 | max_hole_size = hole_size; | |
1561 | } | |
9779b72f | 1562 | |
7bfc837d MX |
1563 | /* |
1564 | * If this free space is greater than which we need, | |
1565 | * it must be the max free space that we have found | |
1566 | * until now, so max_hole_start must point to the start | |
1567 | * of this free space and the length of this free space | |
1568 | * is stored in max_hole_size. Thus, we return | |
1569 | * max_hole_start and max_hole_size and go back to the | |
1570 | * caller. | |
1571 | */ | |
1572 | if (hole_size >= num_bytes) { | |
1573 | ret = 0; | |
1574 | goto out; | |
0b86a832 CM |
1575 | } |
1576 | } | |
0b86a832 | 1577 | |
0b86a832 | 1578 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1579 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1580 | dev_extent); | |
1581 | if (extent_end > search_start) | |
1582 | search_start = extent_end; | |
0b86a832 CM |
1583 | next: |
1584 | path->slots[0]++; | |
1585 | cond_resched(); | |
1586 | } | |
0b86a832 | 1587 | |
38c01b96 | 1588 | /* |
1589 | * At this point, search_start should be the end of | |
1590 | * allocated dev extents, and when shrinking the device, | |
1591 | * search_end may be smaller than search_start. | |
1592 | */ | |
f2ab7618 | 1593 | if (search_end > search_start) { |
38c01b96 | 1594 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1595 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1596 | num_bytes)) { | |
f2ab7618 ZL |
1597 | btrfs_release_path(path); |
1598 | goto again; | |
1599 | } | |
0b86a832 | 1600 | |
f2ab7618 ZL |
1601 | if (hole_size > max_hole_size) { |
1602 | max_hole_start = search_start; | |
1603 | max_hole_size = hole_size; | |
1604 | } | |
6df9a95e JB |
1605 | } |
1606 | ||
7bfc837d | 1607 | /* See above. */ |
f2ab7618 | 1608 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1609 | ret = -ENOSPC; |
1610 | else | |
1611 | ret = 0; | |
1612 | ||
1613 | out: | |
2b82032c | 1614 | btrfs_free_path(path); |
7bfc837d | 1615 | *start = max_hole_start; |
b2117a39 | 1616 | if (len) |
7bfc837d | 1617 | *len = max_hole_size; |
0b86a832 CM |
1618 | return ret; |
1619 | } | |
1620 | ||
60dfdf25 | 1621 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1622 | u64 *start, u64 *len) |
1623 | { | |
499f377f | 1624 | /* FIXME use last free of some kind */ |
60dfdf25 | 1625 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1626 | } |
1627 | ||
b2950863 | 1628 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1629 | struct btrfs_device *device, |
2196d6e8 | 1630 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1631 | { |
0b246afa JM |
1632 | struct btrfs_fs_info *fs_info = device->fs_info; |
1633 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1634 | int ret; |
1635 | struct btrfs_path *path; | |
8f18cf13 | 1636 | struct btrfs_key key; |
a061fc8d CM |
1637 | struct btrfs_key found_key; |
1638 | struct extent_buffer *leaf = NULL; | |
1639 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1640 | |
1641 | path = btrfs_alloc_path(); | |
1642 | if (!path) | |
1643 | return -ENOMEM; | |
1644 | ||
1645 | key.objectid = device->devid; | |
1646 | key.offset = start; | |
1647 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1648 | again: |
8f18cf13 | 1649 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1650 | if (ret > 0) { |
1651 | ret = btrfs_previous_item(root, path, key.objectid, | |
1652 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1653 | if (ret) |
1654 | goto out; | |
a061fc8d CM |
1655 | leaf = path->nodes[0]; |
1656 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1657 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1658 | struct btrfs_dev_extent); | |
1659 | BUG_ON(found_key.offset > start || found_key.offset + | |
1660 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1661 | key = found_key; |
1662 | btrfs_release_path(path); | |
1663 | goto again; | |
a061fc8d CM |
1664 | } else if (ret == 0) { |
1665 | leaf = path->nodes[0]; | |
1666 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1667 | struct btrfs_dev_extent); | |
79787eaa | 1668 | } else { |
0b246afa | 1669 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1670 | goto out; |
a061fc8d | 1671 | } |
8f18cf13 | 1672 | |
2196d6e8 MX |
1673 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1674 | ||
8f18cf13 | 1675 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1676 | if (ret) { |
0b246afa JM |
1677 | btrfs_handle_fs_error(fs_info, ret, |
1678 | "Failed to remove dev extent item"); | |
13212b54 | 1679 | } else { |
3204d33c | 1680 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1681 | } |
b0b802d7 | 1682 | out: |
8f18cf13 CM |
1683 | btrfs_free_path(path); |
1684 | return ret; | |
1685 | } | |
1686 | ||
48a3b636 ES |
1687 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1688 | struct btrfs_device *device, | |
48a3b636 | 1689 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1690 | { |
1691 | int ret; | |
1692 | struct btrfs_path *path; | |
0b246afa JM |
1693 | struct btrfs_fs_info *fs_info = device->fs_info; |
1694 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1695 | struct btrfs_dev_extent *extent; |
1696 | struct extent_buffer *leaf; | |
1697 | struct btrfs_key key; | |
1698 | ||
e12c9621 | 1699 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1700 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1701 | path = btrfs_alloc_path(); |
1702 | if (!path) | |
1703 | return -ENOMEM; | |
1704 | ||
0b86a832 | 1705 | key.objectid = device->devid; |
2b82032c | 1706 | key.offset = start; |
0b86a832 CM |
1707 | key.type = BTRFS_DEV_EXTENT_KEY; |
1708 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1709 | sizeof(*extent)); | |
2cdcecbc MF |
1710 | if (ret) |
1711 | goto out; | |
0b86a832 CM |
1712 | |
1713 | leaf = path->nodes[0]; | |
1714 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1715 | struct btrfs_dev_extent); | |
b5d9071c NB |
1716 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1717 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1718 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1719 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1720 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1721 | ||
0b86a832 CM |
1722 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1723 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1724 | out: |
0b86a832 CM |
1725 | btrfs_free_path(path); |
1726 | return ret; | |
1727 | } | |
1728 | ||
6df9a95e | 1729 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1730 | { |
6df9a95e JB |
1731 | struct extent_map_tree *em_tree; |
1732 | struct extent_map *em; | |
1733 | struct rb_node *n; | |
1734 | u64 ret = 0; | |
0b86a832 | 1735 | |
c8bf1b67 | 1736 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1737 | read_lock(&em_tree->lock); |
07e1ce09 | 1738 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1739 | if (n) { |
1740 | em = rb_entry(n, struct extent_map, rb_node); | |
1741 | ret = em->start + em->len; | |
0b86a832 | 1742 | } |
6df9a95e JB |
1743 | read_unlock(&em_tree->lock); |
1744 | ||
0b86a832 CM |
1745 | return ret; |
1746 | } | |
1747 | ||
53f10659 ID |
1748 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1749 | u64 *devid_ret) | |
0b86a832 CM |
1750 | { |
1751 | int ret; | |
1752 | struct btrfs_key key; | |
1753 | struct btrfs_key found_key; | |
2b82032c YZ |
1754 | struct btrfs_path *path; |
1755 | ||
2b82032c YZ |
1756 | path = btrfs_alloc_path(); |
1757 | if (!path) | |
1758 | return -ENOMEM; | |
0b86a832 CM |
1759 | |
1760 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1761 | key.type = BTRFS_DEV_ITEM_KEY; | |
1762 | key.offset = (u64)-1; | |
1763 | ||
53f10659 | 1764 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1765 | if (ret < 0) |
1766 | goto error; | |
1767 | ||
a06dee4d AJ |
1768 | if (ret == 0) { |
1769 | /* Corruption */ | |
1770 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1771 | ret = -EUCLEAN; | |
1772 | goto error; | |
1773 | } | |
0b86a832 | 1774 | |
53f10659 ID |
1775 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1776 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1777 | BTRFS_DEV_ITEM_KEY); |
1778 | if (ret) { | |
53f10659 | 1779 | *devid_ret = 1; |
0b86a832 CM |
1780 | } else { |
1781 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1782 | path->slots[0]); | |
53f10659 | 1783 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1784 | } |
1785 | ret = 0; | |
1786 | error: | |
2b82032c | 1787 | btrfs_free_path(path); |
0b86a832 CM |
1788 | return ret; |
1789 | } | |
1790 | ||
1791 | /* | |
1792 | * the device information is stored in the chunk root | |
1793 | * the btrfs_device struct should be fully filled in | |
1794 | */ | |
c74a0b02 | 1795 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1796 | struct btrfs_device *device) |
0b86a832 CM |
1797 | { |
1798 | int ret; | |
1799 | struct btrfs_path *path; | |
1800 | struct btrfs_dev_item *dev_item; | |
1801 | struct extent_buffer *leaf; | |
1802 | struct btrfs_key key; | |
1803 | unsigned long ptr; | |
0b86a832 | 1804 | |
0b86a832 CM |
1805 | path = btrfs_alloc_path(); |
1806 | if (!path) | |
1807 | return -ENOMEM; | |
1808 | ||
0b86a832 CM |
1809 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1810 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1811 | key.offset = device->devid; |
0b86a832 | 1812 | |
8e87e856 NB |
1813 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1814 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1815 | if (ret) |
1816 | goto out; | |
1817 | ||
1818 | leaf = path->nodes[0]; | |
1819 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1820 | ||
1821 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1822 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1823 | btrfs_set_device_type(leaf, dev_item, device->type); |
1824 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1825 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1826 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1827 | btrfs_set_device_total_bytes(leaf, dev_item, |
1828 | btrfs_device_get_disk_total_bytes(device)); | |
1829 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1830 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1831 | btrfs_set_device_group(leaf, dev_item, 0); |
1832 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1833 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1834 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1835 | |
410ba3a2 | 1836 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1837 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1838 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1839 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1840 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1841 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1842 | |
2b82032c | 1843 | ret = 0; |
0b86a832 CM |
1844 | out: |
1845 | btrfs_free_path(path); | |
1846 | return ret; | |
1847 | } | |
8f18cf13 | 1848 | |
5a1972bd QW |
1849 | /* |
1850 | * Function to update ctime/mtime for a given device path. | |
1851 | * Mainly used for ctime/mtime based probe like libblkid. | |
1852 | */ | |
da353f6b | 1853 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1854 | { |
1855 | struct file *filp; | |
1856 | ||
1857 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1858 | if (IS_ERR(filp)) |
5a1972bd QW |
1859 | return; |
1860 | file_update_time(filp); | |
1861 | filp_close(filp, NULL); | |
5a1972bd QW |
1862 | } |
1863 | ||
f331a952 | 1864 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1865 | { |
f331a952 | 1866 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1867 | int ret; |
1868 | struct btrfs_path *path; | |
a061fc8d | 1869 | struct btrfs_key key; |
a061fc8d CM |
1870 | struct btrfs_trans_handle *trans; |
1871 | ||
a061fc8d CM |
1872 | path = btrfs_alloc_path(); |
1873 | if (!path) | |
1874 | return -ENOMEM; | |
1875 | ||
a22285a6 | 1876 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1877 | if (IS_ERR(trans)) { |
1878 | btrfs_free_path(path); | |
1879 | return PTR_ERR(trans); | |
1880 | } | |
a061fc8d CM |
1881 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1882 | key.type = BTRFS_DEV_ITEM_KEY; | |
1883 | key.offset = device->devid; | |
1884 | ||
1885 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1886 | if (ret) { |
1887 | if (ret > 0) | |
1888 | ret = -ENOENT; | |
1889 | btrfs_abort_transaction(trans, ret); | |
1890 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1891 | goto out; |
1892 | } | |
1893 | ||
1894 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1895 | if (ret) { |
1896 | btrfs_abort_transaction(trans, ret); | |
1897 | btrfs_end_transaction(trans); | |
1898 | } | |
1899 | ||
a061fc8d CM |
1900 | out: |
1901 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1902 | if (!ret) |
1903 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1904 | return ret; |
1905 | } | |
1906 | ||
3cc31a0d DS |
1907 | /* |
1908 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1909 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1910 | * replace. | |
1911 | */ | |
1912 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1913 | u64 num_devices) | |
a061fc8d | 1914 | { |
a061fc8d | 1915 | u64 all_avail; |
de98ced9 | 1916 | unsigned seq; |
418775a2 | 1917 | int i; |
a061fc8d | 1918 | |
de98ced9 | 1919 | do { |
bd45ffbc | 1920 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1921 | |
bd45ffbc AJ |
1922 | all_avail = fs_info->avail_data_alloc_bits | |
1923 | fs_info->avail_system_alloc_bits | | |
1924 | fs_info->avail_metadata_alloc_bits; | |
1925 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1926 | |
418775a2 | 1927 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1928 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1929 | continue; |
a061fc8d | 1930 | |
418775a2 | 1931 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 1932 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 1933 | |
418775a2 DS |
1934 | if (ret) |
1935 | return ret; | |
1936 | } | |
53b381b3 DW |
1937 | } |
1938 | ||
bd45ffbc | 1939 | return 0; |
f1fa7f26 AJ |
1940 | } |
1941 | ||
c9162bdf OS |
1942 | static struct btrfs_device * btrfs_find_next_active_device( |
1943 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1944 | { |
2b82032c | 1945 | struct btrfs_device *next_device; |
88acff64 AJ |
1946 | |
1947 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1948 | if (next_device != device && | |
e6e674bd AJ |
1949 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1950 | && next_device->bdev) | |
88acff64 AJ |
1951 | return next_device; |
1952 | } | |
1953 | ||
1954 | return NULL; | |
1955 | } | |
1956 | ||
1957 | /* | |
1958 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1959 | * and replace it with the provided or the next active device, in the context | |
1960 | * where this function called, there should be always be another device (or | |
1961 | * this_dev) which is active. | |
1962 | */ | |
b105e927 | 1963 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
d6507cf1 | 1964 | struct btrfs_device *this_dev) |
88acff64 | 1965 | { |
d6507cf1 | 1966 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
1967 | struct btrfs_device *next_device; |
1968 | ||
1969 | if (this_dev) | |
1970 | next_device = this_dev; | |
1971 | else | |
1972 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1973 | device); | |
1974 | ASSERT(next_device); | |
1975 | ||
1976 | if (fs_info->sb->s_bdev && | |
1977 | (fs_info->sb->s_bdev == device->bdev)) | |
1978 | fs_info->sb->s_bdev = next_device->bdev; | |
1979 | ||
1980 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1981 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1982 | } | |
1983 | ||
1da73967 AJ |
1984 | /* |
1985 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1986 | * currently replaced. | |
1987 | */ | |
1988 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
1989 | { | |
1990 | u64 num_devices = fs_info->fs_devices->num_devices; | |
1991 | ||
cb5583dd | 1992 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1993 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
1994 | ASSERT(num_devices > 1); | |
1995 | num_devices--; | |
1996 | } | |
cb5583dd | 1997 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
1998 | |
1999 | return num_devices; | |
2000 | } | |
2001 | ||
8f32380d JT |
2002 | static void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2003 | struct block_device *bdev, | |
6fbceb9f JT |
2004 | const char *device_path) |
2005 | { | |
6fbceb9f JT |
2006 | struct btrfs_super_block *disk_super; |
2007 | int copy_num; | |
2008 | ||
2009 | if (!bdev) | |
2010 | return; | |
2011 | ||
2012 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
8f32380d JT |
2013 | struct page *page; |
2014 | int ret; | |
6fbceb9f | 2015 | |
8f32380d JT |
2016 | disk_super = btrfs_read_dev_one_super(bdev, copy_num); |
2017 | if (IS_ERR(disk_super)) | |
2018 | continue; | |
6fbceb9f JT |
2019 | |
2020 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
8f32380d JT |
2021 | |
2022 | page = virt_to_page(disk_super); | |
2023 | set_page_dirty(page); | |
2024 | lock_page(page); | |
2025 | /* write_on_page() unlocks the page */ | |
2026 | ret = write_one_page(page); | |
2027 | if (ret) | |
2028 | btrfs_warn(fs_info, | |
2029 | "error clearing superblock number %d (%d)", | |
2030 | copy_num, ret); | |
2031 | btrfs_release_disk_super(disk_super); | |
2032 | ||
6fbceb9f JT |
2033 | } |
2034 | ||
2035 | /* Notify udev that device has changed */ | |
2036 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2037 | ||
2038 | /* Update ctime/mtime for device path for libblkid */ | |
2039 | update_dev_time(device_path); | |
2040 | } | |
2041 | ||
da353f6b DS |
2042 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2043 | u64 devid) | |
f1fa7f26 AJ |
2044 | { |
2045 | struct btrfs_device *device; | |
1f78160c | 2046 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2047 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2048 | u64 num_devices; |
a061fc8d CM |
2049 | int ret = 0; |
2050 | ||
a061fc8d CM |
2051 | mutex_lock(&uuid_mutex); |
2052 | ||
1da73967 | 2053 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2054 | |
0b246afa | 2055 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2056 | if (ret) |
a061fc8d | 2057 | goto out; |
a061fc8d | 2058 | |
a27a94c2 NB |
2059 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2060 | ||
2061 | if (IS_ERR(device)) { | |
2062 | if (PTR_ERR(device) == -ENOENT && | |
2063 | strcmp(device_path, "missing") == 0) | |
2064 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2065 | else | |
2066 | ret = PTR_ERR(device); | |
53b381b3 | 2067 | goto out; |
a27a94c2 | 2068 | } |
dfe25020 | 2069 | |
eede2bf3 OS |
2070 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2071 | btrfs_warn_in_rcu(fs_info, | |
2072 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2073 | rcu_str_deref(device->name), device->devid); | |
2074 | ret = -ETXTBSY; | |
2075 | goto out; | |
2076 | } | |
2077 | ||
401e29c1 | 2078 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2079 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2080 | goto out; |
63a212ab SB |
2081 | } |
2082 | ||
ebbede42 AJ |
2083 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2084 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2085 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2086 | goto out; |
2b82032c YZ |
2087 | } |
2088 | ||
ebbede42 | 2089 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2090 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2091 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2092 | device->fs_devices->rw_devices--; |
34441361 | 2093 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2094 | } |
a061fc8d | 2095 | |
d7901554 | 2096 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2097 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2098 | mutex_lock(&uuid_mutex); |
a061fc8d | 2099 | if (ret) |
9b3517e9 | 2100 | goto error_undo; |
a061fc8d | 2101 | |
63a212ab SB |
2102 | /* |
2103 | * TODO: the superblock still includes this device in its num_devices | |
2104 | * counter although write_all_supers() is not locked out. This | |
2105 | * could give a filesystem state which requires a degraded mount. | |
2106 | */ | |
f331a952 | 2107 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2108 | if (ret) |
9b3517e9 | 2109 | goto error_undo; |
a061fc8d | 2110 | |
e12c9621 | 2111 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2112 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2113 | |
2114 | /* | |
2115 | * the device list mutex makes sure that we don't change | |
2116 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2117 | * the device supers. Whoever is writing all supers, should |
2118 | * lock the device list mutex before getting the number of | |
2119 | * devices in the super block (super_copy). Conversely, | |
2120 | * whoever updates the number of devices in the super block | |
2121 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2122 | */ |
1f78160c | 2123 | |
41a52a0f AJ |
2124 | /* |
2125 | * In normal cases the cur_devices == fs_devices. But in case | |
2126 | * of deleting a seed device, the cur_devices should point to | |
2127 | * its own fs_devices listed under the fs_devices->seed. | |
2128 | */ | |
1f78160c | 2129 | cur_devices = device->fs_devices; |
b5185197 | 2130 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2131 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2132 | |
41a52a0f AJ |
2133 | cur_devices->num_devices--; |
2134 | cur_devices->total_devices--; | |
b4993e64 AJ |
2135 | /* Update total_devices of the parent fs_devices if it's seed */ |
2136 | if (cur_devices != fs_devices) | |
2137 | fs_devices->total_devices--; | |
2b82032c | 2138 | |
e6e674bd | 2139 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2140 | cur_devices->missing_devices--; |
cd02dca5 | 2141 | |
d6507cf1 | 2142 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2143 | |
0bfaa9c5 | 2144 | if (device->bdev) { |
41a52a0f | 2145 | cur_devices->open_devices--; |
0bfaa9c5 | 2146 | /* remove sysfs entry */ |
f3cd2c58 | 2147 | btrfs_sysfs_remove_devices_dir(fs_devices, device); |
0bfaa9c5 | 2148 | } |
99994cde | 2149 | |
0b246afa JM |
2150 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2151 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2152 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2153 | |
cea67ab9 JM |
2154 | /* |
2155 | * at this point, the device is zero sized and detached from | |
2156 | * the devices list. All that's left is to zero out the old | |
2157 | * supers and free the device. | |
2158 | */ | |
ebbede42 | 2159 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
8f32380d JT |
2160 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2161 | device->name->str); | |
cea67ab9 JM |
2162 | |
2163 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2164 | synchronize_rcu(); |
2165 | btrfs_free_device(device); | |
cea67ab9 | 2166 | |
1f78160c | 2167 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2168 | while (fs_devices) { |
8321cf25 RS |
2169 | if (fs_devices->seed == cur_devices) { |
2170 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2171 | break; |
8321cf25 | 2172 | } |
e4404d6e | 2173 | fs_devices = fs_devices->seed; |
2b82032c | 2174 | } |
1f78160c | 2175 | cur_devices->seed = NULL; |
0226e0eb | 2176 | close_fs_devices(cur_devices); |
1f78160c | 2177 | free_fs_devices(cur_devices); |
2b82032c YZ |
2178 | } |
2179 | ||
a061fc8d CM |
2180 | out: |
2181 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2182 | return ret; |
24fc572f | 2183 | |
9b3517e9 | 2184 | error_undo: |
ebbede42 | 2185 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2186 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2187 | list_add(&device->dev_alloc_list, |
b5185197 | 2188 | &fs_devices->alloc_list); |
c3929c36 | 2189 | device->fs_devices->rw_devices++; |
34441361 | 2190 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2191 | } |
24fc572f | 2192 | goto out; |
a061fc8d CM |
2193 | } |
2194 | ||
68a9db5f | 2195 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2196 | { |
d51908ce AJ |
2197 | struct btrfs_fs_devices *fs_devices; |
2198 | ||
68a9db5f | 2199 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2200 | |
25e8e911 AJ |
2201 | /* |
2202 | * in case of fs with no seed, srcdev->fs_devices will point | |
2203 | * to fs_devices of fs_info. However when the dev being replaced is | |
2204 | * a seed dev it will point to the seed's local fs_devices. In short | |
2205 | * srcdev will have its correct fs_devices in both the cases. | |
2206 | */ | |
2207 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2208 | |
e93c89c1 | 2209 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2210 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2211 | fs_devices->num_devices--; |
e6e674bd | 2212 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2213 | fs_devices->missing_devices--; |
e93c89c1 | 2214 | |
ebbede42 | 2215 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2216 | fs_devices->rw_devices--; |
1357272f | 2217 | |
82372bc8 | 2218 | if (srcdev->bdev) |
d51908ce | 2219 | fs_devices->open_devices--; |
084b6e7c QW |
2220 | } |
2221 | ||
65237ee3 | 2222 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c | 2223 | { |
65237ee3 | 2224 | struct btrfs_fs_info *fs_info = srcdev->fs_info; |
084b6e7c | 2225 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
e93c89c1 | 2226 | |
ebbede42 | 2227 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 | 2228 | /* zero out the old super if it is writable */ |
8f32380d JT |
2229 | btrfs_scratch_superblocks(fs_info, srcdev->bdev, |
2230 | srcdev->name->str); | |
48b3b9d4 | 2231 | } |
14238819 AJ |
2232 | |
2233 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2234 | synchronize_rcu(); |
2235 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2236 | |
94d5f0c2 AJ |
2237 | /* if this is no devs we rather delete the fs_devices */ |
2238 | if (!fs_devices->num_devices) { | |
2239 | struct btrfs_fs_devices *tmp_fs_devices; | |
2240 | ||
6dd38f81 AJ |
2241 | /* |
2242 | * On a mounted FS, num_devices can't be zero unless it's a | |
2243 | * seed. In case of a seed device being replaced, the replace | |
2244 | * target added to the sprout FS, so there will be no more | |
2245 | * device left under the seed FS. | |
2246 | */ | |
2247 | ASSERT(fs_devices->seeding); | |
2248 | ||
94d5f0c2 AJ |
2249 | tmp_fs_devices = fs_info->fs_devices; |
2250 | while (tmp_fs_devices) { | |
2251 | if (tmp_fs_devices->seed == fs_devices) { | |
2252 | tmp_fs_devices->seed = fs_devices->seed; | |
2253 | break; | |
2254 | } | |
2255 | tmp_fs_devices = tmp_fs_devices->seed; | |
2256 | } | |
2257 | fs_devices->seed = NULL; | |
0226e0eb | 2258 | close_fs_devices(fs_devices); |
8bef8401 | 2259 | free_fs_devices(fs_devices); |
94d5f0c2 | 2260 | } |
e93c89c1 SB |
2261 | } |
2262 | ||
4f5ad7bd | 2263 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2264 | { |
4f5ad7bd | 2265 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2266 | |
d9a071f0 | 2267 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2268 | |
f3cd2c58 | 2269 | btrfs_sysfs_remove_devices_dir(fs_devices, tgtdev); |
d2ff1b20 | 2270 | |
779bf3fe | 2271 | if (tgtdev->bdev) |
d9a071f0 | 2272 | fs_devices->open_devices--; |
779bf3fe | 2273 | |
d9a071f0 | 2274 | fs_devices->num_devices--; |
e93c89c1 | 2275 | |
d6507cf1 | 2276 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2277 | |
e93c89c1 | 2278 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2279 | |
d9a071f0 | 2280 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2281 | |
2282 | /* | |
2283 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2284 | * may lead to a call to btrfs_show_devname() which will try | |
2285 | * to hold device_list_mutex. And here this device | |
2286 | * is already out of device list, so we don't have to hold | |
2287 | * the device_list_mutex lock. | |
2288 | */ | |
8f32380d JT |
2289 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2290 | tgtdev->name->str); | |
14238819 AJ |
2291 | |
2292 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2293 | synchronize_rcu(); |
2294 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2295 | } |
2296 | ||
b444ad46 NB |
2297 | static struct btrfs_device *btrfs_find_device_by_path( |
2298 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2299 | { |
2300 | int ret = 0; | |
2301 | struct btrfs_super_block *disk_super; | |
2302 | u64 devid; | |
2303 | u8 *dev_uuid; | |
2304 | struct block_device *bdev; | |
b444ad46 | 2305 | struct btrfs_device *device; |
7ba15b7d | 2306 | |
7ba15b7d | 2307 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
8f32380d | 2308 | fs_info->bdev_holder, 0, &bdev, &disk_super); |
7ba15b7d | 2309 | if (ret) |
b444ad46 | 2310 | return ERR_PTR(ret); |
8f32380d | 2311 | |
7ba15b7d SB |
2312 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2313 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2314 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2315 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2316 | disk_super->metadata_uuid, true); |
7239ff4b | 2317 | else |
e4319cd9 | 2318 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2319 | disk_super->fsid, true); |
7239ff4b | 2320 | |
8f32380d | 2321 | btrfs_release_disk_super(disk_super); |
b444ad46 NB |
2322 | if (!device) |
2323 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2324 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2325 | return device; |
7ba15b7d SB |
2326 | } |
2327 | ||
5c5c0df0 DS |
2328 | /* |
2329 | * Lookup a device given by device id, or the path if the id is 0. | |
2330 | */ | |
a27a94c2 | 2331 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2332 | struct btrfs_fs_info *fs_info, u64 devid, |
2333 | const char *device_path) | |
24e0474b | 2334 | { |
a27a94c2 | 2335 | struct btrfs_device *device; |
24e0474b | 2336 | |
5c5c0df0 | 2337 | if (devid) { |
e4319cd9 | 2338 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2339 | NULL, true); |
a27a94c2 NB |
2340 | if (!device) |
2341 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2342 | return device; |
2343 | } | |
2344 | ||
2345 | if (!device_path || !device_path[0]) | |
2346 | return ERR_PTR(-EINVAL); | |
2347 | ||
2348 | if (strcmp(device_path, "missing") == 0) { | |
2349 | /* Find first missing device */ | |
2350 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2351 | dev_list) { | |
2352 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2353 | &device->dev_state) && !device->bdev) | |
2354 | return device; | |
d95a830c | 2355 | } |
6e927ceb | 2356 | return ERR_PTR(-ENOENT); |
24e0474b | 2357 | } |
6e927ceb AJ |
2358 | |
2359 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2360 | } |
2361 | ||
2b82032c YZ |
2362 | /* |
2363 | * does all the dirty work required for changing file system's UUID. | |
2364 | */ | |
2ff7e61e | 2365 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2366 | { |
0b246afa | 2367 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2368 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2369 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2370 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2371 | struct btrfs_device *device; |
2372 | u64 super_flags; | |
2373 | ||
a32bf9a3 | 2374 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2375 | if (!fs_devices->seeding) |
2b82032c YZ |
2376 | return -EINVAL; |
2377 | ||
7239ff4b | 2378 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2379 | if (IS_ERR(seed_devices)) |
2380 | return PTR_ERR(seed_devices); | |
2b82032c | 2381 | |
e4404d6e YZ |
2382 | old_devices = clone_fs_devices(fs_devices); |
2383 | if (IS_ERR(old_devices)) { | |
2384 | kfree(seed_devices); | |
2385 | return PTR_ERR(old_devices); | |
2b82032c | 2386 | } |
e4404d6e | 2387 | |
c4babc5e | 2388 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2389 | |
e4404d6e YZ |
2390 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2391 | seed_devices->opened = 1; | |
2392 | INIT_LIST_HEAD(&seed_devices->devices); | |
2393 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2394 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2395 | |
321a4bf7 | 2396 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2397 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2398 | synchronize_rcu); | |
2196d6e8 MX |
2399 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2400 | device->fs_devices = seed_devices; | |
c9513edb | 2401 | |
34441361 | 2402 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2403 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2404 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2405 | |
0395d84f | 2406 | fs_devices->seeding = false; |
2b82032c YZ |
2407 | fs_devices->num_devices = 0; |
2408 | fs_devices->open_devices = 0; | |
69611ac8 | 2409 | fs_devices->missing_devices = 0; |
7f0432d0 | 2410 | fs_devices->rotating = false; |
e4404d6e | 2411 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2412 | |
2413 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2414 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2415 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2416 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2417 | |
2b82032c YZ |
2418 | super_flags = btrfs_super_flags(disk_super) & |
2419 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2420 | btrfs_set_super_flags(disk_super, super_flags); | |
2421 | ||
2422 | return 0; | |
2423 | } | |
2424 | ||
2425 | /* | |
01327610 | 2426 | * Store the expected generation for seed devices in device items. |
2b82032c | 2427 | */ |
5c466629 | 2428 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2429 | { |
5c466629 | 2430 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2431 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2432 | struct btrfs_path *path; |
2433 | struct extent_buffer *leaf; | |
2434 | struct btrfs_dev_item *dev_item; | |
2435 | struct btrfs_device *device; | |
2436 | struct btrfs_key key; | |
44880fdc | 2437 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2438 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2439 | u64 devid; | |
2440 | int ret; | |
2441 | ||
2442 | path = btrfs_alloc_path(); | |
2443 | if (!path) | |
2444 | return -ENOMEM; | |
2445 | ||
2b82032c YZ |
2446 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2447 | key.offset = 0; | |
2448 | key.type = BTRFS_DEV_ITEM_KEY; | |
2449 | ||
2450 | while (1) { | |
2451 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2452 | if (ret < 0) | |
2453 | goto error; | |
2454 | ||
2455 | leaf = path->nodes[0]; | |
2456 | next_slot: | |
2457 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2458 | ret = btrfs_next_leaf(root, path); | |
2459 | if (ret > 0) | |
2460 | break; | |
2461 | if (ret < 0) | |
2462 | goto error; | |
2463 | leaf = path->nodes[0]; | |
2464 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2465 | btrfs_release_path(path); |
2b82032c YZ |
2466 | continue; |
2467 | } | |
2468 | ||
2469 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2470 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2471 | key.type != BTRFS_DEV_ITEM_KEY) | |
2472 | break; | |
2473 | ||
2474 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2475 | struct btrfs_dev_item); | |
2476 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2477 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2478 | BTRFS_UUID_SIZE); |
1473b24e | 2479 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2480 | BTRFS_FSID_SIZE); |
e4319cd9 | 2481 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2482 | fs_uuid, true); |
79787eaa | 2483 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2484 | |
2485 | if (device->fs_devices->seeding) { | |
2486 | btrfs_set_device_generation(leaf, dev_item, | |
2487 | device->generation); | |
2488 | btrfs_mark_buffer_dirty(leaf); | |
2489 | } | |
2490 | ||
2491 | path->slots[0]++; | |
2492 | goto next_slot; | |
2493 | } | |
2494 | ret = 0; | |
2495 | error: | |
2496 | btrfs_free_path(path); | |
2497 | return ret; | |
2498 | } | |
2499 | ||
da353f6b | 2500 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2501 | { |
5112febb | 2502 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2503 | struct request_queue *q; |
788f20eb CM |
2504 | struct btrfs_trans_handle *trans; |
2505 | struct btrfs_device *device; | |
2506 | struct block_device *bdev; | |
0b246afa | 2507 | struct super_block *sb = fs_info->sb; |
606686ee | 2508 | struct rcu_string *name; |
5da54bc1 | 2509 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2510 | u64 orig_super_total_bytes; |
2511 | u64 orig_super_num_devices; | |
2b82032c | 2512 | int seeding_dev = 0; |
788f20eb | 2513 | int ret = 0; |
7132a262 | 2514 | bool unlocked = false; |
788f20eb | 2515 | |
5da54bc1 | 2516 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2517 | return -EROFS; |
788f20eb | 2518 | |
a5d16333 | 2519 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2520 | fs_info->bdev_holder); |
7f59203a JB |
2521 | if (IS_ERR(bdev)) |
2522 | return PTR_ERR(bdev); | |
a2135011 | 2523 | |
5da54bc1 | 2524 | if (fs_devices->seeding) { |
2b82032c YZ |
2525 | seeding_dev = 1; |
2526 | down_write(&sb->s_umount); | |
2527 | mutex_lock(&uuid_mutex); | |
2528 | } | |
2529 | ||
8c8bee1d | 2530 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2531 | |
5da54bc1 | 2532 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2533 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2534 | if (device->bdev == bdev) { |
2535 | ret = -EEXIST; | |
d25628bd | 2536 | mutex_unlock( |
5da54bc1 | 2537 | &fs_devices->device_list_mutex); |
2b82032c | 2538 | goto error; |
788f20eb CM |
2539 | } |
2540 | } | |
5da54bc1 | 2541 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2542 | |
0b246afa | 2543 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2544 | if (IS_ERR(device)) { |
788f20eb | 2545 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2546 | ret = PTR_ERR(device); |
2b82032c | 2547 | goto error; |
788f20eb CM |
2548 | } |
2549 | ||
78f2c9e6 | 2550 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2551 | if (!name) { |
2b82032c | 2552 | ret = -ENOMEM; |
5c4cf6c9 | 2553 | goto error_free_device; |
788f20eb | 2554 | } |
606686ee | 2555 | rcu_assign_pointer(device->name, name); |
2b82032c | 2556 | |
a22285a6 | 2557 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2558 | if (IS_ERR(trans)) { |
98d5dc13 | 2559 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2560 | goto error_free_device; |
98d5dc13 TI |
2561 | } |
2562 | ||
d5e2003c | 2563 | q = bdev_get_queue(bdev); |
ebbede42 | 2564 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2565 | device->generation = trans->transid; |
0b246afa JM |
2566 | device->io_width = fs_info->sectorsize; |
2567 | device->io_align = fs_info->sectorsize; | |
2568 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2569 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2570 | fs_info->sectorsize); | |
2cc3c559 | 2571 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2572 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2573 | device->fs_info = fs_info; |
788f20eb | 2574 | device->bdev = bdev; |
e12c9621 | 2575 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2576 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2577 | device->mode = FMODE_EXCL; |
27087f37 | 2578 | device->dev_stats_valid = 1; |
9f6d2510 | 2579 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2580 | |
2b82032c | 2581 | if (seeding_dev) { |
1751e8a6 | 2582 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2583 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2584 | if (ret) { |
2585 | btrfs_abort_transaction(trans, ret); | |
2586 | goto error_trans; | |
2587 | } | |
2b82032c | 2588 | } |
788f20eb | 2589 | |
5da54bc1 | 2590 | device->fs_devices = fs_devices; |
e5e9a520 | 2591 | |
5da54bc1 | 2592 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2593 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2594 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2595 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2596 | fs_devices->num_devices++; | |
2597 | fs_devices->open_devices++; | |
2598 | fs_devices->rw_devices++; | |
2599 | fs_devices->total_devices++; | |
2600 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2601 | |
a5ed45f8 | 2602 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2603 | |
e884f4f0 | 2604 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2605 | fs_devices->rotating = true; |
c289811c | 2606 | |
39379faa | 2607 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2608 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2609 | round_down(orig_super_total_bytes + device->total_bytes, |
2610 | fs_info->sectorsize)); | |
788f20eb | 2611 | |
39379faa NA |
2612 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2613 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2614 | orig_super_num_devices + 1); | |
0d39376a AJ |
2615 | |
2616 | /* add sysfs device entry */ | |
f3cd2c58 | 2617 | btrfs_sysfs_add_devices_dir(fs_devices, device); |
0d39376a | 2618 | |
2196d6e8 MX |
2619 | /* |
2620 | * we've got more storage, clear any full flags on the space | |
2621 | * infos | |
2622 | */ | |
0b246afa | 2623 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2624 | |
34441361 | 2625 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2626 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2627 | |
2b82032c | 2628 | if (seeding_dev) { |
34441361 | 2629 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2630 | ret = init_first_rw_device(trans); |
34441361 | 2631 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2632 | if (ret) { |
66642832 | 2633 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2634 | goto error_sysfs; |
005d6427 | 2635 | } |
2196d6e8 MX |
2636 | } |
2637 | ||
8e87e856 | 2638 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2639 | if (ret) { |
66642832 | 2640 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2641 | goto error_sysfs; |
2196d6e8 MX |
2642 | } |
2643 | ||
2644 | if (seeding_dev) { | |
5c466629 | 2645 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2646 | if (ret) { |
66642832 | 2647 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2648 | goto error_sysfs; |
005d6427 | 2649 | } |
b2373f25 | 2650 | |
f93c3997 DS |
2651 | btrfs_sysfs_update_sprout_fsid(fs_devices, |
2652 | fs_info->fs_devices->fsid); | |
2b82032c YZ |
2653 | } |
2654 | ||
3a45bb20 | 2655 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2656 | |
2b82032c YZ |
2657 | if (seeding_dev) { |
2658 | mutex_unlock(&uuid_mutex); | |
2659 | up_write(&sb->s_umount); | |
7132a262 | 2660 | unlocked = true; |
788f20eb | 2661 | |
79787eaa JM |
2662 | if (ret) /* transaction commit */ |
2663 | return ret; | |
2664 | ||
2ff7e61e | 2665 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2666 | if (ret < 0) |
0b246afa | 2667 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2668 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2669 | trans = btrfs_attach_transaction(root); |
2670 | if (IS_ERR(trans)) { | |
2671 | if (PTR_ERR(trans) == -ENOENT) | |
2672 | return 0; | |
7132a262 AJ |
2673 | ret = PTR_ERR(trans); |
2674 | trans = NULL; | |
2675 | goto error_sysfs; | |
671415b7 | 2676 | } |
3a45bb20 | 2677 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2678 | } |
c9e9f97b | 2679 | |
7f551d96 AJ |
2680 | /* |
2681 | * Now that we have written a new super block to this device, check all | |
2682 | * other fs_devices list if device_path alienates any other scanned | |
2683 | * device. | |
2684 | * We can ignore the return value as it typically returns -EINVAL and | |
2685 | * only succeeds if the device was an alien. | |
2686 | */ | |
2687 | btrfs_forget_devices(device_path); | |
2688 | ||
2689 | /* Update ctime/mtime for blkid or udev */ | |
5a1972bd | 2690 | update_dev_time(device_path); |
7f551d96 | 2691 | |
2b82032c | 2692 | return ret; |
79787eaa | 2693 | |
d31c32f6 | 2694 | error_sysfs: |
f3cd2c58 | 2695 | btrfs_sysfs_remove_devices_dir(fs_devices, device); |
39379faa NA |
2696 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2697 | mutex_lock(&fs_info->chunk_mutex); | |
2698 | list_del_rcu(&device->dev_list); | |
2699 | list_del(&device->dev_alloc_list); | |
2700 | fs_info->fs_devices->num_devices--; | |
2701 | fs_info->fs_devices->open_devices--; | |
2702 | fs_info->fs_devices->rw_devices--; | |
2703 | fs_info->fs_devices->total_devices--; | |
2704 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2705 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2706 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2707 | orig_super_total_bytes); | |
2708 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2709 | orig_super_num_devices); | |
2710 | mutex_unlock(&fs_info->chunk_mutex); | |
2711 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2712 | error_trans: |
0af2c4bf | 2713 | if (seeding_dev) |
1751e8a6 | 2714 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2715 | if (trans) |
2716 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2717 | error_free_device: |
a425f9d4 | 2718 | btrfs_free_device(device); |
2b82032c | 2719 | error: |
e525fd89 | 2720 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2721 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2722 | mutex_unlock(&uuid_mutex); |
2723 | up_write(&sb->s_umount); | |
2724 | } | |
c9e9f97b | 2725 | return ret; |
788f20eb CM |
2726 | } |
2727 | ||
d397712b CM |
2728 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2729 | struct btrfs_device *device) | |
0b86a832 CM |
2730 | { |
2731 | int ret; | |
2732 | struct btrfs_path *path; | |
0b246afa | 2733 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2734 | struct btrfs_dev_item *dev_item; |
2735 | struct extent_buffer *leaf; | |
2736 | struct btrfs_key key; | |
2737 | ||
0b86a832 CM |
2738 | path = btrfs_alloc_path(); |
2739 | if (!path) | |
2740 | return -ENOMEM; | |
2741 | ||
2742 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2743 | key.type = BTRFS_DEV_ITEM_KEY; | |
2744 | key.offset = device->devid; | |
2745 | ||
2746 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2747 | if (ret < 0) | |
2748 | goto out; | |
2749 | ||
2750 | if (ret > 0) { | |
2751 | ret = -ENOENT; | |
2752 | goto out; | |
2753 | } | |
2754 | ||
2755 | leaf = path->nodes[0]; | |
2756 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2757 | ||
2758 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2759 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2760 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2761 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2762 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2763 | btrfs_set_device_total_bytes(leaf, dev_item, |
2764 | btrfs_device_get_disk_total_bytes(device)); | |
2765 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2766 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2767 | btrfs_mark_buffer_dirty(leaf); |
2768 | ||
2769 | out: | |
2770 | btrfs_free_path(path); | |
2771 | return ret; | |
2772 | } | |
2773 | ||
2196d6e8 | 2774 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2775 | struct btrfs_device *device, u64 new_size) |
2776 | { | |
0b246afa JM |
2777 | struct btrfs_fs_info *fs_info = device->fs_info; |
2778 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2779 | u64 old_total; |
2780 | u64 diff; | |
8f18cf13 | 2781 | |
ebbede42 | 2782 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2783 | return -EACCES; |
2196d6e8 | 2784 | |
7dfb8be1 NB |
2785 | new_size = round_down(new_size, fs_info->sectorsize); |
2786 | ||
34441361 | 2787 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2788 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2789 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2790 | |
63a212ab | 2791 | if (new_size <= device->total_bytes || |
401e29c1 | 2792 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2793 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2794 | return -EINVAL; |
2196d6e8 | 2795 | } |
2b82032c | 2796 | |
7dfb8be1 NB |
2797 | btrfs_set_super_total_bytes(super_copy, |
2798 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2799 | device->fs_devices->total_rw_bytes += diff; |
2800 | ||
7cc8e58d MX |
2801 | btrfs_device_set_total_bytes(device, new_size); |
2802 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2803 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2804 | if (list_empty(&device->post_commit_list)) |
2805 | list_add_tail(&device->post_commit_list, | |
2806 | &trans->transaction->dev_update_list); | |
34441361 | 2807 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2808 | |
8f18cf13 CM |
2809 | return btrfs_update_device(trans, device); |
2810 | } | |
2811 | ||
f4208794 | 2812 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2813 | { |
f4208794 | 2814 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2815 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2816 | int ret; |
2817 | struct btrfs_path *path; | |
2818 | struct btrfs_key key; | |
2819 | ||
8f18cf13 CM |
2820 | path = btrfs_alloc_path(); |
2821 | if (!path) | |
2822 | return -ENOMEM; | |
2823 | ||
408fbf19 | 2824 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2825 | key.offset = chunk_offset; |
2826 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2827 | ||
2828 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2829 | if (ret < 0) |
2830 | goto out; | |
2831 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2832 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2833 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2834 | ret = -ENOENT; |
2835 | goto out; | |
2836 | } | |
8f18cf13 CM |
2837 | |
2838 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2839 | if (ret < 0) |
0b246afa JM |
2840 | btrfs_handle_fs_error(fs_info, ret, |
2841 | "Failed to delete chunk item."); | |
79787eaa | 2842 | out: |
8f18cf13 | 2843 | btrfs_free_path(path); |
65a246c5 | 2844 | return ret; |
8f18cf13 CM |
2845 | } |
2846 | ||
408fbf19 | 2847 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2848 | { |
0b246afa | 2849 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2850 | struct btrfs_disk_key *disk_key; |
2851 | struct btrfs_chunk *chunk; | |
2852 | u8 *ptr; | |
2853 | int ret = 0; | |
2854 | u32 num_stripes; | |
2855 | u32 array_size; | |
2856 | u32 len = 0; | |
2857 | u32 cur; | |
2858 | struct btrfs_key key; | |
2859 | ||
34441361 | 2860 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2861 | array_size = btrfs_super_sys_array_size(super_copy); |
2862 | ||
2863 | ptr = super_copy->sys_chunk_array; | |
2864 | cur = 0; | |
2865 | ||
2866 | while (cur < array_size) { | |
2867 | disk_key = (struct btrfs_disk_key *)ptr; | |
2868 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2869 | ||
2870 | len = sizeof(*disk_key); | |
2871 | ||
2872 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2873 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2874 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2875 | len += btrfs_chunk_item_size(num_stripes); | |
2876 | } else { | |
2877 | ret = -EIO; | |
2878 | break; | |
2879 | } | |
408fbf19 | 2880 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2881 | key.offset == chunk_offset) { |
2882 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2883 | array_size -= len; | |
2884 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2885 | } else { | |
2886 | ptr += len; | |
2887 | cur += len; | |
2888 | } | |
2889 | } | |
34441361 | 2890 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2891 | return ret; |
2892 | } | |
2893 | ||
60ca842e OS |
2894 | /* |
2895 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2896 | * @logical: Logical block offset in bytes. | |
2897 | * @length: Length of extent in bytes. | |
2898 | * | |
2899 | * Return: Chunk mapping or ERR_PTR. | |
2900 | */ | |
2901 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2902 | u64 logical, u64 length) | |
592d92ee LB |
2903 | { |
2904 | struct extent_map_tree *em_tree; | |
2905 | struct extent_map *em; | |
2906 | ||
c8bf1b67 | 2907 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2908 | read_lock(&em_tree->lock); |
2909 | em = lookup_extent_mapping(em_tree, logical, length); | |
2910 | read_unlock(&em_tree->lock); | |
2911 | ||
2912 | if (!em) { | |
2913 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2914 | logical, length); | |
2915 | return ERR_PTR(-EINVAL); | |
2916 | } | |
2917 | ||
2918 | if (em->start > logical || em->start + em->len < logical) { | |
2919 | btrfs_crit(fs_info, | |
2920 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2921 | logical, length, em->start, em->start + em->len); | |
2922 | free_extent_map(em); | |
2923 | return ERR_PTR(-EINVAL); | |
2924 | } | |
2925 | ||
2926 | /* callers are responsible for dropping em's ref. */ | |
2927 | return em; | |
2928 | } | |
2929 | ||
97aff912 | 2930 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2931 | { |
97aff912 | 2932 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2933 | struct extent_map *em; |
2934 | struct map_lookup *map; | |
2196d6e8 | 2935 | u64 dev_extent_len = 0; |
47ab2a6c | 2936 | int i, ret = 0; |
0b246afa | 2937 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2938 | |
60ca842e | 2939 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2940 | if (IS_ERR(em)) { |
47ab2a6c JB |
2941 | /* |
2942 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2943 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2944 | * do anything we still error out. |
2945 | */ | |
2946 | ASSERT(0); | |
592d92ee | 2947 | return PTR_ERR(em); |
47ab2a6c | 2948 | } |
95617d69 | 2949 | map = em->map_lookup; |
34441361 | 2950 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2951 | check_system_chunk(trans, map->type); |
34441361 | 2952 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2953 | |
57ba4cb8 FM |
2954 | /* |
2955 | * Take the device list mutex to prevent races with the final phase of | |
2956 | * a device replace operation that replaces the device object associated | |
2957 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2958 | */ | |
2959 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2960 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2961 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2962 | ret = btrfs_free_dev_extent(trans, device, |
2963 | map->stripes[i].physical, | |
2964 | &dev_extent_len); | |
47ab2a6c | 2965 | if (ret) { |
57ba4cb8 | 2966 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2967 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2968 | goto out; |
2969 | } | |
a061fc8d | 2970 | |
2196d6e8 | 2971 | if (device->bytes_used > 0) { |
34441361 | 2972 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
2973 | btrfs_device_set_bytes_used(device, |
2974 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 2975 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 2976 | btrfs_clear_space_info_full(fs_info); |
34441361 | 2977 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 2978 | } |
a061fc8d | 2979 | |
64bc6c2a NB |
2980 | ret = btrfs_update_device(trans, device); |
2981 | if (ret) { | |
2982 | mutex_unlock(&fs_devices->device_list_mutex); | |
2983 | btrfs_abort_transaction(trans, ret); | |
2984 | goto out; | |
dfe25020 | 2985 | } |
8f18cf13 | 2986 | } |
57ba4cb8 FM |
2987 | mutex_unlock(&fs_devices->device_list_mutex); |
2988 | ||
f4208794 | 2989 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 2990 | if (ret) { |
66642832 | 2991 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2992 | goto out; |
2993 | } | |
8f18cf13 | 2994 | |
6bccf3ab | 2995 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2996 | |
8f18cf13 | 2997 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 2998 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 2999 | if (ret) { |
66642832 | 3000 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3001 | goto out; |
3002 | } | |
8f18cf13 CM |
3003 | } |
3004 | ||
5a98ec01 | 3005 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3006 | if (ret) { |
66642832 | 3007 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3008 | goto out; |
3009 | } | |
2b82032c | 3010 | |
47ab2a6c | 3011 | out: |
2b82032c YZ |
3012 | /* once for us */ |
3013 | free_extent_map(em); | |
47ab2a6c JB |
3014 | return ret; |
3015 | } | |
2b82032c | 3016 | |
5b4aacef | 3017 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3018 | { |
5b4aacef | 3019 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3020 | struct btrfs_trans_handle *trans; |
b0643e59 | 3021 | struct btrfs_block_group *block_group; |
47ab2a6c | 3022 | int ret; |
2b82032c | 3023 | |
67c5e7d4 FM |
3024 | /* |
3025 | * Prevent races with automatic removal of unused block groups. | |
3026 | * After we relocate and before we remove the chunk with offset | |
3027 | * chunk_offset, automatic removal of the block group can kick in, | |
3028 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3029 | * | |
3030 | * Make sure to acquire this mutex before doing a tree search (dev | |
3031 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3032 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3033 | * we release the path used to search the chunk/dev tree and before | |
3034 | * the current task acquires this mutex and calls us. | |
3035 | */ | |
a32bf9a3 | 3036 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3037 | |
47ab2a6c | 3038 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3039 | btrfs_scrub_pause(fs_info); |
0b246afa | 3040 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3041 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3042 | if (ret) |
3043 | return ret; | |
3044 | ||
b0643e59 DZ |
3045 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3046 | if (!block_group) | |
3047 | return -ENOENT; | |
3048 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
3049 | btrfs_put_block_group(block_group); | |
3050 | ||
19c4d2f9 CM |
3051 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3052 | chunk_offset); | |
3053 | if (IS_ERR(trans)) { | |
3054 | ret = PTR_ERR(trans); | |
3055 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3056 | return ret; | |
3057 | } | |
3058 | ||
47ab2a6c | 3059 | /* |
19c4d2f9 CM |
3060 | * step two, delete the device extents and the |
3061 | * chunk tree entries | |
47ab2a6c | 3062 | */ |
97aff912 | 3063 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3064 | btrfs_end_transaction(trans); |
19c4d2f9 | 3065 | return ret; |
2b82032c YZ |
3066 | } |
3067 | ||
2ff7e61e | 3068 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3069 | { |
0b246afa | 3070 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3071 | struct btrfs_path *path; |
3072 | struct extent_buffer *leaf; | |
3073 | struct btrfs_chunk *chunk; | |
3074 | struct btrfs_key key; | |
3075 | struct btrfs_key found_key; | |
2b82032c | 3076 | u64 chunk_type; |
ba1bf481 JB |
3077 | bool retried = false; |
3078 | int failed = 0; | |
2b82032c YZ |
3079 | int ret; |
3080 | ||
3081 | path = btrfs_alloc_path(); | |
3082 | if (!path) | |
3083 | return -ENOMEM; | |
3084 | ||
ba1bf481 | 3085 | again: |
2b82032c YZ |
3086 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3087 | key.offset = (u64)-1; | |
3088 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3089 | ||
3090 | while (1) { | |
0b246afa | 3091 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3092 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3093 | if (ret < 0) { |
0b246afa | 3094 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3095 | goto error; |
67c5e7d4 | 3096 | } |
79787eaa | 3097 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3098 | |
3099 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3100 | key.type); | |
67c5e7d4 | 3101 | if (ret) |
0b246afa | 3102 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3103 | if (ret < 0) |
3104 | goto error; | |
3105 | if (ret > 0) | |
3106 | break; | |
1a40e23b | 3107 | |
2b82032c YZ |
3108 | leaf = path->nodes[0]; |
3109 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3110 | |
2b82032c YZ |
3111 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3112 | struct btrfs_chunk); | |
3113 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3114 | btrfs_release_path(path); |
8f18cf13 | 3115 | |
2b82032c | 3116 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3117 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3118 | if (ret == -ENOSPC) |
3119 | failed++; | |
14586651 HS |
3120 | else |
3121 | BUG_ON(ret); | |
2b82032c | 3122 | } |
0b246afa | 3123 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3124 | |
2b82032c YZ |
3125 | if (found_key.offset == 0) |
3126 | break; | |
3127 | key.offset = found_key.offset - 1; | |
3128 | } | |
3129 | ret = 0; | |
ba1bf481 JB |
3130 | if (failed && !retried) { |
3131 | failed = 0; | |
3132 | retried = true; | |
3133 | goto again; | |
fae7f21c | 3134 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3135 | ret = -ENOSPC; |
3136 | } | |
2b82032c YZ |
3137 | error: |
3138 | btrfs_free_path(path); | |
3139 | return ret; | |
8f18cf13 CM |
3140 | } |
3141 | ||
a6f93c71 LB |
3142 | /* |
3143 | * return 1 : allocate a data chunk successfully, | |
3144 | * return <0: errors during allocating a data chunk, | |
3145 | * return 0 : no need to allocate a data chunk. | |
3146 | */ | |
3147 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3148 | u64 chunk_offset) | |
3149 | { | |
32da5386 | 3150 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3151 | u64 bytes_used; |
3152 | u64 chunk_type; | |
3153 | ||
3154 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3155 | ASSERT(cache); | |
3156 | chunk_type = cache->flags; | |
3157 | btrfs_put_block_group(cache); | |
3158 | ||
5ae21692 JT |
3159 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3160 | return 0; | |
3161 | ||
3162 | spin_lock(&fs_info->data_sinfo->lock); | |
3163 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3164 | spin_unlock(&fs_info->data_sinfo->lock); | |
3165 | ||
3166 | if (!bytes_used) { | |
3167 | struct btrfs_trans_handle *trans; | |
3168 | int ret; | |
3169 | ||
3170 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3171 | if (IS_ERR(trans)) | |
3172 | return PTR_ERR(trans); | |
3173 | ||
3174 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3175 | btrfs_end_transaction(trans); | |
3176 | if (ret < 0) | |
3177 | return ret; | |
3178 | return 1; | |
a6f93c71 | 3179 | } |
5ae21692 | 3180 | |
a6f93c71 LB |
3181 | return 0; |
3182 | } | |
3183 | ||
6bccf3ab | 3184 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3185 | struct btrfs_balance_control *bctl) |
3186 | { | |
6bccf3ab | 3187 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3188 | struct btrfs_trans_handle *trans; |
3189 | struct btrfs_balance_item *item; | |
3190 | struct btrfs_disk_balance_args disk_bargs; | |
3191 | struct btrfs_path *path; | |
3192 | struct extent_buffer *leaf; | |
3193 | struct btrfs_key key; | |
3194 | int ret, err; | |
3195 | ||
3196 | path = btrfs_alloc_path(); | |
3197 | if (!path) | |
3198 | return -ENOMEM; | |
3199 | ||
3200 | trans = btrfs_start_transaction(root, 0); | |
3201 | if (IS_ERR(trans)) { | |
3202 | btrfs_free_path(path); | |
3203 | return PTR_ERR(trans); | |
3204 | } | |
3205 | ||
3206 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3207 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3208 | key.offset = 0; |
3209 | ||
3210 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3211 | sizeof(*item)); | |
3212 | if (ret) | |
3213 | goto out; | |
3214 | ||
3215 | leaf = path->nodes[0]; | |
3216 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3217 | ||
b159fa28 | 3218 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3219 | |
3220 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3221 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3222 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3223 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3224 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3225 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3226 | ||
3227 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3228 | ||
3229 | btrfs_mark_buffer_dirty(leaf); | |
3230 | out: | |
3231 | btrfs_free_path(path); | |
3a45bb20 | 3232 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3233 | if (err && !ret) |
3234 | ret = err; | |
3235 | return ret; | |
3236 | } | |
3237 | ||
6bccf3ab | 3238 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3239 | { |
6bccf3ab | 3240 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3241 | struct btrfs_trans_handle *trans; |
3242 | struct btrfs_path *path; | |
3243 | struct btrfs_key key; | |
3244 | int ret, err; | |
3245 | ||
3246 | path = btrfs_alloc_path(); | |
3247 | if (!path) | |
3248 | return -ENOMEM; | |
3249 | ||
3502a8c0 | 3250 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3251 | if (IS_ERR(trans)) { |
3252 | btrfs_free_path(path); | |
3253 | return PTR_ERR(trans); | |
3254 | } | |
3255 | ||
3256 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3257 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3258 | key.offset = 0; |
3259 | ||
3260 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3261 | if (ret < 0) | |
3262 | goto out; | |
3263 | if (ret > 0) { | |
3264 | ret = -ENOENT; | |
3265 | goto out; | |
3266 | } | |
3267 | ||
3268 | ret = btrfs_del_item(trans, root, path); | |
3269 | out: | |
3270 | btrfs_free_path(path); | |
3a45bb20 | 3271 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3272 | if (err && !ret) |
3273 | ret = err; | |
3274 | return ret; | |
3275 | } | |
3276 | ||
59641015 ID |
3277 | /* |
3278 | * This is a heuristic used to reduce the number of chunks balanced on | |
3279 | * resume after balance was interrupted. | |
3280 | */ | |
3281 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3282 | { | |
3283 | /* | |
3284 | * Turn on soft mode for chunk types that were being converted. | |
3285 | */ | |
3286 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3287 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3288 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3289 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3290 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3291 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3292 | ||
3293 | /* | |
3294 | * Turn on usage filter if is not already used. The idea is | |
3295 | * that chunks that we have already balanced should be | |
3296 | * reasonably full. Don't do it for chunks that are being | |
3297 | * converted - that will keep us from relocating unconverted | |
3298 | * (albeit full) chunks. | |
3299 | */ | |
3300 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3301 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3302 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3303 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3304 | bctl->data.usage = 90; | |
3305 | } | |
3306 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3307 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3308 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3309 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3310 | bctl->sys.usage = 90; | |
3311 | } | |
3312 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3313 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3314 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3315 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3316 | bctl->meta.usage = 90; | |
3317 | } | |
3318 | } | |
3319 | ||
149196a2 DS |
3320 | /* |
3321 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3322 | */ | |
3323 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3324 | { |
3325 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3326 | int ret; |
c9e9f97b ID |
3327 | |
3328 | BUG_ON(!fs_info->balance_ctl); | |
3329 | ||
3330 | spin_lock(&fs_info->balance_lock); | |
3331 | fs_info->balance_ctl = NULL; | |
3332 | spin_unlock(&fs_info->balance_lock); | |
3333 | ||
3334 | kfree(bctl); | |
149196a2 DS |
3335 | ret = del_balance_item(fs_info); |
3336 | if (ret) | |
3337 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3338 | } |
3339 | ||
ed25e9b2 ID |
3340 | /* |
3341 | * Balance filters. Return 1 if chunk should be filtered out | |
3342 | * (should not be balanced). | |
3343 | */ | |
899c81ea | 3344 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3345 | struct btrfs_balance_args *bargs) |
3346 | { | |
899c81ea ID |
3347 | chunk_type = chunk_to_extended(chunk_type) & |
3348 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3349 | |
899c81ea | 3350 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3351 | return 0; |
3352 | ||
3353 | return 1; | |
3354 | } | |
3355 | ||
dba72cb3 | 3356 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3357 | struct btrfs_balance_args *bargs) |
bc309467 | 3358 | { |
32da5386 | 3359 | struct btrfs_block_group *cache; |
bc309467 DS |
3360 | u64 chunk_used; |
3361 | u64 user_thresh_min; | |
3362 | u64 user_thresh_max; | |
3363 | int ret = 1; | |
3364 | ||
3365 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3366 | chunk_used = cache->used; |
bc309467 DS |
3367 | |
3368 | if (bargs->usage_min == 0) | |
3369 | user_thresh_min = 0; | |
3370 | else | |
b3470b5d DS |
3371 | user_thresh_min = div_factor_fine(cache->length, |
3372 | bargs->usage_min); | |
bc309467 DS |
3373 | |
3374 | if (bargs->usage_max == 0) | |
3375 | user_thresh_max = 1; | |
3376 | else if (bargs->usage_max > 100) | |
b3470b5d | 3377 | user_thresh_max = cache->length; |
bc309467 | 3378 | else |
b3470b5d DS |
3379 | user_thresh_max = div_factor_fine(cache->length, |
3380 | bargs->usage_max); | |
bc309467 DS |
3381 | |
3382 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3383 | ret = 0; | |
3384 | ||
3385 | btrfs_put_block_group(cache); | |
3386 | return ret; | |
3387 | } | |
3388 | ||
dba72cb3 | 3389 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3390 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3391 | { |
32da5386 | 3392 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3393 | u64 chunk_used, user_thresh; |
3394 | int ret = 1; | |
3395 | ||
3396 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3397 | chunk_used = cache->used; |
5ce5b3c0 | 3398 | |
bc309467 | 3399 | if (bargs->usage_min == 0) |
3e39cea6 | 3400 | user_thresh = 1; |
a105bb88 | 3401 | else if (bargs->usage > 100) |
b3470b5d | 3402 | user_thresh = cache->length; |
a105bb88 | 3403 | else |
b3470b5d | 3404 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3405 | |
5ce5b3c0 ID |
3406 | if (chunk_used < user_thresh) |
3407 | ret = 0; | |
3408 | ||
3409 | btrfs_put_block_group(cache); | |
3410 | return ret; | |
3411 | } | |
3412 | ||
409d404b ID |
3413 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3414 | struct btrfs_chunk *chunk, | |
3415 | struct btrfs_balance_args *bargs) | |
3416 | { | |
3417 | struct btrfs_stripe *stripe; | |
3418 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3419 | int i; | |
3420 | ||
3421 | for (i = 0; i < num_stripes; i++) { | |
3422 | stripe = btrfs_stripe_nr(chunk, i); | |
3423 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3424 | return 0; | |
3425 | } | |
3426 | ||
3427 | return 1; | |
3428 | } | |
3429 | ||
946c9256 DS |
3430 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3431 | { | |
3432 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3433 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3434 | const int nparity = btrfs_raid_array[index].nparity; | |
3435 | ||
3436 | if (nparity) | |
3437 | return num_stripes - nparity; | |
3438 | else | |
3439 | return num_stripes / ncopies; | |
3440 | } | |
3441 | ||
94e60d5a ID |
3442 | /* [pstart, pend) */ |
3443 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3444 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3445 | struct btrfs_balance_args *bargs) |
3446 | { | |
3447 | struct btrfs_stripe *stripe; | |
3448 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3449 | u64 stripe_offset; | |
3450 | u64 stripe_length; | |
946c9256 | 3451 | u64 type; |
94e60d5a ID |
3452 | int factor; |
3453 | int i; | |
3454 | ||
3455 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3456 | return 0; | |
3457 | ||
946c9256 DS |
3458 | type = btrfs_chunk_type(leaf, chunk); |
3459 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3460 | |
3461 | for (i = 0; i < num_stripes; i++) { | |
3462 | stripe = btrfs_stripe_nr(chunk, i); | |
3463 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3464 | continue; | |
3465 | ||
3466 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3467 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3468 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3469 | |
3470 | if (stripe_offset < bargs->pend && | |
3471 | stripe_offset + stripe_length > bargs->pstart) | |
3472 | return 0; | |
3473 | } | |
3474 | ||
3475 | return 1; | |
3476 | } | |
3477 | ||
ea67176a ID |
3478 | /* [vstart, vend) */ |
3479 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3480 | struct btrfs_chunk *chunk, | |
3481 | u64 chunk_offset, | |
3482 | struct btrfs_balance_args *bargs) | |
3483 | { | |
3484 | if (chunk_offset < bargs->vend && | |
3485 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3486 | /* at least part of the chunk is inside this vrange */ | |
3487 | return 0; | |
3488 | ||
3489 | return 1; | |
3490 | } | |
3491 | ||
dee32d0a GAP |
3492 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3493 | struct btrfs_chunk *chunk, | |
3494 | struct btrfs_balance_args *bargs) | |
3495 | { | |
3496 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3497 | ||
3498 | if (bargs->stripes_min <= num_stripes | |
3499 | && num_stripes <= bargs->stripes_max) | |
3500 | return 0; | |
3501 | ||
3502 | return 1; | |
3503 | } | |
3504 | ||
899c81ea | 3505 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3506 | struct btrfs_balance_args *bargs) |
3507 | { | |
3508 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3509 | return 0; | |
3510 | ||
899c81ea ID |
3511 | chunk_type = chunk_to_extended(chunk_type) & |
3512 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3513 | |
899c81ea | 3514 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3515 | return 1; |
3516 | ||
3517 | return 0; | |
3518 | } | |
3519 | ||
6ec0896c | 3520 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3521 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3522 | { | |
6ec0896c | 3523 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3524 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3525 | struct btrfs_balance_args *bargs = NULL; |
3526 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3527 | ||
3528 | /* type filter */ | |
3529 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3530 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3531 | return 0; | |
3532 | } | |
3533 | ||
3534 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3535 | bargs = &bctl->data; | |
3536 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3537 | bargs = &bctl->sys; | |
3538 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3539 | bargs = &bctl->meta; | |
3540 | ||
ed25e9b2 ID |
3541 | /* profiles filter */ |
3542 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3543 | chunk_profiles_filter(chunk_type, bargs)) { | |
3544 | return 0; | |
5ce5b3c0 ID |
3545 | } |
3546 | ||
3547 | /* usage filter */ | |
3548 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3549 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3550 | return 0; |
bc309467 | 3551 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3552 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3553 | return 0; |
409d404b ID |
3554 | } |
3555 | ||
3556 | /* devid filter */ | |
3557 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3558 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3559 | return 0; | |
94e60d5a ID |
3560 | } |
3561 | ||
3562 | /* drange filter, makes sense only with devid filter */ | |
3563 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3564 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3565 | return 0; |
ea67176a ID |
3566 | } |
3567 | ||
3568 | /* vrange filter */ | |
3569 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3570 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3571 | return 0; | |
ed25e9b2 ID |
3572 | } |
3573 | ||
dee32d0a GAP |
3574 | /* stripes filter */ |
3575 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3576 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3577 | return 0; | |
3578 | } | |
3579 | ||
cfa4c961 ID |
3580 | /* soft profile changing mode */ |
3581 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3582 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3583 | return 0; | |
3584 | } | |
3585 | ||
7d824b6f DS |
3586 | /* |
3587 | * limited by count, must be the last filter | |
3588 | */ | |
3589 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3590 | if (bargs->limit == 0) | |
3591 | return 0; | |
3592 | else | |
3593 | bargs->limit--; | |
12907fc7 DS |
3594 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3595 | /* | |
3596 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3597 | * determined here because we do not have the global information |
12907fc7 DS |
3598 | * about the count of all chunks that satisfy the filters. |
3599 | */ | |
3600 | if (bargs->limit_max == 0) | |
3601 | return 0; | |
3602 | else | |
3603 | bargs->limit_max--; | |
7d824b6f DS |
3604 | } |
3605 | ||
f43ffb60 ID |
3606 | return 1; |
3607 | } | |
3608 | ||
c9e9f97b | 3609 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3610 | { |
19a39dce | 3611 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3612 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3613 | u64 chunk_type; |
f43ffb60 | 3614 | struct btrfs_chunk *chunk; |
5a488b9d | 3615 | struct btrfs_path *path = NULL; |
ec44a35c | 3616 | struct btrfs_key key; |
ec44a35c | 3617 | struct btrfs_key found_key; |
f43ffb60 ID |
3618 | struct extent_buffer *leaf; |
3619 | int slot; | |
c9e9f97b ID |
3620 | int ret; |
3621 | int enospc_errors = 0; | |
19a39dce | 3622 | bool counting = true; |
12907fc7 | 3623 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3624 | u64 limit_data = bctl->data.limit; |
3625 | u64 limit_meta = bctl->meta.limit; | |
3626 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3627 | u32 count_data = 0; |
3628 | u32 count_meta = 0; | |
3629 | u32 count_sys = 0; | |
2c9fe835 | 3630 | int chunk_reserved = 0; |
ec44a35c | 3631 | |
ec44a35c | 3632 | path = btrfs_alloc_path(); |
17e9f796 MF |
3633 | if (!path) { |
3634 | ret = -ENOMEM; | |
3635 | goto error; | |
3636 | } | |
19a39dce ID |
3637 | |
3638 | /* zero out stat counters */ | |
3639 | spin_lock(&fs_info->balance_lock); | |
3640 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3641 | spin_unlock(&fs_info->balance_lock); | |
3642 | again: | |
7d824b6f | 3643 | if (!counting) { |
12907fc7 DS |
3644 | /* |
3645 | * The single value limit and min/max limits use the same bytes | |
3646 | * in the | |
3647 | */ | |
7d824b6f DS |
3648 | bctl->data.limit = limit_data; |
3649 | bctl->meta.limit = limit_meta; | |
3650 | bctl->sys.limit = limit_sys; | |
3651 | } | |
ec44a35c CM |
3652 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3653 | key.offset = (u64)-1; | |
3654 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3655 | ||
d397712b | 3656 | while (1) { |
19a39dce | 3657 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3658 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3659 | ret = -ECANCELED; |
3660 | goto error; | |
3661 | } | |
3662 | ||
67c5e7d4 | 3663 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3664 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3665 | if (ret < 0) { |
3666 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3667 | goto error; |
67c5e7d4 | 3668 | } |
ec44a35c CM |
3669 | |
3670 | /* | |
3671 | * this shouldn't happen, it means the last relocate | |
3672 | * failed | |
3673 | */ | |
3674 | if (ret == 0) | |
c9e9f97b | 3675 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3676 | |
3677 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3678 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3679 | if (ret) { |
67c5e7d4 | 3680 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3681 | ret = 0; |
ec44a35c | 3682 | break; |
c9e9f97b | 3683 | } |
7d9eb12c | 3684 | |
f43ffb60 ID |
3685 | leaf = path->nodes[0]; |
3686 | slot = path->slots[0]; | |
3687 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3688 | |
67c5e7d4 FM |
3689 | if (found_key.objectid != key.objectid) { |
3690 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3691 | break; |
67c5e7d4 | 3692 | } |
7d9eb12c | 3693 | |
f43ffb60 | 3694 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3695 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3696 | |
19a39dce ID |
3697 | if (!counting) { |
3698 | spin_lock(&fs_info->balance_lock); | |
3699 | bctl->stat.considered++; | |
3700 | spin_unlock(&fs_info->balance_lock); | |
3701 | } | |
3702 | ||
6ec0896c | 3703 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3704 | |
b3b4aa74 | 3705 | btrfs_release_path(path); |
67c5e7d4 FM |
3706 | if (!ret) { |
3707 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3708 | goto loop; |
67c5e7d4 | 3709 | } |
f43ffb60 | 3710 | |
19a39dce | 3711 | if (counting) { |
67c5e7d4 | 3712 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3713 | spin_lock(&fs_info->balance_lock); |
3714 | bctl->stat.expected++; | |
3715 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3716 | |
3717 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3718 | count_data++; | |
3719 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3720 | count_sys++; | |
3721 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3722 | count_meta++; | |
3723 | ||
3724 | goto loop; | |
3725 | } | |
3726 | ||
3727 | /* | |
3728 | * Apply limit_min filter, no need to check if the LIMITS | |
3729 | * filter is used, limit_min is 0 by default | |
3730 | */ | |
3731 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3732 | count_data < bctl->data.limit_min) | |
3733 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3734 | count_meta < bctl->meta.limit_min) | |
3735 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3736 | count_sys < bctl->sys.limit_min)) { | |
3737 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3738 | goto loop; |
3739 | } | |
3740 | ||
a6f93c71 LB |
3741 | if (!chunk_reserved) { |
3742 | /* | |
3743 | * We may be relocating the only data chunk we have, | |
3744 | * which could potentially end up with losing data's | |
3745 | * raid profile, so lets allocate an empty one in | |
3746 | * advance. | |
3747 | */ | |
3748 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3749 | found_key.offset); | |
2c9fe835 ZL |
3750 | if (ret < 0) { |
3751 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3752 | goto error; | |
a6f93c71 LB |
3753 | } else if (ret == 1) { |
3754 | chunk_reserved = 1; | |
2c9fe835 | 3755 | } |
2c9fe835 ZL |
3756 | } |
3757 | ||
5b4aacef | 3758 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3759 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3760 | if (ret == -ENOSPC) { |
c9e9f97b | 3761 | enospc_errors++; |
eede2bf3 OS |
3762 | } else if (ret == -ETXTBSY) { |
3763 | btrfs_info(fs_info, | |
3764 | "skipping relocation of block group %llu due to active swapfile", | |
3765 | found_key.offset); | |
3766 | ret = 0; | |
3767 | } else if (ret) { | |
3768 | goto error; | |
19a39dce ID |
3769 | } else { |
3770 | spin_lock(&fs_info->balance_lock); | |
3771 | bctl->stat.completed++; | |
3772 | spin_unlock(&fs_info->balance_lock); | |
3773 | } | |
f43ffb60 | 3774 | loop: |
795a3321 ID |
3775 | if (found_key.offset == 0) |
3776 | break; | |
ba1bf481 | 3777 | key.offset = found_key.offset - 1; |
ec44a35c | 3778 | } |
c9e9f97b | 3779 | |
19a39dce ID |
3780 | if (counting) { |
3781 | btrfs_release_path(path); | |
3782 | counting = false; | |
3783 | goto again; | |
3784 | } | |
ec44a35c CM |
3785 | error: |
3786 | btrfs_free_path(path); | |
c9e9f97b | 3787 | if (enospc_errors) { |
efe120a0 | 3788 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3789 | enospc_errors); |
c9e9f97b ID |
3790 | if (!ret) |
3791 | ret = -ENOSPC; | |
3792 | } | |
3793 | ||
ec44a35c CM |
3794 | return ret; |
3795 | } | |
3796 | ||
0c460c0d ID |
3797 | /** |
3798 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3799 | * @flags: profile to validate | |
3800 | * @extended: if true @flags is treated as an extended profile | |
3801 | */ | |
3802 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3803 | { | |
3804 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3805 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3806 | ||
3807 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3808 | ||
3809 | /* 1) check that all other bits are zeroed */ | |
3810 | if (flags & ~mask) | |
3811 | return 0; | |
3812 | ||
3813 | /* 2) see if profile is reduced */ | |
3814 | if (flags == 0) | |
3815 | return !extended; /* "0" is valid for usual profiles */ | |
3816 | ||
c1499166 | 3817 | return has_single_bit_set(flags); |
0c460c0d ID |
3818 | } |
3819 | ||
837d5b6e ID |
3820 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3821 | { | |
a7e99c69 ID |
3822 | /* cancel requested || normal exit path */ |
3823 | return atomic_read(&fs_info->balance_cancel_req) || | |
3824 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3825 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3826 | } |
3827 | ||
5ba366c3 DS |
3828 | /* |
3829 | * Validate target profile against allowed profiles and return true if it's OK. | |
3830 | * Otherwise print the error message and return false. | |
3831 | */ | |
3832 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
3833 | const struct btrfs_balance_args *bargs, | |
3834 | u64 allowed, const char *type) | |
bdcd3c97 | 3835 | { |
5ba366c3 DS |
3836 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
3837 | return true; | |
3838 | ||
3839 | /* Profile is valid and does not have bits outside of the allowed set */ | |
3840 | if (alloc_profile_is_valid(bargs->target, 1) && | |
3841 | (bargs->target & ~allowed) == 0) | |
3842 | return true; | |
3843 | ||
3844 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
3845 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
3846 | return false; | |
bdcd3c97 AM |
3847 | } |
3848 | ||
56fc37d9 AJ |
3849 | /* |
3850 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3851 | * @size_buf including the terminating null. The output may be trimmed if it | |
3852 | * does not fit into the provided buffer. | |
3853 | */ | |
3854 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3855 | u32 size_buf) | |
3856 | { | |
3857 | int ret; | |
3858 | u32 size_bp = size_buf; | |
3859 | char *bp = buf; | |
3860 | u64 flags = bargs->flags; | |
3861 | char tmp_buf[128] = {'\0'}; | |
3862 | ||
3863 | if (!flags) | |
3864 | return; | |
3865 | ||
3866 | #define CHECK_APPEND_NOARG(a) \ | |
3867 | do { \ | |
3868 | ret = snprintf(bp, size_bp, (a)); \ | |
3869 | if (ret < 0 || ret >= size_bp) \ | |
3870 | goto out_overflow; \ | |
3871 | size_bp -= ret; \ | |
3872 | bp += ret; \ | |
3873 | } while (0) | |
3874 | ||
3875 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3876 | do { \ | |
3877 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3878 | if (ret < 0 || ret >= size_bp) \ | |
3879 | goto out_overflow; \ | |
3880 | size_bp -= ret; \ | |
3881 | bp += ret; \ | |
3882 | } while (0) | |
3883 | ||
3884 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3885 | do { \ | |
3886 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3887 | if (ret < 0 || ret >= size_bp) \ | |
3888 | goto out_overflow; \ | |
3889 | size_bp -= ret; \ | |
3890 | bp += ret; \ | |
3891 | } while (0) | |
3892 | ||
158da513 DS |
3893 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3894 | CHECK_APPEND_1ARG("convert=%s,", | |
3895 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3896 | |
3897 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3898 | CHECK_APPEND_NOARG("soft,"); | |
3899 | ||
3900 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3901 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3902 | sizeof(tmp_buf)); | |
3903 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3904 | } | |
3905 | ||
3906 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3907 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3908 | ||
3909 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3910 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3911 | bargs->usage_min, bargs->usage_max); | |
3912 | ||
3913 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3914 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3915 | ||
3916 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3917 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3918 | bargs->pstart, bargs->pend); | |
3919 | ||
3920 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3921 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3922 | bargs->vstart, bargs->vend); | |
3923 | ||
3924 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3925 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3926 | ||
3927 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3928 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3929 | bargs->limit_min, bargs->limit_max); | |
3930 | ||
3931 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3932 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3933 | bargs->stripes_min, bargs->stripes_max); | |
3934 | ||
3935 | #undef CHECK_APPEND_2ARG | |
3936 | #undef CHECK_APPEND_1ARG | |
3937 | #undef CHECK_APPEND_NOARG | |
3938 | ||
3939 | out_overflow: | |
3940 | ||
3941 | if (size_bp < size_buf) | |
3942 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3943 | else | |
3944 | buf[0] = '\0'; | |
3945 | } | |
3946 | ||
3947 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3948 | { | |
3949 | u32 size_buf = 1024; | |
3950 | char tmp_buf[192] = {'\0'}; | |
3951 | char *buf; | |
3952 | char *bp; | |
3953 | u32 size_bp = size_buf; | |
3954 | int ret; | |
3955 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3956 | ||
3957 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3958 | if (!buf) | |
3959 | return; | |
3960 | ||
3961 | bp = buf; | |
3962 | ||
3963 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3964 | do { \ | |
3965 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3966 | if (ret < 0 || ret >= size_bp) \ | |
3967 | goto out_overflow; \ | |
3968 | size_bp -= ret; \ | |
3969 | bp += ret; \ | |
3970 | } while (0) | |
3971 | ||
3972 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3973 | CHECK_APPEND_1ARG("%s", "-f "); | |
3974 | ||
3975 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3976 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3977 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3978 | } | |
3979 | ||
3980 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
3981 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
3982 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
3983 | } | |
3984 | ||
3985 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
3986 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
3987 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
3988 | } | |
3989 | ||
3990 | #undef CHECK_APPEND_1ARG | |
3991 | ||
3992 | out_overflow: | |
3993 | ||
3994 | if (size_bp < size_buf) | |
3995 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
3996 | btrfs_info(fs_info, "balance: %s %s", | |
3997 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
3998 | "resume" : "start", buf); | |
3999 | ||
4000 | kfree(buf); | |
4001 | } | |
4002 | ||
c9e9f97b | 4003 | /* |
dccdb07b | 4004 | * Should be called with balance mutexe held |
c9e9f97b | 4005 | */ |
6fcf6e2b DS |
4006 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4007 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4008 | struct btrfs_ioctl_balance_args *bargs) |
4009 | { | |
14506127 | 4010 | u64 meta_target, data_target; |
f43ffb60 | 4011 | u64 allowed; |
e4837f8f | 4012 | int mixed = 0; |
c9e9f97b | 4013 | int ret; |
8dabb742 | 4014 | u64 num_devices; |
de98ced9 | 4015 | unsigned seq; |
e62869be | 4016 | bool reducing_redundancy; |
081db89b | 4017 | int i; |
c9e9f97b | 4018 | |
837d5b6e | 4019 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4020 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4021 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4022 | ret = -EINVAL; |
4023 | goto out; | |
4024 | } | |
4025 | ||
e4837f8f ID |
4026 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4027 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4028 | mixed = 1; | |
4029 | ||
f43ffb60 ID |
4030 | /* |
4031 | * In case of mixed groups both data and meta should be picked, | |
4032 | * and identical options should be given for both of them. | |
4033 | */ | |
e4837f8f ID |
4034 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4035 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4036 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4037 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4038 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4039 | btrfs_err(fs_info, |
6dac13f8 | 4040 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4041 | ret = -EINVAL; |
4042 | goto out; | |
4043 | } | |
4044 | } | |
4045 | ||
b35cf1f0 JB |
4046 | /* |
4047 | * rw_devices will not change at the moment, device add/delete/replace | |
4048 | * are excluded by EXCL_OP | |
4049 | */ | |
4050 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4051 | |
4052 | /* | |
4053 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4054 | * special bit for it, to make it easier to distinguish. Thus we need | |
4055 | * to set it manually, or balance would refuse the profile. | |
4056 | */ | |
4057 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4058 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4059 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4060 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4061 | |
5ba366c3 DS |
4062 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4063 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4064 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4065 | ret = -EINVAL; |
4066 | goto out; | |
4067 | } | |
4068 | ||
6079e12c DS |
4069 | /* |
4070 | * Allow to reduce metadata or system integrity only if force set for | |
4071 | * profiles with redundancy (copies, parity) | |
4072 | */ | |
4073 | allowed = 0; | |
4074 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4075 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4076 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4077 | allowed |= btrfs_raid_array[i].bg_flag; | |
4078 | } | |
de98ced9 MX |
4079 | do { |
4080 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4081 | ||
4082 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4083 | (fs_info->avail_system_alloc_bits & allowed) && | |
4084 | !(bctl->sys.target & allowed)) || | |
4085 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4086 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4087 | !(bctl->meta.target & allowed))) |
e62869be | 4088 | reducing_redundancy = true; |
5a8067c0 | 4089 | else |
e62869be | 4090 | reducing_redundancy = false; |
5a8067c0 FM |
4091 | |
4092 | /* if we're not converting, the target field is uninitialized */ | |
4093 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4094 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4095 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4096 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4097 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4098 | |
e62869be | 4099 | if (reducing_redundancy) { |
5a8067c0 FM |
4100 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4101 | btrfs_info(fs_info, | |
e62869be | 4102 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4103 | } else { |
4104 | btrfs_err(fs_info, | |
e62869be | 4105 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4106 | ret = -EINVAL; |
4107 | goto out; | |
4108 | } | |
4109 | } | |
4110 | ||
14506127 AB |
4111 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4112 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4113 | btrfs_warn(fs_info, |
6dac13f8 | 4114 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4115 | btrfs_bg_type_to_raid_name(meta_target), |
4116 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4117 | } |
4118 | ||
9e967495 FM |
4119 | if (fs_info->send_in_progress) { |
4120 | btrfs_warn_rl(fs_info, | |
4121 | "cannot run balance while send operations are in progress (%d in progress)", | |
4122 | fs_info->send_in_progress); | |
4123 | ret = -EAGAIN; | |
4124 | goto out; | |
4125 | } | |
4126 | ||
6bccf3ab | 4127 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4128 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4129 | goto out; |
4130 | ||
59641015 ID |
4131 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4132 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4133 | BUG_ON(fs_info->balance_ctl); |
4134 | spin_lock(&fs_info->balance_lock); | |
4135 | fs_info->balance_ctl = bctl; | |
4136 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4137 | } else { |
4138 | BUG_ON(ret != -EEXIST); | |
4139 | spin_lock(&fs_info->balance_lock); | |
4140 | update_balance_args(bctl); | |
4141 | spin_unlock(&fs_info->balance_lock); | |
4142 | } | |
c9e9f97b | 4143 | |
3009a62f DS |
4144 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4145 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4146 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4147 | mutex_unlock(&fs_info->balance_mutex); |
4148 | ||
4149 | ret = __btrfs_balance(fs_info); | |
4150 | ||
4151 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4152 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4153 | btrfs_info(fs_info, "balance: paused"); | |
44d354ab QW |
4154 | /* |
4155 | * Balance can be canceled by: | |
4156 | * | |
4157 | * - Regular cancel request | |
4158 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4159 | * | |
4160 | * - Fatal signal to "btrfs" process | |
4161 | * Either the signal caught by wait_reserve_ticket() and callers | |
4162 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4163 | * got -ECANCELED. | |
4164 | * Either way, in this case balance_cancel_req = 0, and | |
4165 | * ret == -EINTR or ret == -ECANCELED. | |
4166 | * | |
4167 | * So here we only check the return value to catch canceled balance. | |
4168 | */ | |
4169 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4170 | btrfs_info(fs_info, "balance: canceled"); |
4171 | else | |
4172 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4173 | ||
3009a62f | 4174 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4175 | |
4176 | if (bargs) { | |
4177 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4178 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4179 | } |
4180 | ||
3a01aa7a ID |
4181 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4182 | balance_need_close(fs_info)) { | |
149196a2 | 4183 | reset_balance_state(fs_info); |
a17c95df | 4184 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4185 | } |
4186 | ||
837d5b6e | 4187 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4188 | |
4189 | return ret; | |
4190 | out: | |
59641015 | 4191 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4192 | reset_balance_state(fs_info); |
a17c95df | 4193 | else |
59641015 | 4194 | kfree(bctl); |
a17c95df DS |
4195 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4196 | ||
59641015 ID |
4197 | return ret; |
4198 | } | |
4199 | ||
4200 | static int balance_kthread(void *data) | |
4201 | { | |
2b6ba629 | 4202 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4203 | int ret = 0; |
59641015 | 4204 | |
59641015 | 4205 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4206 | if (fs_info->balance_ctl) |
6fcf6e2b | 4207 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4208 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4209 | |
59641015 ID |
4210 | return ret; |
4211 | } | |
4212 | ||
2b6ba629 ID |
4213 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4214 | { | |
4215 | struct task_struct *tsk; | |
4216 | ||
1354e1a1 | 4217 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4218 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4219 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4220 | return 0; |
4221 | } | |
1354e1a1 | 4222 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4223 | |
3cdde224 | 4224 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4225 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4226 | return 0; |
4227 | } | |
4228 | ||
02ee654d AJ |
4229 | /* |
4230 | * A ro->rw remount sequence should continue with the paused balance | |
4231 | * regardless of who pauses it, system or the user as of now, so set | |
4232 | * the resume flag. | |
4233 | */ | |
4234 | spin_lock(&fs_info->balance_lock); | |
4235 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4236 | spin_unlock(&fs_info->balance_lock); | |
4237 | ||
2b6ba629 | 4238 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4239 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4240 | } |
4241 | ||
68310a5e | 4242 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4243 | { |
59641015 ID |
4244 | struct btrfs_balance_control *bctl; |
4245 | struct btrfs_balance_item *item; | |
4246 | struct btrfs_disk_balance_args disk_bargs; | |
4247 | struct btrfs_path *path; | |
4248 | struct extent_buffer *leaf; | |
4249 | struct btrfs_key key; | |
4250 | int ret; | |
4251 | ||
4252 | path = btrfs_alloc_path(); | |
4253 | if (!path) | |
4254 | return -ENOMEM; | |
4255 | ||
59641015 | 4256 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4257 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4258 | key.offset = 0; |
4259 | ||
68310a5e | 4260 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4261 | if (ret < 0) |
68310a5e | 4262 | goto out; |
59641015 ID |
4263 | if (ret > 0) { /* ret = -ENOENT; */ |
4264 | ret = 0; | |
68310a5e ID |
4265 | goto out; |
4266 | } | |
4267 | ||
4268 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4269 | if (!bctl) { | |
4270 | ret = -ENOMEM; | |
4271 | goto out; | |
59641015 ID |
4272 | } |
4273 | ||
4274 | leaf = path->nodes[0]; | |
4275 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4276 | ||
68310a5e ID |
4277 | bctl->flags = btrfs_balance_flags(leaf, item); |
4278 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4279 | |
4280 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4281 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4282 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4283 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4284 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4285 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4286 | ||
eee95e3f DS |
4287 | /* |
4288 | * This should never happen, as the paused balance state is recovered | |
4289 | * during mount without any chance of other exclusive ops to collide. | |
4290 | * | |
4291 | * This gives the exclusive op status to balance and keeps in paused | |
4292 | * state until user intervention (cancel or umount). If the ownership | |
4293 | * cannot be assigned, show a message but do not fail. The balance | |
4294 | * is in a paused state and must have fs_info::balance_ctl properly | |
4295 | * set up. | |
4296 | */ | |
4297 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4298 | btrfs_warn(fs_info, | |
6dac13f8 | 4299 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4300 | |
68310a5e | 4301 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4302 | BUG_ON(fs_info->balance_ctl); |
4303 | spin_lock(&fs_info->balance_lock); | |
4304 | fs_info->balance_ctl = bctl; | |
4305 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4306 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4307 | out: |
4308 | btrfs_free_path(path); | |
ec44a35c CM |
4309 | return ret; |
4310 | } | |
4311 | ||
837d5b6e ID |
4312 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4313 | { | |
4314 | int ret = 0; | |
4315 | ||
4316 | mutex_lock(&fs_info->balance_mutex); | |
4317 | if (!fs_info->balance_ctl) { | |
4318 | mutex_unlock(&fs_info->balance_mutex); | |
4319 | return -ENOTCONN; | |
4320 | } | |
4321 | ||
3009a62f | 4322 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4323 | atomic_inc(&fs_info->balance_pause_req); |
4324 | mutex_unlock(&fs_info->balance_mutex); | |
4325 | ||
4326 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4327 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4328 | |
4329 | mutex_lock(&fs_info->balance_mutex); | |
4330 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4331 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4332 | atomic_dec(&fs_info->balance_pause_req); |
4333 | } else { | |
4334 | ret = -ENOTCONN; | |
4335 | } | |
4336 | ||
4337 | mutex_unlock(&fs_info->balance_mutex); | |
4338 | return ret; | |
4339 | } | |
4340 | ||
a7e99c69 ID |
4341 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4342 | { | |
4343 | mutex_lock(&fs_info->balance_mutex); | |
4344 | if (!fs_info->balance_ctl) { | |
4345 | mutex_unlock(&fs_info->balance_mutex); | |
4346 | return -ENOTCONN; | |
4347 | } | |
4348 | ||
cf7d20f4 DS |
4349 | /* |
4350 | * A paused balance with the item stored on disk can be resumed at | |
4351 | * mount time if the mount is read-write. Otherwise it's still paused | |
4352 | * and we must not allow cancelling as it deletes the item. | |
4353 | */ | |
4354 | if (sb_rdonly(fs_info->sb)) { | |
4355 | mutex_unlock(&fs_info->balance_mutex); | |
4356 | return -EROFS; | |
4357 | } | |
4358 | ||
a7e99c69 ID |
4359 | atomic_inc(&fs_info->balance_cancel_req); |
4360 | /* | |
4361 | * if we are running just wait and return, balance item is | |
4362 | * deleted in btrfs_balance in this case | |
4363 | */ | |
3009a62f | 4364 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4365 | mutex_unlock(&fs_info->balance_mutex); |
4366 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4367 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4368 | mutex_lock(&fs_info->balance_mutex); |
4369 | } else { | |
a7e99c69 | 4370 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4371 | /* |
4372 | * Lock released to allow other waiters to continue, we'll | |
4373 | * reexamine the status again. | |
4374 | */ | |
a7e99c69 ID |
4375 | mutex_lock(&fs_info->balance_mutex); |
4376 | ||
a17c95df | 4377 | if (fs_info->balance_ctl) { |
149196a2 | 4378 | reset_balance_state(fs_info); |
a17c95df | 4379 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4380 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4381 | } |
a7e99c69 ID |
4382 | } |
4383 | ||
3009a62f DS |
4384 | BUG_ON(fs_info->balance_ctl || |
4385 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4386 | atomic_dec(&fs_info->balance_cancel_req); |
4387 | mutex_unlock(&fs_info->balance_mutex); | |
4388 | return 0; | |
4389 | } | |
4390 | ||
97f4dd09 | 4391 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4392 | { |
4393 | struct btrfs_fs_info *fs_info = data; | |
4394 | struct btrfs_root *root = fs_info->tree_root; | |
4395 | struct btrfs_key key; | |
803b2f54 SB |
4396 | struct btrfs_path *path = NULL; |
4397 | int ret = 0; | |
4398 | struct extent_buffer *eb; | |
4399 | int slot; | |
4400 | struct btrfs_root_item root_item; | |
4401 | u32 item_size; | |
f45388f3 | 4402 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4403 | bool closing = false; |
803b2f54 SB |
4404 | |
4405 | path = btrfs_alloc_path(); | |
4406 | if (!path) { | |
4407 | ret = -ENOMEM; | |
4408 | goto out; | |
4409 | } | |
4410 | ||
4411 | key.objectid = 0; | |
4412 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4413 | key.offset = 0; | |
4414 | ||
803b2f54 | 4415 | while (1) { |
c94bec2c JB |
4416 | if (btrfs_fs_closing(fs_info)) { |
4417 | closing = true; | |
4418 | break; | |
4419 | } | |
7c829b72 AJ |
4420 | ret = btrfs_search_forward(root, &key, path, |
4421 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4422 | if (ret) { |
4423 | if (ret > 0) | |
4424 | ret = 0; | |
4425 | break; | |
4426 | } | |
4427 | ||
4428 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4429 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4430 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4431 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4432 | goto skip; | |
4433 | ||
4434 | eb = path->nodes[0]; | |
4435 | slot = path->slots[0]; | |
4436 | item_size = btrfs_item_size_nr(eb, slot); | |
4437 | if (item_size < sizeof(root_item)) | |
4438 | goto skip; | |
4439 | ||
803b2f54 SB |
4440 | read_extent_buffer(eb, &root_item, |
4441 | btrfs_item_ptr_offset(eb, slot), | |
4442 | (int)sizeof(root_item)); | |
4443 | if (btrfs_root_refs(&root_item) == 0) | |
4444 | goto skip; | |
f45388f3 FDBM |
4445 | |
4446 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4447 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4448 | if (trans) | |
4449 | goto update_tree; | |
4450 | ||
4451 | btrfs_release_path(path); | |
803b2f54 SB |
4452 | /* |
4453 | * 1 - subvol uuid item | |
4454 | * 1 - received_subvol uuid item | |
4455 | */ | |
4456 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4457 | if (IS_ERR(trans)) { | |
4458 | ret = PTR_ERR(trans); | |
4459 | break; | |
4460 | } | |
f45388f3 FDBM |
4461 | continue; |
4462 | } else { | |
4463 | goto skip; | |
4464 | } | |
4465 | update_tree: | |
9771a5cf | 4466 | btrfs_release_path(path); |
f45388f3 | 4467 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4468 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4469 | BTRFS_UUID_KEY_SUBVOL, |
4470 | key.objectid); | |
4471 | if (ret < 0) { | |
efe120a0 | 4472 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4473 | ret); |
803b2f54 SB |
4474 | break; |
4475 | } | |
4476 | } | |
4477 | ||
4478 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4479 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4480 | root_item.received_uuid, |
4481 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4482 | key.objectid); | |
4483 | if (ret < 0) { | |
efe120a0 | 4484 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4485 | ret); |
803b2f54 SB |
4486 | break; |
4487 | } | |
4488 | } | |
4489 | ||
f45388f3 | 4490 | skip: |
9771a5cf | 4491 | btrfs_release_path(path); |
803b2f54 | 4492 | if (trans) { |
3a45bb20 | 4493 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4494 | trans = NULL; |
803b2f54 SB |
4495 | if (ret) |
4496 | break; | |
4497 | } | |
4498 | ||
803b2f54 SB |
4499 | if (key.offset < (u64)-1) { |
4500 | key.offset++; | |
4501 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4502 | key.offset = 0; | |
4503 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4504 | } else if (key.objectid < (u64)-1) { | |
4505 | key.offset = 0; | |
4506 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4507 | key.objectid++; | |
4508 | } else { | |
4509 | break; | |
4510 | } | |
4511 | cond_resched(); | |
4512 | } | |
4513 | ||
4514 | out: | |
4515 | btrfs_free_path(path); | |
f45388f3 | 4516 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4517 | btrfs_end_transaction(trans); |
803b2f54 | 4518 | if (ret) |
efe120a0 | 4519 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4520 | else if (!closing) |
afcdd129 | 4521 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4522 | up(&fs_info->uuid_tree_rescan_sem); |
4523 | return 0; | |
4524 | } | |
4525 | ||
f7a81ea4 SB |
4526 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4527 | { | |
4528 | struct btrfs_trans_handle *trans; | |
4529 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4530 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4531 | struct task_struct *task; |
4532 | int ret; | |
f7a81ea4 SB |
4533 | |
4534 | /* | |
4535 | * 1 - root node | |
4536 | * 1 - root item | |
4537 | */ | |
4538 | trans = btrfs_start_transaction(tree_root, 2); | |
4539 | if (IS_ERR(trans)) | |
4540 | return PTR_ERR(trans); | |
4541 | ||
9b7a2440 | 4542 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4543 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4544 | ret = PTR_ERR(uuid_root); |
66642832 | 4545 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4546 | btrfs_end_transaction(trans); |
6d13f549 | 4547 | return ret; |
f7a81ea4 SB |
4548 | } |
4549 | ||
4550 | fs_info->uuid_root = uuid_root; | |
4551 | ||
3a45bb20 | 4552 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4553 | if (ret) |
4554 | return ret; | |
4555 | ||
4556 | down(&fs_info->uuid_tree_rescan_sem); | |
4557 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4558 | if (IS_ERR(task)) { | |
70f80175 | 4559 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4560 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4561 | up(&fs_info->uuid_tree_rescan_sem); |
4562 | return PTR_ERR(task); | |
4563 | } | |
4564 | ||
4565 | return 0; | |
f7a81ea4 | 4566 | } |
803b2f54 | 4567 | |
8f18cf13 CM |
4568 | /* |
4569 | * shrinking a device means finding all of the device extents past | |
4570 | * the new size, and then following the back refs to the chunks. | |
4571 | * The chunk relocation code actually frees the device extent | |
4572 | */ | |
4573 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4574 | { | |
0b246afa JM |
4575 | struct btrfs_fs_info *fs_info = device->fs_info; |
4576 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4577 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4578 | struct btrfs_dev_extent *dev_extent = NULL; |
4579 | struct btrfs_path *path; | |
4580 | u64 length; | |
8f18cf13 CM |
4581 | u64 chunk_offset; |
4582 | int ret; | |
4583 | int slot; | |
ba1bf481 JB |
4584 | int failed = 0; |
4585 | bool retried = false; | |
8f18cf13 CM |
4586 | struct extent_buffer *l; |
4587 | struct btrfs_key key; | |
0b246afa | 4588 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4589 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4590 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4591 | u64 diff; |
61d0d0d2 | 4592 | u64 start; |
7dfb8be1 NB |
4593 | |
4594 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4595 | start = new_size; |
0e4324a4 | 4596 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4597 | |
401e29c1 | 4598 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4599 | return -EINVAL; |
4600 | ||
8f18cf13 CM |
4601 | path = btrfs_alloc_path(); |
4602 | if (!path) | |
4603 | return -ENOMEM; | |
4604 | ||
0338dff6 | 4605 | path->reada = READA_BACK; |
8f18cf13 | 4606 | |
61d0d0d2 NB |
4607 | trans = btrfs_start_transaction(root, 0); |
4608 | if (IS_ERR(trans)) { | |
4609 | btrfs_free_path(path); | |
4610 | return PTR_ERR(trans); | |
4611 | } | |
4612 | ||
34441361 | 4613 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4614 | |
7cc8e58d | 4615 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4616 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4617 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4618 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4619 | } |
61d0d0d2 NB |
4620 | |
4621 | /* | |
4622 | * Once the device's size has been set to the new size, ensure all | |
4623 | * in-memory chunks are synced to disk so that the loop below sees them | |
4624 | * and relocates them accordingly. | |
4625 | */ | |
1c11b63e | 4626 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4627 | mutex_unlock(&fs_info->chunk_mutex); |
4628 | ret = btrfs_commit_transaction(trans); | |
4629 | if (ret) | |
4630 | goto done; | |
4631 | } else { | |
4632 | mutex_unlock(&fs_info->chunk_mutex); | |
4633 | btrfs_end_transaction(trans); | |
4634 | } | |
8f18cf13 | 4635 | |
ba1bf481 | 4636 | again: |
8f18cf13 CM |
4637 | key.objectid = device->devid; |
4638 | key.offset = (u64)-1; | |
4639 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4640 | ||
213e64da | 4641 | do { |
0b246afa | 4642 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4643 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4644 | if (ret < 0) { |
0b246afa | 4645 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4646 | goto done; |
67c5e7d4 | 4647 | } |
8f18cf13 CM |
4648 | |
4649 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4650 | if (ret) |
0b246afa | 4651 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4652 | if (ret < 0) |
4653 | goto done; | |
4654 | if (ret) { | |
4655 | ret = 0; | |
b3b4aa74 | 4656 | btrfs_release_path(path); |
bf1fb512 | 4657 | break; |
8f18cf13 CM |
4658 | } |
4659 | ||
4660 | l = path->nodes[0]; | |
4661 | slot = path->slots[0]; | |
4662 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4663 | ||
ba1bf481 | 4664 | if (key.objectid != device->devid) { |
0b246afa | 4665 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4666 | btrfs_release_path(path); |
bf1fb512 | 4667 | break; |
ba1bf481 | 4668 | } |
8f18cf13 CM |
4669 | |
4670 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4671 | length = btrfs_dev_extent_length(l, dev_extent); | |
4672 | ||
ba1bf481 | 4673 | if (key.offset + length <= new_size) { |
0b246afa | 4674 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4675 | btrfs_release_path(path); |
d6397bae | 4676 | break; |
ba1bf481 | 4677 | } |
8f18cf13 | 4678 | |
8f18cf13 | 4679 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4680 | btrfs_release_path(path); |
8f18cf13 | 4681 | |
a6f93c71 LB |
4682 | /* |
4683 | * We may be relocating the only data chunk we have, | |
4684 | * which could potentially end up with losing data's | |
4685 | * raid profile, so lets allocate an empty one in | |
4686 | * advance. | |
4687 | */ | |
4688 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4689 | if (ret < 0) { | |
4690 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4691 | goto done; | |
4692 | } | |
4693 | ||
0b246afa JM |
4694 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4695 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4696 | if (ret == -ENOSPC) { |
ba1bf481 | 4697 | failed++; |
eede2bf3 OS |
4698 | } else if (ret) { |
4699 | if (ret == -ETXTBSY) { | |
4700 | btrfs_warn(fs_info, | |
4701 | "could not shrink block group %llu due to active swapfile", | |
4702 | chunk_offset); | |
4703 | } | |
4704 | goto done; | |
4705 | } | |
213e64da | 4706 | } while (key.offset-- > 0); |
ba1bf481 JB |
4707 | |
4708 | if (failed && !retried) { | |
4709 | failed = 0; | |
4710 | retried = true; | |
4711 | goto again; | |
4712 | } else if (failed && retried) { | |
4713 | ret = -ENOSPC; | |
ba1bf481 | 4714 | goto done; |
8f18cf13 CM |
4715 | } |
4716 | ||
d6397bae | 4717 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4718 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4719 | if (IS_ERR(trans)) { |
4720 | ret = PTR_ERR(trans); | |
4721 | goto done; | |
4722 | } | |
4723 | ||
34441361 | 4724 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4725 | /* Clear all state bits beyond the shrunk device size */ |
4726 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4727 | CHUNK_STATE_MASK); | |
4728 | ||
7cc8e58d | 4729 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4730 | if (list_empty(&device->post_commit_list)) |
4731 | list_add_tail(&device->post_commit_list, | |
4732 | &trans->transaction->dev_update_list); | |
d6397bae | 4733 | |
d6397bae | 4734 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4735 | btrfs_set_super_total_bytes(super_copy, |
4736 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4737 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4738 | |
4739 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4740 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4741 | if (ret < 0) { |
4742 | btrfs_abort_transaction(trans, ret); | |
4743 | btrfs_end_transaction(trans); | |
4744 | } else { | |
4745 | ret = btrfs_commit_transaction(trans); | |
4746 | } | |
8f18cf13 CM |
4747 | done: |
4748 | btrfs_free_path(path); | |
53e489bc | 4749 | if (ret) { |
34441361 | 4750 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4751 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4752 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4753 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4754 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4755 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4756 | } |
8f18cf13 CM |
4757 | return ret; |
4758 | } | |
4759 | ||
2ff7e61e | 4760 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4761 | struct btrfs_key *key, |
4762 | struct btrfs_chunk *chunk, int item_size) | |
4763 | { | |
0b246afa | 4764 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4765 | struct btrfs_disk_key disk_key; |
4766 | u32 array_size; | |
4767 | u8 *ptr; | |
4768 | ||
34441361 | 4769 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4770 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4771 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4772 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4773 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4774 | return -EFBIG; |
fe48a5c0 | 4775 | } |
0b86a832 CM |
4776 | |
4777 | ptr = super_copy->sys_chunk_array + array_size; | |
4778 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4779 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4780 | ptr += sizeof(disk_key); | |
4781 | memcpy(ptr, chunk, item_size); | |
4782 | item_size += sizeof(disk_key); | |
4783 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4784 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4785 | |
0b86a832 CM |
4786 | return 0; |
4787 | } | |
4788 | ||
73c5de00 AJ |
4789 | /* |
4790 | * sort the devices in descending order by max_avail, total_avail | |
4791 | */ | |
4792 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4793 | { |
73c5de00 AJ |
4794 | const struct btrfs_device_info *di_a = a; |
4795 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4796 | |
73c5de00 | 4797 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4798 | return -1; |
73c5de00 | 4799 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4800 | return 1; |
73c5de00 AJ |
4801 | if (di_a->total_avail > di_b->total_avail) |
4802 | return -1; | |
4803 | if (di_a->total_avail < di_b->total_avail) | |
4804 | return 1; | |
4805 | return 0; | |
b2117a39 | 4806 | } |
0b86a832 | 4807 | |
53b381b3 DW |
4808 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4809 | { | |
ffe2d203 | 4810 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4811 | return; |
4812 | ||
ceda0864 | 4813 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4814 | } |
4815 | ||
cfbb825c DS |
4816 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4817 | { | |
4818 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4819 | return; | |
4820 | ||
4821 | btrfs_set_fs_incompat(info, RAID1C34); | |
4822 | } | |
4823 | ||
4f2bafe8 NA |
4824 | /* |
4825 | * Structure used internally for __btrfs_alloc_chunk() function. | |
4826 | * Wraps needed parameters. | |
4827 | */ | |
4828 | struct alloc_chunk_ctl { | |
4829 | u64 start; | |
4830 | u64 type; | |
4831 | /* Total number of stripes to allocate */ | |
4832 | int num_stripes; | |
4833 | /* sub_stripes info for map */ | |
4834 | int sub_stripes; | |
4835 | /* Stripes per device */ | |
4836 | int dev_stripes; | |
4837 | /* Maximum number of devices to use */ | |
4838 | int devs_max; | |
4839 | /* Minimum number of devices to use */ | |
4840 | int devs_min; | |
4841 | /* ndevs has to be a multiple of this */ | |
4842 | int devs_increment; | |
4843 | /* Number of copies */ | |
4844 | int ncopies; | |
4845 | /* Number of stripes worth of bytes to store parity information */ | |
4846 | int nparity; | |
4847 | u64 max_stripe_size; | |
4848 | u64 max_chunk_size; | |
6aafb303 | 4849 | u64 dev_extent_min; |
4f2bafe8 NA |
4850 | u64 stripe_size; |
4851 | u64 chunk_size; | |
4852 | int ndevs; | |
4853 | }; | |
4854 | ||
27c314d5 NA |
4855 | static void init_alloc_chunk_ctl_policy_regular( |
4856 | struct btrfs_fs_devices *fs_devices, | |
4857 | struct alloc_chunk_ctl *ctl) | |
4858 | { | |
4859 | u64 type = ctl->type; | |
4860 | ||
4861 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4862 | ctl->max_stripe_size = SZ_1G; | |
4863 | ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; | |
4864 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4865 | /* For larger filesystems, use larger metadata chunks */ | |
4866 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) | |
4867 | ctl->max_stripe_size = SZ_1G; | |
4868 | else | |
4869 | ctl->max_stripe_size = SZ_256M; | |
4870 | ctl->max_chunk_size = ctl->max_stripe_size; | |
4871 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4872 | ctl->max_stripe_size = SZ_32M; | |
4873 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
4874 | ctl->devs_max = min_t(int, ctl->devs_max, | |
4875 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
4876 | } else { | |
4877 | BUG(); | |
4878 | } | |
4879 | ||
4880 | /* We don't want a chunk larger than 10% of writable space */ | |
4881 | ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4882 | ctl->max_chunk_size); | |
6aafb303 | 4883 | ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; |
27c314d5 NA |
4884 | } |
4885 | ||
4886 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, | |
4887 | struct alloc_chunk_ctl *ctl) | |
4888 | { | |
4889 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
4890 | ||
4891 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
4892 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4893 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
4894 | if (!ctl->devs_max) | |
4895 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
4896 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
4897 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
4898 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
4899 | ctl->nparity = btrfs_raid_array[index].nparity; | |
4900 | ctl->ndevs = 0; | |
4901 | ||
4902 | switch (fs_devices->chunk_alloc_policy) { | |
4903 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
4904 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
4905 | break; | |
4906 | default: | |
4907 | BUG(); | |
4908 | } | |
4909 | } | |
4910 | ||
560156cb NA |
4911 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
4912 | struct alloc_chunk_ctl *ctl, | |
4913 | struct btrfs_device_info *devices_info) | |
b2117a39 | 4914 | { |
560156cb | 4915 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 4916 | struct btrfs_device *device; |
73c5de00 | 4917 | u64 total_avail; |
560156cb | 4918 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 4919 | int ret; |
560156cb NA |
4920 | int ndevs = 0; |
4921 | u64 max_avail; | |
4922 | u64 dev_offset; | |
0cad8a11 | 4923 | |
9f680ce0 | 4924 | /* |
73c5de00 AJ |
4925 | * in the first pass through the devices list, we gather information |
4926 | * about the available holes on each device. | |
9f680ce0 | 4927 | */ |
ebcc9301 | 4928 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 4929 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4930 | WARN(1, KERN_ERR |
efe120a0 | 4931 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4932 | continue; |
4933 | } | |
b2117a39 | 4934 | |
e12c9621 AJ |
4935 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4936 | &device->dev_state) || | |
401e29c1 | 4937 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4938 | continue; |
b2117a39 | 4939 | |
73c5de00 AJ |
4940 | if (device->total_bytes > device->bytes_used) |
4941 | total_avail = device->total_bytes - device->bytes_used; | |
4942 | else | |
4943 | total_avail = 0; | |
38c01b96 | 4944 | |
4945 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 4946 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 4947 | continue; |
b2117a39 | 4948 | |
560156cb NA |
4949 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
4950 | &max_avail); | |
73c5de00 | 4951 | if (ret && ret != -ENOSPC) |
560156cb | 4952 | return ret; |
b2117a39 | 4953 | |
73c5de00 | 4954 | if (ret == 0) |
560156cb | 4955 | max_avail = dev_extent_want; |
b2117a39 | 4956 | |
6aafb303 | 4957 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
4958 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
4959 | btrfs_debug(info, | |
560156cb | 4960 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 4961 | __func__, device->devid, max_avail, |
6aafb303 | 4962 | ctl->dev_extent_min); |
73c5de00 | 4963 | continue; |
4117f207 | 4964 | } |
b2117a39 | 4965 | |
063d006f ES |
4966 | if (ndevs == fs_devices->rw_devices) { |
4967 | WARN(1, "%s: found more than %llu devices\n", | |
4968 | __func__, fs_devices->rw_devices); | |
4969 | break; | |
4970 | } | |
73c5de00 AJ |
4971 | devices_info[ndevs].dev_offset = dev_offset; |
4972 | devices_info[ndevs].max_avail = max_avail; | |
4973 | devices_info[ndevs].total_avail = total_avail; | |
4974 | devices_info[ndevs].dev = device; | |
4975 | ++ndevs; | |
4976 | } | |
560156cb | 4977 | ctl->ndevs = ndevs; |
b2117a39 | 4978 | |
73c5de00 AJ |
4979 | /* |
4980 | * now sort the devices by hole size / available space | |
4981 | */ | |
560156cb | 4982 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 4983 | btrfs_cmp_device_info, NULL); |
b2117a39 | 4984 | |
560156cb NA |
4985 | return 0; |
4986 | } | |
4987 | ||
5badf512 NA |
4988 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
4989 | struct btrfs_device_info *devices_info) | |
4990 | { | |
4991 | /* Number of stripes that count for block group size */ | |
4992 | int data_stripes; | |
4993 | ||
4994 | /* | |
4995 | * The primary goal is to maximize the number of stripes, so use as | |
4996 | * many devices as possible, even if the stripes are not maximum sized. | |
4997 | * | |
4998 | * The DUP profile stores more than one stripe per device, the | |
4999 | * max_avail is the total size so we have to adjust. | |
5000 | */ | |
5001 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5002 | ctl->dev_stripes); | |
5003 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5004 | ||
5005 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5006 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5007 | ||
5008 | /* | |
5009 | * Use the number of data stripes to figure out how big this chunk is | |
5010 | * really going to be in terms of logical address space, and compare | |
5011 | * that answer with the max chunk size. If it's higher, we try to | |
5012 | * reduce stripe_size. | |
5013 | */ | |
5014 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5015 | /* | |
5016 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5017 | * then use it, unless it ends up being even bigger than the | |
5018 | * previous value we had already. | |
5019 | */ | |
5020 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5021 | data_stripes), SZ_16M), | |
5022 | ctl->stripe_size); | |
5023 | } | |
5024 | ||
5025 | /* Align to BTRFS_STRIPE_LEN */ | |
5026 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5027 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5028 | ||
5029 | return 0; | |
5030 | } | |
5031 | ||
5032 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, | |
5033 | struct alloc_chunk_ctl *ctl, | |
5034 | struct btrfs_device_info *devices_info) | |
5035 | { | |
5036 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5037 | ||
5038 | /* | |
5039 | * Round down to number of usable stripes, devs_increment can be any | |
5040 | * number so we can't use round_down() that requires power of 2, while | |
5041 | * rounddown is safe. | |
5042 | */ | |
5043 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5044 | ||
5045 | if (ctl->ndevs < ctl->devs_min) { | |
5046 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5047 | btrfs_debug(info, | |
5048 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5049 | __func__, ctl->ndevs, ctl->devs_min); | |
5050 | } | |
5051 | return -ENOSPC; | |
5052 | } | |
5053 | ||
5054 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5055 | ||
5056 | switch (fs_devices->chunk_alloc_policy) { | |
5057 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5058 | return decide_stripe_size_regular(ctl, devices_info); | |
5059 | default: | |
5060 | BUG(); | |
5061 | } | |
5062 | } | |
5063 | ||
dce580ca NA |
5064 | static int create_chunk(struct btrfs_trans_handle *trans, |
5065 | struct alloc_chunk_ctl *ctl, | |
5066 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5067 | { |
5068 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5069 | struct map_lookup *map = NULL; |
5070 | struct extent_map_tree *em_tree; | |
5071 | struct extent_map *em; | |
dce580ca NA |
5072 | u64 start = ctl->start; |
5073 | u64 type = ctl->type; | |
560156cb NA |
5074 | int ret; |
5075 | int i; | |
5076 | int j; | |
5077 | ||
dce580ca NA |
5078 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5079 | if (!map) | |
560156cb | 5080 | return -ENOMEM; |
dce580ca | 5081 | map->num_stripes = ctl->num_stripes; |
560156cb | 5082 | |
dce580ca NA |
5083 | for (i = 0; i < ctl->ndevs; ++i) { |
5084 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5085 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5086 | map->stripes[s].dev = devices_info[i].dev; |
5087 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5088 | j * ctl->stripe_size; |
6324fbf3 | 5089 | } |
6324fbf3 | 5090 | } |
500ceed8 NB |
5091 | map->stripe_len = BTRFS_STRIPE_LEN; |
5092 | map->io_align = BTRFS_STRIPE_LEN; | |
5093 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5094 | map->type = type; |
dce580ca | 5095 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5096 | |
dce580ca | 5097 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5098 | |
172ddd60 | 5099 | em = alloc_extent_map(); |
2b82032c | 5100 | if (!em) { |
298a8f9c | 5101 | kfree(map); |
dce580ca | 5102 | return -ENOMEM; |
593060d7 | 5103 | } |
298a8f9c | 5104 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5105 | em->map_lookup = map; |
2b82032c | 5106 | em->start = start; |
dce580ca | 5107 | em->len = ctl->chunk_size; |
2b82032c YZ |
5108 | em->block_start = 0; |
5109 | em->block_len = em->len; | |
dce580ca | 5110 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5111 | |
c8bf1b67 | 5112 | em_tree = &info->mapping_tree; |
890871be | 5113 | write_lock(&em_tree->lock); |
09a2a8f9 | 5114 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5115 | if (ret) { |
1efb72a3 | 5116 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5117 | free_extent_map(em); |
dce580ca | 5118 | return ret; |
0f5d42b2 | 5119 | } |
1efb72a3 NB |
5120 | write_unlock(&em_tree->lock); |
5121 | ||
dce580ca | 5122 | ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); |
6df9a95e JB |
5123 | if (ret) |
5124 | goto error_del_extent; | |
2b82032c | 5125 | |
bbbf7243 NB |
5126 | for (i = 0; i < map->num_stripes; i++) { |
5127 | struct btrfs_device *dev = map->stripes[i].dev; | |
5128 | ||
4f2bafe8 | 5129 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5130 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5131 | if (list_empty(&dev->post_commit_list)) |
5132 | list_add_tail(&dev->post_commit_list, | |
5133 | &trans->transaction->dev_update_list); | |
5134 | } | |
43530c46 | 5135 | |
dce580ca | 5136 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5137 | &info->free_chunk_space); |
1c116187 | 5138 | |
0f5d42b2 | 5139 | free_extent_map(em); |
0b246afa | 5140 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5141 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5142 | |
2b82032c | 5143 | return 0; |
b2117a39 | 5144 | |
6df9a95e | 5145 | error_del_extent: |
0f5d42b2 JB |
5146 | write_lock(&em_tree->lock); |
5147 | remove_extent_mapping(em_tree, em); | |
5148 | write_unlock(&em_tree->lock); | |
5149 | ||
5150 | /* One for our allocation */ | |
5151 | free_extent_map(em); | |
5152 | /* One for the tree reference */ | |
5153 | free_extent_map(em); | |
dce580ca NA |
5154 | |
5155 | return ret; | |
5156 | } | |
5157 | ||
11c67b1a | 5158 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
dce580ca NA |
5159 | { |
5160 | struct btrfs_fs_info *info = trans->fs_info; | |
5161 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5162 | struct btrfs_device_info *devices_info = NULL; | |
5163 | struct alloc_chunk_ctl ctl; | |
5164 | int ret; | |
5165 | ||
11c67b1a NB |
5166 | lockdep_assert_held(&info->chunk_mutex); |
5167 | ||
dce580ca NA |
5168 | if (!alloc_profile_is_valid(type, 0)) { |
5169 | ASSERT(0); | |
5170 | return -EINVAL; | |
5171 | } | |
5172 | ||
5173 | if (list_empty(&fs_devices->alloc_list)) { | |
5174 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5175 | btrfs_debug(info, "%s: no writable device", __func__); | |
5176 | return -ENOSPC; | |
5177 | } | |
5178 | ||
5179 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5180 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5181 | ASSERT(0); | |
5182 | return -EINVAL; | |
5183 | } | |
5184 | ||
11c67b1a | 5185 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5186 | ctl.type = type; |
5187 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5188 | ||
5189 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5190 | GFP_NOFS); | |
5191 | if (!devices_info) | |
5192 | return -ENOMEM; | |
5193 | ||
5194 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
5195 | if (ret < 0) | |
5196 | goto out; | |
5197 | ||
5198 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
5199 | if (ret < 0) | |
5200 | goto out; | |
5201 | ||
5202 | ret = create_chunk(trans, &ctl, devices_info); | |
5203 | ||
5204 | out: | |
b2117a39 MX |
5205 | kfree(devices_info); |
5206 | return ret; | |
2b82032c YZ |
5207 | } |
5208 | ||
11c67b1a NB |
5209 | /* |
5210 | * Chunk allocation falls into two parts. The first part does work | |
5211 | * that makes the new allocated chunk usable, but does not do any operation | |
5212 | * that modifies the chunk tree. The second part does the work that | |
5213 | * requires modifying the chunk tree. This division is important for the | |
5214 | * bootstrap process of adding storage to a seed btrfs. | |
5215 | */ | |
6df9a95e | 5216 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5217 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5218 | { |
97aff912 | 5219 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5220 | struct btrfs_root *extent_root = fs_info->extent_root; |
5221 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5222 | struct btrfs_key key; |
2b82032c YZ |
5223 | struct btrfs_device *device; |
5224 | struct btrfs_chunk *chunk; | |
5225 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5226 | struct extent_map *em; |
5227 | struct map_lookup *map; | |
5228 | size_t item_size; | |
5229 | u64 dev_offset; | |
5230 | u64 stripe_size; | |
5231 | int i = 0; | |
140e639f | 5232 | int ret = 0; |
2b82032c | 5233 | |
60ca842e | 5234 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5235 | if (IS_ERR(em)) |
5236 | return PTR_ERR(em); | |
6df9a95e | 5237 | |
95617d69 | 5238 | map = em->map_lookup; |
6df9a95e JB |
5239 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5240 | stripe_size = em->orig_block_len; | |
5241 | ||
2b82032c | 5242 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5243 | if (!chunk) { |
5244 | ret = -ENOMEM; | |
5245 | goto out; | |
5246 | } | |
5247 | ||
50460e37 FM |
5248 | /* |
5249 | * Take the device list mutex to prevent races with the final phase of | |
5250 | * a device replace operation that replaces the device object associated | |
5251 | * with the map's stripes, because the device object's id can change | |
5252 | * at any time during that final phase of the device replace operation | |
5253 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5254 | */ | |
0b246afa | 5255 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5256 | for (i = 0; i < map->num_stripes; i++) { |
5257 | device = map->stripes[i].dev; | |
5258 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5259 | |
0b86a832 | 5260 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5261 | if (ret) |
50460e37 | 5262 | break; |
b5d9071c NB |
5263 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5264 | dev_offset, stripe_size); | |
6df9a95e | 5265 | if (ret) |
50460e37 FM |
5266 | break; |
5267 | } | |
5268 | if (ret) { | |
0b246afa | 5269 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5270 | goto out; |
2b82032c YZ |
5271 | } |
5272 | ||
2b82032c | 5273 | stripe = &chunk->stripe; |
6df9a95e JB |
5274 | for (i = 0; i < map->num_stripes; i++) { |
5275 | device = map->stripes[i].dev; | |
5276 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5277 | |
e17cade2 CM |
5278 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5279 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5280 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5281 | stripe++; |
0b86a832 | 5282 | } |
0b246afa | 5283 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5284 | |
2b82032c | 5285 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5286 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5287 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5288 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5289 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5290 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5291 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5292 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5293 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5294 | |
2b82032c YZ |
5295 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5296 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5297 | key.offset = chunk_offset; | |
0b86a832 | 5298 | |
2b82032c | 5299 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5300 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5301 | /* | |
5302 | * TODO: Cleanup of inserted chunk root in case of | |
5303 | * failure. | |
5304 | */ | |
2ff7e61e | 5305 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5306 | } |
1abe9b8a | 5307 | |
6df9a95e | 5308 | out: |
0b86a832 | 5309 | kfree(chunk); |
6df9a95e | 5310 | free_extent_map(em); |
4ed1d16e | 5311 | return ret; |
2b82032c | 5312 | } |
0b86a832 | 5313 | |
6f8e0fc7 | 5314 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5315 | { |
6f8e0fc7 | 5316 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5317 | u64 alloc_profile; |
2b82032c YZ |
5318 | int ret; |
5319 | ||
1b86826d | 5320 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
11c67b1a | 5321 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa JM |
5322 | if (ret) |
5323 | return ret; | |
2b82032c | 5324 | |
1b86826d | 5325 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
11c67b1a | 5326 | ret = btrfs_alloc_chunk(trans, alloc_profile); |
79787eaa | 5327 | return ret; |
2b82032c YZ |
5328 | } |
5329 | ||
d20983b4 MX |
5330 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5331 | { | |
fc9a2ac7 | 5332 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5333 | |
fc9a2ac7 | 5334 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5335 | } |
5336 | ||
2ff7e61e | 5337 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5338 | { |
5339 | struct extent_map *em; | |
5340 | struct map_lookup *map; | |
2b82032c | 5341 | int readonly = 0; |
d20983b4 | 5342 | int miss_ndevs = 0; |
2b82032c YZ |
5343 | int i; |
5344 | ||
60ca842e | 5345 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5346 | if (IS_ERR(em)) |
2b82032c YZ |
5347 | return 1; |
5348 | ||
95617d69 | 5349 | map = em->map_lookup; |
2b82032c | 5350 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5351 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5352 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5353 | miss_ndevs++; |
5354 | continue; | |
5355 | } | |
ebbede42 AJ |
5356 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5357 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5358 | readonly = 1; |
d20983b4 | 5359 | goto end; |
2b82032c YZ |
5360 | } |
5361 | } | |
d20983b4 MX |
5362 | |
5363 | /* | |
5364 | * If the number of missing devices is larger than max errors, | |
5365 | * we can not write the data into that chunk successfully, so | |
5366 | * set it readonly. | |
5367 | */ | |
5368 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5369 | readonly = 1; | |
5370 | end: | |
0b86a832 | 5371 | free_extent_map(em); |
2b82032c | 5372 | return readonly; |
0b86a832 CM |
5373 | } |
5374 | ||
c8bf1b67 | 5375 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5376 | { |
5377 | struct extent_map *em; | |
5378 | ||
d397712b | 5379 | while (1) { |
c8bf1b67 DS |
5380 | write_lock(&tree->lock); |
5381 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5382 | if (em) |
c8bf1b67 DS |
5383 | remove_extent_mapping(tree, em); |
5384 | write_unlock(&tree->lock); | |
0b86a832 CM |
5385 | if (!em) |
5386 | break; | |
0b86a832 CM |
5387 | /* once for us */ |
5388 | free_extent_map(em); | |
5389 | /* once for the tree */ | |
5390 | free_extent_map(em); | |
5391 | } | |
5392 | } | |
5393 | ||
5d964051 | 5394 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5395 | { |
5396 | struct extent_map *em; | |
5397 | struct map_lookup *map; | |
f188591e CM |
5398 | int ret; |
5399 | ||
60ca842e | 5400 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5401 | if (IS_ERR(em)) |
5402 | /* | |
5403 | * We could return errors for these cases, but that could get | |
5404 | * ugly and we'd probably do the same thing which is just not do | |
5405 | * anything else and exit, so return 1 so the callers don't try | |
5406 | * to use other copies. | |
5407 | */ | |
fb7669b5 | 5408 | return 1; |
fb7669b5 | 5409 | |
95617d69 | 5410 | map = em->map_lookup; |
c7369b3f | 5411 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5412 | ret = map->num_stripes; |
321aecc6 CM |
5413 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5414 | ret = map->sub_stripes; | |
53b381b3 DW |
5415 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5416 | ret = 2; | |
5417 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5418 | /* |
5419 | * There could be two corrupted data stripes, we need | |
5420 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5421 | * |
8810f751 LB |
5422 | * Fail a stripe at a time on every retry except the |
5423 | * stripe under reconstruction. | |
5424 | */ | |
5425 | ret = map->num_stripes; | |
f188591e CM |
5426 | else |
5427 | ret = 1; | |
5428 | free_extent_map(em); | |
ad6d620e | 5429 | |
cb5583dd | 5430 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5431 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5432 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5433 | ret++; |
cb5583dd | 5434 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5435 | |
f188591e CM |
5436 | return ret; |
5437 | } | |
5438 | ||
2ff7e61e | 5439 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5440 | u64 logical) |
5441 | { | |
5442 | struct extent_map *em; | |
5443 | struct map_lookup *map; | |
0b246afa | 5444 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5445 | |
60ca842e | 5446 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5447 | |
69f03f13 NB |
5448 | if (!WARN_ON(IS_ERR(em))) { |
5449 | map = em->map_lookup; | |
5450 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5451 | len = map->stripe_len * nr_data_stripes(map); | |
5452 | free_extent_map(em); | |
5453 | } | |
53b381b3 DW |
5454 | return len; |
5455 | } | |
5456 | ||
e4ff5fb5 | 5457 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5458 | { |
5459 | struct extent_map *em; | |
5460 | struct map_lookup *map; | |
53b381b3 DW |
5461 | int ret = 0; |
5462 | ||
60ca842e | 5463 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5464 | |
69f03f13 NB |
5465 | if(!WARN_ON(IS_ERR(em))) { |
5466 | map = em->map_lookup; | |
5467 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5468 | ret = 1; | |
5469 | free_extent_map(em); | |
5470 | } | |
53b381b3 DW |
5471 | return ret; |
5472 | } | |
5473 | ||
30d9861f | 5474 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5475 | struct map_lookup *map, int first, |
8ba0ae78 | 5476 | int dev_replace_is_ongoing) |
dfe25020 CM |
5477 | { |
5478 | int i; | |
99f92a7c | 5479 | int num_stripes; |
8ba0ae78 | 5480 | int preferred_mirror; |
30d9861f SB |
5481 | int tolerance; |
5482 | struct btrfs_device *srcdev; | |
5483 | ||
99f92a7c | 5484 | ASSERT((map->type & |
c7369b3f | 5485 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5486 | |
5487 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5488 | num_stripes = map->sub_stripes; | |
5489 | else | |
5490 | num_stripes = map->num_stripes; | |
5491 | ||
8ba0ae78 AJ |
5492 | preferred_mirror = first + current->pid % num_stripes; |
5493 | ||
30d9861f SB |
5494 | if (dev_replace_is_ongoing && |
5495 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5496 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5497 | srcdev = fs_info->dev_replace.srcdev; | |
5498 | else | |
5499 | srcdev = NULL; | |
5500 | ||
5501 | /* | |
5502 | * try to avoid the drive that is the source drive for a | |
5503 | * dev-replace procedure, only choose it if no other non-missing | |
5504 | * mirror is available | |
5505 | */ | |
5506 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5507 | if (map->stripes[preferred_mirror].dev->bdev && |
5508 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5509 | return preferred_mirror; | |
99f92a7c | 5510 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5511 | if (map->stripes[i].dev->bdev && |
5512 | (tolerance || map->stripes[i].dev != srcdev)) | |
5513 | return i; | |
5514 | } | |
dfe25020 | 5515 | } |
30d9861f | 5516 | |
dfe25020 CM |
5517 | /* we couldn't find one that doesn't fail. Just return something |
5518 | * and the io error handling code will clean up eventually | |
5519 | */ | |
8ba0ae78 | 5520 | return preferred_mirror; |
dfe25020 CM |
5521 | } |
5522 | ||
53b381b3 | 5523 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ |
8e5cfb55 | 5524 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 | 5525 | { |
53b381b3 | 5526 | int i; |
53b381b3 DW |
5527 | int again = 1; |
5528 | ||
5529 | while (again) { | |
5530 | again = 0; | |
cc7539ed | 5531 | for (i = 0; i < num_stripes - 1; i++) { |
eeb6f172 DS |
5532 | /* Swap if parity is on a smaller index */ |
5533 | if (bbio->raid_map[i] > bbio->raid_map[i + 1]) { | |
5534 | swap(bbio->stripes[i], bbio->stripes[i + 1]); | |
5535 | swap(bbio->raid_map[i], bbio->raid_map[i + 1]); | |
53b381b3 DW |
5536 | again = 1; |
5537 | } | |
5538 | } | |
5539 | } | |
5540 | } | |
5541 | ||
6e9606d2 ZL |
5542 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5543 | { | |
5544 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5545 | /* the size of the btrfs_bio */ |
6e9606d2 | 5546 | sizeof(struct btrfs_bio) + |
e57cf21e | 5547 | /* plus the variable array for the stripes */ |
6e9606d2 | 5548 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5549 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5550 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5551 | /* |
5552 | * plus the raid_map, which includes both the tgt dev | |
5553 | * and the stripes | |
5554 | */ | |
5555 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5556 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5557 | |
5558 | atomic_set(&bbio->error, 0); | |
140475ae | 5559 | refcount_set(&bbio->refs, 1); |
6e9606d2 | 5560 | |
608769a4 NB |
5561 | bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes); |
5562 | bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes); | |
5563 | ||
6e9606d2 ZL |
5564 | return bbio; |
5565 | } | |
5566 | ||
5567 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5568 | { | |
140475ae ER |
5569 | WARN_ON(!refcount_read(&bbio->refs)); |
5570 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5571 | } |
5572 | ||
5573 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5574 | { | |
5575 | if (!bbio) | |
5576 | return; | |
140475ae | 5577 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5578 | kfree(bbio); |
5579 | } | |
5580 | ||
0b3d4cd3 LB |
5581 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5582 | /* | |
5583 | * Please note that, discard won't be sent to target device of device | |
5584 | * replace. | |
5585 | */ | |
5586 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5587 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5588 | struct btrfs_bio **bbio_ret) |
5589 | { | |
5590 | struct extent_map *em; | |
5591 | struct map_lookup *map; | |
5592 | struct btrfs_bio *bbio; | |
6b7faadd | 5593 | u64 length = *length_ret; |
0b3d4cd3 LB |
5594 | u64 offset; |
5595 | u64 stripe_nr; | |
5596 | u64 stripe_nr_end; | |
5597 | u64 stripe_end_offset; | |
5598 | u64 stripe_cnt; | |
5599 | u64 stripe_len; | |
5600 | u64 stripe_offset; | |
5601 | u64 num_stripes; | |
5602 | u32 stripe_index; | |
5603 | u32 factor = 0; | |
5604 | u32 sub_stripes = 0; | |
5605 | u64 stripes_per_dev = 0; | |
5606 | u32 remaining_stripes = 0; | |
5607 | u32 last_stripe = 0; | |
5608 | int ret = 0; | |
5609 | int i; | |
5610 | ||
5611 | /* discard always return a bbio */ | |
5612 | ASSERT(bbio_ret); | |
5613 | ||
60ca842e | 5614 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5615 | if (IS_ERR(em)) |
5616 | return PTR_ERR(em); | |
5617 | ||
5618 | map = em->map_lookup; | |
5619 | /* we don't discard raid56 yet */ | |
5620 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5621 | ret = -EOPNOTSUPP; | |
5622 | goto out; | |
5623 | } | |
5624 | ||
5625 | offset = logical - em->start; | |
2d974619 | 5626 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5627 | *length_ret = length; |
0b3d4cd3 LB |
5628 | |
5629 | stripe_len = map->stripe_len; | |
5630 | /* | |
5631 | * stripe_nr counts the total number of stripes we have to stride | |
5632 | * to get to this block | |
5633 | */ | |
5634 | stripe_nr = div64_u64(offset, stripe_len); | |
5635 | ||
5636 | /* stripe_offset is the offset of this block in its stripe */ | |
5637 | stripe_offset = offset - stripe_nr * stripe_len; | |
5638 | ||
5639 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5640 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5641 | stripe_cnt = stripe_nr_end - stripe_nr; |
5642 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5643 | (offset + length); | |
5644 | /* | |
5645 | * after this, stripe_nr is the number of stripes on this | |
5646 | * device we have to walk to find the data, and stripe_index is | |
5647 | * the number of our device in the stripe array | |
5648 | */ | |
5649 | num_stripes = 1; | |
5650 | stripe_index = 0; | |
5651 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5652 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5653 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5654 | sub_stripes = 1; | |
5655 | else | |
5656 | sub_stripes = map->sub_stripes; | |
5657 | ||
5658 | factor = map->num_stripes / sub_stripes; | |
5659 | num_stripes = min_t(u64, map->num_stripes, | |
5660 | sub_stripes * stripe_cnt); | |
5661 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5662 | stripe_index *= sub_stripes; | |
5663 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5664 | &remaining_stripes); | |
5665 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5666 | last_stripe *= sub_stripes; | |
c7369b3f | 5667 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5668 | BTRFS_BLOCK_GROUP_DUP)) { |
5669 | num_stripes = map->num_stripes; | |
5670 | } else { | |
5671 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5672 | &stripe_index); | |
5673 | } | |
5674 | ||
5675 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5676 | if (!bbio) { | |
5677 | ret = -ENOMEM; | |
5678 | goto out; | |
5679 | } | |
5680 | ||
5681 | for (i = 0; i < num_stripes; i++) { | |
5682 | bbio->stripes[i].physical = | |
5683 | map->stripes[stripe_index].physical + | |
5684 | stripe_offset + stripe_nr * map->stripe_len; | |
5685 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5686 | ||
5687 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5688 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5689 | bbio->stripes[i].length = stripes_per_dev * | |
5690 | map->stripe_len; | |
5691 | ||
5692 | if (i / sub_stripes < remaining_stripes) | |
5693 | bbio->stripes[i].length += | |
5694 | map->stripe_len; | |
5695 | ||
5696 | /* | |
5697 | * Special for the first stripe and | |
5698 | * the last stripe: | |
5699 | * | |
5700 | * |-------|...|-------| | |
5701 | * |----------| | |
5702 | * off end_off | |
5703 | */ | |
5704 | if (i < sub_stripes) | |
5705 | bbio->stripes[i].length -= | |
5706 | stripe_offset; | |
5707 | ||
5708 | if (stripe_index >= last_stripe && | |
5709 | stripe_index <= (last_stripe + | |
5710 | sub_stripes - 1)) | |
5711 | bbio->stripes[i].length -= | |
5712 | stripe_end_offset; | |
5713 | ||
5714 | if (i == sub_stripes - 1) | |
5715 | stripe_offset = 0; | |
5716 | } else { | |
5717 | bbio->stripes[i].length = length; | |
5718 | } | |
5719 | ||
5720 | stripe_index++; | |
5721 | if (stripe_index == map->num_stripes) { | |
5722 | stripe_index = 0; | |
5723 | stripe_nr++; | |
5724 | } | |
5725 | } | |
5726 | ||
5727 | *bbio_ret = bbio; | |
5728 | bbio->map_type = map->type; | |
5729 | bbio->num_stripes = num_stripes; | |
5730 | out: | |
5731 | free_extent_map(em); | |
5732 | return ret; | |
5733 | } | |
5734 | ||
5ab56090 LB |
5735 | /* |
5736 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5737 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5738 | * left cursor can also be read from the target drive. | |
5739 | * | |
5740 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5741 | * array of stripes. | |
5742 | * For READ, it also needs to be supported using the same mirror number. | |
5743 | * | |
5744 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5745 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5746 | * case. | |
5747 | */ | |
5748 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5749 | u64 logical, u64 length, | |
5750 | u64 srcdev_devid, int *mirror_num, | |
5751 | u64 *physical) | |
5752 | { | |
5753 | struct btrfs_bio *bbio = NULL; | |
5754 | int num_stripes; | |
5755 | int index_srcdev = 0; | |
5756 | int found = 0; | |
5757 | u64 physical_of_found = 0; | |
5758 | int i; | |
5759 | int ret = 0; | |
5760 | ||
5761 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5762 | logical, &length, &bbio, 0, 0); | |
5763 | if (ret) { | |
5764 | ASSERT(bbio == NULL); | |
5765 | return ret; | |
5766 | } | |
5767 | ||
5768 | num_stripes = bbio->num_stripes; | |
5769 | if (*mirror_num > num_stripes) { | |
5770 | /* | |
5771 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5772 | * that means that the requested area is not left of the left | |
5773 | * cursor | |
5774 | */ | |
5775 | btrfs_put_bbio(bbio); | |
5776 | return -EIO; | |
5777 | } | |
5778 | ||
5779 | /* | |
5780 | * process the rest of the function using the mirror_num of the source | |
5781 | * drive. Therefore look it up first. At the end, patch the device | |
5782 | * pointer to the one of the target drive. | |
5783 | */ | |
5784 | for (i = 0; i < num_stripes; i++) { | |
5785 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5786 | continue; | |
5787 | ||
5788 | /* | |
5789 | * In case of DUP, in order to keep it simple, only add the | |
5790 | * mirror with the lowest physical address | |
5791 | */ | |
5792 | if (found && | |
5793 | physical_of_found <= bbio->stripes[i].physical) | |
5794 | continue; | |
5795 | ||
5796 | index_srcdev = i; | |
5797 | found = 1; | |
5798 | physical_of_found = bbio->stripes[i].physical; | |
5799 | } | |
5800 | ||
5801 | btrfs_put_bbio(bbio); | |
5802 | ||
5803 | ASSERT(found); | |
5804 | if (!found) | |
5805 | return -EIO; | |
5806 | ||
5807 | *mirror_num = index_srcdev + 1; | |
5808 | *physical = physical_of_found; | |
5809 | return ret; | |
5810 | } | |
5811 | ||
73c0f228 LB |
5812 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5813 | struct btrfs_bio **bbio_ret, | |
5814 | struct btrfs_dev_replace *dev_replace, | |
5815 | int *num_stripes_ret, int *max_errors_ret) | |
5816 | { | |
5817 | struct btrfs_bio *bbio = *bbio_ret; | |
5818 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5819 | int tgtdev_indexes = 0; | |
5820 | int num_stripes = *num_stripes_ret; | |
5821 | int max_errors = *max_errors_ret; | |
5822 | int i; | |
5823 | ||
5824 | if (op == BTRFS_MAP_WRITE) { | |
5825 | int index_where_to_add; | |
5826 | ||
5827 | /* | |
5828 | * duplicate the write operations while the dev replace | |
5829 | * procedure is running. Since the copying of the old disk to | |
5830 | * the new disk takes place at run time while the filesystem is | |
5831 | * mounted writable, the regular write operations to the old | |
5832 | * disk have to be duplicated to go to the new disk as well. | |
5833 | * | |
5834 | * Note that device->missing is handled by the caller, and that | |
5835 | * the write to the old disk is already set up in the stripes | |
5836 | * array. | |
5837 | */ | |
5838 | index_where_to_add = num_stripes; | |
5839 | for (i = 0; i < num_stripes; i++) { | |
5840 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5841 | /* write to new disk, too */ | |
5842 | struct btrfs_bio_stripe *new = | |
5843 | bbio->stripes + index_where_to_add; | |
5844 | struct btrfs_bio_stripe *old = | |
5845 | bbio->stripes + i; | |
5846 | ||
5847 | new->physical = old->physical; | |
5848 | new->length = old->length; | |
5849 | new->dev = dev_replace->tgtdev; | |
5850 | bbio->tgtdev_map[i] = index_where_to_add; | |
5851 | index_where_to_add++; | |
5852 | max_errors++; | |
5853 | tgtdev_indexes++; | |
5854 | } | |
5855 | } | |
5856 | num_stripes = index_where_to_add; | |
5857 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5858 | int index_srcdev = 0; | |
5859 | int found = 0; | |
5860 | u64 physical_of_found = 0; | |
5861 | ||
5862 | /* | |
5863 | * During the dev-replace procedure, the target drive can also | |
5864 | * be used to read data in case it is needed to repair a corrupt | |
5865 | * block elsewhere. This is possible if the requested area is | |
5866 | * left of the left cursor. In this area, the target drive is a | |
5867 | * full copy of the source drive. | |
5868 | */ | |
5869 | for (i = 0; i < num_stripes; i++) { | |
5870 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5871 | /* | |
5872 | * In case of DUP, in order to keep it simple, | |
5873 | * only add the mirror with the lowest physical | |
5874 | * address | |
5875 | */ | |
5876 | if (found && | |
5877 | physical_of_found <= | |
5878 | bbio->stripes[i].physical) | |
5879 | continue; | |
5880 | index_srcdev = i; | |
5881 | found = 1; | |
5882 | physical_of_found = bbio->stripes[i].physical; | |
5883 | } | |
5884 | } | |
5885 | if (found) { | |
5886 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5887 | bbio->stripes + num_stripes; | |
5888 | ||
5889 | tgtdev_stripe->physical = physical_of_found; | |
5890 | tgtdev_stripe->length = | |
5891 | bbio->stripes[index_srcdev].length; | |
5892 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5893 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5894 | ||
5895 | tgtdev_indexes++; | |
5896 | num_stripes++; | |
5897 | } | |
5898 | } | |
5899 | ||
5900 | *num_stripes_ret = num_stripes; | |
5901 | *max_errors_ret = max_errors; | |
5902 | bbio->num_tgtdevs = tgtdev_indexes; | |
5903 | *bbio_ret = bbio; | |
5904 | } | |
5905 | ||
2b19a1fe LB |
5906 | static bool need_full_stripe(enum btrfs_map_op op) |
5907 | { | |
5908 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5909 | } | |
5910 | ||
5f141126 NB |
5911 | /* |
5912 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5913 | * tuple. This information is used to calculate how big a | |
5914 | * particular bio can get before it straddles a stripe. | |
5915 | * | |
5916 | * @fs_info - the filesystem | |
5917 | * @logical - address that we want to figure out the geometry of | |
5918 | * @len - the length of IO we are going to perform, starting at @logical | |
5919 | * @op - type of operation - write or read | |
5920 | * @io_geom - pointer used to return values | |
5921 | * | |
5922 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5923 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5924 | */ | |
5925 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5926 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5927 | { |
5928 | struct extent_map *em; | |
5929 | struct map_lookup *map; | |
5930 | u64 offset; | |
5931 | u64 stripe_offset; | |
5932 | u64 stripe_nr; | |
5933 | u64 stripe_len; | |
5934 | u64 raid56_full_stripe_start = (u64)-1; | |
5935 | int data_stripes; | |
373c3b80 | 5936 | int ret = 0; |
5f141126 NB |
5937 | |
5938 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5939 | ||
5940 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5941 | if (IS_ERR(em)) | |
5942 | return PTR_ERR(em); | |
5943 | ||
5944 | map = em->map_lookup; | |
5945 | /* Offset of this logical address in the chunk */ | |
5946 | offset = logical - em->start; | |
5947 | /* Len of a stripe in a chunk */ | |
5948 | stripe_len = map->stripe_len; | |
5949 | /* Stripe wher this block falls in */ | |
5950 | stripe_nr = div64_u64(offset, stripe_len); | |
5951 | /* Offset of stripe in the chunk */ | |
5952 | stripe_offset = stripe_nr * stripe_len; | |
5953 | if (offset < stripe_offset) { | |
5954 | btrfs_crit(fs_info, | |
5955 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5956 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5957 | ret = -EINVAL; |
5958 | goto out; | |
5f141126 NB |
5959 | } |
5960 | ||
5961 | /* stripe_offset is the offset of this block in its stripe */ | |
5962 | stripe_offset = offset - stripe_offset; | |
5963 | data_stripes = nr_data_stripes(map); | |
5964 | ||
5965 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5966 | u64 max_len = stripe_len - stripe_offset; | |
5967 | ||
5968 | /* | |
5969 | * In case of raid56, we need to know the stripe aligned start | |
5970 | */ | |
5971 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5972 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5973 | raid56_full_stripe_start = offset; | |
5974 | ||
5975 | /* | |
5976 | * Allow a write of a full stripe, but make sure we | |
5977 | * don't allow straddling of stripes | |
5978 | */ | |
5979 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5980 | full_stripe_len); | |
5981 | raid56_full_stripe_start *= full_stripe_len; | |
5982 | ||
5983 | /* | |
5984 | * For writes to RAID[56], allow a full stripeset across | |
5985 | * all disks. For other RAID types and for RAID[56] | |
5986 | * reads, just allow a single stripe (on a single disk). | |
5987 | */ | |
5988 | if (op == BTRFS_MAP_WRITE) { | |
5989 | max_len = stripe_len * data_stripes - | |
5990 | (offset - raid56_full_stripe_start); | |
5991 | } | |
5992 | } | |
5993 | len = min_t(u64, em->len - offset, max_len); | |
5994 | } else { | |
5995 | len = em->len - offset; | |
5996 | } | |
5997 | ||
5998 | io_geom->len = len; | |
5999 | io_geom->offset = offset; | |
6000 | io_geom->stripe_len = stripe_len; | |
6001 | io_geom->stripe_nr = stripe_nr; | |
6002 | io_geom->stripe_offset = stripe_offset; | |
6003 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
6004 | ||
373c3b80 JT |
6005 | out: |
6006 | /* once for us */ | |
6007 | free_extent_map(em); | |
6008 | return ret; | |
5f141126 NB |
6009 | } |
6010 | ||
cf8cddd3 CH |
6011 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
6012 | enum btrfs_map_op op, | |
f2d8d74d | 6013 | u64 logical, u64 *length, |
a1d3c478 | 6014 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6015 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6016 | { |
6017 | struct extent_map *em; | |
6018 | struct map_lookup *map; | |
593060d7 CM |
6019 | u64 stripe_offset; |
6020 | u64 stripe_nr; | |
53b381b3 | 6021 | u64 stripe_len; |
9d644a62 | 6022 | u32 stripe_index; |
cff82672 | 6023 | int data_stripes; |
cea9e445 | 6024 | int i; |
de11cc12 | 6025 | int ret = 0; |
f2d8d74d | 6026 | int num_stripes; |
a236aed1 | 6027 | int max_errors = 0; |
2c8cdd6e | 6028 | int tgtdev_indexes = 0; |
a1d3c478 | 6029 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6030 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6031 | int dev_replace_is_ongoing = 0; | |
6032 | int num_alloc_stripes; | |
ad6d620e SB |
6033 | int patch_the_first_stripe_for_dev_replace = 0; |
6034 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6035 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6036 | struct btrfs_io_geometry geom; |
6037 | ||
6038 | ASSERT(bbio_ret); | |
75fb2e9e | 6039 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6040 | |
89b798ad NB |
6041 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
6042 | if (ret < 0) | |
6043 | return ret; | |
0b86a832 | 6044 | |
89b798ad | 6045 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 6046 | ASSERT(!IS_ERR(em)); |
95617d69 | 6047 | map = em->map_lookup; |
593060d7 | 6048 | |
89b798ad | 6049 | *length = geom.len; |
89b798ad NB |
6050 | stripe_len = geom.stripe_len; |
6051 | stripe_nr = geom.stripe_nr; | |
6052 | stripe_offset = geom.stripe_offset; | |
6053 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6054 | data_stripes = nr_data_stripes(map); |
593060d7 | 6055 | |
cb5583dd | 6056 | down_read(&dev_replace->rwsem); |
472262f3 | 6057 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6058 | /* |
6059 | * Hold the semaphore for read during the whole operation, write is | |
6060 | * requested at commit time but must wait. | |
6061 | */ | |
472262f3 | 6062 | if (!dev_replace_is_ongoing) |
cb5583dd | 6063 | up_read(&dev_replace->rwsem); |
472262f3 | 6064 | |
ad6d620e | 6065 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6066 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6067 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6068 | dev_replace->srcdev->devid, | |
6069 | &mirror_num, | |
6070 | &physical_to_patch_in_first_stripe); | |
6071 | if (ret) | |
ad6d620e | 6072 | goto out; |
5ab56090 LB |
6073 | else |
6074 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6075 | } else if (mirror_num > map->num_stripes) { |
6076 | mirror_num = 0; | |
6077 | } | |
6078 | ||
f2d8d74d | 6079 | num_stripes = 1; |
cea9e445 | 6080 | stripe_index = 0; |
fce3bb9a | 6081 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6082 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6083 | &stripe_index); | |
de483734 | 6084 | if (!need_full_stripe(op)) |
28e1cc7d | 6085 | mirror_num = 1; |
c7369b3f | 6086 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6087 | if (need_full_stripe(op)) |
f2d8d74d | 6088 | num_stripes = map->num_stripes; |
2fff734f | 6089 | else if (mirror_num) |
f188591e | 6090 | stripe_index = mirror_num - 1; |
dfe25020 | 6091 | else { |
30d9861f | 6092 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6093 | dev_replace_is_ongoing); |
a1d3c478 | 6094 | mirror_num = stripe_index + 1; |
dfe25020 | 6095 | } |
2fff734f | 6096 | |
611f0e00 | 6097 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6098 | if (need_full_stripe(op)) { |
f2d8d74d | 6099 | num_stripes = map->num_stripes; |
a1d3c478 | 6100 | } else if (mirror_num) { |
f188591e | 6101 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6102 | } else { |
6103 | mirror_num = 1; | |
6104 | } | |
2fff734f | 6105 | |
321aecc6 | 6106 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6107 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6108 | |
47c5713f | 6109 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6110 | stripe_index *= map->sub_stripes; |
6111 | ||
de483734 | 6112 | if (need_full_stripe(op)) |
f2d8d74d | 6113 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6114 | else if (mirror_num) |
6115 | stripe_index += mirror_num - 1; | |
dfe25020 | 6116 | else { |
3e74317a | 6117 | int old_stripe_index = stripe_index; |
30d9861f SB |
6118 | stripe_index = find_live_mirror(fs_info, map, |
6119 | stripe_index, | |
30d9861f | 6120 | dev_replace_is_ongoing); |
3e74317a | 6121 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6122 | } |
53b381b3 | 6123 | |
ffe2d203 | 6124 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6125 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6126 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6127 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6128 | stripe_len * data_stripes); |
53b381b3 DW |
6129 | |
6130 | /* RAID[56] write or recovery. Return all stripes */ | |
6131 | num_stripes = map->num_stripes; | |
6132 | max_errors = nr_parity_stripes(map); | |
6133 | ||
53b381b3 DW |
6134 | *length = map->stripe_len; |
6135 | stripe_index = 0; | |
6136 | stripe_offset = 0; | |
6137 | } else { | |
6138 | /* | |
6139 | * Mirror #0 or #1 means the original data block. | |
6140 | * Mirror #2 is RAID5 parity block. | |
6141 | * Mirror #3 is RAID6 Q block. | |
6142 | */ | |
47c5713f | 6143 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6144 | data_stripes, &stripe_index); |
53b381b3 | 6145 | if (mirror_num > 1) |
cff82672 | 6146 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6147 | |
6148 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6149 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6150 | &stripe_index); | |
de483734 | 6151 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6152 | mirror_num = 1; |
53b381b3 | 6153 | } |
8790d502 CM |
6154 | } else { |
6155 | /* | |
47c5713f DS |
6156 | * after this, stripe_nr is the number of stripes on this |
6157 | * device we have to walk to find the data, and stripe_index is | |
6158 | * the number of our device in the stripe array | |
8790d502 | 6159 | */ |
47c5713f DS |
6160 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6161 | &stripe_index); | |
a1d3c478 | 6162 | mirror_num = stripe_index + 1; |
8790d502 | 6163 | } |
e042d1ec | 6164 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6165 | btrfs_crit(fs_info, |
6166 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6167 | stripe_index, map->num_stripes); |
6168 | ret = -EINVAL; | |
6169 | goto out; | |
6170 | } | |
cea9e445 | 6171 | |
472262f3 | 6172 | num_alloc_stripes = num_stripes; |
6fad823f | 6173 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6174 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6175 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6176 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6177 | num_alloc_stripes++; |
2c8cdd6e | 6178 | tgtdev_indexes = num_stripes; |
ad6d620e | 6179 | } |
2c8cdd6e | 6180 | |
6e9606d2 | 6181 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6182 | if (!bbio) { |
6183 | ret = -ENOMEM; | |
6184 | goto out; | |
6185 | } | |
608769a4 NB |
6186 | |
6187 | for (i = 0; i < num_stripes; i++) { | |
6188 | bbio->stripes[i].physical = map->stripes[stripe_index].physical + | |
6189 | stripe_offset + stripe_nr * map->stripe_len; | |
6190 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
6191 | stripe_index++; | |
6192 | } | |
de11cc12 | 6193 | |
8e5cfb55 | 6194 | /* build raid_map */ |
2b19a1fe LB |
6195 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6196 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6197 | u64 tmp; |
9d644a62 | 6198 | unsigned rot; |
8e5cfb55 | 6199 | |
8e5cfb55 | 6200 | /* Work out the disk rotation on this stripe-set */ |
47c5713f | 6201 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6202 | |
6203 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6204 | tmp = stripe_nr * data_stripes; |
6205 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6206 | bbio->raid_map[(i+rot) % num_stripes] = |
6207 | em->start + (tmp + i) * map->stripe_len; | |
6208 | ||
6209 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6210 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6211 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6212 | RAID6_Q_STRIPE; | |
8e5cfb55 | 6213 | |
608769a4 | 6214 | sort_parity_stripes(bbio, num_stripes); |
593060d7 | 6215 | } |
de11cc12 | 6216 | |
2b19a1fe | 6217 | if (need_full_stripe(op)) |
d20983b4 | 6218 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6219 | |
73c0f228 | 6220 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6221 | need_full_stripe(op)) { |
73c0f228 LB |
6222 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6223 | &max_errors); | |
472262f3 SB |
6224 | } |
6225 | ||
de11cc12 | 6226 | *bbio_ret = bbio; |
10f11900 | 6227 | bbio->map_type = map->type; |
de11cc12 LZ |
6228 | bbio->num_stripes = num_stripes; |
6229 | bbio->max_errors = max_errors; | |
6230 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6231 | |
6232 | /* | |
6233 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6234 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6235 | * available as a mirror | |
6236 | */ | |
6237 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6238 | WARN_ON(num_stripes > 1); | |
6239 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6240 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6241 | bbio->mirror_num = map->num_stripes + 1; | |
6242 | } | |
cea9e445 | 6243 | out: |
73beece9 | 6244 | if (dev_replace_is_ongoing) { |
53176dde DS |
6245 | lockdep_assert_held(&dev_replace->rwsem); |
6246 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6247 | up_read(&dev_replace->rwsem); |
73beece9 | 6248 | } |
0b86a832 | 6249 | free_extent_map(em); |
de11cc12 | 6250 | return ret; |
0b86a832 CM |
6251 | } |
6252 | ||
cf8cddd3 | 6253 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6254 | u64 logical, u64 *length, |
a1d3c478 | 6255 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6256 | { |
75fb2e9e DS |
6257 | if (op == BTRFS_MAP_DISCARD) |
6258 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6259 | length, bbio_ret); | |
6260 | ||
b3d3fa51 | 6261 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6262 | mirror_num, 0); |
f2d8d74d CM |
6263 | } |
6264 | ||
af8e2d1d | 6265 | /* For Scrub/replace */ |
cf8cddd3 | 6266 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6267 | u64 logical, u64 *length, |
825ad4c9 | 6268 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6269 | { |
825ad4c9 | 6270 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6271 | } |
6272 | ||
4246a0b6 | 6273 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6274 | { |
326e1dbb MS |
6275 | bio->bi_private = bbio->private; |
6276 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6277 | bio_endio(bio); |
326e1dbb | 6278 | |
6e9606d2 | 6279 | btrfs_put_bbio(bbio); |
8408c716 MX |
6280 | } |
6281 | ||
4246a0b6 | 6282 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6283 | { |
9be3395b | 6284 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6285 | int is_orig_bio = 0; |
8790d502 | 6286 | |
4e4cbee9 | 6287 | if (bio->bi_status) { |
a1d3c478 | 6288 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6289 | if (bio->bi_status == BLK_STS_IOERR || |
6290 | bio->bi_status == BLK_STS_TARGET) { | |
c31efbdf | 6291 | struct btrfs_device *dev = btrfs_io_bio(bio)->device; |
442a4f63 | 6292 | |
3eee86c8 NB |
6293 | ASSERT(dev->bdev); |
6294 | if (bio_op(bio) == REQ_OP_WRITE) | |
6295 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6296 | BTRFS_DEV_STAT_WRITE_ERRS); |
3eee86c8 NB |
6297 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
6298 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6299 | BTRFS_DEV_STAT_READ_ERRS); |
3eee86c8 NB |
6300 | if (bio->bi_opf & REQ_PREFLUSH) |
6301 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6302 | BTRFS_DEV_STAT_FLUSH_ERRS); |
442a4f63 SB |
6303 | } |
6304 | } | |
8790d502 | 6305 | |
a1d3c478 | 6306 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6307 | is_orig_bio = 1; |
6308 | ||
c404e0dc MX |
6309 | btrfs_bio_counter_dec(bbio->fs_info); |
6310 | ||
a1d3c478 | 6311 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6312 | if (!is_orig_bio) { |
6313 | bio_put(bio); | |
a1d3c478 | 6314 | bio = bbio->orig_bio; |
7d2b4daa | 6315 | } |
c7b22bb1 | 6316 | |
9be3395b | 6317 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6318 | /* only send an error to the higher layers if it is |
53b381b3 | 6319 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6320 | */ |
a1d3c478 | 6321 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6322 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6323 | } else { |
1259ab75 CM |
6324 | /* |
6325 | * this bio is actually up to date, we didn't | |
6326 | * go over the max number of errors | |
6327 | */ | |
2dbe0c77 | 6328 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6329 | } |
c55f1396 | 6330 | |
4246a0b6 | 6331 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6332 | } else if (!is_orig_bio) { |
8790d502 CM |
6333 | bio_put(bio); |
6334 | } | |
8790d502 CM |
6335 | } |
6336 | ||
2ff7e61e | 6337 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
c31efbdf | 6338 | u64 physical, struct btrfs_device *dev) |
de1ee92a | 6339 | { |
2ff7e61e | 6340 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6341 | |
6342 | bio->bi_private = bbio; | |
c31efbdf | 6343 | btrfs_io_bio(bio)->device = dev; |
de1ee92a | 6344 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6345 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6346 | btrfs_debug_in_rcu(fs_info, |
6347 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6348 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
1db45a35 DS |
6349 | (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), |
6350 | dev->devid, bio->bi_iter.bi_size); | |
74d46992 | 6351 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6352 | |
2ff7e61e | 6353 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6354 | |
08635bae | 6355 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6356 | } |
6357 | ||
de1ee92a JB |
6358 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6359 | { | |
6360 | atomic_inc(&bbio->error); | |
6361 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6362 | /* Should be the original bio. */ |
8408c716 MX |
6363 | WARN_ON(bio != bbio->orig_bio); |
6364 | ||
9be3395b | 6365 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6366 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6367 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6368 | bio->bi_status = BLK_STS_IOERR; | |
6369 | else | |
6370 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6371 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6372 | } |
6373 | } | |
6374 | ||
58efbc9f | 6375 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6376 | int mirror_num) |
0b86a832 | 6377 | { |
0b86a832 | 6378 | struct btrfs_device *dev; |
8790d502 | 6379 | struct bio *first_bio = bio; |
4f024f37 | 6380 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6381 | u64 length = 0; |
6382 | u64 map_length; | |
0b86a832 | 6383 | int ret; |
08da757d ZL |
6384 | int dev_nr; |
6385 | int total_devs; | |
a1d3c478 | 6386 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6387 | |
4f024f37 | 6388 | length = bio->bi_iter.bi_size; |
0b86a832 | 6389 | map_length = length; |
cea9e445 | 6390 | |
0b246afa | 6391 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6392 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6393 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6394 | if (ret) { |
0b246afa | 6395 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6396 | return errno_to_blk_status(ret); |
c404e0dc | 6397 | } |
cea9e445 | 6398 | |
a1d3c478 | 6399 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6400 | bbio->orig_bio = first_bio; |
6401 | bbio->private = first_bio->bi_private; | |
6402 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6403 | bbio->fs_info = fs_info; |
53b381b3 DW |
6404 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6405 | ||
ad1ba2a0 | 6406 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6407 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6408 | /* In this case, map_length has been set to the length of |
6409 | a single stripe; not the whole write */ | |
37226b21 | 6410 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6411 | ret = raid56_parity_write(fs_info, bio, bbio, |
6412 | map_length); | |
53b381b3 | 6413 | } else { |
2ff7e61e JM |
6414 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6415 | map_length, mirror_num, 1); | |
53b381b3 | 6416 | } |
4245215d | 6417 | |
0b246afa | 6418 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6419 | return errno_to_blk_status(ret); |
53b381b3 DW |
6420 | } |
6421 | ||
cea9e445 | 6422 | if (map_length < length) { |
0b246afa | 6423 | btrfs_crit(fs_info, |
5d163e0e JM |
6424 | "mapping failed logical %llu bio len %llu len %llu", |
6425 | logical, length, map_length); | |
cea9e445 CM |
6426 | BUG(); |
6427 | } | |
a1d3c478 | 6428 | |
08da757d | 6429 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6430 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6431 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6432 | &dev->dev_state) || | |
ebbede42 AJ |
6433 | (bio_op(first_bio) == REQ_OP_WRITE && |
6434 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6435 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6436 | continue; |
6437 | } | |
6438 | ||
3aa8e074 | 6439 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6440 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6441 | else |
a1d3c478 | 6442 | bio = first_bio; |
de1ee92a | 6443 | |
c31efbdf | 6444 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev); |
8790d502 | 6445 | } |
0b246afa | 6446 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6447 | return BLK_STS_OK; |
0b86a832 CM |
6448 | } |
6449 | ||
09ba3bc9 AJ |
6450 | /* |
6451 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6452 | * return NULL. | |
6453 | * | |
6454 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6455 | * only devid is used. | |
6456 | * | |
6457 | * If @seed is true, traverse through the seed devices. | |
6458 | */ | |
e4319cd9 | 6459 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6460 | u64 devid, u8 *uuid, u8 *fsid, |
6461 | bool seed) | |
0b86a832 | 6462 | { |
2b82032c | 6463 | struct btrfs_device *device; |
2b82032c | 6464 | |
e4319cd9 | 6465 | while (fs_devices) { |
2b82032c | 6466 | if (!fsid || |
e4319cd9 | 6467 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6468 | list_for_each_entry(device, &fs_devices->devices, |
6469 | dev_list) { | |
6470 | if (device->devid == devid && | |
6471 | (!uuid || memcmp(device->uuid, uuid, | |
6472 | BTRFS_UUID_SIZE) == 0)) | |
6473 | return device; | |
6474 | } | |
2b82032c | 6475 | } |
09ba3bc9 AJ |
6476 | if (seed) |
6477 | fs_devices = fs_devices->seed; | |
6478 | else | |
6479 | return NULL; | |
2b82032c YZ |
6480 | } |
6481 | return NULL; | |
0b86a832 CM |
6482 | } |
6483 | ||
2ff7e61e | 6484 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6485 | u64 devid, u8 *dev_uuid) |
6486 | { | |
6487 | struct btrfs_device *device; | |
fccc0007 | 6488 | unsigned int nofs_flag; |
dfe25020 | 6489 | |
fccc0007 JB |
6490 | /* |
6491 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6492 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6493 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6494 | * places. | |
6495 | */ | |
6496 | nofs_flag = memalloc_nofs_save(); | |
12bd2fc0 | 6497 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
fccc0007 | 6498 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6499 | if (IS_ERR(device)) |
adfb69af | 6500 | return device; |
12bd2fc0 ID |
6501 | |
6502 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6503 | device->fs_devices = fs_devices; |
dfe25020 | 6504 | fs_devices->num_devices++; |
12bd2fc0 | 6505 | |
e6e674bd | 6506 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6507 | fs_devices->missing_devices++; |
12bd2fc0 | 6508 | |
dfe25020 CM |
6509 | return device; |
6510 | } | |
6511 | ||
12bd2fc0 ID |
6512 | /** |
6513 | * btrfs_alloc_device - allocate struct btrfs_device | |
6514 | * @fs_info: used only for generating a new devid, can be NULL if | |
6515 | * devid is provided (i.e. @devid != NULL). | |
6516 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6517 | * is generated. | |
6518 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6519 | * is generated. | |
6520 | * | |
6521 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6522 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6523 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6524 | */ |
6525 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6526 | const u64 *devid, | |
6527 | const u8 *uuid) | |
6528 | { | |
6529 | struct btrfs_device *dev; | |
6530 | u64 tmp; | |
6531 | ||
fae7f21c | 6532 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6533 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6534 | |
6535 | dev = __alloc_device(); | |
6536 | if (IS_ERR(dev)) | |
6537 | return dev; | |
6538 | ||
6539 | if (devid) | |
6540 | tmp = *devid; | |
6541 | else { | |
6542 | int ret; | |
6543 | ||
6544 | ret = find_next_devid(fs_info, &tmp); | |
6545 | if (ret) { | |
a425f9d4 | 6546 | btrfs_free_device(dev); |
12bd2fc0 ID |
6547 | return ERR_PTR(ret); |
6548 | } | |
6549 | } | |
6550 | dev->devid = tmp; | |
6551 | ||
6552 | if (uuid) | |
6553 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6554 | else | |
6555 | generate_random_uuid(dev->uuid); | |
6556 | ||
12bd2fc0 ID |
6557 | return dev; |
6558 | } | |
6559 | ||
5a2b8e60 | 6560 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6561 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6562 | { |
2b902dfc AJ |
6563 | if (error) |
6564 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6565 | devid, uuid); | |
6566 | else | |
6567 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6568 | devid, uuid); | |
5a2b8e60 AJ |
6569 | } |
6570 | ||
39e264a4 NB |
6571 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6572 | { | |
6573 | int index = btrfs_bg_flags_to_raid_index(type); | |
6574 | int ncopies = btrfs_raid_array[index].ncopies; | |
e4f6c6be | 6575 | const int nparity = btrfs_raid_array[index].nparity; |
39e264a4 NB |
6576 | int data_stripes; |
6577 | ||
e4f6c6be DS |
6578 | if (nparity) |
6579 | data_stripes = num_stripes - nparity; | |
6580 | else | |
39e264a4 | 6581 | data_stripes = num_stripes / ncopies; |
e4f6c6be | 6582 | |
39e264a4 NB |
6583 | return div_u64(chunk_len, data_stripes); |
6584 | } | |
6585 | ||
9690ac09 | 6586 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6587 | struct btrfs_chunk *chunk) |
6588 | { | |
9690ac09 | 6589 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6590 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6591 | struct map_lookup *map; |
6592 | struct extent_map *em; | |
6593 | u64 logical; | |
6594 | u64 length; | |
e06cd3dd LB |
6595 | u64 devid; |
6596 | u8 uuid[BTRFS_UUID_SIZE]; | |
6597 | int num_stripes; | |
6598 | int ret; | |
6599 | int i; | |
6600 | ||
6601 | logical = key->offset; | |
6602 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6603 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6604 | ||
075cb3c7 QW |
6605 | /* |
6606 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6607 | * as chunk item in tree block is already verified by tree-checker. | |
6608 | */ | |
6609 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6610 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6611 | if (ret) |
6612 | return ret; | |
6613 | } | |
a061fc8d | 6614 | |
c8bf1b67 DS |
6615 | read_lock(&map_tree->lock); |
6616 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6617 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6618 | |
6619 | /* already mapped? */ | |
6620 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6621 | free_extent_map(em); | |
0b86a832 CM |
6622 | return 0; |
6623 | } else if (em) { | |
6624 | free_extent_map(em); | |
6625 | } | |
0b86a832 | 6626 | |
172ddd60 | 6627 | em = alloc_extent_map(); |
0b86a832 CM |
6628 | if (!em) |
6629 | return -ENOMEM; | |
593060d7 | 6630 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6631 | if (!map) { |
6632 | free_extent_map(em); | |
6633 | return -ENOMEM; | |
6634 | } | |
6635 | ||
298a8f9c | 6636 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6637 | em->map_lookup = map; |
0b86a832 CM |
6638 | em->start = logical; |
6639 | em->len = length; | |
70c8a91c | 6640 | em->orig_start = 0; |
0b86a832 | 6641 | em->block_start = 0; |
c8b97818 | 6642 | em->block_len = em->len; |
0b86a832 | 6643 | |
593060d7 CM |
6644 | map->num_stripes = num_stripes; |
6645 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6646 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6647 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6648 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6649 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6650 | map->verified_stripes = 0; |
39e264a4 NB |
6651 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6652 | map->num_stripes); | |
593060d7 CM |
6653 | for (i = 0; i < num_stripes; i++) { |
6654 | map->stripes[i].physical = | |
6655 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6656 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6657 | read_extent_buffer(leaf, uuid, (unsigned long) |
6658 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6659 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6660 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6661 | devid, uuid, NULL, true); |
3cdde224 | 6662 | if (!map->stripes[i].dev && |
0b246afa | 6663 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6664 | free_extent_map(em); |
2b902dfc | 6665 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6666 | return -ENOENT; |
593060d7 | 6667 | } |
dfe25020 CM |
6668 | if (!map->stripes[i].dev) { |
6669 | map->stripes[i].dev = | |
2ff7e61e JM |
6670 | add_missing_dev(fs_info->fs_devices, devid, |
6671 | uuid); | |
adfb69af | 6672 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6673 | free_extent_map(em); |
adfb69af AJ |
6674 | btrfs_err(fs_info, |
6675 | "failed to init missing dev %llu: %ld", | |
6676 | devid, PTR_ERR(map->stripes[i].dev)); | |
6677 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6678 | } |
2b902dfc | 6679 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6680 | } |
e12c9621 AJ |
6681 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6682 | &(map->stripes[i].dev->dev_state)); | |
6683 | ||
0b86a832 CM |
6684 | } |
6685 | ||
c8bf1b67 DS |
6686 | write_lock(&map_tree->lock); |
6687 | ret = add_extent_mapping(map_tree, em, 0); | |
6688 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6689 | if (ret < 0) { |
6690 | btrfs_err(fs_info, | |
6691 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6692 | em->start, em->len, ret); | |
6693 | } | |
0b86a832 CM |
6694 | free_extent_map(em); |
6695 | ||
64f64f43 | 6696 | return ret; |
0b86a832 CM |
6697 | } |
6698 | ||
143bede5 | 6699 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6700 | struct btrfs_dev_item *dev_item, |
6701 | struct btrfs_device *device) | |
6702 | { | |
6703 | unsigned long ptr; | |
0b86a832 CM |
6704 | |
6705 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6706 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6707 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6708 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6709 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6710 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6711 | device->type = btrfs_device_type(leaf, dev_item); |
6712 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6713 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6714 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6715 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6716 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6717 | |
410ba3a2 | 6718 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6719 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6720 | } |
6721 | ||
2ff7e61e | 6722 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6723 | u8 *fsid) |
2b82032c YZ |
6724 | { |
6725 | struct btrfs_fs_devices *fs_devices; | |
6726 | int ret; | |
6727 | ||
a32bf9a3 | 6728 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6729 | ASSERT(fsid); |
2b82032c | 6730 | |
0b246afa | 6731 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6732 | while (fs_devices) { |
44880fdc | 6733 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6734 | return fs_devices; |
6735 | ||
2b82032c YZ |
6736 | fs_devices = fs_devices->seed; |
6737 | } | |
6738 | ||
7239ff4b | 6739 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6740 | if (!fs_devices) { |
0b246afa | 6741 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6742 | return ERR_PTR(-ENOENT); |
6743 | ||
7239ff4b | 6744 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6745 | if (IS_ERR(fs_devices)) |
6746 | return fs_devices; | |
6747 | ||
0395d84f | 6748 | fs_devices->seeding = true; |
5f375835 MX |
6749 | fs_devices->opened = 1; |
6750 | return fs_devices; | |
2b82032c | 6751 | } |
e4404d6e YZ |
6752 | |
6753 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6754 | if (IS_ERR(fs_devices)) |
6755 | return fs_devices; | |
2b82032c | 6756 | |
897fb573 | 6757 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6758 | if (ret) { |
6759 | free_fs_devices(fs_devices); | |
5f375835 | 6760 | fs_devices = ERR_PTR(ret); |
2b82032c | 6761 | goto out; |
48d28232 | 6762 | } |
2b82032c YZ |
6763 | |
6764 | if (!fs_devices->seeding) { | |
0226e0eb | 6765 | close_fs_devices(fs_devices); |
e4404d6e | 6766 | free_fs_devices(fs_devices); |
5f375835 | 6767 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6768 | goto out; |
6769 | } | |
6770 | ||
0b246afa JM |
6771 | fs_devices->seed = fs_info->fs_devices->seed; |
6772 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6773 | out: |
5f375835 | 6774 | return fs_devices; |
2b82032c YZ |
6775 | } |
6776 | ||
17850759 | 6777 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6778 | struct btrfs_dev_item *dev_item) |
6779 | { | |
17850759 | 6780 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6781 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6782 | struct btrfs_device *device; |
6783 | u64 devid; | |
6784 | int ret; | |
44880fdc | 6785 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6786 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6787 | ||
0b86a832 | 6788 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6789 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6790 | BTRFS_UUID_SIZE); |
1473b24e | 6791 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6792 | BTRFS_FSID_SIZE); |
2b82032c | 6793 | |
de37aa51 | 6794 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6795 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6796 | if (IS_ERR(fs_devices)) |
6797 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6798 | } |
6799 | ||
e4319cd9 | 6800 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6801 | fs_uuid, true); |
5f375835 | 6802 | if (!device) { |
c5502451 | 6803 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6804 | btrfs_report_missing_device(fs_info, devid, |
6805 | dev_uuid, true); | |
45dbdbc9 | 6806 | return -ENOENT; |
c5502451 | 6807 | } |
2b82032c | 6808 | |
2ff7e61e | 6809 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6810 | if (IS_ERR(device)) { |
6811 | btrfs_err(fs_info, | |
6812 | "failed to add missing dev %llu: %ld", | |
6813 | devid, PTR_ERR(device)); | |
6814 | return PTR_ERR(device); | |
6815 | } | |
2b902dfc | 6816 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6817 | } else { |
c5502451 | 6818 | if (!device->bdev) { |
2b902dfc AJ |
6819 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6820 | btrfs_report_missing_device(fs_info, | |
6821 | devid, dev_uuid, true); | |
45dbdbc9 | 6822 | return -ENOENT; |
2b902dfc AJ |
6823 | } |
6824 | btrfs_report_missing_device(fs_info, devid, | |
6825 | dev_uuid, false); | |
c5502451 | 6826 | } |
5f375835 | 6827 | |
e6e674bd AJ |
6828 | if (!device->bdev && |
6829 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6830 | /* |
6831 | * this happens when a device that was properly setup | |
6832 | * in the device info lists suddenly goes bad. | |
6833 | * device->bdev is NULL, and so we have to set | |
6834 | * device->missing to one here | |
6835 | */ | |
5f375835 | 6836 | device->fs_devices->missing_devices++; |
e6e674bd | 6837 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6838 | } |
5f375835 MX |
6839 | |
6840 | /* Move the device to its own fs_devices */ | |
6841 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6842 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6843 | &device->dev_state)); | |
5f375835 MX |
6844 | |
6845 | list_move(&device->dev_list, &fs_devices->devices); | |
6846 | device->fs_devices->num_devices--; | |
6847 | fs_devices->num_devices++; | |
6848 | ||
6849 | device->fs_devices->missing_devices--; | |
6850 | fs_devices->missing_devices++; | |
6851 | ||
6852 | device->fs_devices = fs_devices; | |
6853 | } | |
2b82032c YZ |
6854 | } |
6855 | ||
0b246afa | 6856 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6857 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6858 | if (device->generation != |
6859 | btrfs_device_generation(leaf, dev_item)) | |
6860 | return -EINVAL; | |
6324fbf3 | 6861 | } |
0b86a832 CM |
6862 | |
6863 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6864 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6865 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6866 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6867 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6868 | atomic64_add(device->total_bytes - device->bytes_used, |
6869 | &fs_info->free_chunk_space); | |
2bf64758 | 6870 | } |
0b86a832 | 6871 | ret = 0; |
0b86a832 CM |
6872 | return ret; |
6873 | } | |
6874 | ||
6bccf3ab | 6875 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6876 | { |
6bccf3ab | 6877 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6878 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6879 | struct extent_buffer *sb; |
0b86a832 | 6880 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6881 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6882 | u8 *array_ptr; |
6883 | unsigned long sb_array_offset; | |
84eed90f | 6884 | int ret = 0; |
0b86a832 CM |
6885 | u32 num_stripes; |
6886 | u32 array_size; | |
6887 | u32 len = 0; | |
1ffb22cf | 6888 | u32 cur_offset; |
e06cd3dd | 6889 | u64 type; |
84eed90f | 6890 | struct btrfs_key key; |
0b86a832 | 6891 | |
0b246afa | 6892 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6893 | /* |
6894 | * This will create extent buffer of nodesize, superblock size is | |
6895 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6896 | * overallocate but we can keep it as-is, only the first page is used. | |
6897 | */ | |
2ff7e61e | 6898 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
6899 | if (IS_ERR(sb)) |
6900 | return PTR_ERR(sb); | |
4db8c528 | 6901 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6902 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6903 | /* |
01327610 | 6904 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6905 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6906 | * pages up-to-date when the page is larger: extent does not cover the |
6907 | * whole page and consequently check_page_uptodate does not find all | |
6908 | * the page's extents up-to-date (the hole beyond sb), | |
6909 | * write_extent_buffer then triggers a WARN_ON. | |
6910 | * | |
6911 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6912 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6913 | * to silence the warning eg. on PowerPC 64. | |
6914 | */ | |
09cbfeaf | 6915 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6916 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6917 | |
a061fc8d | 6918 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6919 | array_size = btrfs_super_sys_array_size(super_copy); |
6920 | ||
1ffb22cf DS |
6921 | array_ptr = super_copy->sys_chunk_array; |
6922 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6923 | cur_offset = 0; | |
0b86a832 | 6924 | |
1ffb22cf DS |
6925 | while (cur_offset < array_size) { |
6926 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6927 | len = sizeof(*disk_key); |
6928 | if (cur_offset + len > array_size) | |
6929 | goto out_short_read; | |
6930 | ||
0b86a832 CM |
6931 | btrfs_disk_key_to_cpu(&key, disk_key); |
6932 | ||
1ffb22cf DS |
6933 | array_ptr += len; |
6934 | sb_array_offset += len; | |
6935 | cur_offset += len; | |
0b86a832 | 6936 | |
32ab3d1b JT |
6937 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
6938 | btrfs_err(fs_info, | |
6939 | "unexpected item type %u in sys_array at offset %u", | |
6940 | (u32)key.type, cur_offset); | |
6941 | ret = -EIO; | |
6942 | break; | |
6943 | } | |
f5cdedd7 | 6944 | |
32ab3d1b JT |
6945 | chunk = (struct btrfs_chunk *)sb_array_offset; |
6946 | /* | |
6947 | * At least one btrfs_chunk with one stripe must be present, | |
6948 | * exact stripe count check comes afterwards | |
6949 | */ | |
6950 | len = btrfs_chunk_item_size(1); | |
6951 | if (cur_offset + len > array_size) | |
6952 | goto out_short_read; | |
e06cd3dd | 6953 | |
32ab3d1b JT |
6954 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
6955 | if (!num_stripes) { | |
6956 | btrfs_err(fs_info, | |
6957 | "invalid number of stripes %u in sys_array at offset %u", | |
6958 | num_stripes, cur_offset); | |
6959 | ret = -EIO; | |
6960 | break; | |
6961 | } | |
e3540eab | 6962 | |
32ab3d1b JT |
6963 | type = btrfs_chunk_type(sb, chunk); |
6964 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6965 | btrfs_err(fs_info, |
32ab3d1b JT |
6966 | "invalid chunk type %llu in sys_array at offset %u", |
6967 | type, cur_offset); | |
84eed90f CM |
6968 | ret = -EIO; |
6969 | break; | |
0b86a832 | 6970 | } |
32ab3d1b JT |
6971 | |
6972 | len = btrfs_chunk_item_size(num_stripes); | |
6973 | if (cur_offset + len > array_size) | |
6974 | goto out_short_read; | |
6975 | ||
6976 | ret = read_one_chunk(&key, sb, chunk); | |
6977 | if (ret) | |
6978 | break; | |
6979 | ||
1ffb22cf DS |
6980 | array_ptr += len; |
6981 | sb_array_offset += len; | |
6982 | cur_offset += len; | |
0b86a832 | 6983 | } |
d865177a | 6984 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6985 | free_extent_buffer_stale(sb); |
84eed90f | 6986 | return ret; |
e3540eab DS |
6987 | |
6988 | out_short_read: | |
ab8d0fc4 | 6989 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6990 | len, cur_offset); |
d865177a | 6991 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6992 | free_extent_buffer_stale(sb); |
e3540eab | 6993 | return -EIO; |
0b86a832 CM |
6994 | } |
6995 | ||
21634a19 QW |
6996 | /* |
6997 | * Check if all chunks in the fs are OK for read-write degraded mount | |
6998 | * | |
6528b99d AJ |
6999 | * If the @failing_dev is specified, it's accounted as missing. |
7000 | * | |
21634a19 QW |
7001 | * Return true if all chunks meet the minimal RW mount requirements. |
7002 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7003 | */ | |
6528b99d AJ |
7004 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7005 | struct btrfs_device *failing_dev) | |
21634a19 | 7006 | { |
c8bf1b67 | 7007 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7008 | struct extent_map *em; |
7009 | u64 next_start = 0; | |
7010 | bool ret = true; | |
7011 | ||
c8bf1b67 DS |
7012 | read_lock(&map_tree->lock); |
7013 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7014 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7015 | /* No chunk at all? Return false anyway */ |
7016 | if (!em) { | |
7017 | ret = false; | |
7018 | goto out; | |
7019 | } | |
7020 | while (em) { | |
7021 | struct map_lookup *map; | |
7022 | int missing = 0; | |
7023 | int max_tolerated; | |
7024 | int i; | |
7025 | ||
7026 | map = em->map_lookup; | |
7027 | max_tolerated = | |
7028 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7029 | map->type); | |
7030 | for (i = 0; i < map->num_stripes; i++) { | |
7031 | struct btrfs_device *dev = map->stripes[i].dev; | |
7032 | ||
e6e674bd AJ |
7033 | if (!dev || !dev->bdev || |
7034 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7035 | dev->last_flush_error) |
7036 | missing++; | |
6528b99d AJ |
7037 | else if (failing_dev && failing_dev == dev) |
7038 | missing++; | |
21634a19 QW |
7039 | } |
7040 | if (missing > max_tolerated) { | |
6528b99d AJ |
7041 | if (!failing_dev) |
7042 | btrfs_warn(fs_info, | |
52042d8e | 7043 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7044 | em->start, missing, max_tolerated); |
7045 | free_extent_map(em); | |
7046 | ret = false; | |
7047 | goto out; | |
7048 | } | |
7049 | next_start = extent_map_end(em); | |
7050 | free_extent_map(em); | |
7051 | ||
c8bf1b67 DS |
7052 | read_lock(&map_tree->lock); |
7053 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7054 | (u64)(-1) - next_start); |
c8bf1b67 | 7055 | read_unlock(&map_tree->lock); |
21634a19 QW |
7056 | } |
7057 | out: | |
7058 | return ret; | |
7059 | } | |
7060 | ||
d85327b1 DS |
7061 | static void readahead_tree_node_children(struct extent_buffer *node) |
7062 | { | |
7063 | int i; | |
7064 | const int nr_items = btrfs_header_nritems(node); | |
7065 | ||
7066 | for (i = 0; i < nr_items; i++) { | |
7067 | u64 start; | |
7068 | ||
7069 | start = btrfs_node_blockptr(node, i); | |
7070 | readahead_tree_block(node->fs_info, start); | |
7071 | } | |
7072 | } | |
7073 | ||
5b4aacef | 7074 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7075 | { |
5b4aacef | 7076 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7077 | struct btrfs_path *path; |
7078 | struct extent_buffer *leaf; | |
7079 | struct btrfs_key key; | |
7080 | struct btrfs_key found_key; | |
7081 | int ret; | |
7082 | int slot; | |
99e3ecfc | 7083 | u64 total_dev = 0; |
d85327b1 | 7084 | u64 last_ra_node = 0; |
0b86a832 | 7085 | |
0b86a832 CM |
7086 | path = btrfs_alloc_path(); |
7087 | if (!path) | |
7088 | return -ENOMEM; | |
7089 | ||
3dd0f7a3 AJ |
7090 | /* |
7091 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7092 | * otherwise we don't need it. | |
7093 | */ | |
b367e47f | 7094 | mutex_lock(&uuid_mutex); |
b367e47f | 7095 | |
48cfa61b BB |
7096 | /* |
7097 | * It is possible for mount and umount to race in such a way that | |
7098 | * we execute this code path, but open_fs_devices failed to clear | |
7099 | * total_rw_bytes. We certainly want it cleared before reading the | |
7100 | * device items, so clear it here. | |
7101 | */ | |
7102 | fs_info->fs_devices->total_rw_bytes = 0; | |
7103 | ||
395927a9 FDBM |
7104 | /* |
7105 | * Read all device items, and then all the chunk items. All | |
7106 | * device items are found before any chunk item (their object id | |
7107 | * is smaller than the lowest possible object id for a chunk | |
7108 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7109 | */ |
7110 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7111 | key.offset = 0; | |
7112 | key.type = 0; | |
0b86a832 | 7113 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7114 | if (ret < 0) |
7115 | goto error; | |
d397712b | 7116 | while (1) { |
d85327b1 DS |
7117 | struct extent_buffer *node; |
7118 | ||
0b86a832 CM |
7119 | leaf = path->nodes[0]; |
7120 | slot = path->slots[0]; | |
7121 | if (slot >= btrfs_header_nritems(leaf)) { | |
7122 | ret = btrfs_next_leaf(root, path); | |
7123 | if (ret == 0) | |
7124 | continue; | |
7125 | if (ret < 0) | |
7126 | goto error; | |
7127 | break; | |
7128 | } | |
d85327b1 DS |
7129 | /* |
7130 | * The nodes on level 1 are not locked but we don't need to do | |
7131 | * that during mount time as nothing else can access the tree | |
7132 | */ | |
7133 | node = path->nodes[1]; | |
7134 | if (node) { | |
7135 | if (last_ra_node != node->start) { | |
7136 | readahead_tree_node_children(node); | |
7137 | last_ra_node = node->start; | |
7138 | } | |
7139 | } | |
0b86a832 | 7140 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
395927a9 FDBM |
7141 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7142 | struct btrfs_dev_item *dev_item; | |
7143 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7144 | struct btrfs_dev_item); |
17850759 | 7145 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7146 | if (ret) |
7147 | goto error; | |
99e3ecfc | 7148 | total_dev++; |
0b86a832 CM |
7149 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7150 | struct btrfs_chunk *chunk; | |
7151 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
01d01caf | 7152 | mutex_lock(&fs_info->chunk_mutex); |
9690ac09 | 7153 | ret = read_one_chunk(&found_key, leaf, chunk); |
01d01caf | 7154 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c YZ |
7155 | if (ret) |
7156 | goto error; | |
0b86a832 CM |
7157 | } |
7158 | path->slots[0]++; | |
7159 | } | |
99e3ecfc LB |
7160 | |
7161 | /* | |
7162 | * After loading chunk tree, we've got all device information, | |
7163 | * do another round of validation checks. | |
7164 | */ | |
0b246afa JM |
7165 | if (total_dev != fs_info->fs_devices->total_devices) { |
7166 | btrfs_err(fs_info, | |
99e3ecfc | 7167 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7168 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7169 | total_dev); |
7170 | ret = -EINVAL; | |
7171 | goto error; | |
7172 | } | |
0b246afa JM |
7173 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7174 | fs_info->fs_devices->total_rw_bytes) { | |
7175 | btrfs_err(fs_info, | |
99e3ecfc | 7176 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7177 | btrfs_super_total_bytes(fs_info->super_copy), |
7178 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7179 | ret = -EINVAL; |
7180 | goto error; | |
7181 | } | |
0b86a832 CM |
7182 | ret = 0; |
7183 | error: | |
b367e47f LZ |
7184 | mutex_unlock(&uuid_mutex); |
7185 | ||
2b82032c | 7186 | btrfs_free_path(path); |
0b86a832 CM |
7187 | return ret; |
7188 | } | |
442a4f63 | 7189 | |
cb517eab MX |
7190 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7191 | { | |
7192 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7193 | struct btrfs_device *device; | |
7194 | ||
29cc83f6 LB |
7195 | while (fs_devices) { |
7196 | mutex_lock(&fs_devices->device_list_mutex); | |
7197 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7198 | device->fs_info = fs_info; |
29cc83f6 LB |
7199 | mutex_unlock(&fs_devices->device_list_mutex); |
7200 | ||
7201 | fs_devices = fs_devices->seed; | |
7202 | } | |
cb517eab MX |
7203 | } |
7204 | ||
1dc990df DS |
7205 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7206 | const struct btrfs_dev_stats_item *ptr, | |
7207 | int index) | |
7208 | { | |
7209 | u64 val; | |
7210 | ||
7211 | read_extent_buffer(eb, &val, | |
7212 | offsetof(struct btrfs_dev_stats_item, values) + | |
7213 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7214 | sizeof(val)); | |
7215 | return val; | |
7216 | } | |
7217 | ||
7218 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7219 | struct btrfs_dev_stats_item *ptr, | |
7220 | int index, u64 val) | |
7221 | { | |
7222 | write_extent_buffer(eb, &val, | |
7223 | offsetof(struct btrfs_dev_stats_item, values) + | |
7224 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7225 | sizeof(val)); | |
7226 | } | |
7227 | ||
733f4fbb SB |
7228 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7229 | { | |
7230 | struct btrfs_key key; | |
733f4fbb SB |
7231 | struct btrfs_root *dev_root = fs_info->dev_root; |
7232 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7233 | struct extent_buffer *eb; | |
7234 | int slot; | |
7235 | int ret = 0; | |
7236 | struct btrfs_device *device; | |
7237 | struct btrfs_path *path = NULL; | |
7238 | int i; | |
7239 | ||
7240 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7241 | if (!path) |
7242 | return -ENOMEM; | |
733f4fbb SB |
7243 | |
7244 | mutex_lock(&fs_devices->device_list_mutex); | |
7245 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7246 | int item_size; | |
7247 | struct btrfs_dev_stats_item *ptr; | |
7248 | ||
242e2956 DS |
7249 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7250 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7251 | key.offset = device->devid; |
7252 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7253 | if (ret) { | |
ae4b9b4c AJ |
7254 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7255 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7256 | device->dev_stats_valid = 1; |
7257 | btrfs_release_path(path); | |
7258 | continue; | |
7259 | } | |
7260 | slot = path->slots[0]; | |
7261 | eb = path->nodes[0]; | |
733f4fbb SB |
7262 | item_size = btrfs_item_size_nr(eb, slot); |
7263 | ||
7264 | ptr = btrfs_item_ptr(eb, slot, | |
7265 | struct btrfs_dev_stats_item); | |
7266 | ||
7267 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7268 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7269 | btrfs_dev_stat_set(device, i, | |
7270 | btrfs_dev_stats_value(eb, ptr, i)); | |
7271 | else | |
4e411a7d | 7272 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7273 | } |
7274 | ||
7275 | device->dev_stats_valid = 1; | |
7276 | btrfs_dev_stat_print_on_load(device); | |
7277 | btrfs_release_path(path); | |
7278 | } | |
7279 | mutex_unlock(&fs_devices->device_list_mutex); | |
7280 | ||
733f4fbb SB |
7281 | btrfs_free_path(path); |
7282 | return ret < 0 ? ret : 0; | |
7283 | } | |
7284 | ||
7285 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7286 | struct btrfs_device *device) |
7287 | { | |
5495f195 | 7288 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7289 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7290 | struct btrfs_path *path; |
7291 | struct btrfs_key key; | |
7292 | struct extent_buffer *eb; | |
7293 | struct btrfs_dev_stats_item *ptr; | |
7294 | int ret; | |
7295 | int i; | |
7296 | ||
242e2956 DS |
7297 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7298 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7299 | key.offset = device->devid; |
7300 | ||
7301 | path = btrfs_alloc_path(); | |
fa252992 DS |
7302 | if (!path) |
7303 | return -ENOMEM; | |
733f4fbb SB |
7304 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7305 | if (ret < 0) { | |
0b246afa | 7306 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7307 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7308 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7309 | goto out; |
7310 | } | |
7311 | ||
7312 | if (ret == 0 && | |
7313 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7314 | /* need to delete old one and insert a new one */ | |
7315 | ret = btrfs_del_item(trans, dev_root, path); | |
7316 | if (ret != 0) { | |
0b246afa | 7317 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7318 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7319 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7320 | goto out; |
7321 | } | |
7322 | ret = 1; | |
7323 | } | |
7324 | ||
7325 | if (ret == 1) { | |
7326 | /* need to insert a new item */ | |
7327 | btrfs_release_path(path); | |
7328 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7329 | &key, sizeof(*ptr)); | |
7330 | if (ret < 0) { | |
0b246afa | 7331 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7332 | "insert dev_stats item for device %s failed %d", |
7333 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7334 | goto out; |
7335 | } | |
7336 | } | |
7337 | ||
7338 | eb = path->nodes[0]; | |
7339 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7340 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7341 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7342 | btrfs_dev_stat_read(device, i)); | |
7343 | btrfs_mark_buffer_dirty(eb); | |
7344 | ||
7345 | out: | |
7346 | btrfs_free_path(path); | |
7347 | return ret; | |
7348 | } | |
7349 | ||
7350 | /* | |
7351 | * called from commit_transaction. Writes all changed device stats to disk. | |
7352 | */ | |
196c9d8d | 7353 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7354 | { |
196c9d8d | 7355 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7356 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7357 | struct btrfs_device *device; | |
addc3fa7 | 7358 | int stats_cnt; |
733f4fbb SB |
7359 | int ret = 0; |
7360 | ||
7361 | mutex_lock(&fs_devices->device_list_mutex); | |
7362 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7363 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7364 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7365 | continue; |
7366 | ||
9deae968 NB |
7367 | |
7368 | /* | |
7369 | * There is a LOAD-LOAD control dependency between the value of | |
7370 | * dev_stats_ccnt and updating the on-disk values which requires | |
7371 | * reading the in-memory counters. Such control dependencies | |
7372 | * require explicit read memory barriers. | |
7373 | * | |
7374 | * This memory barriers pairs with smp_mb__before_atomic in | |
7375 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7376 | * barrier implied by atomic_xchg in | |
7377 | * btrfs_dev_stats_read_and_reset | |
7378 | */ | |
7379 | smp_rmb(); | |
7380 | ||
5495f195 | 7381 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7382 | if (!ret) |
addc3fa7 | 7383 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7384 | } |
7385 | mutex_unlock(&fs_devices->device_list_mutex); | |
7386 | ||
7387 | return ret; | |
7388 | } | |
7389 | ||
442a4f63 SB |
7390 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7391 | { | |
7392 | btrfs_dev_stat_inc(dev, index); | |
7393 | btrfs_dev_stat_print_on_error(dev); | |
7394 | } | |
7395 | ||
48a3b636 | 7396 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7397 | { |
733f4fbb SB |
7398 | if (!dev->dev_stats_valid) |
7399 | return; | |
fb456252 | 7400 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7401 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7402 | rcu_str_deref(dev->name), |
442a4f63 SB |
7403 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7404 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7405 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7406 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7407 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7408 | } |
c11d2c23 | 7409 | |
733f4fbb SB |
7410 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7411 | { | |
a98cdb85 SB |
7412 | int i; |
7413 | ||
7414 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7415 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7416 | break; | |
7417 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7418 | return; /* all values == 0, suppress message */ | |
7419 | ||
fb456252 | 7420 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7421 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7422 | rcu_str_deref(dev->name), |
733f4fbb SB |
7423 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7424 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7425 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7426 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7427 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7428 | } | |
7429 | ||
2ff7e61e | 7430 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7431 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7432 | { |
7433 | struct btrfs_device *dev; | |
0b246afa | 7434 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7435 | int i; |
7436 | ||
7437 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7438 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7439 | true); | |
c11d2c23 SB |
7440 | mutex_unlock(&fs_devices->device_list_mutex); |
7441 | ||
7442 | if (!dev) { | |
0b246afa | 7443 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7444 | return -ENODEV; |
733f4fbb | 7445 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7446 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7447 | return -ENODEV; |
b27f7c0c | 7448 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7449 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7450 | if (stats->nr_items > i) | |
7451 | stats->values[i] = | |
7452 | btrfs_dev_stat_read_and_reset(dev, i); | |
7453 | else | |
4e411a7d | 7454 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7455 | } |
a69976bc AJ |
7456 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7457 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7458 | } else { |
7459 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7460 | if (stats->nr_items > i) | |
7461 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7462 | } | |
7463 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7464 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7465 | return 0; | |
7466 | } | |
a8a6dab7 | 7467 | |
935e5cc9 | 7468 | /* |
bbbf7243 NB |
7469 | * Update the size and bytes used for each device where it changed. This is |
7470 | * delayed since we would otherwise get errors while writing out the | |
7471 | * superblocks. | |
7472 | * | |
7473 | * Must be invoked during transaction commit. | |
935e5cc9 | 7474 | */ |
bbbf7243 | 7475 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7476 | { |
935e5cc9 MX |
7477 | struct btrfs_device *curr, *next; |
7478 | ||
bbbf7243 | 7479 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7480 | |
bbbf7243 | 7481 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7482 | return; |
7483 | ||
bbbf7243 NB |
7484 | /* |
7485 | * We don't need the device_list_mutex here. This list is owned by the | |
7486 | * transaction and the transaction must complete before the device is | |
7487 | * released. | |
7488 | */ | |
7489 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7490 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7491 | post_commit_list) { | |
7492 | list_del_init(&curr->post_commit_list); | |
7493 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7494 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7495 | } |
bbbf7243 | 7496 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7497 | } |
5a13f430 AJ |
7498 | |
7499 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7500 | { | |
7501 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7502 | while (fs_devices) { | |
7503 | fs_devices->fs_info = fs_info; | |
7504 | fs_devices = fs_devices->seed; | |
7505 | } | |
7506 | } | |
7507 | ||
7508 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7509 | { | |
7510 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7511 | while (fs_devices) { | |
7512 | fs_devices->fs_info = NULL; | |
7513 | fs_devices = fs_devices->seed; | |
7514 | } | |
7515 | } | |
46df06b8 DS |
7516 | |
7517 | /* | |
7518 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7519 | */ | |
7520 | int btrfs_bg_type_to_factor(u64 flags) | |
7521 | { | |
44b28ada DS |
7522 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7523 | ||
7524 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7525 | } |
cf90d884 QW |
7526 | |
7527 | ||
cf90d884 QW |
7528 | |
7529 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7530 | u64 chunk_offset, u64 devid, | |
7531 | u64 physical_offset, u64 physical_len) | |
7532 | { | |
c8bf1b67 | 7533 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7534 | struct extent_map *em; |
7535 | struct map_lookup *map; | |
05a37c48 | 7536 | struct btrfs_device *dev; |
cf90d884 QW |
7537 | u64 stripe_len; |
7538 | bool found = false; | |
7539 | int ret = 0; | |
7540 | int i; | |
7541 | ||
7542 | read_lock(&em_tree->lock); | |
7543 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7544 | read_unlock(&em_tree->lock); | |
7545 | ||
7546 | if (!em) { | |
7547 | btrfs_err(fs_info, | |
7548 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7549 | physical_offset, devid); | |
7550 | ret = -EUCLEAN; | |
7551 | goto out; | |
7552 | } | |
7553 | ||
7554 | map = em->map_lookup; | |
7555 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7556 | if (physical_len != stripe_len) { | |
7557 | btrfs_err(fs_info, | |
7558 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7559 | physical_offset, devid, em->start, physical_len, | |
7560 | stripe_len); | |
7561 | ret = -EUCLEAN; | |
7562 | goto out; | |
7563 | } | |
7564 | ||
7565 | for (i = 0; i < map->num_stripes; i++) { | |
7566 | if (map->stripes[i].dev->devid == devid && | |
7567 | map->stripes[i].physical == physical_offset) { | |
7568 | found = true; | |
7569 | if (map->verified_stripes >= map->num_stripes) { | |
7570 | btrfs_err(fs_info, | |
7571 | "too many dev extents for chunk %llu found", | |
7572 | em->start); | |
7573 | ret = -EUCLEAN; | |
7574 | goto out; | |
7575 | } | |
7576 | map->verified_stripes++; | |
7577 | break; | |
7578 | } | |
7579 | } | |
7580 | if (!found) { | |
7581 | btrfs_err(fs_info, | |
7582 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7583 | physical_offset, devid); | |
7584 | ret = -EUCLEAN; | |
7585 | } | |
05a37c48 QW |
7586 | |
7587 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7588 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7589 | if (!dev) { |
7590 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7591 | ret = -EUCLEAN; | |
7592 | goto out; | |
7593 | } | |
1b3922a8 QW |
7594 | |
7595 | /* It's possible this device is a dummy for seed device */ | |
7596 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7597 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7598 | NULL, false); | |
1b3922a8 QW |
7599 | if (!dev) { |
7600 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7601 | devid); | |
7602 | ret = -EUCLEAN; | |
7603 | goto out; | |
7604 | } | |
7605 | } | |
7606 | ||
05a37c48 QW |
7607 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7608 | btrfs_err(fs_info, | |
7609 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7610 | devid, physical_offset, physical_len, | |
7611 | dev->disk_total_bytes); | |
7612 | ret = -EUCLEAN; | |
7613 | goto out; | |
7614 | } | |
cf90d884 QW |
7615 | out: |
7616 | free_extent_map(em); | |
7617 | return ret; | |
7618 | } | |
7619 | ||
7620 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7621 | { | |
c8bf1b67 | 7622 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7623 | struct extent_map *em; |
7624 | struct rb_node *node; | |
7625 | int ret = 0; | |
7626 | ||
7627 | read_lock(&em_tree->lock); | |
07e1ce09 | 7628 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7629 | em = rb_entry(node, struct extent_map, rb_node); |
7630 | if (em->map_lookup->num_stripes != | |
7631 | em->map_lookup->verified_stripes) { | |
7632 | btrfs_err(fs_info, | |
7633 | "chunk %llu has missing dev extent, have %d expect %d", | |
7634 | em->start, em->map_lookup->verified_stripes, | |
7635 | em->map_lookup->num_stripes); | |
7636 | ret = -EUCLEAN; | |
7637 | goto out; | |
7638 | } | |
7639 | } | |
7640 | out: | |
7641 | read_unlock(&em_tree->lock); | |
7642 | return ret; | |
7643 | } | |
7644 | ||
7645 | /* | |
7646 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7647 | * later chunk allocation/free would cause unexpected behavior. | |
7648 | * | |
7649 | * NOTE: This will iterate through the whole device tree, which should be of | |
7650 | * the same size level as the chunk tree. This slightly increases mount time. | |
7651 | */ | |
7652 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7653 | { | |
7654 | struct btrfs_path *path; | |
7655 | struct btrfs_root *root = fs_info->dev_root; | |
7656 | struct btrfs_key key; | |
5eb19381 QW |
7657 | u64 prev_devid = 0; |
7658 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7659 | int ret = 0; |
7660 | ||
7661 | key.objectid = 1; | |
7662 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7663 | key.offset = 0; | |
7664 | ||
7665 | path = btrfs_alloc_path(); | |
7666 | if (!path) | |
7667 | return -ENOMEM; | |
7668 | ||
7669 | path->reada = READA_FORWARD; | |
7670 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7671 | if (ret < 0) | |
7672 | goto out; | |
7673 | ||
7674 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7675 | ret = btrfs_next_item(root, path); | |
7676 | if (ret < 0) | |
7677 | goto out; | |
7678 | /* No dev extents at all? Not good */ | |
7679 | if (ret > 0) { | |
7680 | ret = -EUCLEAN; | |
7681 | goto out; | |
7682 | } | |
7683 | } | |
7684 | while (1) { | |
7685 | struct extent_buffer *leaf = path->nodes[0]; | |
7686 | struct btrfs_dev_extent *dext; | |
7687 | int slot = path->slots[0]; | |
7688 | u64 chunk_offset; | |
7689 | u64 physical_offset; | |
7690 | u64 physical_len; | |
7691 | u64 devid; | |
7692 | ||
7693 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7694 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7695 | break; | |
7696 | devid = key.objectid; | |
7697 | physical_offset = key.offset; | |
7698 | ||
7699 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7700 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7701 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7702 | ||
5eb19381 QW |
7703 | /* Check if this dev extent overlaps with the previous one */ |
7704 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7705 | btrfs_err(fs_info, | |
7706 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7707 | devid, physical_offset, prev_dev_ext_end); | |
7708 | ret = -EUCLEAN; | |
7709 | goto out; | |
7710 | } | |
7711 | ||
cf90d884 QW |
7712 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7713 | physical_offset, physical_len); | |
7714 | if (ret < 0) | |
7715 | goto out; | |
5eb19381 QW |
7716 | prev_devid = devid; |
7717 | prev_dev_ext_end = physical_offset + physical_len; | |
7718 | ||
cf90d884 QW |
7719 | ret = btrfs_next_item(root, path); |
7720 | if (ret < 0) | |
7721 | goto out; | |
7722 | if (ret > 0) { | |
7723 | ret = 0; | |
7724 | break; | |
7725 | } | |
7726 | } | |
7727 | ||
7728 | /* Ensure all chunks have corresponding dev extents */ | |
7729 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7730 | out: | |
7731 | btrfs_free_path(path); | |
7732 | return ret; | |
7733 | } | |
eede2bf3 OS |
7734 | |
7735 | /* | |
7736 | * Check whether the given block group or device is pinned by any inode being | |
7737 | * used as a swapfile. | |
7738 | */ | |
7739 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7740 | { | |
7741 | struct btrfs_swapfile_pin *sp; | |
7742 | struct rb_node *node; | |
7743 | ||
7744 | spin_lock(&fs_info->swapfile_pins_lock); | |
7745 | node = fs_info->swapfile_pins.rb_node; | |
7746 | while (node) { | |
7747 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7748 | if (ptr < sp->ptr) | |
7749 | node = node->rb_left; | |
7750 | else if (ptr > sp->ptr) | |
7751 | node = node->rb_right; | |
7752 | else | |
7753 | break; | |
7754 | } | |
7755 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7756 | return node != NULL; | |
7757 | } |