Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 CM |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
8a4b83cc | 9 | #include <linux/buffer_head.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
0b86a832 CM |
17 | #include "ctree.h" |
18 | #include "extent_map.h" | |
19 | #include "disk-io.h" | |
20 | #include "transaction.h" | |
21 | #include "print-tree.h" | |
22 | #include "volumes.h" | |
53b381b3 | 23 | #include "raid56.h" |
8b712842 | 24 | #include "async-thread.h" |
21adbd5c | 25 | #include "check-integrity.h" |
606686ee | 26 | #include "rcu-string.h" |
3fed40cc | 27 | #include "math.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
0b86a832 | 33 | |
af902047 ZL |
34 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
35 | [BTRFS_RAID_RAID10] = { | |
36 | .sub_stripes = 2, | |
37 | .dev_stripes = 1, | |
38 | .devs_max = 0, /* 0 == as many as possible */ | |
39 | .devs_min = 4, | |
8789f4fe | 40 | .tolerated_failures = 1, |
af902047 ZL |
41 | .devs_increment = 2, |
42 | .ncopies = 2, | |
b50836ed | 43 | .nparity = 0, |
ed23467b | 44 | .raid_name = "raid10", |
41a6e891 | 45 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 46 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
47 | }, |
48 | [BTRFS_RAID_RAID1] = { | |
49 | .sub_stripes = 1, | |
50 | .dev_stripes = 1, | |
51 | .devs_max = 2, | |
52 | .devs_min = 2, | |
8789f4fe | 53 | .tolerated_failures = 1, |
af902047 ZL |
54 | .devs_increment = 2, |
55 | .ncopies = 2, | |
b50836ed | 56 | .nparity = 0, |
ed23467b | 57 | .raid_name = "raid1", |
41a6e891 | 58 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 59 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 ZL |
60 | }, |
61 | [BTRFS_RAID_DUP] = { | |
62 | .sub_stripes = 1, | |
63 | .dev_stripes = 2, | |
64 | .devs_max = 1, | |
65 | .devs_min = 1, | |
8789f4fe | 66 | .tolerated_failures = 0, |
af902047 ZL |
67 | .devs_increment = 1, |
68 | .ncopies = 2, | |
b50836ed | 69 | .nparity = 0, |
ed23467b | 70 | .raid_name = "dup", |
41a6e891 | 71 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 72 | .mindev_error = 0, |
af902047 ZL |
73 | }, |
74 | [BTRFS_RAID_RAID0] = { | |
75 | .sub_stripes = 1, | |
76 | .dev_stripes = 1, | |
77 | .devs_max = 0, | |
78 | .devs_min = 2, | |
8789f4fe | 79 | .tolerated_failures = 0, |
af902047 ZL |
80 | .devs_increment = 1, |
81 | .ncopies = 1, | |
b50836ed | 82 | .nparity = 0, |
ed23467b | 83 | .raid_name = "raid0", |
41a6e891 | 84 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 85 | .mindev_error = 0, |
af902047 ZL |
86 | }, |
87 | [BTRFS_RAID_SINGLE] = { | |
88 | .sub_stripes = 1, | |
89 | .dev_stripes = 1, | |
90 | .devs_max = 1, | |
91 | .devs_min = 1, | |
8789f4fe | 92 | .tolerated_failures = 0, |
af902047 ZL |
93 | .devs_increment = 1, |
94 | .ncopies = 1, | |
b50836ed | 95 | .nparity = 0, |
ed23467b | 96 | .raid_name = "single", |
41a6e891 | 97 | .bg_flag = 0, |
f9fbcaa2 | 98 | .mindev_error = 0, |
af902047 ZL |
99 | }, |
100 | [BTRFS_RAID_RAID5] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 2, | |
8789f4fe | 105 | .tolerated_failures = 1, |
af902047 | 106 | .devs_increment = 1, |
da612e31 | 107 | .ncopies = 1, |
b50836ed | 108 | .nparity = 1, |
ed23467b | 109 | .raid_name = "raid5", |
41a6e891 | 110 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 111 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
112 | }, |
113 | [BTRFS_RAID_RAID6] = { | |
114 | .sub_stripes = 1, | |
115 | .dev_stripes = 1, | |
116 | .devs_max = 0, | |
117 | .devs_min = 3, | |
8789f4fe | 118 | .tolerated_failures = 2, |
af902047 | 119 | .devs_increment = 1, |
da612e31 | 120 | .ncopies = 1, |
b50836ed | 121 | .nparity = 2, |
ed23467b | 122 | .raid_name = "raid6", |
41a6e891 | 123 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 124 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
125 | }, |
126 | }; | |
127 | ||
158da513 | 128 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 129 | { |
158da513 DS |
130 | const int index = btrfs_bg_flags_to_raid_index(flags); |
131 | ||
132 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
133 | return NULL; |
134 | ||
158da513 | 135 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
136 | } |
137 | ||
f89e09cf AJ |
138 | /* |
139 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
140 | * bytes including terminating null byte. | |
141 | */ | |
142 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
143 | { | |
144 | int i; | |
145 | int ret; | |
146 | char *bp = buf; | |
147 | u64 flags = bg_flags; | |
148 | u32 size_bp = size_buf; | |
149 | ||
150 | if (!flags) { | |
151 | strcpy(bp, "NONE"); | |
152 | return; | |
153 | } | |
154 | ||
155 | #define DESCRIBE_FLAG(flag, desc) \ | |
156 | do { \ | |
157 | if (flags & (flag)) { \ | |
158 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
159 | if (ret < 0 || ret >= size_bp) \ | |
160 | goto out_overflow; \ | |
161 | size_bp -= ret; \ | |
162 | bp += ret; \ | |
163 | flags &= ~(flag); \ | |
164 | } \ | |
165 | } while (0) | |
166 | ||
167 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
168 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
169 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
170 | ||
171 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
172 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
173 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
174 | btrfs_raid_array[i].raid_name); | |
175 | #undef DESCRIBE_FLAG | |
176 | ||
177 | if (flags) { | |
178 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
179 | size_bp -= ret; | |
180 | } | |
181 | ||
182 | if (size_bp < size_buf) | |
183 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
184 | ||
185 | /* | |
186 | * The text is trimmed, it's up to the caller to provide sufficiently | |
187 | * large buffer | |
188 | */ | |
189 | out_overflow:; | |
190 | } | |
191 | ||
6f8e0fc7 | 192 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 193 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 194 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 195 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
196 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
197 | enum btrfs_map_op op, | |
198 | u64 logical, u64 *length, | |
199 | struct btrfs_bio **bbio_ret, | |
200 | int mirror_num, int need_raid_map); | |
2b82032c | 201 | |
9c6b1c4d DS |
202 | /* |
203 | * Device locking | |
204 | * ============== | |
205 | * | |
206 | * There are several mutexes that protect manipulation of devices and low-level | |
207 | * structures like chunks but not block groups, extents or files | |
208 | * | |
209 | * uuid_mutex (global lock) | |
210 | * ------------------------ | |
211 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
212 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
213 | * device) or requested by the device= mount option | |
214 | * | |
215 | * the mutex can be very coarse and can cover long-running operations | |
216 | * | |
217 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 218 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
219 | * |
220 | * global::fs_devs - add, remove, updates to the global list | |
221 | * | |
222 | * does not protect: manipulation of the fs_devices::devices list! | |
223 | * | |
224 | * btrfs_device::name - renames (write side), read is RCU | |
225 | * | |
226 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
227 | * ------------------------------------------------ | |
228 | * protects updates to fs_devices::devices, ie. adding and deleting | |
229 | * | |
230 | * simple list traversal with read-only actions can be done with RCU protection | |
231 | * | |
232 | * may be used to exclude some operations from running concurrently without any | |
233 | * modifications to the list (see write_all_supers) | |
234 | * | |
9c6b1c4d DS |
235 | * balance_mutex |
236 | * ------------- | |
237 | * protects balance structures (status, state) and context accessed from | |
238 | * several places (internally, ioctl) | |
239 | * | |
240 | * chunk_mutex | |
241 | * ----------- | |
242 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
243 | * device is added/removed. Additionally it also protects post_commit_list of |
244 | * individual devices, since they can be added to the transaction's | |
245 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
246 | * |
247 | * cleaner_mutex | |
248 | * ------------- | |
249 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
250 | * cleaning together with relocation or delayed iputs | |
251 | * | |
252 | * | |
253 | * Lock nesting | |
254 | * ============ | |
255 | * | |
256 | * uuid_mutex | |
257 | * volume_mutex | |
258 | * device_list_mutex | |
259 | * chunk_mutex | |
260 | * balance_mutex | |
89595e80 AJ |
261 | * |
262 | * | |
263 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
264 | * ====================================== | |
265 | * | |
266 | * Maintains the exclusivity of the following operations that apply to the | |
267 | * whole filesystem and cannot run in parallel. | |
268 | * | |
269 | * - Balance (*) | |
270 | * - Device add | |
271 | * - Device remove | |
272 | * - Device replace (*) | |
273 | * - Resize | |
274 | * | |
275 | * The device operations (as above) can be in one of the following states: | |
276 | * | |
277 | * - Running state | |
278 | * - Paused state | |
279 | * - Completed state | |
280 | * | |
281 | * Only device operations marked with (*) can go into the Paused state for the | |
282 | * following reasons: | |
283 | * | |
284 | * - ioctl (only Balance can be Paused through ioctl) | |
285 | * - filesystem remounted as read-only | |
286 | * - filesystem unmounted and mounted as read-only | |
287 | * - system power-cycle and filesystem mounted as read-only | |
288 | * - filesystem or device errors leading to forced read-only | |
289 | * | |
290 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
291 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
292 | * A device operation in Paused or Running state can be canceled or resumed | |
293 | * either by ioctl (Balance only) or when remounted as read-write. | |
294 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
295 | * completed. | |
9c6b1c4d DS |
296 | */ |
297 | ||
67a2c45e | 298 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 299 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
300 | struct list_head *btrfs_get_fs_uuids(void) |
301 | { | |
302 | return &fs_uuids; | |
303 | } | |
8a4b83cc | 304 | |
2dfeca9b DS |
305 | /* |
306 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
307 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
308 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
309 | * |
310 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
311 | * The returned struct is not linked onto any lists and can be destroyed with | |
312 | * kfree() right away. | |
313 | */ | |
7239ff4b NB |
314 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
315 | const u8 *metadata_fsid) | |
2208a378 ID |
316 | { |
317 | struct btrfs_fs_devices *fs_devs; | |
318 | ||
78f2c9e6 | 319 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
320 | if (!fs_devs) |
321 | return ERR_PTR(-ENOMEM); | |
322 | ||
323 | mutex_init(&fs_devs->device_list_mutex); | |
324 | ||
325 | INIT_LIST_HEAD(&fs_devs->devices); | |
326 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 327 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
328 | if (fsid) |
329 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 330 | |
7239ff4b NB |
331 | if (metadata_fsid) |
332 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
333 | else if (fsid) | |
334 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
335 | ||
2208a378 ID |
336 | return fs_devs; |
337 | } | |
338 | ||
a425f9d4 | 339 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 340 | { |
bbbf7243 | 341 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 342 | rcu_string_free(device->name); |
1c11b63e | 343 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
344 | bio_put(device->flush_bio); |
345 | kfree(device); | |
346 | } | |
347 | ||
e4404d6e YZ |
348 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
349 | { | |
350 | struct btrfs_device *device; | |
351 | WARN_ON(fs_devices->opened); | |
352 | while (!list_empty(&fs_devices->devices)) { | |
353 | device = list_entry(fs_devices->devices.next, | |
354 | struct btrfs_device, dev_list); | |
355 | list_del(&device->dev_list); | |
a425f9d4 | 356 | btrfs_free_device(device); |
e4404d6e YZ |
357 | } |
358 | kfree(fs_devices); | |
359 | } | |
360 | ||
ffc5a379 | 361 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
362 | { |
363 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 364 | |
2b82032c YZ |
365 | while (!list_empty(&fs_uuids)) { |
366 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
367 | struct btrfs_fs_devices, fs_list); |
368 | list_del(&fs_devices->fs_list); | |
e4404d6e | 369 | free_fs_devices(fs_devices); |
8a4b83cc | 370 | } |
8a4b83cc CM |
371 | } |
372 | ||
48dae9cf DS |
373 | /* |
374 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
375 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 376 | * btrfs_free_device. |
48dae9cf | 377 | */ |
12bd2fc0 ID |
378 | static struct btrfs_device *__alloc_device(void) |
379 | { | |
380 | struct btrfs_device *dev; | |
381 | ||
78f2c9e6 | 382 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
383 | if (!dev) |
384 | return ERR_PTR(-ENOMEM); | |
385 | ||
e0ae9994 DS |
386 | /* |
387 | * Preallocate a bio that's always going to be used for flushing device | |
388 | * barriers and matches the device lifespan | |
389 | */ | |
390 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
391 | if (!dev->flush_bio) { | |
392 | kfree(dev); | |
393 | return ERR_PTR(-ENOMEM); | |
394 | } | |
e0ae9994 | 395 | |
12bd2fc0 ID |
396 | INIT_LIST_HEAD(&dev->dev_list); |
397 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 398 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 ID |
399 | |
400 | spin_lock_init(&dev->io_lock); | |
401 | ||
12bd2fc0 | 402 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 403 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 404 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 405 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 406 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
1c11b63e | 407 | extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL); |
12bd2fc0 ID |
408 | |
409 | return dev; | |
410 | } | |
411 | ||
7239ff4b NB |
412 | static noinline struct btrfs_fs_devices *find_fsid( |
413 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 414 | { |
8a4b83cc CM |
415 | struct btrfs_fs_devices *fs_devices; |
416 | ||
7239ff4b NB |
417 | ASSERT(fsid); |
418 | ||
7a62d0f0 NB |
419 | if (metadata_fsid) { |
420 | /* | |
421 | * Handle scanned device having completed its fsid change but | |
422 | * belonging to a fs_devices that was created by first scanning | |
423 | * a device which didn't have its fsid/metadata_uuid changed | |
424 | * at all and the CHANGING_FSID_V2 flag set. | |
425 | */ | |
426 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
427 | if (fs_devices->fsid_change && | |
428 | memcmp(metadata_fsid, fs_devices->fsid, | |
429 | BTRFS_FSID_SIZE) == 0 && | |
430 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
431 | BTRFS_FSID_SIZE) == 0) { | |
432 | return fs_devices; | |
433 | } | |
434 | } | |
cc5de4e7 NB |
435 | /* |
436 | * Handle scanned device having completed its fsid change but | |
437 | * belonging to a fs_devices that was created by a device that | |
438 | * has an outdated pair of fsid/metadata_uuid and | |
439 | * CHANGING_FSID_V2 flag set. | |
440 | */ | |
441 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
442 | if (fs_devices->fsid_change && | |
443 | memcmp(fs_devices->metadata_uuid, | |
444 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
445 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
446 | BTRFS_FSID_SIZE) == 0) { | |
447 | return fs_devices; | |
448 | } | |
449 | } | |
7a62d0f0 NB |
450 | } |
451 | ||
452 | /* Handle non-split brain cases */ | |
c4babc5e | 453 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
454 | if (metadata_fsid) { |
455 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
456 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
457 | BTRFS_FSID_SIZE) == 0) | |
458 | return fs_devices; | |
459 | } else { | |
460 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
461 | return fs_devices; | |
462 | } | |
8a4b83cc CM |
463 | } |
464 | return NULL; | |
465 | } | |
466 | ||
beaf8ab3 SB |
467 | static int |
468 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
469 | int flush, struct block_device **bdev, | |
470 | struct buffer_head **bh) | |
471 | { | |
472 | int ret; | |
473 | ||
474 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
475 | ||
476 | if (IS_ERR(*bdev)) { | |
477 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
478 | goto error; |
479 | } | |
480 | ||
481 | if (flush) | |
482 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 483 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
484 | if (ret) { |
485 | blkdev_put(*bdev, flags); | |
486 | goto error; | |
487 | } | |
488 | invalidate_bdev(*bdev); | |
489 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
490 | if (IS_ERR(*bh)) { |
491 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
492 | blkdev_put(*bdev, flags); |
493 | goto error; | |
494 | } | |
495 | ||
496 | return 0; | |
497 | ||
498 | error: | |
499 | *bdev = NULL; | |
500 | *bh = NULL; | |
501 | return ret; | |
502 | } | |
503 | ||
ffbd517d CM |
504 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
505 | struct bio *head, struct bio *tail) | |
506 | { | |
507 | ||
508 | struct bio *old_head; | |
509 | ||
510 | old_head = pending_bios->head; | |
511 | pending_bios->head = head; | |
512 | if (pending_bios->tail) | |
513 | tail->bi_next = old_head; | |
514 | else | |
515 | pending_bios->tail = tail; | |
516 | } | |
517 | ||
8b712842 CM |
518 | /* |
519 | * we try to collect pending bios for a device so we don't get a large | |
520 | * number of procs sending bios down to the same device. This greatly | |
521 | * improves the schedulers ability to collect and merge the bios. | |
522 | * | |
523 | * But, it also turns into a long list of bios to process and that is sure | |
524 | * to eventually make the worker thread block. The solution here is to | |
525 | * make some progress and then put this work struct back at the end of | |
526 | * the list if the block device is congested. This way, multiple devices | |
527 | * can make progress from a single worker thread. | |
528 | */ | |
143bede5 | 529 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 530 | { |
0b246afa | 531 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
532 | struct bio *pending; |
533 | struct backing_dev_info *bdi; | |
ffbd517d | 534 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
535 | struct bio *tail; |
536 | struct bio *cur; | |
537 | int again = 0; | |
ffbd517d | 538 | unsigned long num_run; |
d644d8a1 | 539 | unsigned long batch_run = 0; |
b765ead5 | 540 | unsigned long last_waited = 0; |
d84275c9 | 541 | int force_reg = 0; |
0e588859 | 542 | int sync_pending = 0; |
211588ad CM |
543 | struct blk_plug plug; |
544 | ||
545 | /* | |
546 | * this function runs all the bios we've collected for | |
547 | * a particular device. We don't want to wander off to | |
548 | * another device without first sending all of these down. | |
549 | * So, setup a plug here and finish it off before we return | |
550 | */ | |
551 | blk_start_plug(&plug); | |
8b712842 | 552 | |
efa7c9f9 | 553 | bdi = device->bdev->bd_bdi; |
b64a2851 | 554 | |
8b712842 CM |
555 | loop: |
556 | spin_lock(&device->io_lock); | |
557 | ||
a6837051 | 558 | loop_lock: |
d84275c9 | 559 | num_run = 0; |
ffbd517d | 560 | |
8b712842 CM |
561 | /* take all the bios off the list at once and process them |
562 | * later on (without the lock held). But, remember the | |
563 | * tail and other pointers so the bios can be properly reinserted | |
564 | * into the list if we hit congestion | |
565 | */ | |
d84275c9 | 566 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 567 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
568 | force_reg = 1; |
569 | } else { | |
ffbd517d | 570 | pending_bios = &device->pending_bios; |
d84275c9 CM |
571 | force_reg = 0; |
572 | } | |
ffbd517d CM |
573 | |
574 | pending = pending_bios->head; | |
575 | tail = pending_bios->tail; | |
8b712842 | 576 | WARN_ON(pending && !tail); |
8b712842 CM |
577 | |
578 | /* | |
579 | * if pending was null this time around, no bios need processing | |
580 | * at all and we can stop. Otherwise it'll loop back up again | |
581 | * and do an additional check so no bios are missed. | |
582 | * | |
583 | * device->running_pending is used to synchronize with the | |
584 | * schedule_bio code. | |
585 | */ | |
ffbd517d CM |
586 | if (device->pending_sync_bios.head == NULL && |
587 | device->pending_bios.head == NULL) { | |
8b712842 CM |
588 | again = 0; |
589 | device->running_pending = 0; | |
ffbd517d CM |
590 | } else { |
591 | again = 1; | |
592 | device->running_pending = 1; | |
8b712842 | 593 | } |
ffbd517d CM |
594 | |
595 | pending_bios->head = NULL; | |
596 | pending_bios->tail = NULL; | |
597 | ||
8b712842 CM |
598 | spin_unlock(&device->io_lock); |
599 | ||
d397712b | 600 | while (pending) { |
ffbd517d CM |
601 | |
602 | rmb(); | |
d84275c9 CM |
603 | /* we want to work on both lists, but do more bios on the |
604 | * sync list than the regular list | |
605 | */ | |
606 | if ((num_run > 32 && | |
607 | pending_bios != &device->pending_sync_bios && | |
608 | device->pending_sync_bios.head) || | |
609 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
610 | device->pending_bios.head)) { | |
ffbd517d CM |
611 | spin_lock(&device->io_lock); |
612 | requeue_list(pending_bios, pending, tail); | |
613 | goto loop_lock; | |
614 | } | |
615 | ||
8b712842 CM |
616 | cur = pending; |
617 | pending = pending->bi_next; | |
618 | cur->bi_next = NULL; | |
b64a2851 | 619 | |
dac56212 | 620 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 621 | |
2ab1ba68 CM |
622 | /* |
623 | * if we're doing the sync list, record that our | |
624 | * plug has some sync requests on it | |
625 | * | |
626 | * If we're doing the regular list and there are | |
627 | * sync requests sitting around, unplug before | |
628 | * we add more | |
629 | */ | |
630 | if (pending_bios == &device->pending_sync_bios) { | |
631 | sync_pending = 1; | |
632 | } else if (sync_pending) { | |
633 | blk_finish_plug(&plug); | |
634 | blk_start_plug(&plug); | |
635 | sync_pending = 0; | |
636 | } | |
637 | ||
4e49ea4a | 638 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
639 | num_run++; |
640 | batch_run++; | |
853d8ec4 DS |
641 | |
642 | cond_resched(); | |
8b712842 CM |
643 | |
644 | /* | |
645 | * we made progress, there is more work to do and the bdi | |
646 | * is now congested. Back off and let other work structs | |
647 | * run instead | |
648 | */ | |
57fd5a5f | 649 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 650 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 651 | struct io_context *ioc; |
8b712842 | 652 | |
b765ead5 CM |
653 | ioc = current->io_context; |
654 | ||
655 | /* | |
656 | * the main goal here is that we don't want to | |
657 | * block if we're going to be able to submit | |
658 | * more requests without blocking. | |
659 | * | |
660 | * This code does two great things, it pokes into | |
661 | * the elevator code from a filesystem _and_ | |
662 | * it makes assumptions about how batching works. | |
663 | */ | |
664 | if (ioc && ioc->nr_batch_requests > 0 && | |
665 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
666 | (last_waited == 0 || | |
667 | ioc->last_waited == last_waited)) { | |
668 | /* | |
669 | * we want to go through our batch of | |
670 | * requests and stop. So, we copy out | |
671 | * the ioc->last_waited time and test | |
672 | * against it before looping | |
673 | */ | |
674 | last_waited = ioc->last_waited; | |
853d8ec4 | 675 | cond_resched(); |
b765ead5 CM |
676 | continue; |
677 | } | |
8b712842 | 678 | spin_lock(&device->io_lock); |
ffbd517d | 679 | requeue_list(pending_bios, pending, tail); |
a6837051 | 680 | device->running_pending = 1; |
8b712842 CM |
681 | |
682 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
683 | btrfs_queue_work(fs_info->submit_workers, |
684 | &device->work); | |
8b712842 CM |
685 | goto done; |
686 | } | |
687 | } | |
ffbd517d | 688 | |
51684082 CM |
689 | cond_resched(); |
690 | if (again) | |
691 | goto loop; | |
692 | ||
693 | spin_lock(&device->io_lock); | |
694 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
695 | goto loop_lock; | |
696 | spin_unlock(&device->io_lock); | |
697 | ||
8b712842 | 698 | done: |
211588ad | 699 | blk_finish_plug(&plug); |
8b712842 CM |
700 | } |
701 | ||
b2950863 | 702 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
703 | { |
704 | struct btrfs_device *device; | |
705 | ||
706 | device = container_of(work, struct btrfs_device, work); | |
707 | run_scheduled_bios(device); | |
708 | } | |
709 | ||
70bc7088 AJ |
710 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
711 | { | |
712 | int found; | |
713 | ||
714 | rcu_read_lock(); | |
715 | found = strcmp(rcu_str_deref(device->name), path); | |
716 | rcu_read_unlock(); | |
717 | ||
718 | return found == 0; | |
719 | } | |
720 | ||
d8367db3 AJ |
721 | /* |
722 | * Search and remove all stale (devices which are not mounted) devices. | |
723 | * When both inputs are NULL, it will search and release all stale devices. | |
724 | * path: Optional. When provided will it release all unmounted devices | |
725 | * matching this path only. | |
726 | * skip_dev: Optional. Will skip this device when searching for the stale | |
727 | * devices. | |
70bc7088 AJ |
728 | * Return: 0 for success or if @path is NULL. |
729 | * -EBUSY if @path is a mounted device. | |
730 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 731 | */ |
70bc7088 | 732 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 733 | struct btrfs_device *skip_device) |
4fde46f0 | 734 | { |
fa6d2ae5 AJ |
735 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
736 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
737 | int ret = 0; |
738 | ||
739 | if (path) | |
740 | ret = -ENOENT; | |
4fde46f0 | 741 | |
fa6d2ae5 | 742 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 743 | |
70bc7088 | 744 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
745 | list_for_each_entry_safe(device, tmp_device, |
746 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 747 | if (skip_device && skip_device == device) |
d8367db3 | 748 | continue; |
fa6d2ae5 | 749 | if (path && !device->name) |
4fde46f0 | 750 | continue; |
70bc7088 | 751 | if (path && !device_path_matched(path, device)) |
38cf665d | 752 | continue; |
70bc7088 AJ |
753 | if (fs_devices->opened) { |
754 | /* for an already deleted device return 0 */ | |
755 | if (path && ret != 0) | |
756 | ret = -EBUSY; | |
757 | break; | |
758 | } | |
4fde46f0 | 759 | |
4fde46f0 | 760 | /* delete the stale device */ |
7bcb8164 AJ |
761 | fs_devices->num_devices--; |
762 | list_del(&device->dev_list); | |
763 | btrfs_free_device(device); | |
764 | ||
70bc7088 | 765 | ret = 0; |
7bcb8164 | 766 | if (fs_devices->num_devices == 0) |
fd649f10 | 767 | break; |
7bcb8164 AJ |
768 | } |
769 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 770 | |
7bcb8164 AJ |
771 | if (fs_devices->num_devices == 0) { |
772 | btrfs_sysfs_remove_fsid(fs_devices); | |
773 | list_del(&fs_devices->fs_list); | |
774 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
775 | } |
776 | } | |
70bc7088 AJ |
777 | |
778 | return ret; | |
4fde46f0 AJ |
779 | } |
780 | ||
0fb08bcc AJ |
781 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
782 | struct btrfs_device *device, fmode_t flags, | |
783 | void *holder) | |
784 | { | |
785 | struct request_queue *q; | |
786 | struct block_device *bdev; | |
787 | struct buffer_head *bh; | |
788 | struct btrfs_super_block *disk_super; | |
789 | u64 devid; | |
790 | int ret; | |
791 | ||
792 | if (device->bdev) | |
793 | return -EINVAL; | |
794 | if (!device->name) | |
795 | return -EINVAL; | |
796 | ||
797 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
798 | &bdev, &bh); | |
799 | if (ret) | |
800 | return ret; | |
801 | ||
802 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
803 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
804 | if (devid != device->devid) | |
805 | goto error_brelse; | |
806 | ||
807 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
808 | goto error_brelse; | |
809 | ||
810 | device->generation = btrfs_super_generation(disk_super); | |
811 | ||
812 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
813 | if (btrfs_super_incompat_flags(disk_super) & |
814 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
815 | pr_err( | |
816 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
817 | goto error_brelse; | |
818 | } | |
819 | ||
ebbede42 | 820 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
821 | fs_devices->seeding = 1; |
822 | } else { | |
ebbede42 AJ |
823 | if (bdev_read_only(bdev)) |
824 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
825 | else | |
826 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
827 | } |
828 | ||
829 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
830 | if (!blk_queue_nonrot(q)) |
831 | fs_devices->rotating = 1; | |
832 | ||
833 | device->bdev = bdev; | |
e12c9621 | 834 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
835 | device->mode = flags; |
836 | ||
837 | fs_devices->open_devices++; | |
ebbede42 AJ |
838 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
839 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 840 | fs_devices->rw_devices++; |
b1b8e386 | 841 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
842 | } |
843 | brelse(bh); | |
844 | ||
845 | return 0; | |
846 | ||
847 | error_brelse: | |
848 | brelse(bh); | |
849 | blkdev_put(bdev, flags); | |
850 | ||
851 | return -EINVAL; | |
852 | } | |
853 | ||
7a62d0f0 NB |
854 | /* |
855 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
856 | * being created with a disk that has already completed its fsid change. | |
857 | */ | |
858 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
859 | struct btrfs_super_block *disk_super) | |
860 | { | |
861 | struct btrfs_fs_devices *fs_devices; | |
862 | ||
863 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
864 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
865 | BTRFS_FSID_SIZE) != 0 && | |
866 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
867 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
868 | return fs_devices; | |
869 | } | |
870 | } | |
871 | ||
872 | return NULL; | |
873 | } | |
874 | ||
cc5de4e7 NB |
875 | |
876 | static struct btrfs_fs_devices *find_fsid_changed( | |
877 | struct btrfs_super_block *disk_super) | |
878 | { | |
879 | struct btrfs_fs_devices *fs_devices; | |
880 | ||
881 | /* | |
882 | * Handles the case where scanned device is part of an fs that had | |
883 | * multiple successful changes of FSID but curently device didn't | |
884 | * observe it. Meaning our fsid will be different than theirs. | |
885 | */ | |
886 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
887 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
888 | BTRFS_FSID_SIZE) != 0 && | |
889 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
890 | BTRFS_FSID_SIZE) == 0 && | |
891 | memcmp(fs_devices->fsid, disk_super->fsid, | |
892 | BTRFS_FSID_SIZE) != 0) { | |
893 | return fs_devices; | |
894 | } | |
895 | } | |
896 | ||
897 | return NULL; | |
898 | } | |
60999ca4 DS |
899 | /* |
900 | * Add new device to list of registered devices | |
901 | * | |
902 | * Returns: | |
e124ece5 AJ |
903 | * device pointer which was just added or updated when successful |
904 | * error pointer when failed | |
60999ca4 | 905 | */ |
e124ece5 | 906 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
907 | struct btrfs_super_block *disk_super, |
908 | bool *new_device_added) | |
8a4b83cc CM |
909 | { |
910 | struct btrfs_device *device; | |
7a62d0f0 | 911 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 912 | struct rcu_string *name; |
8a4b83cc | 913 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 914 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
915 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
916 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
917 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
918 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 919 | |
cc5de4e7 NB |
920 | if (fsid_change_in_progress) { |
921 | if (!has_metadata_uuid) { | |
922 | /* | |
923 | * When we have an image which has CHANGING_FSID_V2 set | |
924 | * it might belong to either a filesystem which has | |
925 | * disks with completed fsid change or it might belong | |
926 | * to fs with no UUID changes in effect, handle both. | |
927 | */ | |
928 | fs_devices = find_fsid_inprogress(disk_super); | |
929 | if (!fs_devices) | |
930 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
931 | } else { | |
932 | fs_devices = find_fsid_changed(disk_super); | |
933 | } | |
7a62d0f0 NB |
934 | } else if (has_metadata_uuid) { |
935 | fs_devices = find_fsid(disk_super->fsid, | |
936 | disk_super->metadata_uuid); | |
937 | } else { | |
7239ff4b | 938 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
939 | } |
940 | ||
8a4b83cc | 941 | |
8a4b83cc | 942 | if (!fs_devices) { |
7239ff4b NB |
943 | if (has_metadata_uuid) |
944 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
945 | disk_super->metadata_uuid); | |
946 | else | |
947 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
948 | ||
2208a378 | 949 | if (IS_ERR(fs_devices)) |
e124ece5 | 950 | return ERR_CAST(fs_devices); |
2208a378 | 951 | |
92900e51 AV |
952 | fs_devices->fsid_change = fsid_change_in_progress; |
953 | ||
9c6d173e | 954 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 955 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 956 | |
8a4b83cc CM |
957 | device = NULL; |
958 | } else { | |
9c6d173e | 959 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
960 | device = btrfs_find_device(fs_devices, devid, |
961 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
962 | |
963 | /* | |
964 | * If this disk has been pulled into an fs devices created by | |
965 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
966 | * metadata_uuid/fsid values of the fs_devices. | |
967 | */ | |
968 | if (has_metadata_uuid && fs_devices->fsid_change && | |
969 | found_transid > fs_devices->latest_generation) { | |
970 | memcpy(fs_devices->fsid, disk_super->fsid, | |
971 | BTRFS_FSID_SIZE); | |
972 | memcpy(fs_devices->metadata_uuid, | |
973 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
974 | ||
975 | fs_devices->fsid_change = false; | |
976 | } | |
8a4b83cc | 977 | } |
443f24fe | 978 | |
8a4b83cc | 979 | if (!device) { |
9c6d173e AJ |
980 | if (fs_devices->opened) { |
981 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 982 | return ERR_PTR(-EBUSY); |
9c6d173e | 983 | } |
2b82032c | 984 | |
12bd2fc0 ID |
985 | device = btrfs_alloc_device(NULL, &devid, |
986 | disk_super->dev_item.uuid); | |
987 | if (IS_ERR(device)) { | |
9c6d173e | 988 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 989 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 990 | return device; |
8a4b83cc | 991 | } |
606686ee JB |
992 | |
993 | name = rcu_string_strdup(path, GFP_NOFS); | |
994 | if (!name) { | |
a425f9d4 | 995 | btrfs_free_device(device); |
9c6d173e | 996 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 997 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 998 | } |
606686ee | 999 | rcu_assign_pointer(device->name, name); |
90519d66 | 1000 | |
1f78160c | 1001 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 1002 | fs_devices->num_devices++; |
e5e9a520 | 1003 | |
2b82032c | 1004 | device->fs_devices = fs_devices; |
4306a974 | 1005 | *new_device_added = true; |
327f18cc AJ |
1006 | |
1007 | if (disk_super->label[0]) | |
1008 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
1009 | disk_super->label, devid, found_transid, path); | |
1010 | else | |
1011 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
1012 | disk_super->fsid, devid, found_transid, path); | |
1013 | ||
606686ee | 1014 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
1015 | /* |
1016 | * When FS is already mounted. | |
1017 | * 1. If you are here and if the device->name is NULL that | |
1018 | * means this device was missing at time of FS mount. | |
1019 | * 2. If you are here and if the device->name is different | |
1020 | * from 'path' that means either | |
1021 | * a. The same device disappeared and reappeared with | |
1022 | * different name. or | |
1023 | * b. The missing-disk-which-was-replaced, has | |
1024 | * reappeared now. | |
1025 | * | |
1026 | * We must allow 1 and 2a above. But 2b would be a spurious | |
1027 | * and unintentional. | |
1028 | * | |
1029 | * Further in case of 1 and 2a above, the disk at 'path' | |
1030 | * would have missed some transaction when it was away and | |
1031 | * in case of 2a the stale bdev has to be updated as well. | |
1032 | * 2b must not be allowed at all time. | |
1033 | */ | |
1034 | ||
1035 | /* | |
0f23ae74 CM |
1036 | * For now, we do allow update to btrfs_fs_device through the |
1037 | * btrfs dev scan cli after FS has been mounted. We're still | |
1038 | * tracking a problem where systems fail mount by subvolume id | |
1039 | * when we reject replacement on a mounted FS. | |
b96de000 | 1040 | */ |
0f23ae74 | 1041 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
1042 | /* |
1043 | * That is if the FS is _not_ mounted and if you | |
1044 | * are here, that means there is more than one | |
1045 | * disk with same uuid and devid.We keep the one | |
1046 | * with larger generation number or the last-in if | |
1047 | * generation are equal. | |
1048 | */ | |
9c6d173e | 1049 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1050 | return ERR_PTR(-EEXIST); |
77bdae4d | 1051 | } |
b96de000 | 1052 | |
a9261d41 AJ |
1053 | /* |
1054 | * We are going to replace the device path for a given devid, | |
1055 | * make sure it's the same device if the device is mounted | |
1056 | */ | |
1057 | if (device->bdev) { | |
1058 | struct block_device *path_bdev; | |
1059 | ||
1060 | path_bdev = lookup_bdev(path); | |
1061 | if (IS_ERR(path_bdev)) { | |
1062 | mutex_unlock(&fs_devices->device_list_mutex); | |
1063 | return ERR_CAST(path_bdev); | |
1064 | } | |
1065 | ||
1066 | if (device->bdev != path_bdev) { | |
1067 | bdput(path_bdev); | |
1068 | mutex_unlock(&fs_devices->device_list_mutex); | |
1069 | btrfs_warn_in_rcu(device->fs_info, | |
1070 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1071 | disk_super->fsid, devid, | |
1072 | rcu_str_deref(device->name), path); | |
1073 | return ERR_PTR(-EEXIST); | |
1074 | } | |
1075 | bdput(path_bdev); | |
1076 | btrfs_info_in_rcu(device->fs_info, | |
1077 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1078 | disk_super->fsid, devid, | |
1079 | rcu_str_deref(device->name), path); | |
1080 | } | |
1081 | ||
606686ee | 1082 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1083 | if (!name) { |
1084 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1085 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1086 | } |
606686ee JB |
1087 | rcu_string_free(device->name); |
1088 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1089 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1090 | fs_devices->missing_devices--; |
e6e674bd | 1091 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1092 | } |
8a4b83cc CM |
1093 | } |
1094 | ||
77bdae4d AJ |
1095 | /* |
1096 | * Unmount does not free the btrfs_device struct but would zero | |
1097 | * generation along with most of the other members. So just update | |
1098 | * it back. We need it to pick the disk with largest generation | |
1099 | * (as above). | |
1100 | */ | |
d1a63002 | 1101 | if (!fs_devices->opened) { |
77bdae4d | 1102 | device->generation = found_transid; |
d1a63002 NB |
1103 | fs_devices->latest_generation = max_t(u64, found_transid, |
1104 | fs_devices->latest_generation); | |
1105 | } | |
77bdae4d | 1106 | |
f2788d2f AJ |
1107 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1108 | ||
9c6d173e | 1109 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1110 | return device; |
8a4b83cc CM |
1111 | } |
1112 | ||
e4404d6e YZ |
1113 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1114 | { | |
1115 | struct btrfs_fs_devices *fs_devices; | |
1116 | struct btrfs_device *device; | |
1117 | struct btrfs_device *orig_dev; | |
1118 | ||
7239ff4b | 1119 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1120 | if (IS_ERR(fs_devices)) |
1121 | return fs_devices; | |
e4404d6e | 1122 | |
adbbb863 | 1123 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1124 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1125 | |
1126 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1127 | struct rcu_string *name; |
1128 | ||
12bd2fc0 ID |
1129 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1130 | orig_dev->uuid); | |
1131 | if (IS_ERR(device)) | |
e4404d6e YZ |
1132 | goto error; |
1133 | ||
606686ee JB |
1134 | /* |
1135 | * This is ok to do without rcu read locked because we hold the | |
1136 | * uuid mutex so nothing we touch in here is going to disappear. | |
1137 | */ | |
e755f780 | 1138 | if (orig_dev->name) { |
78f2c9e6 DS |
1139 | name = rcu_string_strdup(orig_dev->name->str, |
1140 | GFP_KERNEL); | |
e755f780 | 1141 | if (!name) { |
a425f9d4 | 1142 | btrfs_free_device(device); |
e755f780 AJ |
1143 | goto error; |
1144 | } | |
1145 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1146 | } |
e4404d6e | 1147 | |
e4404d6e YZ |
1148 | list_add(&device->dev_list, &fs_devices->devices); |
1149 | device->fs_devices = fs_devices; | |
1150 | fs_devices->num_devices++; | |
1151 | } | |
adbbb863 | 1152 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1153 | return fs_devices; |
1154 | error: | |
adbbb863 | 1155 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1156 | free_fs_devices(fs_devices); |
1157 | return ERR_PTR(-ENOMEM); | |
1158 | } | |
1159 | ||
9b99b115 AJ |
1160 | /* |
1161 | * After we have read the system tree and know devids belonging to | |
1162 | * this filesystem, remove the device which does not belong there. | |
1163 | */ | |
1164 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1165 | { |
c6e30871 | 1166 | struct btrfs_device *device, *next; |
443f24fe | 1167 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1168 | |
dfe25020 CM |
1169 | mutex_lock(&uuid_mutex); |
1170 | again: | |
46224705 | 1171 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1172 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1173 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1174 | &device->dev_state)) { | |
401e29c1 AJ |
1175 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1176 | &device->dev_state) && | |
1177 | (!latest_dev || | |
1178 | device->generation > latest_dev->generation)) { | |
443f24fe | 1179 | latest_dev = device; |
a6b0d5c8 | 1180 | } |
2b82032c | 1181 | continue; |
a6b0d5c8 | 1182 | } |
2b82032c | 1183 | |
8dabb742 SB |
1184 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1185 | /* | |
1186 | * In the first step, keep the device which has | |
1187 | * the correct fsid and the devid that is used | |
1188 | * for the dev_replace procedure. | |
1189 | * In the second step, the dev_replace state is | |
1190 | * read from the device tree and it is known | |
1191 | * whether the procedure is really active or | |
1192 | * not, which means whether this device is | |
1193 | * used or whether it should be removed. | |
1194 | */ | |
401e29c1 AJ |
1195 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1196 | &device->dev_state)) { | |
8dabb742 SB |
1197 | continue; |
1198 | } | |
1199 | } | |
2b82032c | 1200 | if (device->bdev) { |
d4d77629 | 1201 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1202 | device->bdev = NULL; |
1203 | fs_devices->open_devices--; | |
1204 | } | |
ebbede42 | 1205 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1206 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1207 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1208 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1209 | &device->dev_state)) | |
8dabb742 | 1210 | fs_devices->rw_devices--; |
2b82032c | 1211 | } |
e4404d6e YZ |
1212 | list_del_init(&device->dev_list); |
1213 | fs_devices->num_devices--; | |
a425f9d4 | 1214 | btrfs_free_device(device); |
dfe25020 | 1215 | } |
2b82032c YZ |
1216 | |
1217 | if (fs_devices->seed) { | |
1218 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1219 | goto again; |
1220 | } | |
1221 | ||
443f24fe | 1222 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1223 | |
dfe25020 | 1224 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1225 | } |
a0af469b | 1226 | |
14238819 AJ |
1227 | static void btrfs_close_bdev(struct btrfs_device *device) |
1228 | { | |
08ffcae8 DS |
1229 | if (!device->bdev) |
1230 | return; | |
1231 | ||
ebbede42 | 1232 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1233 | sync_blockdev(device->bdev); |
1234 | invalidate_bdev(device->bdev); | |
1235 | } | |
1236 | ||
08ffcae8 | 1237 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1238 | } |
1239 | ||
959b1c04 | 1240 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1241 | { |
1242 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1243 | struct btrfs_device *new_device; | |
1244 | struct rcu_string *name; | |
1245 | ||
1246 | if (device->bdev) | |
1247 | fs_devices->open_devices--; | |
1248 | ||
ebbede42 | 1249 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1250 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1251 | list_del_init(&device->dev_alloc_list); | |
1252 | fs_devices->rw_devices--; | |
1253 | } | |
1254 | ||
e6e674bd | 1255 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1256 | fs_devices->missing_devices--; |
1257 | ||
959b1c04 NB |
1258 | btrfs_close_bdev(device); |
1259 | ||
f448341a AJ |
1260 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1261 | device->uuid); | |
1262 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1263 | ||
1264 | /* Safe because we are under uuid_mutex */ | |
1265 | if (device->name) { | |
1266 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1267 | BUG_ON(!name); /* -ENOMEM */ | |
1268 | rcu_assign_pointer(new_device->name, name); | |
1269 | } | |
1270 | ||
1271 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1272 | new_device->fs_devices = device->fs_devices; | |
959b1c04 | 1273 | |
8e75fd89 NB |
1274 | synchronize_rcu(); |
1275 | btrfs_free_device(device); | |
f448341a AJ |
1276 | } |
1277 | ||
0226e0eb | 1278 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1279 | { |
2037a093 | 1280 | struct btrfs_device *device, *tmp; |
e4404d6e | 1281 | |
2b82032c YZ |
1282 | if (--fs_devices->opened > 0) |
1283 | return 0; | |
8a4b83cc | 1284 | |
c9513edb | 1285 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1286 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1287 | btrfs_close_one_device(device); |
8a4b83cc | 1288 | } |
c9513edb XG |
1289 | mutex_unlock(&fs_devices->device_list_mutex); |
1290 | ||
e4404d6e YZ |
1291 | WARN_ON(fs_devices->open_devices); |
1292 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1293 | fs_devices->opened = 0; |
1294 | fs_devices->seeding = 0; | |
2b82032c | 1295 | |
8a4b83cc CM |
1296 | return 0; |
1297 | } | |
1298 | ||
2b82032c YZ |
1299 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1300 | { | |
e4404d6e | 1301 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1302 | int ret; |
1303 | ||
1304 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1305 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1306 | if (!fs_devices->opened) { |
1307 | seed_devices = fs_devices->seed; | |
1308 | fs_devices->seed = NULL; | |
1309 | } | |
2b82032c | 1310 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1311 | |
1312 | while (seed_devices) { | |
1313 | fs_devices = seed_devices; | |
1314 | seed_devices = fs_devices->seed; | |
0226e0eb | 1315 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1316 | free_fs_devices(fs_devices); |
1317 | } | |
2b82032c YZ |
1318 | return ret; |
1319 | } | |
1320 | ||
897fb573 | 1321 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1322 | fmode_t flags, void *holder) |
8a4b83cc | 1323 | { |
8a4b83cc | 1324 | struct btrfs_device *device; |
443f24fe | 1325 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1326 | int ret = 0; |
8a4b83cc | 1327 | |
d4d77629 TH |
1328 | flags |= FMODE_EXCL; |
1329 | ||
f117e290 | 1330 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1331 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1332 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1333 | continue; |
a0af469b | 1334 | |
9f050db4 AJ |
1335 | if (!latest_dev || |
1336 | device->generation > latest_dev->generation) | |
1337 | latest_dev = device; | |
8a4b83cc | 1338 | } |
a0af469b | 1339 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1340 | ret = -EINVAL; |
a0af469b CM |
1341 | goto out; |
1342 | } | |
2b82032c | 1343 | fs_devices->opened = 1; |
443f24fe | 1344 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1345 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1346 | out: |
2b82032c YZ |
1347 | return ret; |
1348 | } | |
1349 | ||
f8e10cd3 AJ |
1350 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1351 | { | |
1352 | struct btrfs_device *dev1, *dev2; | |
1353 | ||
1354 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1355 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1356 | ||
1357 | if (dev1->devid < dev2->devid) | |
1358 | return -1; | |
1359 | else if (dev1->devid > dev2->devid) | |
1360 | return 1; | |
1361 | return 0; | |
1362 | } | |
1363 | ||
2b82032c | 1364 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1365 | fmode_t flags, void *holder) |
2b82032c YZ |
1366 | { |
1367 | int ret; | |
1368 | ||
f5194e34 DS |
1369 | lockdep_assert_held(&uuid_mutex); |
1370 | ||
542c5908 | 1371 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1372 | if (fs_devices->opened) { |
e4404d6e YZ |
1373 | fs_devices->opened++; |
1374 | ret = 0; | |
2b82032c | 1375 | } else { |
f8e10cd3 | 1376 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1377 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1378 | } |
542c5908 AJ |
1379 | mutex_unlock(&fs_devices->device_list_mutex); |
1380 | ||
8a4b83cc CM |
1381 | return ret; |
1382 | } | |
1383 | ||
c9162bdf | 1384 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1385 | { |
1386 | kunmap(page); | |
1387 | put_page(page); | |
1388 | } | |
1389 | ||
c9162bdf OS |
1390 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1391 | struct page **page, | |
1392 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1393 | { |
1394 | void *p; | |
1395 | pgoff_t index; | |
1396 | ||
1397 | /* make sure our super fits in the device */ | |
1398 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1399 | return 1; | |
1400 | ||
1401 | /* make sure our super fits in the page */ | |
1402 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1403 | return 1; | |
1404 | ||
1405 | /* make sure our super doesn't straddle pages on disk */ | |
1406 | index = bytenr >> PAGE_SHIFT; | |
1407 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1408 | return 1; | |
1409 | ||
1410 | /* pull in the page with our super */ | |
1411 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1412 | index, GFP_KERNEL); | |
1413 | ||
1414 | if (IS_ERR_OR_NULL(*page)) | |
1415 | return 1; | |
1416 | ||
1417 | p = kmap(*page); | |
1418 | ||
1419 | /* align our pointer to the offset of the super block */ | |
7073017a | 1420 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1421 | |
1422 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1423 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1424 | btrfs_release_disk_super(*page); | |
1425 | return 1; | |
1426 | } | |
1427 | ||
1428 | if ((*disk_super)->label[0] && | |
1429 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1430 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1431 | ||
1432 | return 0; | |
1433 | } | |
1434 | ||
228a73ab AJ |
1435 | int btrfs_forget_devices(const char *path) |
1436 | { | |
1437 | int ret; | |
1438 | ||
1439 | mutex_lock(&uuid_mutex); | |
1440 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1441 | mutex_unlock(&uuid_mutex); | |
1442 | ||
1443 | return ret; | |
1444 | } | |
1445 | ||
6f60cbd3 DS |
1446 | /* |
1447 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1448 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1449 | * is read via pagecache | |
1450 | */ | |
36350e95 GJ |
1451 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1452 | void *holder) | |
8a4b83cc CM |
1453 | { |
1454 | struct btrfs_super_block *disk_super; | |
4306a974 | 1455 | bool new_device_added = false; |
36350e95 | 1456 | struct btrfs_device *device = NULL; |
8a4b83cc | 1457 | struct block_device *bdev; |
6f60cbd3 | 1458 | struct page *page; |
6f60cbd3 | 1459 | u64 bytenr; |
8a4b83cc | 1460 | |
899f9307 DS |
1461 | lockdep_assert_held(&uuid_mutex); |
1462 | ||
6f60cbd3 DS |
1463 | /* |
1464 | * we would like to check all the supers, but that would make | |
1465 | * a btrfs mount succeed after a mkfs from a different FS. | |
1466 | * So, we need to add a special mount option to scan for | |
1467 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1468 | */ | |
1469 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1470 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1471 | |
1472 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1473 | if (IS_ERR(bdev)) |
36350e95 | 1474 | return ERR_CAST(bdev); |
6f60cbd3 | 1475 | |
05a5c55d | 1476 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1477 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1478 | goto error_bdev_put; |
05a5c55d | 1479 | } |
6f60cbd3 | 1480 | |
4306a974 | 1481 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1482 | if (!IS_ERR(device)) { |
4306a974 AJ |
1483 | if (new_device_added) |
1484 | btrfs_free_stale_devices(path, device); | |
1485 | } | |
6f60cbd3 | 1486 | |
6cf86a00 | 1487 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1488 | |
1489 | error_bdev_put: | |
d4d77629 | 1490 | blkdev_put(bdev, flags); |
b6ed73bc | 1491 | |
36350e95 | 1492 | return device; |
8a4b83cc | 1493 | } |
0b86a832 | 1494 | |
1c11b63e JM |
1495 | /* |
1496 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1497 | * such is found, record the end of it in *start | |
1498 | */ | |
1c11b63e JM |
1499 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1500 | u64 len) | |
6df9a95e | 1501 | { |
1c11b63e | 1502 | u64 physical_start, physical_end; |
6df9a95e | 1503 | |
1c11b63e | 1504 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1505 | |
1c11b63e JM |
1506 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1507 | &physical_start, &physical_end, | |
1508 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1509 | |
1c11b63e JM |
1510 | if (in_range(physical_start, *start, len) || |
1511 | in_range(*start, physical_start, | |
1512 | physical_end - physical_start)) { | |
1513 | *start = physical_end + 1; | |
1514 | return true; | |
6df9a95e JB |
1515 | } |
1516 | } | |
1c11b63e | 1517 | return false; |
6df9a95e JB |
1518 | } |
1519 | ||
1520 | ||
0b86a832 | 1521 | /* |
499f377f JM |
1522 | * find_free_dev_extent_start - find free space in the specified device |
1523 | * @device: the device which we search the free space in | |
1524 | * @num_bytes: the size of the free space that we need | |
1525 | * @search_start: the position from which to begin the search | |
1526 | * @start: store the start of the free space. | |
1527 | * @len: the size of the free space. that we find, or the size | |
1528 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1529 | * |
0b86a832 CM |
1530 | * this uses a pretty simple search, the expectation is that it is |
1531 | * called very infrequently and that a given device has a small number | |
1532 | * of extents | |
7bfc837d MX |
1533 | * |
1534 | * @start is used to store the start of the free space if we find. But if we | |
1535 | * don't find suitable free space, it will be used to store the start position | |
1536 | * of the max free space. | |
1537 | * | |
1538 | * @len is used to store the size of the free space that we find. | |
1539 | * But if we don't find suitable free space, it is used to store the size of | |
1540 | * the max free space. | |
135da976 QW |
1541 | * |
1542 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1543 | * check to ensure dev extents are not double allocated. | |
1544 | * This makes the function safe to allocate dev extents but may not report | |
1545 | * correct usable device space, as device extent freed in current transaction | |
1546 | * is not reported as avaiable. | |
0b86a832 | 1547 | */ |
9e3246a5 QW |
1548 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1549 | u64 num_bytes, u64 search_start, u64 *start, | |
1550 | u64 *len) | |
0b86a832 | 1551 | { |
0b246afa JM |
1552 | struct btrfs_fs_info *fs_info = device->fs_info; |
1553 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1554 | struct btrfs_key key; |
7bfc837d | 1555 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1556 | struct btrfs_path *path; |
7bfc837d MX |
1557 | u64 hole_size; |
1558 | u64 max_hole_start; | |
1559 | u64 max_hole_size; | |
1560 | u64 extent_end; | |
0b86a832 CM |
1561 | u64 search_end = device->total_bytes; |
1562 | int ret; | |
7bfc837d | 1563 | int slot; |
0b86a832 | 1564 | struct extent_buffer *l; |
8cdc7c5b FM |
1565 | |
1566 | /* | |
1567 | * We don't want to overwrite the superblock on the drive nor any area | |
1568 | * used by the boot loader (grub for example), so we make sure to start | |
1569 | * at an offset of at least 1MB. | |
1570 | */ | |
0d0c71b3 | 1571 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1572 | |
6df9a95e JB |
1573 | path = btrfs_alloc_path(); |
1574 | if (!path) | |
1575 | return -ENOMEM; | |
f2ab7618 | 1576 | |
7bfc837d MX |
1577 | max_hole_start = search_start; |
1578 | max_hole_size = 0; | |
1579 | ||
f2ab7618 | 1580 | again: |
401e29c1 AJ |
1581 | if (search_start >= search_end || |
1582 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1583 | ret = -ENOSPC; |
6df9a95e | 1584 | goto out; |
7bfc837d MX |
1585 | } |
1586 | ||
e4058b54 | 1587 | path->reada = READA_FORWARD; |
6df9a95e JB |
1588 | path->search_commit_root = 1; |
1589 | path->skip_locking = 1; | |
7bfc837d | 1590 | |
0b86a832 CM |
1591 | key.objectid = device->devid; |
1592 | key.offset = search_start; | |
1593 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1594 | |
125ccb0a | 1595 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1596 | if (ret < 0) |
7bfc837d | 1597 | goto out; |
1fcbac58 YZ |
1598 | if (ret > 0) { |
1599 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1600 | if (ret < 0) | |
7bfc837d | 1601 | goto out; |
1fcbac58 | 1602 | } |
7bfc837d | 1603 | |
0b86a832 CM |
1604 | while (1) { |
1605 | l = path->nodes[0]; | |
1606 | slot = path->slots[0]; | |
1607 | if (slot >= btrfs_header_nritems(l)) { | |
1608 | ret = btrfs_next_leaf(root, path); | |
1609 | if (ret == 0) | |
1610 | continue; | |
1611 | if (ret < 0) | |
7bfc837d MX |
1612 | goto out; |
1613 | ||
1614 | break; | |
0b86a832 CM |
1615 | } |
1616 | btrfs_item_key_to_cpu(l, &key, slot); | |
1617 | ||
1618 | if (key.objectid < device->devid) | |
1619 | goto next; | |
1620 | ||
1621 | if (key.objectid > device->devid) | |
7bfc837d | 1622 | break; |
0b86a832 | 1623 | |
962a298f | 1624 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1625 | goto next; |
9779b72f | 1626 | |
7bfc837d MX |
1627 | if (key.offset > search_start) { |
1628 | hole_size = key.offset - search_start; | |
9779b72f | 1629 | |
6df9a95e JB |
1630 | /* |
1631 | * Have to check before we set max_hole_start, otherwise | |
1632 | * we could end up sending back this offset anyway. | |
1633 | */ | |
1c11b63e | 1634 | if (contains_pending_extent(device, &search_start, |
1b984508 | 1635 | hole_size)) { |
1c11b63e | 1636 | if (key.offset >= search_start) |
1b984508 | 1637 | hole_size = key.offset - search_start; |
1c11b63e | 1638 | else |
1b984508 | 1639 | hole_size = 0; |
1b984508 | 1640 | } |
6df9a95e | 1641 | |
7bfc837d MX |
1642 | if (hole_size > max_hole_size) { |
1643 | max_hole_start = search_start; | |
1644 | max_hole_size = hole_size; | |
1645 | } | |
9779b72f | 1646 | |
7bfc837d MX |
1647 | /* |
1648 | * If this free space is greater than which we need, | |
1649 | * it must be the max free space that we have found | |
1650 | * until now, so max_hole_start must point to the start | |
1651 | * of this free space and the length of this free space | |
1652 | * is stored in max_hole_size. Thus, we return | |
1653 | * max_hole_start and max_hole_size and go back to the | |
1654 | * caller. | |
1655 | */ | |
1656 | if (hole_size >= num_bytes) { | |
1657 | ret = 0; | |
1658 | goto out; | |
0b86a832 CM |
1659 | } |
1660 | } | |
0b86a832 | 1661 | |
0b86a832 | 1662 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1663 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1664 | dev_extent); | |
1665 | if (extent_end > search_start) | |
1666 | search_start = extent_end; | |
0b86a832 CM |
1667 | next: |
1668 | path->slots[0]++; | |
1669 | cond_resched(); | |
1670 | } | |
0b86a832 | 1671 | |
38c01b96 | 1672 | /* |
1673 | * At this point, search_start should be the end of | |
1674 | * allocated dev extents, and when shrinking the device, | |
1675 | * search_end may be smaller than search_start. | |
1676 | */ | |
f2ab7618 | 1677 | if (search_end > search_start) { |
38c01b96 | 1678 | hole_size = search_end - search_start; |
1679 | ||
1c11b63e | 1680 | if (contains_pending_extent(device, &search_start, hole_size)) { |
f2ab7618 ZL |
1681 | btrfs_release_path(path); |
1682 | goto again; | |
1683 | } | |
0b86a832 | 1684 | |
f2ab7618 ZL |
1685 | if (hole_size > max_hole_size) { |
1686 | max_hole_start = search_start; | |
1687 | max_hole_size = hole_size; | |
1688 | } | |
6df9a95e JB |
1689 | } |
1690 | ||
7bfc837d | 1691 | /* See above. */ |
f2ab7618 | 1692 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1693 | ret = -ENOSPC; |
1694 | else | |
1695 | ret = 0; | |
1696 | ||
1697 | out: | |
2b82032c | 1698 | btrfs_free_path(path); |
7bfc837d | 1699 | *start = max_hole_start; |
b2117a39 | 1700 | if (len) |
7bfc837d | 1701 | *len = max_hole_size; |
0b86a832 CM |
1702 | return ret; |
1703 | } | |
1704 | ||
60dfdf25 | 1705 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1706 | u64 *start, u64 *len) |
1707 | { | |
499f377f | 1708 | /* FIXME use last free of some kind */ |
60dfdf25 | 1709 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1710 | } |
1711 | ||
b2950863 | 1712 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1713 | struct btrfs_device *device, |
2196d6e8 | 1714 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1715 | { |
0b246afa JM |
1716 | struct btrfs_fs_info *fs_info = device->fs_info; |
1717 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1718 | int ret; |
1719 | struct btrfs_path *path; | |
8f18cf13 | 1720 | struct btrfs_key key; |
a061fc8d CM |
1721 | struct btrfs_key found_key; |
1722 | struct extent_buffer *leaf = NULL; | |
1723 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1724 | |
1725 | path = btrfs_alloc_path(); | |
1726 | if (!path) | |
1727 | return -ENOMEM; | |
1728 | ||
1729 | key.objectid = device->devid; | |
1730 | key.offset = start; | |
1731 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1732 | again: |
8f18cf13 | 1733 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1734 | if (ret > 0) { |
1735 | ret = btrfs_previous_item(root, path, key.objectid, | |
1736 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1737 | if (ret) |
1738 | goto out; | |
a061fc8d CM |
1739 | leaf = path->nodes[0]; |
1740 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1741 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1742 | struct btrfs_dev_extent); | |
1743 | BUG_ON(found_key.offset > start || found_key.offset + | |
1744 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1745 | key = found_key; |
1746 | btrfs_release_path(path); | |
1747 | goto again; | |
a061fc8d CM |
1748 | } else if (ret == 0) { |
1749 | leaf = path->nodes[0]; | |
1750 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1751 | struct btrfs_dev_extent); | |
79787eaa | 1752 | } else { |
0b246afa | 1753 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1754 | goto out; |
a061fc8d | 1755 | } |
8f18cf13 | 1756 | |
2196d6e8 MX |
1757 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1758 | ||
8f18cf13 | 1759 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1760 | if (ret) { |
0b246afa JM |
1761 | btrfs_handle_fs_error(fs_info, ret, |
1762 | "Failed to remove dev extent item"); | |
13212b54 | 1763 | } else { |
3204d33c | 1764 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1765 | } |
b0b802d7 | 1766 | out: |
8f18cf13 CM |
1767 | btrfs_free_path(path); |
1768 | return ret; | |
1769 | } | |
1770 | ||
48a3b636 ES |
1771 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1772 | struct btrfs_device *device, | |
48a3b636 | 1773 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1774 | { |
1775 | int ret; | |
1776 | struct btrfs_path *path; | |
0b246afa JM |
1777 | struct btrfs_fs_info *fs_info = device->fs_info; |
1778 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1779 | struct btrfs_dev_extent *extent; |
1780 | struct extent_buffer *leaf; | |
1781 | struct btrfs_key key; | |
1782 | ||
e12c9621 | 1783 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1784 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1785 | path = btrfs_alloc_path(); |
1786 | if (!path) | |
1787 | return -ENOMEM; | |
1788 | ||
0b86a832 | 1789 | key.objectid = device->devid; |
2b82032c | 1790 | key.offset = start; |
0b86a832 CM |
1791 | key.type = BTRFS_DEV_EXTENT_KEY; |
1792 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1793 | sizeof(*extent)); | |
2cdcecbc MF |
1794 | if (ret) |
1795 | goto out; | |
0b86a832 CM |
1796 | |
1797 | leaf = path->nodes[0]; | |
1798 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1799 | struct btrfs_dev_extent); | |
b5d9071c NB |
1800 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1801 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1802 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1803 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1804 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1805 | ||
0b86a832 CM |
1806 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1807 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1808 | out: |
0b86a832 CM |
1809 | btrfs_free_path(path); |
1810 | return ret; | |
1811 | } | |
1812 | ||
6df9a95e | 1813 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1814 | { |
6df9a95e JB |
1815 | struct extent_map_tree *em_tree; |
1816 | struct extent_map *em; | |
1817 | struct rb_node *n; | |
1818 | u64 ret = 0; | |
0b86a832 | 1819 | |
c8bf1b67 | 1820 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1821 | read_lock(&em_tree->lock); |
07e1ce09 | 1822 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1823 | if (n) { |
1824 | em = rb_entry(n, struct extent_map, rb_node); | |
1825 | ret = em->start + em->len; | |
0b86a832 | 1826 | } |
6df9a95e JB |
1827 | read_unlock(&em_tree->lock); |
1828 | ||
0b86a832 CM |
1829 | return ret; |
1830 | } | |
1831 | ||
53f10659 ID |
1832 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1833 | u64 *devid_ret) | |
0b86a832 CM |
1834 | { |
1835 | int ret; | |
1836 | struct btrfs_key key; | |
1837 | struct btrfs_key found_key; | |
2b82032c YZ |
1838 | struct btrfs_path *path; |
1839 | ||
2b82032c YZ |
1840 | path = btrfs_alloc_path(); |
1841 | if (!path) | |
1842 | return -ENOMEM; | |
0b86a832 CM |
1843 | |
1844 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1845 | key.type = BTRFS_DEV_ITEM_KEY; | |
1846 | key.offset = (u64)-1; | |
1847 | ||
53f10659 | 1848 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1849 | if (ret < 0) |
1850 | goto error; | |
1851 | ||
a06dee4d AJ |
1852 | if (ret == 0) { |
1853 | /* Corruption */ | |
1854 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1855 | ret = -EUCLEAN; | |
1856 | goto error; | |
1857 | } | |
0b86a832 | 1858 | |
53f10659 ID |
1859 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1860 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1861 | BTRFS_DEV_ITEM_KEY); |
1862 | if (ret) { | |
53f10659 | 1863 | *devid_ret = 1; |
0b86a832 CM |
1864 | } else { |
1865 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1866 | path->slots[0]); | |
53f10659 | 1867 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1868 | } |
1869 | ret = 0; | |
1870 | error: | |
2b82032c | 1871 | btrfs_free_path(path); |
0b86a832 CM |
1872 | return ret; |
1873 | } | |
1874 | ||
1875 | /* | |
1876 | * the device information is stored in the chunk root | |
1877 | * the btrfs_device struct should be fully filled in | |
1878 | */ | |
c74a0b02 | 1879 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1880 | struct btrfs_device *device) |
0b86a832 CM |
1881 | { |
1882 | int ret; | |
1883 | struct btrfs_path *path; | |
1884 | struct btrfs_dev_item *dev_item; | |
1885 | struct extent_buffer *leaf; | |
1886 | struct btrfs_key key; | |
1887 | unsigned long ptr; | |
0b86a832 | 1888 | |
0b86a832 CM |
1889 | path = btrfs_alloc_path(); |
1890 | if (!path) | |
1891 | return -ENOMEM; | |
1892 | ||
0b86a832 CM |
1893 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1894 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1895 | key.offset = device->devid; |
0b86a832 | 1896 | |
8e87e856 NB |
1897 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1898 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1899 | if (ret) |
1900 | goto out; | |
1901 | ||
1902 | leaf = path->nodes[0]; | |
1903 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1904 | ||
1905 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1906 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1907 | btrfs_set_device_type(leaf, dev_item, device->type); |
1908 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1909 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1910 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1911 | btrfs_set_device_total_bytes(leaf, dev_item, |
1912 | btrfs_device_get_disk_total_bytes(device)); | |
1913 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1914 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1915 | btrfs_set_device_group(leaf, dev_item, 0); |
1916 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1917 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1918 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1919 | |
410ba3a2 | 1920 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1921 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1922 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1923 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1924 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1925 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1926 | |
2b82032c | 1927 | ret = 0; |
0b86a832 CM |
1928 | out: |
1929 | btrfs_free_path(path); | |
1930 | return ret; | |
1931 | } | |
8f18cf13 | 1932 | |
5a1972bd QW |
1933 | /* |
1934 | * Function to update ctime/mtime for a given device path. | |
1935 | * Mainly used for ctime/mtime based probe like libblkid. | |
1936 | */ | |
da353f6b | 1937 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1938 | { |
1939 | struct file *filp; | |
1940 | ||
1941 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1942 | if (IS_ERR(filp)) |
5a1972bd QW |
1943 | return; |
1944 | file_update_time(filp); | |
1945 | filp_close(filp, NULL); | |
5a1972bd QW |
1946 | } |
1947 | ||
f331a952 | 1948 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1949 | { |
f331a952 | 1950 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1951 | int ret; |
1952 | struct btrfs_path *path; | |
a061fc8d | 1953 | struct btrfs_key key; |
a061fc8d CM |
1954 | struct btrfs_trans_handle *trans; |
1955 | ||
a061fc8d CM |
1956 | path = btrfs_alloc_path(); |
1957 | if (!path) | |
1958 | return -ENOMEM; | |
1959 | ||
a22285a6 | 1960 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1961 | if (IS_ERR(trans)) { |
1962 | btrfs_free_path(path); | |
1963 | return PTR_ERR(trans); | |
1964 | } | |
a061fc8d CM |
1965 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1966 | key.type = BTRFS_DEV_ITEM_KEY; | |
1967 | key.offset = device->devid; | |
1968 | ||
1969 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1970 | if (ret) { |
1971 | if (ret > 0) | |
1972 | ret = -ENOENT; | |
1973 | btrfs_abort_transaction(trans, ret); | |
1974 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1975 | goto out; |
1976 | } | |
1977 | ||
1978 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1979 | if (ret) { |
1980 | btrfs_abort_transaction(trans, ret); | |
1981 | btrfs_end_transaction(trans); | |
1982 | } | |
1983 | ||
a061fc8d CM |
1984 | out: |
1985 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1986 | if (!ret) |
1987 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1988 | return ret; |
1989 | } | |
1990 | ||
3cc31a0d DS |
1991 | /* |
1992 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1993 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1994 | * replace. | |
1995 | */ | |
1996 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1997 | u64 num_devices) | |
a061fc8d | 1998 | { |
a061fc8d | 1999 | u64 all_avail; |
de98ced9 | 2000 | unsigned seq; |
418775a2 | 2001 | int i; |
a061fc8d | 2002 | |
de98ced9 | 2003 | do { |
bd45ffbc | 2004 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2005 | |
bd45ffbc AJ |
2006 | all_avail = fs_info->avail_data_alloc_bits | |
2007 | fs_info->avail_system_alloc_bits | | |
2008 | fs_info->avail_metadata_alloc_bits; | |
2009 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2010 | |
418775a2 | 2011 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2012 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2013 | continue; |
a061fc8d | 2014 | |
418775a2 | 2015 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2016 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2017 | |
418775a2 DS |
2018 | if (ret) |
2019 | return ret; | |
2020 | } | |
53b381b3 DW |
2021 | } |
2022 | ||
bd45ffbc | 2023 | return 0; |
f1fa7f26 AJ |
2024 | } |
2025 | ||
c9162bdf OS |
2026 | static struct btrfs_device * btrfs_find_next_active_device( |
2027 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2028 | { |
2b82032c | 2029 | struct btrfs_device *next_device; |
88acff64 AJ |
2030 | |
2031 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2032 | if (next_device != device && | |
e6e674bd AJ |
2033 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2034 | && next_device->bdev) | |
88acff64 AJ |
2035 | return next_device; |
2036 | } | |
2037 | ||
2038 | return NULL; | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2043 | * and replace it with the provided or the next active device, in the context | |
2044 | * where this function called, there should be always be another device (or | |
2045 | * this_dev) which is active. | |
2046 | */ | |
d6507cf1 NB |
2047 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2048 | struct btrfs_device *this_dev) | |
88acff64 | 2049 | { |
d6507cf1 | 2050 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2051 | struct btrfs_device *next_device; |
2052 | ||
2053 | if (this_dev) | |
2054 | next_device = this_dev; | |
2055 | else | |
2056 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2057 | device); | |
2058 | ASSERT(next_device); | |
2059 | ||
2060 | if (fs_info->sb->s_bdev && | |
2061 | (fs_info->sb->s_bdev == device->bdev)) | |
2062 | fs_info->sb->s_bdev = next_device->bdev; | |
2063 | ||
2064 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2065 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2066 | } | |
2067 | ||
1da73967 AJ |
2068 | /* |
2069 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2070 | * currently replaced. | |
2071 | */ | |
2072 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2073 | { | |
2074 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2075 | ||
cb5583dd | 2076 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2077 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2078 | ASSERT(num_devices > 1); | |
2079 | num_devices--; | |
2080 | } | |
cb5583dd | 2081 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2082 | |
2083 | return num_devices; | |
2084 | } | |
2085 | ||
da353f6b DS |
2086 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2087 | u64 devid) | |
f1fa7f26 AJ |
2088 | { |
2089 | struct btrfs_device *device; | |
1f78160c | 2090 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2091 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2092 | u64 num_devices; |
a061fc8d CM |
2093 | int ret = 0; |
2094 | ||
a061fc8d CM |
2095 | mutex_lock(&uuid_mutex); |
2096 | ||
1da73967 | 2097 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2098 | |
0b246afa | 2099 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2100 | if (ret) |
a061fc8d | 2101 | goto out; |
a061fc8d | 2102 | |
a27a94c2 NB |
2103 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2104 | ||
2105 | if (IS_ERR(device)) { | |
2106 | if (PTR_ERR(device) == -ENOENT && | |
2107 | strcmp(device_path, "missing") == 0) | |
2108 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2109 | else | |
2110 | ret = PTR_ERR(device); | |
53b381b3 | 2111 | goto out; |
a27a94c2 | 2112 | } |
dfe25020 | 2113 | |
eede2bf3 OS |
2114 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2115 | btrfs_warn_in_rcu(fs_info, | |
2116 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2117 | rcu_str_deref(device->name), device->devid); | |
2118 | ret = -ETXTBSY; | |
2119 | goto out; | |
2120 | } | |
2121 | ||
401e29c1 | 2122 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2123 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2124 | goto out; |
63a212ab SB |
2125 | } |
2126 | ||
ebbede42 AJ |
2127 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2128 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2129 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2130 | goto out; |
2b82032c YZ |
2131 | } |
2132 | ||
ebbede42 | 2133 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2134 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2135 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2136 | device->fs_devices->rw_devices--; |
34441361 | 2137 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2138 | } |
a061fc8d | 2139 | |
d7901554 | 2140 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2141 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2142 | mutex_lock(&uuid_mutex); |
a061fc8d | 2143 | if (ret) |
9b3517e9 | 2144 | goto error_undo; |
a061fc8d | 2145 | |
63a212ab SB |
2146 | /* |
2147 | * TODO: the superblock still includes this device in its num_devices | |
2148 | * counter although write_all_supers() is not locked out. This | |
2149 | * could give a filesystem state which requires a degraded mount. | |
2150 | */ | |
f331a952 | 2151 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2152 | if (ret) |
9b3517e9 | 2153 | goto error_undo; |
a061fc8d | 2154 | |
e12c9621 | 2155 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2156 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2157 | |
2158 | /* | |
2159 | * the device list mutex makes sure that we don't change | |
2160 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2161 | * the device supers. Whoever is writing all supers, should |
2162 | * lock the device list mutex before getting the number of | |
2163 | * devices in the super block (super_copy). Conversely, | |
2164 | * whoever updates the number of devices in the super block | |
2165 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2166 | */ |
1f78160c | 2167 | |
41a52a0f AJ |
2168 | /* |
2169 | * In normal cases the cur_devices == fs_devices. But in case | |
2170 | * of deleting a seed device, the cur_devices should point to | |
2171 | * its own fs_devices listed under the fs_devices->seed. | |
2172 | */ | |
1f78160c | 2173 | cur_devices = device->fs_devices; |
b5185197 | 2174 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2175 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2176 | |
41a52a0f AJ |
2177 | cur_devices->num_devices--; |
2178 | cur_devices->total_devices--; | |
b4993e64 AJ |
2179 | /* Update total_devices of the parent fs_devices if it's seed */ |
2180 | if (cur_devices != fs_devices) | |
2181 | fs_devices->total_devices--; | |
2b82032c | 2182 | |
e6e674bd | 2183 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2184 | cur_devices->missing_devices--; |
cd02dca5 | 2185 | |
d6507cf1 | 2186 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2187 | |
0bfaa9c5 | 2188 | if (device->bdev) { |
41a52a0f | 2189 | cur_devices->open_devices--; |
0bfaa9c5 | 2190 | /* remove sysfs entry */ |
b5185197 | 2191 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2192 | } |
99994cde | 2193 | |
0b246afa JM |
2194 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2195 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2196 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2197 | |
cea67ab9 JM |
2198 | /* |
2199 | * at this point, the device is zero sized and detached from | |
2200 | * the devices list. All that's left is to zero out the old | |
2201 | * supers and free the device. | |
2202 | */ | |
ebbede42 | 2203 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2204 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2205 | ||
2206 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2207 | synchronize_rcu(); |
2208 | btrfs_free_device(device); | |
cea67ab9 | 2209 | |
1f78160c | 2210 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2211 | while (fs_devices) { |
8321cf25 RS |
2212 | if (fs_devices->seed == cur_devices) { |
2213 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2214 | break; |
8321cf25 | 2215 | } |
e4404d6e | 2216 | fs_devices = fs_devices->seed; |
2b82032c | 2217 | } |
1f78160c | 2218 | cur_devices->seed = NULL; |
0226e0eb | 2219 | close_fs_devices(cur_devices); |
1f78160c | 2220 | free_fs_devices(cur_devices); |
2b82032c YZ |
2221 | } |
2222 | ||
a061fc8d CM |
2223 | out: |
2224 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2225 | return ret; |
24fc572f | 2226 | |
9b3517e9 | 2227 | error_undo: |
ebbede42 | 2228 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2229 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2230 | list_add(&device->dev_alloc_list, |
b5185197 | 2231 | &fs_devices->alloc_list); |
c3929c36 | 2232 | device->fs_devices->rw_devices++; |
34441361 | 2233 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2234 | } |
24fc572f | 2235 | goto out; |
a061fc8d CM |
2236 | } |
2237 | ||
68a9db5f | 2238 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2239 | { |
d51908ce AJ |
2240 | struct btrfs_fs_devices *fs_devices; |
2241 | ||
68a9db5f | 2242 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2243 | |
25e8e911 AJ |
2244 | /* |
2245 | * in case of fs with no seed, srcdev->fs_devices will point | |
2246 | * to fs_devices of fs_info. However when the dev being replaced is | |
2247 | * a seed dev it will point to the seed's local fs_devices. In short | |
2248 | * srcdev will have its correct fs_devices in both the cases. | |
2249 | */ | |
2250 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2251 | |
e93c89c1 | 2252 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2253 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2254 | fs_devices->num_devices--; |
e6e674bd | 2255 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2256 | fs_devices->missing_devices--; |
e93c89c1 | 2257 | |
ebbede42 | 2258 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2259 | fs_devices->rw_devices--; |
1357272f | 2260 | |
82372bc8 | 2261 | if (srcdev->bdev) |
d51908ce | 2262 | fs_devices->open_devices--; |
084b6e7c QW |
2263 | } |
2264 | ||
65237ee3 | 2265 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c | 2266 | { |
65237ee3 | 2267 | struct btrfs_fs_info *fs_info = srcdev->fs_info; |
084b6e7c | 2268 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
e93c89c1 | 2269 | |
ebbede42 | 2270 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2271 | /* zero out the old super if it is writable */ |
2272 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2273 | } | |
14238819 AJ |
2274 | |
2275 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2276 | synchronize_rcu(); |
2277 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2278 | |
94d5f0c2 AJ |
2279 | /* if this is no devs we rather delete the fs_devices */ |
2280 | if (!fs_devices->num_devices) { | |
2281 | struct btrfs_fs_devices *tmp_fs_devices; | |
2282 | ||
6dd38f81 AJ |
2283 | /* |
2284 | * On a mounted FS, num_devices can't be zero unless it's a | |
2285 | * seed. In case of a seed device being replaced, the replace | |
2286 | * target added to the sprout FS, so there will be no more | |
2287 | * device left under the seed FS. | |
2288 | */ | |
2289 | ASSERT(fs_devices->seeding); | |
2290 | ||
94d5f0c2 AJ |
2291 | tmp_fs_devices = fs_info->fs_devices; |
2292 | while (tmp_fs_devices) { | |
2293 | if (tmp_fs_devices->seed == fs_devices) { | |
2294 | tmp_fs_devices->seed = fs_devices->seed; | |
2295 | break; | |
2296 | } | |
2297 | tmp_fs_devices = tmp_fs_devices->seed; | |
2298 | } | |
2299 | fs_devices->seed = NULL; | |
0226e0eb | 2300 | close_fs_devices(fs_devices); |
8bef8401 | 2301 | free_fs_devices(fs_devices); |
94d5f0c2 | 2302 | } |
e93c89c1 SB |
2303 | } |
2304 | ||
4f5ad7bd | 2305 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2306 | { |
4f5ad7bd | 2307 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2308 | |
e93c89c1 | 2309 | WARN_ON(!tgtdev); |
d9a071f0 | 2310 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2311 | |
d9a071f0 | 2312 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2313 | |
779bf3fe | 2314 | if (tgtdev->bdev) |
d9a071f0 | 2315 | fs_devices->open_devices--; |
779bf3fe | 2316 | |
d9a071f0 | 2317 | fs_devices->num_devices--; |
e93c89c1 | 2318 | |
d6507cf1 | 2319 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2320 | |
e93c89c1 | 2321 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2322 | |
d9a071f0 | 2323 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2324 | |
2325 | /* | |
2326 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2327 | * may lead to a call to btrfs_show_devname() which will try | |
2328 | * to hold device_list_mutex. And here this device | |
2329 | * is already out of device list, so we don't have to hold | |
2330 | * the device_list_mutex lock. | |
2331 | */ | |
2332 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2333 | |
2334 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2335 | synchronize_rcu(); |
2336 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2337 | } |
2338 | ||
b444ad46 NB |
2339 | static struct btrfs_device *btrfs_find_device_by_path( |
2340 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2341 | { |
2342 | int ret = 0; | |
2343 | struct btrfs_super_block *disk_super; | |
2344 | u64 devid; | |
2345 | u8 *dev_uuid; | |
2346 | struct block_device *bdev; | |
2347 | struct buffer_head *bh; | |
b444ad46 | 2348 | struct btrfs_device *device; |
7ba15b7d | 2349 | |
7ba15b7d | 2350 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2351 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2352 | if (ret) |
b444ad46 | 2353 | return ERR_PTR(ret); |
7ba15b7d SB |
2354 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2355 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2356 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2357 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2358 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2359 | disk_super->metadata_uuid, true); |
7239ff4b | 2360 | else |
e4319cd9 | 2361 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2362 | disk_super->fsid, true); |
7239ff4b | 2363 | |
7ba15b7d | 2364 | brelse(bh); |
b444ad46 NB |
2365 | if (!device) |
2366 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2367 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2368 | return device; |
7ba15b7d SB |
2369 | } |
2370 | ||
5c5c0df0 DS |
2371 | /* |
2372 | * Lookup a device given by device id, or the path if the id is 0. | |
2373 | */ | |
a27a94c2 | 2374 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2375 | struct btrfs_fs_info *fs_info, u64 devid, |
2376 | const char *device_path) | |
24e0474b | 2377 | { |
a27a94c2 | 2378 | struct btrfs_device *device; |
24e0474b | 2379 | |
5c5c0df0 | 2380 | if (devid) { |
e4319cd9 | 2381 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2382 | NULL, true); |
a27a94c2 NB |
2383 | if (!device) |
2384 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2385 | return device; |
2386 | } | |
2387 | ||
2388 | if (!device_path || !device_path[0]) | |
2389 | return ERR_PTR(-EINVAL); | |
2390 | ||
2391 | if (strcmp(device_path, "missing") == 0) { | |
2392 | /* Find first missing device */ | |
2393 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2394 | dev_list) { | |
2395 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2396 | &device->dev_state) && !device->bdev) | |
2397 | return device; | |
d95a830c | 2398 | } |
6e927ceb | 2399 | return ERR_PTR(-ENOENT); |
24e0474b | 2400 | } |
6e927ceb AJ |
2401 | |
2402 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2403 | } |
2404 | ||
2b82032c YZ |
2405 | /* |
2406 | * does all the dirty work required for changing file system's UUID. | |
2407 | */ | |
2ff7e61e | 2408 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2409 | { |
0b246afa | 2410 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2411 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2412 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2413 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2414 | struct btrfs_device *device; |
2415 | u64 super_flags; | |
2416 | ||
a32bf9a3 | 2417 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2418 | if (!fs_devices->seeding) |
2b82032c YZ |
2419 | return -EINVAL; |
2420 | ||
7239ff4b | 2421 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2422 | if (IS_ERR(seed_devices)) |
2423 | return PTR_ERR(seed_devices); | |
2b82032c | 2424 | |
e4404d6e YZ |
2425 | old_devices = clone_fs_devices(fs_devices); |
2426 | if (IS_ERR(old_devices)) { | |
2427 | kfree(seed_devices); | |
2428 | return PTR_ERR(old_devices); | |
2b82032c | 2429 | } |
e4404d6e | 2430 | |
c4babc5e | 2431 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2432 | |
e4404d6e YZ |
2433 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2434 | seed_devices->opened = 1; | |
2435 | INIT_LIST_HEAD(&seed_devices->devices); | |
2436 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2437 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2438 | |
321a4bf7 | 2439 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2440 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2441 | synchronize_rcu); | |
2196d6e8 MX |
2442 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2443 | device->fs_devices = seed_devices; | |
c9513edb | 2444 | |
34441361 | 2445 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2446 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2447 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2448 | |
2b82032c YZ |
2449 | fs_devices->seeding = 0; |
2450 | fs_devices->num_devices = 0; | |
2451 | fs_devices->open_devices = 0; | |
69611ac8 | 2452 | fs_devices->missing_devices = 0; |
69611ac8 | 2453 | fs_devices->rotating = 0; |
e4404d6e | 2454 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2455 | |
2456 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2457 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2458 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2459 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2460 | |
2b82032c YZ |
2461 | super_flags = btrfs_super_flags(disk_super) & |
2462 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2463 | btrfs_set_super_flags(disk_super, super_flags); | |
2464 | ||
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | /* | |
01327610 | 2469 | * Store the expected generation for seed devices in device items. |
2b82032c | 2470 | */ |
5c466629 | 2471 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2472 | { |
5c466629 | 2473 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2474 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2475 | struct btrfs_path *path; |
2476 | struct extent_buffer *leaf; | |
2477 | struct btrfs_dev_item *dev_item; | |
2478 | struct btrfs_device *device; | |
2479 | struct btrfs_key key; | |
44880fdc | 2480 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2481 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2482 | u64 devid; | |
2483 | int ret; | |
2484 | ||
2485 | path = btrfs_alloc_path(); | |
2486 | if (!path) | |
2487 | return -ENOMEM; | |
2488 | ||
2b82032c YZ |
2489 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2490 | key.offset = 0; | |
2491 | key.type = BTRFS_DEV_ITEM_KEY; | |
2492 | ||
2493 | while (1) { | |
2494 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2495 | if (ret < 0) | |
2496 | goto error; | |
2497 | ||
2498 | leaf = path->nodes[0]; | |
2499 | next_slot: | |
2500 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2501 | ret = btrfs_next_leaf(root, path); | |
2502 | if (ret > 0) | |
2503 | break; | |
2504 | if (ret < 0) | |
2505 | goto error; | |
2506 | leaf = path->nodes[0]; | |
2507 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2508 | btrfs_release_path(path); |
2b82032c YZ |
2509 | continue; |
2510 | } | |
2511 | ||
2512 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2513 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2514 | key.type != BTRFS_DEV_ITEM_KEY) | |
2515 | break; | |
2516 | ||
2517 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2518 | struct btrfs_dev_item); | |
2519 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2520 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2521 | BTRFS_UUID_SIZE); |
1473b24e | 2522 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2523 | BTRFS_FSID_SIZE); |
e4319cd9 | 2524 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2525 | fs_uuid, true); |
79787eaa | 2526 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2527 | |
2528 | if (device->fs_devices->seeding) { | |
2529 | btrfs_set_device_generation(leaf, dev_item, | |
2530 | device->generation); | |
2531 | btrfs_mark_buffer_dirty(leaf); | |
2532 | } | |
2533 | ||
2534 | path->slots[0]++; | |
2535 | goto next_slot; | |
2536 | } | |
2537 | ret = 0; | |
2538 | error: | |
2539 | btrfs_free_path(path); | |
2540 | return ret; | |
2541 | } | |
2542 | ||
da353f6b | 2543 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2544 | { |
5112febb | 2545 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2546 | struct request_queue *q; |
788f20eb CM |
2547 | struct btrfs_trans_handle *trans; |
2548 | struct btrfs_device *device; | |
2549 | struct block_device *bdev; | |
0b246afa | 2550 | struct super_block *sb = fs_info->sb; |
606686ee | 2551 | struct rcu_string *name; |
5da54bc1 | 2552 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2553 | u64 orig_super_total_bytes; |
2554 | u64 orig_super_num_devices; | |
2b82032c | 2555 | int seeding_dev = 0; |
788f20eb | 2556 | int ret = 0; |
7132a262 | 2557 | bool unlocked = false; |
788f20eb | 2558 | |
5da54bc1 | 2559 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2560 | return -EROFS; |
788f20eb | 2561 | |
a5d16333 | 2562 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2563 | fs_info->bdev_holder); |
7f59203a JB |
2564 | if (IS_ERR(bdev)) |
2565 | return PTR_ERR(bdev); | |
a2135011 | 2566 | |
5da54bc1 | 2567 | if (fs_devices->seeding) { |
2b82032c YZ |
2568 | seeding_dev = 1; |
2569 | down_write(&sb->s_umount); | |
2570 | mutex_lock(&uuid_mutex); | |
2571 | } | |
2572 | ||
8c8bee1d | 2573 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2574 | |
5da54bc1 | 2575 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2576 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2577 | if (device->bdev == bdev) { |
2578 | ret = -EEXIST; | |
d25628bd | 2579 | mutex_unlock( |
5da54bc1 | 2580 | &fs_devices->device_list_mutex); |
2b82032c | 2581 | goto error; |
788f20eb CM |
2582 | } |
2583 | } | |
5da54bc1 | 2584 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2585 | |
0b246afa | 2586 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2587 | if (IS_ERR(device)) { |
788f20eb | 2588 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2589 | ret = PTR_ERR(device); |
2b82032c | 2590 | goto error; |
788f20eb CM |
2591 | } |
2592 | ||
78f2c9e6 | 2593 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2594 | if (!name) { |
2b82032c | 2595 | ret = -ENOMEM; |
5c4cf6c9 | 2596 | goto error_free_device; |
788f20eb | 2597 | } |
606686ee | 2598 | rcu_assign_pointer(device->name, name); |
2b82032c | 2599 | |
a22285a6 | 2600 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2601 | if (IS_ERR(trans)) { |
98d5dc13 | 2602 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2603 | goto error_free_device; |
98d5dc13 TI |
2604 | } |
2605 | ||
d5e2003c | 2606 | q = bdev_get_queue(bdev); |
ebbede42 | 2607 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2608 | device->generation = trans->transid; |
0b246afa JM |
2609 | device->io_width = fs_info->sectorsize; |
2610 | device->io_align = fs_info->sectorsize; | |
2611 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2612 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2613 | fs_info->sectorsize); | |
2cc3c559 | 2614 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2615 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2616 | device->fs_info = fs_info; |
788f20eb | 2617 | device->bdev = bdev; |
e12c9621 | 2618 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2619 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2620 | device->mode = FMODE_EXCL; |
27087f37 | 2621 | device->dev_stats_valid = 1; |
9f6d2510 | 2622 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2623 | |
2b82032c | 2624 | if (seeding_dev) { |
1751e8a6 | 2625 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2626 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2627 | if (ret) { |
2628 | btrfs_abort_transaction(trans, ret); | |
2629 | goto error_trans; | |
2630 | } | |
2b82032c | 2631 | } |
788f20eb | 2632 | |
5da54bc1 | 2633 | device->fs_devices = fs_devices; |
e5e9a520 | 2634 | |
5da54bc1 | 2635 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2636 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2637 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2638 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2639 | fs_devices->num_devices++; | |
2640 | fs_devices->open_devices++; | |
2641 | fs_devices->rw_devices++; | |
2642 | fs_devices->total_devices++; | |
2643 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2644 | |
a5ed45f8 | 2645 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2646 | |
e884f4f0 | 2647 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2648 | fs_devices->rotating = 1; |
c289811c | 2649 | |
39379faa | 2650 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2651 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2652 | round_down(orig_super_total_bytes + device->total_bytes, |
2653 | fs_info->sectorsize)); | |
788f20eb | 2654 | |
39379faa NA |
2655 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2656 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2657 | orig_super_num_devices + 1); | |
0d39376a AJ |
2658 | |
2659 | /* add sysfs device entry */ | |
5da54bc1 | 2660 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2661 | |
2196d6e8 MX |
2662 | /* |
2663 | * we've got more storage, clear any full flags on the space | |
2664 | * infos | |
2665 | */ | |
0b246afa | 2666 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2667 | |
34441361 | 2668 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2669 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2670 | |
2b82032c | 2671 | if (seeding_dev) { |
34441361 | 2672 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2673 | ret = init_first_rw_device(trans); |
34441361 | 2674 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2675 | if (ret) { |
66642832 | 2676 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2677 | goto error_sysfs; |
005d6427 | 2678 | } |
2196d6e8 MX |
2679 | } |
2680 | ||
8e87e856 | 2681 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2682 | if (ret) { |
66642832 | 2683 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2684 | goto error_sysfs; |
2196d6e8 MX |
2685 | } |
2686 | ||
2687 | if (seeding_dev) { | |
5c466629 | 2688 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2689 | if (ret) { |
66642832 | 2690 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2691 | goto error_sysfs; |
005d6427 | 2692 | } |
b2373f25 | 2693 | |
f93c3997 DS |
2694 | btrfs_sysfs_update_sprout_fsid(fs_devices, |
2695 | fs_info->fs_devices->fsid); | |
2b82032c YZ |
2696 | } |
2697 | ||
3a45bb20 | 2698 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2699 | |
2b82032c YZ |
2700 | if (seeding_dev) { |
2701 | mutex_unlock(&uuid_mutex); | |
2702 | up_write(&sb->s_umount); | |
7132a262 | 2703 | unlocked = true; |
788f20eb | 2704 | |
79787eaa JM |
2705 | if (ret) /* transaction commit */ |
2706 | return ret; | |
2707 | ||
2ff7e61e | 2708 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2709 | if (ret < 0) |
0b246afa | 2710 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2711 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2712 | trans = btrfs_attach_transaction(root); |
2713 | if (IS_ERR(trans)) { | |
2714 | if (PTR_ERR(trans) == -ENOENT) | |
2715 | return 0; | |
7132a262 AJ |
2716 | ret = PTR_ERR(trans); |
2717 | trans = NULL; | |
2718 | goto error_sysfs; | |
671415b7 | 2719 | } |
3a45bb20 | 2720 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2721 | } |
c9e9f97b | 2722 | |
5a1972bd QW |
2723 | /* Update ctime/mtime for libblkid */ |
2724 | update_dev_time(device_path); | |
2b82032c | 2725 | return ret; |
79787eaa | 2726 | |
d31c32f6 | 2727 | error_sysfs: |
5da54bc1 | 2728 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2729 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2730 | mutex_lock(&fs_info->chunk_mutex); | |
2731 | list_del_rcu(&device->dev_list); | |
2732 | list_del(&device->dev_alloc_list); | |
2733 | fs_info->fs_devices->num_devices--; | |
2734 | fs_info->fs_devices->open_devices--; | |
2735 | fs_info->fs_devices->rw_devices--; | |
2736 | fs_info->fs_devices->total_devices--; | |
2737 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2738 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2739 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2740 | orig_super_total_bytes); | |
2741 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2742 | orig_super_num_devices); | |
2743 | mutex_unlock(&fs_info->chunk_mutex); | |
2744 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2745 | error_trans: |
0af2c4bf | 2746 | if (seeding_dev) |
1751e8a6 | 2747 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2748 | if (trans) |
2749 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2750 | error_free_device: |
a425f9d4 | 2751 | btrfs_free_device(device); |
2b82032c | 2752 | error: |
e525fd89 | 2753 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2754 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2755 | mutex_unlock(&uuid_mutex); |
2756 | up_write(&sb->s_umount); | |
2757 | } | |
c9e9f97b | 2758 | return ret; |
788f20eb CM |
2759 | } |
2760 | ||
d397712b CM |
2761 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2762 | struct btrfs_device *device) | |
0b86a832 CM |
2763 | { |
2764 | int ret; | |
2765 | struct btrfs_path *path; | |
0b246afa | 2766 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2767 | struct btrfs_dev_item *dev_item; |
2768 | struct extent_buffer *leaf; | |
2769 | struct btrfs_key key; | |
2770 | ||
0b86a832 CM |
2771 | path = btrfs_alloc_path(); |
2772 | if (!path) | |
2773 | return -ENOMEM; | |
2774 | ||
2775 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2776 | key.type = BTRFS_DEV_ITEM_KEY; | |
2777 | key.offset = device->devid; | |
2778 | ||
2779 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2780 | if (ret < 0) | |
2781 | goto out; | |
2782 | ||
2783 | if (ret > 0) { | |
2784 | ret = -ENOENT; | |
2785 | goto out; | |
2786 | } | |
2787 | ||
2788 | leaf = path->nodes[0]; | |
2789 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2790 | ||
2791 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2792 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2793 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2794 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2795 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2796 | btrfs_set_device_total_bytes(leaf, dev_item, |
2797 | btrfs_device_get_disk_total_bytes(device)); | |
2798 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2799 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2800 | btrfs_mark_buffer_dirty(leaf); |
2801 | ||
2802 | out: | |
2803 | btrfs_free_path(path); | |
2804 | return ret; | |
2805 | } | |
2806 | ||
2196d6e8 | 2807 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2808 | struct btrfs_device *device, u64 new_size) |
2809 | { | |
0b246afa JM |
2810 | struct btrfs_fs_info *fs_info = device->fs_info; |
2811 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2812 | u64 old_total; |
2813 | u64 diff; | |
8f18cf13 | 2814 | |
ebbede42 | 2815 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2816 | return -EACCES; |
2196d6e8 | 2817 | |
7dfb8be1 NB |
2818 | new_size = round_down(new_size, fs_info->sectorsize); |
2819 | ||
34441361 | 2820 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2821 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2822 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2823 | |
63a212ab | 2824 | if (new_size <= device->total_bytes || |
401e29c1 | 2825 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2826 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2827 | return -EINVAL; |
2196d6e8 | 2828 | } |
2b82032c | 2829 | |
7dfb8be1 NB |
2830 | btrfs_set_super_total_bytes(super_copy, |
2831 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2832 | device->fs_devices->total_rw_bytes += diff; |
2833 | ||
7cc8e58d MX |
2834 | btrfs_device_set_total_bytes(device, new_size); |
2835 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2836 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2837 | if (list_empty(&device->post_commit_list)) |
2838 | list_add_tail(&device->post_commit_list, | |
2839 | &trans->transaction->dev_update_list); | |
34441361 | 2840 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2841 | |
8f18cf13 CM |
2842 | return btrfs_update_device(trans, device); |
2843 | } | |
2844 | ||
f4208794 | 2845 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2846 | { |
f4208794 | 2847 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2848 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2849 | int ret; |
2850 | struct btrfs_path *path; | |
2851 | struct btrfs_key key; | |
2852 | ||
8f18cf13 CM |
2853 | path = btrfs_alloc_path(); |
2854 | if (!path) | |
2855 | return -ENOMEM; | |
2856 | ||
408fbf19 | 2857 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2858 | key.offset = chunk_offset; |
2859 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2860 | ||
2861 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2862 | if (ret < 0) |
2863 | goto out; | |
2864 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2865 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2866 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2867 | ret = -ENOENT; |
2868 | goto out; | |
2869 | } | |
8f18cf13 CM |
2870 | |
2871 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2872 | if (ret < 0) |
0b246afa JM |
2873 | btrfs_handle_fs_error(fs_info, ret, |
2874 | "Failed to delete chunk item."); | |
79787eaa | 2875 | out: |
8f18cf13 | 2876 | btrfs_free_path(path); |
65a246c5 | 2877 | return ret; |
8f18cf13 CM |
2878 | } |
2879 | ||
408fbf19 | 2880 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2881 | { |
0b246afa | 2882 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2883 | struct btrfs_disk_key *disk_key; |
2884 | struct btrfs_chunk *chunk; | |
2885 | u8 *ptr; | |
2886 | int ret = 0; | |
2887 | u32 num_stripes; | |
2888 | u32 array_size; | |
2889 | u32 len = 0; | |
2890 | u32 cur; | |
2891 | struct btrfs_key key; | |
2892 | ||
34441361 | 2893 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2894 | array_size = btrfs_super_sys_array_size(super_copy); |
2895 | ||
2896 | ptr = super_copy->sys_chunk_array; | |
2897 | cur = 0; | |
2898 | ||
2899 | while (cur < array_size) { | |
2900 | disk_key = (struct btrfs_disk_key *)ptr; | |
2901 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2902 | ||
2903 | len = sizeof(*disk_key); | |
2904 | ||
2905 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2906 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2907 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2908 | len += btrfs_chunk_item_size(num_stripes); | |
2909 | } else { | |
2910 | ret = -EIO; | |
2911 | break; | |
2912 | } | |
408fbf19 | 2913 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2914 | key.offset == chunk_offset) { |
2915 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2916 | array_size -= len; | |
2917 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2918 | } else { | |
2919 | ptr += len; | |
2920 | cur += len; | |
2921 | } | |
2922 | } | |
34441361 | 2923 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2924 | return ret; |
2925 | } | |
2926 | ||
60ca842e OS |
2927 | /* |
2928 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2929 | * @logical: Logical block offset in bytes. | |
2930 | * @length: Length of extent in bytes. | |
2931 | * | |
2932 | * Return: Chunk mapping or ERR_PTR. | |
2933 | */ | |
2934 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2935 | u64 logical, u64 length) | |
592d92ee LB |
2936 | { |
2937 | struct extent_map_tree *em_tree; | |
2938 | struct extent_map *em; | |
2939 | ||
c8bf1b67 | 2940 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2941 | read_lock(&em_tree->lock); |
2942 | em = lookup_extent_mapping(em_tree, logical, length); | |
2943 | read_unlock(&em_tree->lock); | |
2944 | ||
2945 | if (!em) { | |
2946 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2947 | logical, length); | |
2948 | return ERR_PTR(-EINVAL); | |
2949 | } | |
2950 | ||
2951 | if (em->start > logical || em->start + em->len < logical) { | |
2952 | btrfs_crit(fs_info, | |
2953 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2954 | logical, length, em->start, em->start + em->len); | |
2955 | free_extent_map(em); | |
2956 | return ERR_PTR(-EINVAL); | |
2957 | } | |
2958 | ||
2959 | /* callers are responsible for dropping em's ref. */ | |
2960 | return em; | |
2961 | } | |
2962 | ||
97aff912 | 2963 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2964 | { |
97aff912 | 2965 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2966 | struct extent_map *em; |
2967 | struct map_lookup *map; | |
2196d6e8 | 2968 | u64 dev_extent_len = 0; |
47ab2a6c | 2969 | int i, ret = 0; |
0b246afa | 2970 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2971 | |
60ca842e | 2972 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2973 | if (IS_ERR(em)) { |
47ab2a6c JB |
2974 | /* |
2975 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2976 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2977 | * do anything we still error out. |
2978 | */ | |
2979 | ASSERT(0); | |
592d92ee | 2980 | return PTR_ERR(em); |
47ab2a6c | 2981 | } |
95617d69 | 2982 | map = em->map_lookup; |
34441361 | 2983 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2984 | check_system_chunk(trans, map->type); |
34441361 | 2985 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2986 | |
57ba4cb8 FM |
2987 | /* |
2988 | * Take the device list mutex to prevent races with the final phase of | |
2989 | * a device replace operation that replaces the device object associated | |
2990 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2991 | */ | |
2992 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2993 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2994 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2995 | ret = btrfs_free_dev_extent(trans, device, |
2996 | map->stripes[i].physical, | |
2997 | &dev_extent_len); | |
47ab2a6c | 2998 | if (ret) { |
57ba4cb8 | 2999 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3000 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3001 | goto out; |
3002 | } | |
a061fc8d | 3003 | |
2196d6e8 | 3004 | if (device->bytes_used > 0) { |
34441361 | 3005 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3006 | btrfs_device_set_bytes_used(device, |
3007 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3008 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3009 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3010 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3011 | } |
a061fc8d | 3012 | |
64bc6c2a NB |
3013 | ret = btrfs_update_device(trans, device); |
3014 | if (ret) { | |
3015 | mutex_unlock(&fs_devices->device_list_mutex); | |
3016 | btrfs_abort_transaction(trans, ret); | |
3017 | goto out; | |
dfe25020 | 3018 | } |
8f18cf13 | 3019 | } |
57ba4cb8 FM |
3020 | mutex_unlock(&fs_devices->device_list_mutex); |
3021 | ||
f4208794 | 3022 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3023 | if (ret) { |
66642832 | 3024 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3025 | goto out; |
3026 | } | |
8f18cf13 | 3027 | |
6bccf3ab | 3028 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3029 | |
8f18cf13 | 3030 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3031 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3032 | if (ret) { |
66642832 | 3033 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3034 | goto out; |
3035 | } | |
8f18cf13 CM |
3036 | } |
3037 | ||
5a98ec01 | 3038 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3039 | if (ret) { |
66642832 | 3040 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3041 | goto out; |
3042 | } | |
2b82032c | 3043 | |
47ab2a6c | 3044 | out: |
2b82032c YZ |
3045 | /* once for us */ |
3046 | free_extent_map(em); | |
47ab2a6c JB |
3047 | return ret; |
3048 | } | |
2b82032c | 3049 | |
5b4aacef | 3050 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3051 | { |
5b4aacef | 3052 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3053 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3054 | int ret; |
2b82032c | 3055 | |
67c5e7d4 FM |
3056 | /* |
3057 | * Prevent races with automatic removal of unused block groups. | |
3058 | * After we relocate and before we remove the chunk with offset | |
3059 | * chunk_offset, automatic removal of the block group can kick in, | |
3060 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3061 | * | |
3062 | * Make sure to acquire this mutex before doing a tree search (dev | |
3063 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3064 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3065 | * we release the path used to search the chunk/dev tree and before | |
3066 | * the current task acquires this mutex and calls us. | |
3067 | */ | |
a32bf9a3 | 3068 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3069 | |
47ab2a6c | 3070 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3071 | btrfs_scrub_pause(fs_info); |
0b246afa | 3072 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3073 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3074 | if (ret) |
3075 | return ret; | |
3076 | ||
19c4d2f9 CM |
3077 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3078 | chunk_offset); | |
3079 | if (IS_ERR(trans)) { | |
3080 | ret = PTR_ERR(trans); | |
3081 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3082 | return ret; | |
3083 | } | |
3084 | ||
47ab2a6c | 3085 | /* |
19c4d2f9 CM |
3086 | * step two, delete the device extents and the |
3087 | * chunk tree entries | |
47ab2a6c | 3088 | */ |
97aff912 | 3089 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3090 | btrfs_end_transaction(trans); |
19c4d2f9 | 3091 | return ret; |
2b82032c YZ |
3092 | } |
3093 | ||
2ff7e61e | 3094 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3095 | { |
0b246afa | 3096 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3097 | struct btrfs_path *path; |
3098 | struct extent_buffer *leaf; | |
3099 | struct btrfs_chunk *chunk; | |
3100 | struct btrfs_key key; | |
3101 | struct btrfs_key found_key; | |
2b82032c | 3102 | u64 chunk_type; |
ba1bf481 JB |
3103 | bool retried = false; |
3104 | int failed = 0; | |
2b82032c YZ |
3105 | int ret; |
3106 | ||
3107 | path = btrfs_alloc_path(); | |
3108 | if (!path) | |
3109 | return -ENOMEM; | |
3110 | ||
ba1bf481 | 3111 | again: |
2b82032c YZ |
3112 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3113 | key.offset = (u64)-1; | |
3114 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3115 | ||
3116 | while (1) { | |
0b246afa | 3117 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3118 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3119 | if (ret < 0) { |
0b246afa | 3120 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3121 | goto error; |
67c5e7d4 | 3122 | } |
79787eaa | 3123 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3124 | |
3125 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3126 | key.type); | |
67c5e7d4 | 3127 | if (ret) |
0b246afa | 3128 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3129 | if (ret < 0) |
3130 | goto error; | |
3131 | if (ret > 0) | |
3132 | break; | |
1a40e23b | 3133 | |
2b82032c YZ |
3134 | leaf = path->nodes[0]; |
3135 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3136 | |
2b82032c YZ |
3137 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3138 | struct btrfs_chunk); | |
3139 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3140 | btrfs_release_path(path); |
8f18cf13 | 3141 | |
2b82032c | 3142 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3143 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3144 | if (ret == -ENOSPC) |
3145 | failed++; | |
14586651 HS |
3146 | else |
3147 | BUG_ON(ret); | |
2b82032c | 3148 | } |
0b246afa | 3149 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3150 | |
2b82032c YZ |
3151 | if (found_key.offset == 0) |
3152 | break; | |
3153 | key.offset = found_key.offset - 1; | |
3154 | } | |
3155 | ret = 0; | |
ba1bf481 JB |
3156 | if (failed && !retried) { |
3157 | failed = 0; | |
3158 | retried = true; | |
3159 | goto again; | |
fae7f21c | 3160 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3161 | ret = -ENOSPC; |
3162 | } | |
2b82032c YZ |
3163 | error: |
3164 | btrfs_free_path(path); | |
3165 | return ret; | |
8f18cf13 CM |
3166 | } |
3167 | ||
a6f93c71 LB |
3168 | /* |
3169 | * return 1 : allocate a data chunk successfully, | |
3170 | * return <0: errors during allocating a data chunk, | |
3171 | * return 0 : no need to allocate a data chunk. | |
3172 | */ | |
3173 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3174 | u64 chunk_offset) | |
3175 | { | |
3176 | struct btrfs_block_group_cache *cache; | |
3177 | u64 bytes_used; | |
3178 | u64 chunk_type; | |
3179 | ||
3180 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3181 | ASSERT(cache); | |
3182 | chunk_type = cache->flags; | |
3183 | btrfs_put_block_group(cache); | |
3184 | ||
3185 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3186 | spin_lock(&fs_info->data_sinfo->lock); | |
3187 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3188 | spin_unlock(&fs_info->data_sinfo->lock); | |
3189 | ||
3190 | if (!bytes_used) { | |
3191 | struct btrfs_trans_handle *trans; | |
3192 | int ret; | |
3193 | ||
3194 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3195 | if (IS_ERR(trans)) | |
3196 | return PTR_ERR(trans); | |
3197 | ||
43a7e99d | 3198 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3199 | BTRFS_BLOCK_GROUP_DATA); |
3200 | btrfs_end_transaction(trans); | |
3201 | if (ret < 0) | |
3202 | return ret; | |
a6f93c71 LB |
3203 | return 1; |
3204 | } | |
3205 | } | |
3206 | return 0; | |
3207 | } | |
3208 | ||
6bccf3ab | 3209 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3210 | struct btrfs_balance_control *bctl) |
3211 | { | |
6bccf3ab | 3212 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3213 | struct btrfs_trans_handle *trans; |
3214 | struct btrfs_balance_item *item; | |
3215 | struct btrfs_disk_balance_args disk_bargs; | |
3216 | struct btrfs_path *path; | |
3217 | struct extent_buffer *leaf; | |
3218 | struct btrfs_key key; | |
3219 | int ret, err; | |
3220 | ||
3221 | path = btrfs_alloc_path(); | |
3222 | if (!path) | |
3223 | return -ENOMEM; | |
3224 | ||
3225 | trans = btrfs_start_transaction(root, 0); | |
3226 | if (IS_ERR(trans)) { | |
3227 | btrfs_free_path(path); | |
3228 | return PTR_ERR(trans); | |
3229 | } | |
3230 | ||
3231 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3232 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3233 | key.offset = 0; |
3234 | ||
3235 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3236 | sizeof(*item)); | |
3237 | if (ret) | |
3238 | goto out; | |
3239 | ||
3240 | leaf = path->nodes[0]; | |
3241 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3242 | ||
b159fa28 | 3243 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3244 | |
3245 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3246 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3247 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3248 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3249 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3250 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3251 | ||
3252 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3253 | ||
3254 | btrfs_mark_buffer_dirty(leaf); | |
3255 | out: | |
3256 | btrfs_free_path(path); | |
3a45bb20 | 3257 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3258 | if (err && !ret) |
3259 | ret = err; | |
3260 | return ret; | |
3261 | } | |
3262 | ||
6bccf3ab | 3263 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3264 | { |
6bccf3ab | 3265 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3266 | struct btrfs_trans_handle *trans; |
3267 | struct btrfs_path *path; | |
3268 | struct btrfs_key key; | |
3269 | int ret, err; | |
3270 | ||
3271 | path = btrfs_alloc_path(); | |
3272 | if (!path) | |
3273 | return -ENOMEM; | |
3274 | ||
3275 | trans = btrfs_start_transaction(root, 0); | |
3276 | if (IS_ERR(trans)) { | |
3277 | btrfs_free_path(path); | |
3278 | return PTR_ERR(trans); | |
3279 | } | |
3280 | ||
3281 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3282 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3283 | key.offset = 0; |
3284 | ||
3285 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3286 | if (ret < 0) | |
3287 | goto out; | |
3288 | if (ret > 0) { | |
3289 | ret = -ENOENT; | |
3290 | goto out; | |
3291 | } | |
3292 | ||
3293 | ret = btrfs_del_item(trans, root, path); | |
3294 | out: | |
3295 | btrfs_free_path(path); | |
3a45bb20 | 3296 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3297 | if (err && !ret) |
3298 | ret = err; | |
3299 | return ret; | |
3300 | } | |
3301 | ||
59641015 ID |
3302 | /* |
3303 | * This is a heuristic used to reduce the number of chunks balanced on | |
3304 | * resume after balance was interrupted. | |
3305 | */ | |
3306 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3307 | { | |
3308 | /* | |
3309 | * Turn on soft mode for chunk types that were being converted. | |
3310 | */ | |
3311 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3312 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3313 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3314 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3315 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3316 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3317 | ||
3318 | /* | |
3319 | * Turn on usage filter if is not already used. The idea is | |
3320 | * that chunks that we have already balanced should be | |
3321 | * reasonably full. Don't do it for chunks that are being | |
3322 | * converted - that will keep us from relocating unconverted | |
3323 | * (albeit full) chunks. | |
3324 | */ | |
3325 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3326 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3327 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3328 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3329 | bctl->data.usage = 90; | |
3330 | } | |
3331 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3332 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3333 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3334 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3335 | bctl->sys.usage = 90; | |
3336 | } | |
3337 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3338 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3339 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3340 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3341 | bctl->meta.usage = 90; | |
3342 | } | |
3343 | } | |
3344 | ||
149196a2 DS |
3345 | /* |
3346 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3347 | */ | |
3348 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3349 | { |
3350 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3351 | int ret; |
c9e9f97b ID |
3352 | |
3353 | BUG_ON(!fs_info->balance_ctl); | |
3354 | ||
3355 | spin_lock(&fs_info->balance_lock); | |
3356 | fs_info->balance_ctl = NULL; | |
3357 | spin_unlock(&fs_info->balance_lock); | |
3358 | ||
3359 | kfree(bctl); | |
149196a2 DS |
3360 | ret = del_balance_item(fs_info); |
3361 | if (ret) | |
3362 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3363 | } |
3364 | ||
ed25e9b2 ID |
3365 | /* |
3366 | * Balance filters. Return 1 if chunk should be filtered out | |
3367 | * (should not be balanced). | |
3368 | */ | |
899c81ea | 3369 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3370 | struct btrfs_balance_args *bargs) |
3371 | { | |
899c81ea ID |
3372 | chunk_type = chunk_to_extended(chunk_type) & |
3373 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3374 | |
899c81ea | 3375 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3376 | return 0; |
3377 | ||
3378 | return 1; | |
3379 | } | |
3380 | ||
dba72cb3 | 3381 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3382 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3383 | { |
3384 | struct btrfs_block_group_cache *cache; | |
3385 | u64 chunk_used; | |
3386 | u64 user_thresh_min; | |
3387 | u64 user_thresh_max; | |
3388 | int ret = 1; | |
3389 | ||
3390 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3391 | chunk_used = btrfs_block_group_used(&cache->item); | |
3392 | ||
3393 | if (bargs->usage_min == 0) | |
3394 | user_thresh_min = 0; | |
3395 | else | |
3396 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3397 | bargs->usage_min); | |
3398 | ||
3399 | if (bargs->usage_max == 0) | |
3400 | user_thresh_max = 1; | |
3401 | else if (bargs->usage_max > 100) | |
3402 | user_thresh_max = cache->key.offset; | |
3403 | else | |
3404 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3405 | bargs->usage_max); | |
3406 | ||
3407 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3408 | ret = 0; | |
3409 | ||
3410 | btrfs_put_block_group(cache); | |
3411 | return ret; | |
3412 | } | |
3413 | ||
dba72cb3 | 3414 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3415 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3416 | { |
3417 | struct btrfs_block_group_cache *cache; | |
3418 | u64 chunk_used, user_thresh; | |
3419 | int ret = 1; | |
3420 | ||
3421 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3422 | chunk_used = btrfs_block_group_used(&cache->item); | |
3423 | ||
bc309467 | 3424 | if (bargs->usage_min == 0) |
3e39cea6 | 3425 | user_thresh = 1; |
a105bb88 ID |
3426 | else if (bargs->usage > 100) |
3427 | user_thresh = cache->key.offset; | |
3428 | else | |
3429 | user_thresh = div_factor_fine(cache->key.offset, | |
3430 | bargs->usage); | |
3431 | ||
5ce5b3c0 ID |
3432 | if (chunk_used < user_thresh) |
3433 | ret = 0; | |
3434 | ||
3435 | btrfs_put_block_group(cache); | |
3436 | return ret; | |
3437 | } | |
3438 | ||
409d404b ID |
3439 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3440 | struct btrfs_chunk *chunk, | |
3441 | struct btrfs_balance_args *bargs) | |
3442 | { | |
3443 | struct btrfs_stripe *stripe; | |
3444 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3445 | int i; | |
3446 | ||
3447 | for (i = 0; i < num_stripes; i++) { | |
3448 | stripe = btrfs_stripe_nr(chunk, i); | |
3449 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3450 | return 0; | |
3451 | } | |
3452 | ||
3453 | return 1; | |
3454 | } | |
3455 | ||
946c9256 DS |
3456 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3457 | { | |
3458 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3459 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3460 | const int nparity = btrfs_raid_array[index].nparity; | |
3461 | ||
3462 | if (nparity) | |
3463 | return num_stripes - nparity; | |
3464 | else | |
3465 | return num_stripes / ncopies; | |
3466 | } | |
3467 | ||
94e60d5a ID |
3468 | /* [pstart, pend) */ |
3469 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3470 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3471 | struct btrfs_balance_args *bargs) |
3472 | { | |
3473 | struct btrfs_stripe *stripe; | |
3474 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3475 | u64 stripe_offset; | |
3476 | u64 stripe_length; | |
946c9256 | 3477 | u64 type; |
94e60d5a ID |
3478 | int factor; |
3479 | int i; | |
3480 | ||
3481 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3482 | return 0; | |
3483 | ||
946c9256 DS |
3484 | type = btrfs_chunk_type(leaf, chunk); |
3485 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3486 | |
3487 | for (i = 0; i < num_stripes; i++) { | |
3488 | stripe = btrfs_stripe_nr(chunk, i); | |
3489 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3490 | continue; | |
3491 | ||
3492 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3493 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3494 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3495 | |
3496 | if (stripe_offset < bargs->pend && | |
3497 | stripe_offset + stripe_length > bargs->pstart) | |
3498 | return 0; | |
3499 | } | |
3500 | ||
3501 | return 1; | |
3502 | } | |
3503 | ||
ea67176a ID |
3504 | /* [vstart, vend) */ |
3505 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3506 | struct btrfs_chunk *chunk, | |
3507 | u64 chunk_offset, | |
3508 | struct btrfs_balance_args *bargs) | |
3509 | { | |
3510 | if (chunk_offset < bargs->vend && | |
3511 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3512 | /* at least part of the chunk is inside this vrange */ | |
3513 | return 0; | |
3514 | ||
3515 | return 1; | |
3516 | } | |
3517 | ||
dee32d0a GAP |
3518 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3519 | struct btrfs_chunk *chunk, | |
3520 | struct btrfs_balance_args *bargs) | |
3521 | { | |
3522 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3523 | ||
3524 | if (bargs->stripes_min <= num_stripes | |
3525 | && num_stripes <= bargs->stripes_max) | |
3526 | return 0; | |
3527 | ||
3528 | return 1; | |
3529 | } | |
3530 | ||
899c81ea | 3531 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3532 | struct btrfs_balance_args *bargs) |
3533 | { | |
3534 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3535 | return 0; | |
3536 | ||
899c81ea ID |
3537 | chunk_type = chunk_to_extended(chunk_type) & |
3538 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3539 | |
899c81ea | 3540 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3541 | return 1; |
3542 | ||
3543 | return 0; | |
3544 | } | |
3545 | ||
6ec0896c | 3546 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3547 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3548 | { | |
6ec0896c | 3549 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3550 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3551 | struct btrfs_balance_args *bargs = NULL; |
3552 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3553 | ||
3554 | /* type filter */ | |
3555 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3556 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3557 | return 0; | |
3558 | } | |
3559 | ||
3560 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3561 | bargs = &bctl->data; | |
3562 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3563 | bargs = &bctl->sys; | |
3564 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3565 | bargs = &bctl->meta; | |
3566 | ||
ed25e9b2 ID |
3567 | /* profiles filter */ |
3568 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3569 | chunk_profiles_filter(chunk_type, bargs)) { | |
3570 | return 0; | |
5ce5b3c0 ID |
3571 | } |
3572 | ||
3573 | /* usage filter */ | |
3574 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3575 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3576 | return 0; |
bc309467 | 3577 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3578 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3579 | return 0; |
409d404b ID |
3580 | } |
3581 | ||
3582 | /* devid filter */ | |
3583 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3584 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3585 | return 0; | |
94e60d5a ID |
3586 | } |
3587 | ||
3588 | /* drange filter, makes sense only with devid filter */ | |
3589 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3590 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3591 | return 0; |
ea67176a ID |
3592 | } |
3593 | ||
3594 | /* vrange filter */ | |
3595 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3596 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3597 | return 0; | |
ed25e9b2 ID |
3598 | } |
3599 | ||
dee32d0a GAP |
3600 | /* stripes filter */ |
3601 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3602 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3603 | return 0; | |
3604 | } | |
3605 | ||
cfa4c961 ID |
3606 | /* soft profile changing mode */ |
3607 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3608 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3609 | return 0; | |
3610 | } | |
3611 | ||
7d824b6f DS |
3612 | /* |
3613 | * limited by count, must be the last filter | |
3614 | */ | |
3615 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3616 | if (bargs->limit == 0) | |
3617 | return 0; | |
3618 | else | |
3619 | bargs->limit--; | |
12907fc7 DS |
3620 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3621 | /* | |
3622 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3623 | * determined here because we do not have the global information |
12907fc7 DS |
3624 | * about the count of all chunks that satisfy the filters. |
3625 | */ | |
3626 | if (bargs->limit_max == 0) | |
3627 | return 0; | |
3628 | else | |
3629 | bargs->limit_max--; | |
7d824b6f DS |
3630 | } |
3631 | ||
f43ffb60 ID |
3632 | return 1; |
3633 | } | |
3634 | ||
c9e9f97b | 3635 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3636 | { |
19a39dce | 3637 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3638 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3639 | u64 chunk_type; |
f43ffb60 | 3640 | struct btrfs_chunk *chunk; |
5a488b9d | 3641 | struct btrfs_path *path = NULL; |
ec44a35c | 3642 | struct btrfs_key key; |
ec44a35c | 3643 | struct btrfs_key found_key; |
f43ffb60 ID |
3644 | struct extent_buffer *leaf; |
3645 | int slot; | |
c9e9f97b ID |
3646 | int ret; |
3647 | int enospc_errors = 0; | |
19a39dce | 3648 | bool counting = true; |
12907fc7 | 3649 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3650 | u64 limit_data = bctl->data.limit; |
3651 | u64 limit_meta = bctl->meta.limit; | |
3652 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3653 | u32 count_data = 0; |
3654 | u32 count_meta = 0; | |
3655 | u32 count_sys = 0; | |
2c9fe835 | 3656 | int chunk_reserved = 0; |
ec44a35c | 3657 | |
ec44a35c | 3658 | path = btrfs_alloc_path(); |
17e9f796 MF |
3659 | if (!path) { |
3660 | ret = -ENOMEM; | |
3661 | goto error; | |
3662 | } | |
19a39dce ID |
3663 | |
3664 | /* zero out stat counters */ | |
3665 | spin_lock(&fs_info->balance_lock); | |
3666 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3667 | spin_unlock(&fs_info->balance_lock); | |
3668 | again: | |
7d824b6f | 3669 | if (!counting) { |
12907fc7 DS |
3670 | /* |
3671 | * The single value limit and min/max limits use the same bytes | |
3672 | * in the | |
3673 | */ | |
7d824b6f DS |
3674 | bctl->data.limit = limit_data; |
3675 | bctl->meta.limit = limit_meta; | |
3676 | bctl->sys.limit = limit_sys; | |
3677 | } | |
ec44a35c CM |
3678 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3679 | key.offset = (u64)-1; | |
3680 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3681 | ||
d397712b | 3682 | while (1) { |
19a39dce | 3683 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3684 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3685 | ret = -ECANCELED; |
3686 | goto error; | |
3687 | } | |
3688 | ||
67c5e7d4 | 3689 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3690 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3691 | if (ret < 0) { |
3692 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3693 | goto error; |
67c5e7d4 | 3694 | } |
ec44a35c CM |
3695 | |
3696 | /* | |
3697 | * this shouldn't happen, it means the last relocate | |
3698 | * failed | |
3699 | */ | |
3700 | if (ret == 0) | |
c9e9f97b | 3701 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3702 | |
3703 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3704 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3705 | if (ret) { |
67c5e7d4 | 3706 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3707 | ret = 0; |
ec44a35c | 3708 | break; |
c9e9f97b | 3709 | } |
7d9eb12c | 3710 | |
f43ffb60 ID |
3711 | leaf = path->nodes[0]; |
3712 | slot = path->slots[0]; | |
3713 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3714 | |
67c5e7d4 FM |
3715 | if (found_key.objectid != key.objectid) { |
3716 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3717 | break; |
67c5e7d4 | 3718 | } |
7d9eb12c | 3719 | |
f43ffb60 | 3720 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3721 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3722 | |
19a39dce ID |
3723 | if (!counting) { |
3724 | spin_lock(&fs_info->balance_lock); | |
3725 | bctl->stat.considered++; | |
3726 | spin_unlock(&fs_info->balance_lock); | |
3727 | } | |
3728 | ||
6ec0896c | 3729 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3730 | |
b3b4aa74 | 3731 | btrfs_release_path(path); |
67c5e7d4 FM |
3732 | if (!ret) { |
3733 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3734 | goto loop; |
67c5e7d4 | 3735 | } |
f43ffb60 | 3736 | |
19a39dce | 3737 | if (counting) { |
67c5e7d4 | 3738 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3739 | spin_lock(&fs_info->balance_lock); |
3740 | bctl->stat.expected++; | |
3741 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3742 | |
3743 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3744 | count_data++; | |
3745 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3746 | count_sys++; | |
3747 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3748 | count_meta++; | |
3749 | ||
3750 | goto loop; | |
3751 | } | |
3752 | ||
3753 | /* | |
3754 | * Apply limit_min filter, no need to check if the LIMITS | |
3755 | * filter is used, limit_min is 0 by default | |
3756 | */ | |
3757 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3758 | count_data < bctl->data.limit_min) | |
3759 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3760 | count_meta < bctl->meta.limit_min) | |
3761 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3762 | count_sys < bctl->sys.limit_min)) { | |
3763 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3764 | goto loop; |
3765 | } | |
3766 | ||
a6f93c71 LB |
3767 | if (!chunk_reserved) { |
3768 | /* | |
3769 | * We may be relocating the only data chunk we have, | |
3770 | * which could potentially end up with losing data's | |
3771 | * raid profile, so lets allocate an empty one in | |
3772 | * advance. | |
3773 | */ | |
3774 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3775 | found_key.offset); | |
2c9fe835 ZL |
3776 | if (ret < 0) { |
3777 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3778 | goto error; | |
a6f93c71 LB |
3779 | } else if (ret == 1) { |
3780 | chunk_reserved = 1; | |
2c9fe835 | 3781 | } |
2c9fe835 ZL |
3782 | } |
3783 | ||
5b4aacef | 3784 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3785 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3786 | if (ret == -ENOSPC) { |
c9e9f97b | 3787 | enospc_errors++; |
eede2bf3 OS |
3788 | } else if (ret == -ETXTBSY) { |
3789 | btrfs_info(fs_info, | |
3790 | "skipping relocation of block group %llu due to active swapfile", | |
3791 | found_key.offset); | |
3792 | ret = 0; | |
3793 | } else if (ret) { | |
3794 | goto error; | |
19a39dce ID |
3795 | } else { |
3796 | spin_lock(&fs_info->balance_lock); | |
3797 | bctl->stat.completed++; | |
3798 | spin_unlock(&fs_info->balance_lock); | |
3799 | } | |
f43ffb60 | 3800 | loop: |
795a3321 ID |
3801 | if (found_key.offset == 0) |
3802 | break; | |
ba1bf481 | 3803 | key.offset = found_key.offset - 1; |
ec44a35c | 3804 | } |
c9e9f97b | 3805 | |
19a39dce ID |
3806 | if (counting) { |
3807 | btrfs_release_path(path); | |
3808 | counting = false; | |
3809 | goto again; | |
3810 | } | |
ec44a35c CM |
3811 | error: |
3812 | btrfs_free_path(path); | |
c9e9f97b | 3813 | if (enospc_errors) { |
efe120a0 | 3814 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3815 | enospc_errors); |
c9e9f97b ID |
3816 | if (!ret) |
3817 | ret = -ENOSPC; | |
3818 | } | |
3819 | ||
ec44a35c CM |
3820 | return ret; |
3821 | } | |
3822 | ||
0c460c0d ID |
3823 | /** |
3824 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3825 | * @flags: profile to validate | |
3826 | * @extended: if true @flags is treated as an extended profile | |
3827 | */ | |
3828 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3829 | { | |
3830 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3831 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3832 | ||
3833 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3834 | ||
3835 | /* 1) check that all other bits are zeroed */ | |
3836 | if (flags & ~mask) | |
3837 | return 0; | |
3838 | ||
3839 | /* 2) see if profile is reduced */ | |
3840 | if (flags == 0) | |
3841 | return !extended; /* "0" is valid for usual profiles */ | |
3842 | ||
3843 | /* true if exactly one bit set */ | |
818255fe | 3844 | return is_power_of_2(flags); |
0c460c0d ID |
3845 | } |
3846 | ||
837d5b6e ID |
3847 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3848 | { | |
a7e99c69 ID |
3849 | /* cancel requested || normal exit path */ |
3850 | return atomic_read(&fs_info->balance_cancel_req) || | |
3851 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3852 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3853 | } |
3854 | ||
bdcd3c97 AM |
3855 | /* Non-zero return value signifies invalidity */ |
3856 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3857 | u64 allowed) | |
3858 | { | |
3859 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3860 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3861 | (bctl_arg->target & ~allowed))); | |
3862 | } | |
3863 | ||
56fc37d9 AJ |
3864 | /* |
3865 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3866 | * @size_buf including the terminating null. The output may be trimmed if it | |
3867 | * does not fit into the provided buffer. | |
3868 | */ | |
3869 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3870 | u32 size_buf) | |
3871 | { | |
3872 | int ret; | |
3873 | u32 size_bp = size_buf; | |
3874 | char *bp = buf; | |
3875 | u64 flags = bargs->flags; | |
3876 | char tmp_buf[128] = {'\0'}; | |
3877 | ||
3878 | if (!flags) | |
3879 | return; | |
3880 | ||
3881 | #define CHECK_APPEND_NOARG(a) \ | |
3882 | do { \ | |
3883 | ret = snprintf(bp, size_bp, (a)); \ | |
3884 | if (ret < 0 || ret >= size_bp) \ | |
3885 | goto out_overflow; \ | |
3886 | size_bp -= ret; \ | |
3887 | bp += ret; \ | |
3888 | } while (0) | |
3889 | ||
3890 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3891 | do { \ | |
3892 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3893 | if (ret < 0 || ret >= size_bp) \ | |
3894 | goto out_overflow; \ | |
3895 | size_bp -= ret; \ | |
3896 | bp += ret; \ | |
3897 | } while (0) | |
3898 | ||
3899 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3900 | do { \ | |
3901 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3902 | if (ret < 0 || ret >= size_bp) \ | |
3903 | goto out_overflow; \ | |
3904 | size_bp -= ret; \ | |
3905 | bp += ret; \ | |
3906 | } while (0) | |
3907 | ||
158da513 DS |
3908 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3909 | CHECK_APPEND_1ARG("convert=%s,", | |
3910 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3911 | |
3912 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3913 | CHECK_APPEND_NOARG("soft,"); | |
3914 | ||
3915 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3916 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3917 | sizeof(tmp_buf)); | |
3918 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3919 | } | |
3920 | ||
3921 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3922 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3923 | ||
3924 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3925 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3926 | bargs->usage_min, bargs->usage_max); | |
3927 | ||
3928 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3929 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3930 | ||
3931 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3932 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3933 | bargs->pstart, bargs->pend); | |
3934 | ||
3935 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3936 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3937 | bargs->vstart, bargs->vend); | |
3938 | ||
3939 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3940 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3941 | ||
3942 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3943 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3944 | bargs->limit_min, bargs->limit_max); | |
3945 | ||
3946 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3947 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3948 | bargs->stripes_min, bargs->stripes_max); | |
3949 | ||
3950 | #undef CHECK_APPEND_2ARG | |
3951 | #undef CHECK_APPEND_1ARG | |
3952 | #undef CHECK_APPEND_NOARG | |
3953 | ||
3954 | out_overflow: | |
3955 | ||
3956 | if (size_bp < size_buf) | |
3957 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3958 | else | |
3959 | buf[0] = '\0'; | |
3960 | } | |
3961 | ||
3962 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3963 | { | |
3964 | u32 size_buf = 1024; | |
3965 | char tmp_buf[192] = {'\0'}; | |
3966 | char *buf; | |
3967 | char *bp; | |
3968 | u32 size_bp = size_buf; | |
3969 | int ret; | |
3970 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3971 | ||
3972 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3973 | if (!buf) | |
3974 | return; | |
3975 | ||
3976 | bp = buf; | |
3977 | ||
3978 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3979 | do { \ | |
3980 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3981 | if (ret < 0 || ret >= size_bp) \ | |
3982 | goto out_overflow; \ | |
3983 | size_bp -= ret; \ | |
3984 | bp += ret; \ | |
3985 | } while (0) | |
3986 | ||
3987 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3988 | CHECK_APPEND_1ARG("%s", "-f "); | |
3989 | ||
3990 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3991 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3992 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3993 | } | |
3994 | ||
3995 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
3996 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
3997 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
3998 | } | |
3999 | ||
4000 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4001 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4002 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4003 | } | |
4004 | ||
4005 | #undef CHECK_APPEND_1ARG | |
4006 | ||
4007 | out_overflow: | |
4008 | ||
4009 | if (size_bp < size_buf) | |
4010 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4011 | btrfs_info(fs_info, "balance: %s %s", | |
4012 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4013 | "resume" : "start", buf); | |
4014 | ||
4015 | kfree(buf); | |
4016 | } | |
4017 | ||
c9e9f97b | 4018 | /* |
dccdb07b | 4019 | * Should be called with balance mutexe held |
c9e9f97b | 4020 | */ |
6fcf6e2b DS |
4021 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4022 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4023 | struct btrfs_ioctl_balance_args *bargs) |
4024 | { | |
14506127 | 4025 | u64 meta_target, data_target; |
f43ffb60 | 4026 | u64 allowed; |
e4837f8f | 4027 | int mixed = 0; |
c9e9f97b | 4028 | int ret; |
8dabb742 | 4029 | u64 num_devices; |
de98ced9 | 4030 | unsigned seq; |
5a8067c0 | 4031 | bool reducing_integrity; |
081db89b | 4032 | int i; |
c9e9f97b | 4033 | |
837d5b6e | 4034 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4035 | atomic_read(&fs_info->balance_pause_req) || |
4036 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4037 | ret = -EINVAL; |
4038 | goto out; | |
4039 | } | |
4040 | ||
e4837f8f ID |
4041 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4042 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4043 | mixed = 1; | |
4044 | ||
f43ffb60 ID |
4045 | /* |
4046 | * In case of mixed groups both data and meta should be picked, | |
4047 | * and identical options should be given for both of them. | |
4048 | */ | |
e4837f8f ID |
4049 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4050 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4051 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4052 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4053 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4054 | btrfs_err(fs_info, |
6dac13f8 | 4055 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4056 | ret = -EINVAL; |
4057 | goto out; | |
4058 | } | |
4059 | } | |
4060 | ||
1da73967 | 4061 | num_devices = btrfs_num_devices(fs_info); |
081db89b DS |
4062 | allowed = 0; |
4063 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) | |
4064 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4065 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4066 | |
bdcd3c97 | 4067 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e | 4068 | btrfs_err(fs_info, |
6dac13f8 | 4069 | "balance: invalid convert data profile %s", |
158da513 | 4070 | btrfs_bg_type_to_raid_name(bctl->data.target)); |
e4d8ec0f ID |
4071 | ret = -EINVAL; |
4072 | goto out; | |
4073 | } | |
bdcd3c97 | 4074 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 4075 | btrfs_err(fs_info, |
6dac13f8 | 4076 | "balance: invalid convert metadata profile %s", |
158da513 | 4077 | btrfs_bg_type_to_raid_name(bctl->meta.target)); |
e4d8ec0f ID |
4078 | ret = -EINVAL; |
4079 | goto out; | |
4080 | } | |
bdcd3c97 | 4081 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 4082 | btrfs_err(fs_info, |
6dac13f8 | 4083 | "balance: invalid convert system profile %s", |
158da513 | 4084 | btrfs_bg_type_to_raid_name(bctl->sys.target)); |
e4d8ec0f ID |
4085 | ret = -EINVAL; |
4086 | goto out; | |
4087 | } | |
4088 | ||
6079e12c DS |
4089 | /* |
4090 | * Allow to reduce metadata or system integrity only if force set for | |
4091 | * profiles with redundancy (copies, parity) | |
4092 | */ | |
4093 | allowed = 0; | |
4094 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4095 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4096 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4097 | allowed |= btrfs_raid_array[i].bg_flag; | |
4098 | } | |
de98ced9 MX |
4099 | do { |
4100 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4101 | ||
4102 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4103 | (fs_info->avail_system_alloc_bits & allowed) && | |
4104 | !(bctl->sys.target & allowed)) || | |
4105 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4106 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4107 | !(bctl->meta.target & allowed))) |
4108 | reducing_integrity = true; | |
4109 | else | |
4110 | reducing_integrity = false; | |
4111 | ||
4112 | /* if we're not converting, the target field is uninitialized */ | |
4113 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4114 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4115 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4116 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4117 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4118 | |
5a8067c0 FM |
4119 | if (reducing_integrity) { |
4120 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4121 | btrfs_info(fs_info, | |
4122 | "balance: force reducing metadata integrity"); | |
4123 | } else { | |
4124 | btrfs_err(fs_info, | |
4125 | "balance: reduces metadata integrity, use --force if you want this"); | |
4126 | ret = -EINVAL; | |
4127 | goto out; | |
4128 | } | |
4129 | } | |
4130 | ||
14506127 AB |
4131 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4132 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4133 | btrfs_warn(fs_info, |
6dac13f8 | 4134 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4135 | btrfs_bg_type_to_raid_name(meta_target), |
4136 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4137 | } |
4138 | ||
9e967495 FM |
4139 | if (fs_info->send_in_progress) { |
4140 | btrfs_warn_rl(fs_info, | |
4141 | "cannot run balance while send operations are in progress (%d in progress)", | |
4142 | fs_info->send_in_progress); | |
4143 | ret = -EAGAIN; | |
4144 | goto out; | |
4145 | } | |
4146 | ||
6bccf3ab | 4147 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4148 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4149 | goto out; |
4150 | ||
59641015 ID |
4151 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4152 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4153 | BUG_ON(fs_info->balance_ctl); |
4154 | spin_lock(&fs_info->balance_lock); | |
4155 | fs_info->balance_ctl = bctl; | |
4156 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4157 | } else { |
4158 | BUG_ON(ret != -EEXIST); | |
4159 | spin_lock(&fs_info->balance_lock); | |
4160 | update_balance_args(bctl); | |
4161 | spin_unlock(&fs_info->balance_lock); | |
4162 | } | |
c9e9f97b | 4163 | |
3009a62f DS |
4164 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4165 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4166 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4167 | mutex_unlock(&fs_info->balance_mutex); |
4168 | ||
4169 | ret = __btrfs_balance(fs_info); | |
4170 | ||
4171 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4172 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4173 | btrfs_info(fs_info, "balance: paused"); | |
4174 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4175 | btrfs_info(fs_info, "balance: canceled"); | |
4176 | else | |
4177 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4178 | ||
3009a62f | 4179 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4180 | |
4181 | if (bargs) { | |
4182 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4183 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4184 | } |
4185 | ||
3a01aa7a ID |
4186 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4187 | balance_need_close(fs_info)) { | |
149196a2 | 4188 | reset_balance_state(fs_info); |
a17c95df | 4189 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4190 | } |
4191 | ||
837d5b6e | 4192 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4193 | |
4194 | return ret; | |
4195 | out: | |
59641015 | 4196 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4197 | reset_balance_state(fs_info); |
a17c95df | 4198 | else |
59641015 | 4199 | kfree(bctl); |
a17c95df DS |
4200 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4201 | ||
59641015 ID |
4202 | return ret; |
4203 | } | |
4204 | ||
4205 | static int balance_kthread(void *data) | |
4206 | { | |
2b6ba629 | 4207 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4208 | int ret = 0; |
59641015 | 4209 | |
59641015 | 4210 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4211 | if (fs_info->balance_ctl) |
6fcf6e2b | 4212 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4213 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4214 | |
59641015 ID |
4215 | return ret; |
4216 | } | |
4217 | ||
2b6ba629 ID |
4218 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4219 | { | |
4220 | struct task_struct *tsk; | |
4221 | ||
1354e1a1 | 4222 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4223 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4224 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4225 | return 0; |
4226 | } | |
1354e1a1 | 4227 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4228 | |
3cdde224 | 4229 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4230 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4231 | return 0; |
4232 | } | |
4233 | ||
02ee654d AJ |
4234 | /* |
4235 | * A ro->rw remount sequence should continue with the paused balance | |
4236 | * regardless of who pauses it, system or the user as of now, so set | |
4237 | * the resume flag. | |
4238 | */ | |
4239 | spin_lock(&fs_info->balance_lock); | |
4240 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4241 | spin_unlock(&fs_info->balance_lock); | |
4242 | ||
2b6ba629 | 4243 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4244 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4245 | } |
4246 | ||
68310a5e | 4247 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4248 | { |
59641015 ID |
4249 | struct btrfs_balance_control *bctl; |
4250 | struct btrfs_balance_item *item; | |
4251 | struct btrfs_disk_balance_args disk_bargs; | |
4252 | struct btrfs_path *path; | |
4253 | struct extent_buffer *leaf; | |
4254 | struct btrfs_key key; | |
4255 | int ret; | |
4256 | ||
4257 | path = btrfs_alloc_path(); | |
4258 | if (!path) | |
4259 | return -ENOMEM; | |
4260 | ||
59641015 | 4261 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4262 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4263 | key.offset = 0; |
4264 | ||
68310a5e | 4265 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4266 | if (ret < 0) |
68310a5e | 4267 | goto out; |
59641015 ID |
4268 | if (ret > 0) { /* ret = -ENOENT; */ |
4269 | ret = 0; | |
68310a5e ID |
4270 | goto out; |
4271 | } | |
4272 | ||
4273 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4274 | if (!bctl) { | |
4275 | ret = -ENOMEM; | |
4276 | goto out; | |
59641015 ID |
4277 | } |
4278 | ||
4279 | leaf = path->nodes[0]; | |
4280 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4281 | ||
68310a5e ID |
4282 | bctl->flags = btrfs_balance_flags(leaf, item); |
4283 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4284 | |
4285 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4286 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4287 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4288 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4289 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4290 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4291 | ||
eee95e3f DS |
4292 | /* |
4293 | * This should never happen, as the paused balance state is recovered | |
4294 | * during mount without any chance of other exclusive ops to collide. | |
4295 | * | |
4296 | * This gives the exclusive op status to balance and keeps in paused | |
4297 | * state until user intervention (cancel or umount). If the ownership | |
4298 | * cannot be assigned, show a message but do not fail. The balance | |
4299 | * is in a paused state and must have fs_info::balance_ctl properly | |
4300 | * set up. | |
4301 | */ | |
4302 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4303 | btrfs_warn(fs_info, | |
6dac13f8 | 4304 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4305 | |
68310a5e | 4306 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4307 | BUG_ON(fs_info->balance_ctl); |
4308 | spin_lock(&fs_info->balance_lock); | |
4309 | fs_info->balance_ctl = bctl; | |
4310 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4311 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4312 | out: |
4313 | btrfs_free_path(path); | |
ec44a35c CM |
4314 | return ret; |
4315 | } | |
4316 | ||
837d5b6e ID |
4317 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4318 | { | |
4319 | int ret = 0; | |
4320 | ||
4321 | mutex_lock(&fs_info->balance_mutex); | |
4322 | if (!fs_info->balance_ctl) { | |
4323 | mutex_unlock(&fs_info->balance_mutex); | |
4324 | return -ENOTCONN; | |
4325 | } | |
4326 | ||
3009a62f | 4327 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4328 | atomic_inc(&fs_info->balance_pause_req); |
4329 | mutex_unlock(&fs_info->balance_mutex); | |
4330 | ||
4331 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4332 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4333 | |
4334 | mutex_lock(&fs_info->balance_mutex); | |
4335 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4336 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4337 | atomic_dec(&fs_info->balance_pause_req); |
4338 | } else { | |
4339 | ret = -ENOTCONN; | |
4340 | } | |
4341 | ||
4342 | mutex_unlock(&fs_info->balance_mutex); | |
4343 | return ret; | |
4344 | } | |
4345 | ||
a7e99c69 ID |
4346 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4347 | { | |
4348 | mutex_lock(&fs_info->balance_mutex); | |
4349 | if (!fs_info->balance_ctl) { | |
4350 | mutex_unlock(&fs_info->balance_mutex); | |
4351 | return -ENOTCONN; | |
4352 | } | |
4353 | ||
cf7d20f4 DS |
4354 | /* |
4355 | * A paused balance with the item stored on disk can be resumed at | |
4356 | * mount time if the mount is read-write. Otherwise it's still paused | |
4357 | * and we must not allow cancelling as it deletes the item. | |
4358 | */ | |
4359 | if (sb_rdonly(fs_info->sb)) { | |
4360 | mutex_unlock(&fs_info->balance_mutex); | |
4361 | return -EROFS; | |
4362 | } | |
4363 | ||
a7e99c69 ID |
4364 | atomic_inc(&fs_info->balance_cancel_req); |
4365 | /* | |
4366 | * if we are running just wait and return, balance item is | |
4367 | * deleted in btrfs_balance in this case | |
4368 | */ | |
3009a62f | 4369 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4370 | mutex_unlock(&fs_info->balance_mutex); |
4371 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4372 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4373 | mutex_lock(&fs_info->balance_mutex); |
4374 | } else { | |
a7e99c69 | 4375 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4376 | /* |
4377 | * Lock released to allow other waiters to continue, we'll | |
4378 | * reexamine the status again. | |
4379 | */ | |
a7e99c69 ID |
4380 | mutex_lock(&fs_info->balance_mutex); |
4381 | ||
a17c95df | 4382 | if (fs_info->balance_ctl) { |
149196a2 | 4383 | reset_balance_state(fs_info); |
a17c95df | 4384 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4385 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4386 | } |
a7e99c69 ID |
4387 | } |
4388 | ||
3009a62f DS |
4389 | BUG_ON(fs_info->balance_ctl || |
4390 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4391 | atomic_dec(&fs_info->balance_cancel_req); |
4392 | mutex_unlock(&fs_info->balance_mutex); | |
4393 | return 0; | |
4394 | } | |
4395 | ||
803b2f54 SB |
4396 | static int btrfs_uuid_scan_kthread(void *data) |
4397 | { | |
4398 | struct btrfs_fs_info *fs_info = data; | |
4399 | struct btrfs_root *root = fs_info->tree_root; | |
4400 | struct btrfs_key key; | |
803b2f54 SB |
4401 | struct btrfs_path *path = NULL; |
4402 | int ret = 0; | |
4403 | struct extent_buffer *eb; | |
4404 | int slot; | |
4405 | struct btrfs_root_item root_item; | |
4406 | u32 item_size; | |
f45388f3 | 4407 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4408 | |
4409 | path = btrfs_alloc_path(); | |
4410 | if (!path) { | |
4411 | ret = -ENOMEM; | |
4412 | goto out; | |
4413 | } | |
4414 | ||
4415 | key.objectid = 0; | |
4416 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4417 | key.offset = 0; | |
4418 | ||
803b2f54 | 4419 | while (1) { |
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: | |
4466 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4467 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4468 | BTRFS_UUID_KEY_SUBVOL, |
4469 | key.objectid); | |
4470 | if (ret < 0) { | |
efe120a0 | 4471 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4472 | ret); |
803b2f54 SB |
4473 | break; |
4474 | } | |
4475 | } | |
4476 | ||
4477 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4478 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4479 | root_item.received_uuid, |
4480 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4481 | key.objectid); | |
4482 | if (ret < 0) { | |
efe120a0 | 4483 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4484 | ret); |
803b2f54 SB |
4485 | break; |
4486 | } | |
4487 | } | |
4488 | ||
f45388f3 | 4489 | skip: |
803b2f54 | 4490 | if (trans) { |
3a45bb20 | 4491 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4492 | trans = NULL; |
803b2f54 SB |
4493 | if (ret) |
4494 | break; | |
4495 | } | |
4496 | ||
803b2f54 SB |
4497 | btrfs_release_path(path); |
4498 | if (key.offset < (u64)-1) { | |
4499 | key.offset++; | |
4500 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4501 | key.offset = 0; | |
4502 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4503 | } else if (key.objectid < (u64)-1) { | |
4504 | key.offset = 0; | |
4505 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4506 | key.objectid++; | |
4507 | } else { | |
4508 | break; | |
4509 | } | |
4510 | cond_resched(); | |
4511 | } | |
4512 | ||
4513 | out: | |
4514 | btrfs_free_path(path); | |
f45388f3 | 4515 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4516 | btrfs_end_transaction(trans); |
803b2f54 | 4517 | if (ret) |
efe120a0 | 4518 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4519 | else |
afcdd129 | 4520 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4521 | up(&fs_info->uuid_tree_rescan_sem); |
4522 | return 0; | |
4523 | } | |
4524 | ||
70f80175 SB |
4525 | /* |
4526 | * Callback for btrfs_uuid_tree_iterate(). | |
4527 | * returns: | |
4528 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4529 | * < 0 if an error occurred. |
70f80175 SB |
4530 | * > 0 if the check failed, which means the caller shall remove the entry. |
4531 | */ | |
4532 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4533 | u8 *uuid, u8 type, u64 subid) | |
4534 | { | |
4535 | struct btrfs_key key; | |
4536 | int ret = 0; | |
4537 | struct btrfs_root *subvol_root; | |
4538 | ||
4539 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4540 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4541 | goto out; | |
4542 | ||
4543 | key.objectid = subid; | |
4544 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4545 | key.offset = (u64)-1; | |
4546 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4547 | if (IS_ERR(subvol_root)) { | |
4548 | ret = PTR_ERR(subvol_root); | |
4549 | if (ret == -ENOENT) | |
4550 | ret = 1; | |
4551 | goto out; | |
4552 | } | |
4553 | ||
4554 | switch (type) { | |
4555 | case BTRFS_UUID_KEY_SUBVOL: | |
4556 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4557 | ret = 1; | |
4558 | break; | |
4559 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4560 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4561 | BTRFS_UUID_SIZE)) | |
4562 | ret = 1; | |
4563 | break; | |
4564 | } | |
4565 | ||
4566 | out: | |
4567 | return ret; | |
4568 | } | |
4569 | ||
4570 | static int btrfs_uuid_rescan_kthread(void *data) | |
4571 | { | |
4572 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4573 | int ret; | |
4574 | ||
4575 | /* | |
4576 | * 1st step is to iterate through the existing UUID tree and | |
4577 | * to delete all entries that contain outdated data. | |
4578 | * 2nd step is to add all missing entries to the UUID tree. | |
4579 | */ | |
4580 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4581 | if (ret < 0) { | |
efe120a0 | 4582 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4583 | up(&fs_info->uuid_tree_rescan_sem); |
4584 | return ret; | |
4585 | } | |
4586 | return btrfs_uuid_scan_kthread(data); | |
4587 | } | |
4588 | ||
f7a81ea4 SB |
4589 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4590 | { | |
4591 | struct btrfs_trans_handle *trans; | |
4592 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4593 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4594 | struct task_struct *task; |
4595 | int ret; | |
f7a81ea4 SB |
4596 | |
4597 | /* | |
4598 | * 1 - root node | |
4599 | * 1 - root item | |
4600 | */ | |
4601 | trans = btrfs_start_transaction(tree_root, 2); | |
4602 | if (IS_ERR(trans)) | |
4603 | return PTR_ERR(trans); | |
4604 | ||
9b7a2440 | 4605 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4606 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4607 | ret = PTR_ERR(uuid_root); |
66642832 | 4608 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4609 | btrfs_end_transaction(trans); |
6d13f549 | 4610 | return ret; |
f7a81ea4 SB |
4611 | } |
4612 | ||
4613 | fs_info->uuid_root = uuid_root; | |
4614 | ||
3a45bb20 | 4615 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4616 | if (ret) |
4617 | return ret; | |
4618 | ||
4619 | down(&fs_info->uuid_tree_rescan_sem); | |
4620 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4621 | if (IS_ERR(task)) { | |
70f80175 | 4622 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4623 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4624 | up(&fs_info->uuid_tree_rescan_sem); |
4625 | return PTR_ERR(task); | |
4626 | } | |
4627 | ||
4628 | return 0; | |
f7a81ea4 | 4629 | } |
803b2f54 | 4630 | |
70f80175 SB |
4631 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4632 | { | |
4633 | struct task_struct *task; | |
4634 | ||
4635 | down(&fs_info->uuid_tree_rescan_sem); | |
4636 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4637 | if (IS_ERR(task)) { | |
4638 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4639 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4640 | up(&fs_info->uuid_tree_rescan_sem); |
4641 | return PTR_ERR(task); | |
4642 | } | |
4643 | ||
4644 | return 0; | |
4645 | } | |
4646 | ||
8f18cf13 CM |
4647 | /* |
4648 | * shrinking a device means finding all of the device extents past | |
4649 | * the new size, and then following the back refs to the chunks. | |
4650 | * The chunk relocation code actually frees the device extent | |
4651 | */ | |
4652 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4653 | { | |
0b246afa JM |
4654 | struct btrfs_fs_info *fs_info = device->fs_info; |
4655 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4656 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4657 | struct btrfs_dev_extent *dev_extent = NULL; |
4658 | struct btrfs_path *path; | |
4659 | u64 length; | |
8f18cf13 CM |
4660 | u64 chunk_offset; |
4661 | int ret; | |
4662 | int slot; | |
ba1bf481 JB |
4663 | int failed = 0; |
4664 | bool retried = false; | |
8f18cf13 CM |
4665 | struct extent_buffer *l; |
4666 | struct btrfs_key key; | |
0b246afa | 4667 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4668 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4669 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4670 | u64 diff; |
61d0d0d2 | 4671 | u64 start; |
7dfb8be1 NB |
4672 | |
4673 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4674 | start = new_size; |
0e4324a4 | 4675 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4676 | |
401e29c1 | 4677 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4678 | return -EINVAL; |
4679 | ||
8f18cf13 CM |
4680 | path = btrfs_alloc_path(); |
4681 | if (!path) | |
4682 | return -ENOMEM; | |
4683 | ||
0338dff6 | 4684 | path->reada = READA_BACK; |
8f18cf13 | 4685 | |
61d0d0d2 NB |
4686 | trans = btrfs_start_transaction(root, 0); |
4687 | if (IS_ERR(trans)) { | |
4688 | btrfs_free_path(path); | |
4689 | return PTR_ERR(trans); | |
4690 | } | |
4691 | ||
34441361 | 4692 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4693 | |
7cc8e58d | 4694 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4695 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4696 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4697 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4698 | } |
61d0d0d2 NB |
4699 | |
4700 | /* | |
4701 | * Once the device's size has been set to the new size, ensure all | |
4702 | * in-memory chunks are synced to disk so that the loop below sees them | |
4703 | * and relocates them accordingly. | |
4704 | */ | |
1c11b63e | 4705 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4706 | mutex_unlock(&fs_info->chunk_mutex); |
4707 | ret = btrfs_commit_transaction(trans); | |
4708 | if (ret) | |
4709 | goto done; | |
4710 | } else { | |
4711 | mutex_unlock(&fs_info->chunk_mutex); | |
4712 | btrfs_end_transaction(trans); | |
4713 | } | |
8f18cf13 | 4714 | |
ba1bf481 | 4715 | again: |
8f18cf13 CM |
4716 | key.objectid = device->devid; |
4717 | key.offset = (u64)-1; | |
4718 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4719 | ||
213e64da | 4720 | do { |
0b246afa | 4721 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4722 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4723 | if (ret < 0) { |
0b246afa | 4724 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4725 | goto done; |
67c5e7d4 | 4726 | } |
8f18cf13 CM |
4727 | |
4728 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4729 | if (ret) |
0b246afa | 4730 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4731 | if (ret < 0) |
4732 | goto done; | |
4733 | if (ret) { | |
4734 | ret = 0; | |
b3b4aa74 | 4735 | btrfs_release_path(path); |
bf1fb512 | 4736 | break; |
8f18cf13 CM |
4737 | } |
4738 | ||
4739 | l = path->nodes[0]; | |
4740 | slot = path->slots[0]; | |
4741 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4742 | ||
ba1bf481 | 4743 | if (key.objectid != device->devid) { |
0b246afa | 4744 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4745 | btrfs_release_path(path); |
bf1fb512 | 4746 | break; |
ba1bf481 | 4747 | } |
8f18cf13 CM |
4748 | |
4749 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4750 | length = btrfs_dev_extent_length(l, dev_extent); | |
4751 | ||
ba1bf481 | 4752 | if (key.offset + length <= new_size) { |
0b246afa | 4753 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4754 | btrfs_release_path(path); |
d6397bae | 4755 | break; |
ba1bf481 | 4756 | } |
8f18cf13 | 4757 | |
8f18cf13 | 4758 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4759 | btrfs_release_path(path); |
8f18cf13 | 4760 | |
a6f93c71 LB |
4761 | /* |
4762 | * We may be relocating the only data chunk we have, | |
4763 | * which could potentially end up with losing data's | |
4764 | * raid profile, so lets allocate an empty one in | |
4765 | * advance. | |
4766 | */ | |
4767 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4768 | if (ret < 0) { | |
4769 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4770 | goto done; | |
4771 | } | |
4772 | ||
0b246afa JM |
4773 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4774 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4775 | if (ret == -ENOSPC) { |
ba1bf481 | 4776 | failed++; |
eede2bf3 OS |
4777 | } else if (ret) { |
4778 | if (ret == -ETXTBSY) { | |
4779 | btrfs_warn(fs_info, | |
4780 | "could not shrink block group %llu due to active swapfile", | |
4781 | chunk_offset); | |
4782 | } | |
4783 | goto done; | |
4784 | } | |
213e64da | 4785 | } while (key.offset-- > 0); |
ba1bf481 JB |
4786 | |
4787 | if (failed && !retried) { | |
4788 | failed = 0; | |
4789 | retried = true; | |
4790 | goto again; | |
4791 | } else if (failed && retried) { | |
4792 | ret = -ENOSPC; | |
ba1bf481 | 4793 | goto done; |
8f18cf13 CM |
4794 | } |
4795 | ||
d6397bae | 4796 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4797 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4798 | if (IS_ERR(trans)) { |
4799 | ret = PTR_ERR(trans); | |
4800 | goto done; | |
4801 | } | |
4802 | ||
34441361 | 4803 | mutex_lock(&fs_info->chunk_mutex); |
7cc8e58d | 4804 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4805 | if (list_empty(&device->post_commit_list)) |
4806 | list_add_tail(&device->post_commit_list, | |
4807 | &trans->transaction->dev_update_list); | |
d6397bae | 4808 | |
d6397bae | 4809 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4810 | btrfs_set_super_total_bytes(super_copy, |
4811 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4812 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4813 | |
4814 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4815 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4816 | if (ret < 0) { |
4817 | btrfs_abort_transaction(trans, ret); | |
4818 | btrfs_end_transaction(trans); | |
4819 | } else { | |
4820 | ret = btrfs_commit_transaction(trans); | |
4821 | } | |
8f18cf13 CM |
4822 | done: |
4823 | btrfs_free_path(path); | |
53e489bc | 4824 | if (ret) { |
34441361 | 4825 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4826 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4827 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4828 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4829 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4830 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4831 | } |
8f18cf13 CM |
4832 | return ret; |
4833 | } | |
4834 | ||
2ff7e61e | 4835 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4836 | struct btrfs_key *key, |
4837 | struct btrfs_chunk *chunk, int item_size) | |
4838 | { | |
0b246afa | 4839 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4840 | struct btrfs_disk_key disk_key; |
4841 | u32 array_size; | |
4842 | u8 *ptr; | |
4843 | ||
34441361 | 4844 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4845 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4846 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4847 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4848 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4849 | return -EFBIG; |
fe48a5c0 | 4850 | } |
0b86a832 CM |
4851 | |
4852 | ptr = super_copy->sys_chunk_array + array_size; | |
4853 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4854 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4855 | ptr += sizeof(disk_key); | |
4856 | memcpy(ptr, chunk, item_size); | |
4857 | item_size += sizeof(disk_key); | |
4858 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4859 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4860 | |
0b86a832 CM |
4861 | return 0; |
4862 | } | |
4863 | ||
73c5de00 AJ |
4864 | /* |
4865 | * sort the devices in descending order by max_avail, total_avail | |
4866 | */ | |
4867 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4868 | { |
73c5de00 AJ |
4869 | const struct btrfs_device_info *di_a = a; |
4870 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4871 | |
73c5de00 | 4872 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4873 | return -1; |
73c5de00 | 4874 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4875 | return 1; |
73c5de00 AJ |
4876 | if (di_a->total_avail > di_b->total_avail) |
4877 | return -1; | |
4878 | if (di_a->total_avail < di_b->total_avail) | |
4879 | return 1; | |
4880 | return 0; | |
b2117a39 | 4881 | } |
0b86a832 | 4882 | |
53b381b3 DW |
4883 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4884 | { | |
ffe2d203 | 4885 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4886 | return; |
4887 | ||
ceda0864 | 4888 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4889 | } |
4890 | ||
73c5de00 | 4891 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4892 | u64 start, u64 type) |
b2117a39 | 4893 | { |
2ff7e61e | 4894 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4895 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4896 | struct btrfs_device *device; |
73c5de00 AJ |
4897 | struct map_lookup *map = NULL; |
4898 | struct extent_map_tree *em_tree; | |
4899 | struct extent_map *em; | |
4900 | struct btrfs_device_info *devices_info = NULL; | |
4901 | u64 total_avail; | |
4902 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4903 | int data_stripes; /* number of stripes that count for |
4904 | block group size */ | |
73c5de00 AJ |
4905 | int sub_stripes; /* sub_stripes info for map */ |
4906 | int dev_stripes; /* stripes per dev */ | |
4907 | int devs_max; /* max devs to use */ | |
4908 | int devs_min; /* min devs needed */ | |
4909 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4910 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4911 | int nparity; /* number of stripes worth of bytes to |
4912 | store parity information */ | |
73c5de00 AJ |
4913 | int ret; |
4914 | u64 max_stripe_size; | |
4915 | u64 max_chunk_size; | |
4916 | u64 stripe_size; | |
23f0ff1e | 4917 | u64 chunk_size; |
73c5de00 AJ |
4918 | int ndevs; |
4919 | int i; | |
4920 | int j; | |
31e50229 | 4921 | int index; |
593060d7 | 4922 | |
0c460c0d | 4923 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4924 | |
4117f207 QW |
4925 | if (list_empty(&fs_devices->alloc_list)) { |
4926 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4927 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4928 | return -ENOSPC; |
4117f207 | 4929 | } |
b2117a39 | 4930 | |
3e72ee88 | 4931 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4932 | |
31e50229 LB |
4933 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4934 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4935 | devs_max = btrfs_raid_array[index].devs_max; | |
e3ecdb3f DS |
4936 | if (!devs_max) |
4937 | devs_max = BTRFS_MAX_DEVS(info); | |
31e50229 LB |
4938 | devs_min = btrfs_raid_array[index].devs_min; |
4939 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4940 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4941 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4942 | |
9b3f68b9 | 4943 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4944 | max_stripe_size = SZ_1G; |
fce466ea | 4945 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
9b3f68b9 | 4946 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4947 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4948 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4949 | max_stripe_size = SZ_1G; | |
1100373f | 4950 | else |
ee22184b | 4951 | max_stripe_size = SZ_256M; |
73c5de00 | 4952 | max_chunk_size = max_stripe_size; |
a40a90a0 | 4953 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4954 | max_stripe_size = SZ_32M; |
73c5de00 AJ |
4955 | max_chunk_size = 2 * max_stripe_size; |
4956 | } else { | |
351fd353 | 4957 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 4958 | type); |
290342f6 | 4959 | BUG(); |
9b3f68b9 CM |
4960 | } |
4961 | ||
52042d8e | 4962 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
4963 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
4964 | max_chunk_size); | |
9b3f68b9 | 4965 | |
31e818fe | 4966 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4967 | GFP_NOFS); |
4968 | if (!devices_info) | |
4969 | return -ENOMEM; | |
0cad8a11 | 4970 | |
9f680ce0 | 4971 | /* |
73c5de00 AJ |
4972 | * in the first pass through the devices list, we gather information |
4973 | * about the available holes on each device. | |
9f680ce0 | 4974 | */ |
73c5de00 | 4975 | ndevs = 0; |
ebcc9301 | 4976 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4977 | u64 max_avail; |
4978 | u64 dev_offset; | |
b2117a39 | 4979 | |
ebbede42 | 4980 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4981 | WARN(1, KERN_ERR |
efe120a0 | 4982 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4983 | continue; |
4984 | } | |
b2117a39 | 4985 | |
e12c9621 AJ |
4986 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4987 | &device->dev_state) || | |
401e29c1 | 4988 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4989 | continue; |
b2117a39 | 4990 | |
73c5de00 AJ |
4991 | if (device->total_bytes > device->bytes_used) |
4992 | total_avail = device->total_bytes - device->bytes_used; | |
4993 | else | |
4994 | total_avail = 0; | |
38c01b96 | 4995 | |
4996 | /* If there is no space on this device, skip it. */ | |
4997 | if (total_avail == 0) | |
4998 | continue; | |
b2117a39 | 4999 | |
60dfdf25 | 5000 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
5001 | max_stripe_size * dev_stripes, |
5002 | &dev_offset, &max_avail); | |
5003 | if (ret && ret != -ENOSPC) | |
5004 | goto error; | |
b2117a39 | 5005 | |
73c5de00 AJ |
5006 | if (ret == 0) |
5007 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5008 | |
4117f207 QW |
5009 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5010 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5011 | btrfs_debug(info, | |
5012 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5013 | __func__, device->devid, max_avail, | |
5014 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5015 | continue; |
4117f207 | 5016 | } |
b2117a39 | 5017 | |
063d006f ES |
5018 | if (ndevs == fs_devices->rw_devices) { |
5019 | WARN(1, "%s: found more than %llu devices\n", | |
5020 | __func__, fs_devices->rw_devices); | |
5021 | break; | |
5022 | } | |
73c5de00 AJ |
5023 | devices_info[ndevs].dev_offset = dev_offset; |
5024 | devices_info[ndevs].max_avail = max_avail; | |
5025 | devices_info[ndevs].total_avail = total_avail; | |
5026 | devices_info[ndevs].dev = device; | |
5027 | ++ndevs; | |
5028 | } | |
b2117a39 | 5029 | |
73c5de00 AJ |
5030 | /* |
5031 | * now sort the devices by hole size / available space | |
5032 | */ | |
5033 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5034 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5035 | |
73c5de00 | 5036 | /* round down to number of usable stripes */ |
e5600fd6 | 5037 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5038 | |
ba89b802 | 5039 | if (ndevs < devs_min) { |
73c5de00 | 5040 | ret = -ENOSPC; |
4117f207 QW |
5041 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5042 | btrfs_debug(info, | |
5043 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5044 | __func__, ndevs, devs_min); |
4117f207 | 5045 | } |
73c5de00 | 5046 | goto error; |
b2117a39 | 5047 | } |
9f680ce0 | 5048 | |
f148ef4d NB |
5049 | ndevs = min(ndevs, devs_max); |
5050 | ||
73c5de00 | 5051 | /* |
92e222df HK |
5052 | * The primary goal is to maximize the number of stripes, so use as |
5053 | * many devices as possible, even if the stripes are not maximum sized. | |
5054 | * | |
5055 | * The DUP profile stores more than one stripe per device, the | |
5056 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5057 | */ |
92e222df | 5058 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5059 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5060 | |
53b381b3 DW |
5061 | /* |
5062 | * this will have to be fixed for RAID1 and RAID10 over | |
5063 | * more drives | |
5064 | */ | |
b50836ed | 5065 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5066 | |
5067 | /* | |
5068 | * Use the number of data stripes to figure out how big this chunk | |
5069 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5070 | * and compare that answer with the max chunk size. If it's higher, |
5071 | * we try to reduce stripe_size. | |
86db2578 CM |
5072 | */ |
5073 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5074 | /* |
baf92114 HK |
5075 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5076 | * then use it, unless it ends up being even bigger than the | |
5077 | * previous value we had already. | |
86db2578 | 5078 | */ |
baf92114 HK |
5079 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5080 | data_stripes), SZ_16M), | |
793ff2c8 | 5081 | stripe_size); |
86db2578 CM |
5082 | } |
5083 | ||
37db63a4 | 5084 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5085 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5086 | |
5087 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5088 | if (!map) { | |
5089 | ret = -ENOMEM; | |
5090 | goto error; | |
5091 | } | |
5092 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5093 | |
73c5de00 AJ |
5094 | for (i = 0; i < ndevs; ++i) { |
5095 | for (j = 0; j < dev_stripes; ++j) { | |
5096 | int s = i * dev_stripes + j; | |
5097 | map->stripes[s].dev = devices_info[i].dev; | |
5098 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5099 | j * stripe_size; | |
6324fbf3 | 5100 | } |
6324fbf3 | 5101 | } |
500ceed8 NB |
5102 | map->stripe_len = BTRFS_STRIPE_LEN; |
5103 | map->io_align = BTRFS_STRIPE_LEN; | |
5104 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5105 | map->type = type; |
2b82032c | 5106 | map->sub_stripes = sub_stripes; |
0b86a832 | 5107 | |
23f0ff1e | 5108 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5109 | |
23f0ff1e | 5110 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5111 | |
172ddd60 | 5112 | em = alloc_extent_map(); |
2b82032c | 5113 | if (!em) { |
298a8f9c | 5114 | kfree(map); |
b2117a39 MX |
5115 | ret = -ENOMEM; |
5116 | goto error; | |
593060d7 | 5117 | } |
298a8f9c | 5118 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5119 | em->map_lookup = map; |
2b82032c | 5120 | em->start = start; |
23f0ff1e | 5121 | em->len = chunk_size; |
2b82032c YZ |
5122 | em->block_start = 0; |
5123 | em->block_len = em->len; | |
6df9a95e | 5124 | em->orig_block_len = stripe_size; |
593060d7 | 5125 | |
c8bf1b67 | 5126 | em_tree = &info->mapping_tree; |
890871be | 5127 | write_lock(&em_tree->lock); |
09a2a8f9 | 5128 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5129 | if (ret) { |
1efb72a3 | 5130 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5131 | free_extent_map(em); |
1dd4602f | 5132 | goto error; |
0f5d42b2 | 5133 | } |
1efb72a3 NB |
5134 | write_unlock(&em_tree->lock); |
5135 | ||
23f0ff1e | 5136 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5137 | if (ret) |
5138 | goto error_del_extent; | |
2b82032c | 5139 | |
bbbf7243 NB |
5140 | for (i = 0; i < map->num_stripes; i++) { |
5141 | struct btrfs_device *dev = map->stripes[i].dev; | |
5142 | ||
5143 | btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size); | |
5144 | if (list_empty(&dev->post_commit_list)) | |
5145 | list_add_tail(&dev->post_commit_list, | |
5146 | &trans->transaction->dev_update_list); | |
5147 | } | |
43530c46 | 5148 | |
a5ed45f8 | 5149 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5150 | |
0f5d42b2 | 5151 | free_extent_map(em); |
0b246afa | 5152 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5153 | |
b2117a39 | 5154 | kfree(devices_info); |
2b82032c | 5155 | return 0; |
b2117a39 | 5156 | |
6df9a95e | 5157 | error_del_extent: |
0f5d42b2 JB |
5158 | write_lock(&em_tree->lock); |
5159 | remove_extent_mapping(em_tree, em); | |
5160 | write_unlock(&em_tree->lock); | |
5161 | ||
5162 | /* One for our allocation */ | |
5163 | free_extent_map(em); | |
5164 | /* One for the tree reference */ | |
5165 | free_extent_map(em); | |
b2117a39 | 5166 | error: |
b2117a39 MX |
5167 | kfree(devices_info); |
5168 | return ret; | |
2b82032c YZ |
5169 | } |
5170 | ||
6df9a95e | 5171 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5172 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5173 | { |
97aff912 | 5174 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5175 | struct btrfs_root *extent_root = fs_info->extent_root; |
5176 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5177 | struct btrfs_key key; |
2b82032c YZ |
5178 | struct btrfs_device *device; |
5179 | struct btrfs_chunk *chunk; | |
5180 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5181 | struct extent_map *em; |
5182 | struct map_lookup *map; | |
5183 | size_t item_size; | |
5184 | u64 dev_offset; | |
5185 | u64 stripe_size; | |
5186 | int i = 0; | |
140e639f | 5187 | int ret = 0; |
2b82032c | 5188 | |
60ca842e | 5189 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5190 | if (IS_ERR(em)) |
5191 | return PTR_ERR(em); | |
6df9a95e | 5192 | |
95617d69 | 5193 | map = em->map_lookup; |
6df9a95e JB |
5194 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5195 | stripe_size = em->orig_block_len; | |
5196 | ||
2b82032c | 5197 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5198 | if (!chunk) { |
5199 | ret = -ENOMEM; | |
5200 | goto out; | |
5201 | } | |
5202 | ||
50460e37 FM |
5203 | /* |
5204 | * Take the device list mutex to prevent races with the final phase of | |
5205 | * a device replace operation that replaces the device object associated | |
5206 | * with the map's stripes, because the device object's id can change | |
5207 | * at any time during that final phase of the device replace operation | |
5208 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5209 | */ | |
0b246afa | 5210 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5211 | for (i = 0; i < map->num_stripes; i++) { |
5212 | device = map->stripes[i].dev; | |
5213 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5214 | |
0b86a832 | 5215 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5216 | if (ret) |
50460e37 | 5217 | break; |
b5d9071c NB |
5218 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5219 | dev_offset, stripe_size); | |
6df9a95e | 5220 | if (ret) |
50460e37 FM |
5221 | break; |
5222 | } | |
5223 | if (ret) { | |
0b246afa | 5224 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5225 | goto out; |
2b82032c YZ |
5226 | } |
5227 | ||
2b82032c | 5228 | stripe = &chunk->stripe; |
6df9a95e JB |
5229 | for (i = 0; i < map->num_stripes; i++) { |
5230 | device = map->stripes[i].dev; | |
5231 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5232 | |
e17cade2 CM |
5233 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5234 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5235 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5236 | stripe++; |
0b86a832 | 5237 | } |
0b246afa | 5238 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5239 | |
2b82032c | 5240 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5241 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5242 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5243 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5244 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5245 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5246 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5247 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5248 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5249 | |
2b82032c YZ |
5250 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5251 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5252 | key.offset = chunk_offset; | |
0b86a832 | 5253 | |
2b82032c | 5254 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5255 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5256 | /* | |
5257 | * TODO: Cleanup of inserted chunk root in case of | |
5258 | * failure. | |
5259 | */ | |
2ff7e61e | 5260 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5261 | } |
1abe9b8a | 5262 | |
6df9a95e | 5263 | out: |
0b86a832 | 5264 | kfree(chunk); |
6df9a95e | 5265 | free_extent_map(em); |
4ed1d16e | 5266 | return ret; |
2b82032c | 5267 | } |
0b86a832 | 5268 | |
2b82032c | 5269 | /* |
52042d8e AG |
5270 | * Chunk allocation falls into two parts. The first part does work |
5271 | * that makes the new allocated chunk usable, but does not do any operation | |
5272 | * that modifies the chunk tree. The second part does the work that | |
5273 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5274 | * bootstrap process of adding storage to a seed btrfs. |
5275 | */ | |
c216b203 | 5276 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5277 | { |
5278 | u64 chunk_offset; | |
2b82032c | 5279 | |
c216b203 NB |
5280 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5281 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5282 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5283 | } |
5284 | ||
6f8e0fc7 | 5285 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5286 | { |
6f8e0fc7 | 5287 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c YZ |
5288 | u64 chunk_offset; |
5289 | u64 sys_chunk_offset; | |
2b82032c | 5290 | u64 alloc_profile; |
2b82032c YZ |
5291 | int ret; |
5292 | ||
6df9a95e | 5293 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5294 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5295 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5296 | if (ret) |
5297 | return ret; | |
2b82032c | 5298 | |
0b246afa | 5299 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5300 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5301 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5302 | return ret; |
2b82032c YZ |
5303 | } |
5304 | ||
d20983b4 MX |
5305 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5306 | { | |
fc9a2ac7 | 5307 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5308 | |
fc9a2ac7 | 5309 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5310 | } |
5311 | ||
2ff7e61e | 5312 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5313 | { |
5314 | struct extent_map *em; | |
5315 | struct map_lookup *map; | |
2b82032c | 5316 | int readonly = 0; |
d20983b4 | 5317 | int miss_ndevs = 0; |
2b82032c YZ |
5318 | int i; |
5319 | ||
60ca842e | 5320 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5321 | if (IS_ERR(em)) |
2b82032c YZ |
5322 | return 1; |
5323 | ||
95617d69 | 5324 | map = em->map_lookup; |
2b82032c | 5325 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5326 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5327 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5328 | miss_ndevs++; |
5329 | continue; | |
5330 | } | |
ebbede42 AJ |
5331 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5332 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5333 | readonly = 1; |
d20983b4 | 5334 | goto end; |
2b82032c YZ |
5335 | } |
5336 | } | |
d20983b4 MX |
5337 | |
5338 | /* | |
5339 | * If the number of missing devices is larger than max errors, | |
5340 | * we can not write the data into that chunk successfully, so | |
5341 | * set it readonly. | |
5342 | */ | |
5343 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5344 | readonly = 1; | |
5345 | end: | |
0b86a832 | 5346 | free_extent_map(em); |
2b82032c | 5347 | return readonly; |
0b86a832 CM |
5348 | } |
5349 | ||
c8bf1b67 | 5350 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5351 | { |
5352 | struct extent_map *em; | |
5353 | ||
d397712b | 5354 | while (1) { |
c8bf1b67 DS |
5355 | write_lock(&tree->lock); |
5356 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5357 | if (em) |
c8bf1b67 DS |
5358 | remove_extent_mapping(tree, em); |
5359 | write_unlock(&tree->lock); | |
0b86a832 CM |
5360 | if (!em) |
5361 | break; | |
0b86a832 CM |
5362 | /* once for us */ |
5363 | free_extent_map(em); | |
5364 | /* once for the tree */ | |
5365 | free_extent_map(em); | |
5366 | } | |
5367 | } | |
5368 | ||
5d964051 | 5369 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5370 | { |
5371 | struct extent_map *em; | |
5372 | struct map_lookup *map; | |
f188591e CM |
5373 | int ret; |
5374 | ||
60ca842e | 5375 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5376 | if (IS_ERR(em)) |
5377 | /* | |
5378 | * We could return errors for these cases, but that could get | |
5379 | * ugly and we'd probably do the same thing which is just not do | |
5380 | * anything else and exit, so return 1 so the callers don't try | |
5381 | * to use other copies. | |
5382 | */ | |
fb7669b5 | 5383 | return 1; |
fb7669b5 | 5384 | |
95617d69 | 5385 | map = em->map_lookup; |
c7369b3f | 5386 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5387 | ret = map->num_stripes; |
321aecc6 CM |
5388 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5389 | ret = map->sub_stripes; | |
53b381b3 DW |
5390 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5391 | ret = 2; | |
5392 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5393 | /* |
5394 | * There could be two corrupted data stripes, we need | |
5395 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5396 | * |
8810f751 LB |
5397 | * Fail a stripe at a time on every retry except the |
5398 | * stripe under reconstruction. | |
5399 | */ | |
5400 | ret = map->num_stripes; | |
f188591e CM |
5401 | else |
5402 | ret = 1; | |
5403 | free_extent_map(em); | |
ad6d620e | 5404 | |
cb5583dd | 5405 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5406 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5407 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5408 | ret++; |
cb5583dd | 5409 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5410 | |
f188591e CM |
5411 | return ret; |
5412 | } | |
5413 | ||
2ff7e61e | 5414 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5415 | u64 logical) |
5416 | { | |
5417 | struct extent_map *em; | |
5418 | struct map_lookup *map; | |
0b246afa | 5419 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5420 | |
60ca842e | 5421 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5422 | |
69f03f13 NB |
5423 | if (!WARN_ON(IS_ERR(em))) { |
5424 | map = em->map_lookup; | |
5425 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5426 | len = map->stripe_len * nr_data_stripes(map); | |
5427 | free_extent_map(em); | |
5428 | } | |
53b381b3 DW |
5429 | return len; |
5430 | } | |
5431 | ||
e4ff5fb5 | 5432 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5433 | { |
5434 | struct extent_map *em; | |
5435 | struct map_lookup *map; | |
53b381b3 DW |
5436 | int ret = 0; |
5437 | ||
60ca842e | 5438 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5439 | |
69f03f13 NB |
5440 | if(!WARN_ON(IS_ERR(em))) { |
5441 | map = em->map_lookup; | |
5442 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5443 | ret = 1; | |
5444 | free_extent_map(em); | |
5445 | } | |
53b381b3 DW |
5446 | return ret; |
5447 | } | |
5448 | ||
30d9861f | 5449 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5450 | struct map_lookup *map, int first, |
8ba0ae78 | 5451 | int dev_replace_is_ongoing) |
dfe25020 CM |
5452 | { |
5453 | int i; | |
99f92a7c | 5454 | int num_stripes; |
8ba0ae78 | 5455 | int preferred_mirror; |
30d9861f SB |
5456 | int tolerance; |
5457 | struct btrfs_device *srcdev; | |
5458 | ||
99f92a7c | 5459 | ASSERT((map->type & |
c7369b3f | 5460 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5461 | |
5462 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5463 | num_stripes = map->sub_stripes; | |
5464 | else | |
5465 | num_stripes = map->num_stripes; | |
5466 | ||
8ba0ae78 AJ |
5467 | preferred_mirror = first + current->pid % num_stripes; |
5468 | ||
30d9861f SB |
5469 | if (dev_replace_is_ongoing && |
5470 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5471 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5472 | srcdev = fs_info->dev_replace.srcdev; | |
5473 | else | |
5474 | srcdev = NULL; | |
5475 | ||
5476 | /* | |
5477 | * try to avoid the drive that is the source drive for a | |
5478 | * dev-replace procedure, only choose it if no other non-missing | |
5479 | * mirror is available | |
5480 | */ | |
5481 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5482 | if (map->stripes[preferred_mirror].dev->bdev && |
5483 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5484 | return preferred_mirror; | |
99f92a7c | 5485 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5486 | if (map->stripes[i].dev->bdev && |
5487 | (tolerance || map->stripes[i].dev != srcdev)) | |
5488 | return i; | |
5489 | } | |
dfe25020 | 5490 | } |
30d9861f | 5491 | |
dfe25020 CM |
5492 | /* we couldn't find one that doesn't fail. Just return something |
5493 | * and the io error handling code will clean up eventually | |
5494 | */ | |
8ba0ae78 | 5495 | return preferred_mirror; |
dfe25020 CM |
5496 | } |
5497 | ||
53b381b3 DW |
5498 | static inline int parity_smaller(u64 a, u64 b) |
5499 | { | |
5500 | return a > b; | |
5501 | } | |
5502 | ||
5503 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5504 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5505 | { |
5506 | struct btrfs_bio_stripe s; | |
5507 | int i; | |
5508 | u64 l; | |
5509 | int again = 1; | |
5510 | ||
5511 | while (again) { | |
5512 | again = 0; | |
cc7539ed | 5513 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5514 | if (parity_smaller(bbio->raid_map[i], |
5515 | bbio->raid_map[i+1])) { | |
53b381b3 | 5516 | s = bbio->stripes[i]; |
8e5cfb55 | 5517 | l = bbio->raid_map[i]; |
53b381b3 | 5518 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5519 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5520 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5521 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5522 | |
53b381b3 DW |
5523 | again = 1; |
5524 | } | |
5525 | } | |
5526 | } | |
5527 | } | |
5528 | ||
6e9606d2 ZL |
5529 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5530 | { | |
5531 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5532 | /* the size of the btrfs_bio */ |
6e9606d2 | 5533 | sizeof(struct btrfs_bio) + |
e57cf21e | 5534 | /* plus the variable array for the stripes */ |
6e9606d2 | 5535 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5536 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5537 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5538 | /* |
5539 | * plus the raid_map, which includes both the tgt dev | |
5540 | * and the stripes | |
5541 | */ | |
5542 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5543 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5544 | |
5545 | atomic_set(&bbio->error, 0); | |
140475ae | 5546 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5547 | |
5548 | return bbio; | |
5549 | } | |
5550 | ||
5551 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5552 | { | |
140475ae ER |
5553 | WARN_ON(!refcount_read(&bbio->refs)); |
5554 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5555 | } |
5556 | ||
5557 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5558 | { | |
5559 | if (!bbio) | |
5560 | return; | |
140475ae | 5561 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5562 | kfree(bbio); |
5563 | } | |
5564 | ||
0b3d4cd3 LB |
5565 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5566 | /* | |
5567 | * Please note that, discard won't be sent to target device of device | |
5568 | * replace. | |
5569 | */ | |
5570 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5571 | u64 logical, u64 length, | |
5572 | struct btrfs_bio **bbio_ret) | |
5573 | { | |
5574 | struct extent_map *em; | |
5575 | struct map_lookup *map; | |
5576 | struct btrfs_bio *bbio; | |
5577 | u64 offset; | |
5578 | u64 stripe_nr; | |
5579 | u64 stripe_nr_end; | |
5580 | u64 stripe_end_offset; | |
5581 | u64 stripe_cnt; | |
5582 | u64 stripe_len; | |
5583 | u64 stripe_offset; | |
5584 | u64 num_stripes; | |
5585 | u32 stripe_index; | |
5586 | u32 factor = 0; | |
5587 | u32 sub_stripes = 0; | |
5588 | u64 stripes_per_dev = 0; | |
5589 | u32 remaining_stripes = 0; | |
5590 | u32 last_stripe = 0; | |
5591 | int ret = 0; | |
5592 | int i; | |
5593 | ||
5594 | /* discard always return a bbio */ | |
5595 | ASSERT(bbio_ret); | |
5596 | ||
60ca842e | 5597 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5598 | if (IS_ERR(em)) |
5599 | return PTR_ERR(em); | |
5600 | ||
5601 | map = em->map_lookup; | |
5602 | /* we don't discard raid56 yet */ | |
5603 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5604 | ret = -EOPNOTSUPP; | |
5605 | goto out; | |
5606 | } | |
5607 | ||
5608 | offset = logical - em->start; | |
5609 | length = min_t(u64, em->len - offset, length); | |
5610 | ||
5611 | stripe_len = map->stripe_len; | |
5612 | /* | |
5613 | * stripe_nr counts the total number of stripes we have to stride | |
5614 | * to get to this block | |
5615 | */ | |
5616 | stripe_nr = div64_u64(offset, stripe_len); | |
5617 | ||
5618 | /* stripe_offset is the offset of this block in its stripe */ | |
5619 | stripe_offset = offset - stripe_nr * stripe_len; | |
5620 | ||
5621 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5622 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5623 | stripe_cnt = stripe_nr_end - stripe_nr; |
5624 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5625 | (offset + length); | |
5626 | /* | |
5627 | * after this, stripe_nr is the number of stripes on this | |
5628 | * device we have to walk to find the data, and stripe_index is | |
5629 | * the number of our device in the stripe array | |
5630 | */ | |
5631 | num_stripes = 1; | |
5632 | stripe_index = 0; | |
5633 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5634 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5635 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5636 | sub_stripes = 1; | |
5637 | else | |
5638 | sub_stripes = map->sub_stripes; | |
5639 | ||
5640 | factor = map->num_stripes / sub_stripes; | |
5641 | num_stripes = min_t(u64, map->num_stripes, | |
5642 | sub_stripes * stripe_cnt); | |
5643 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5644 | stripe_index *= sub_stripes; | |
5645 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5646 | &remaining_stripes); | |
5647 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5648 | last_stripe *= sub_stripes; | |
c7369b3f | 5649 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5650 | BTRFS_BLOCK_GROUP_DUP)) { |
5651 | num_stripes = map->num_stripes; | |
5652 | } else { | |
5653 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5654 | &stripe_index); | |
5655 | } | |
5656 | ||
5657 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5658 | if (!bbio) { | |
5659 | ret = -ENOMEM; | |
5660 | goto out; | |
5661 | } | |
5662 | ||
5663 | for (i = 0; i < num_stripes; i++) { | |
5664 | bbio->stripes[i].physical = | |
5665 | map->stripes[stripe_index].physical + | |
5666 | stripe_offset + stripe_nr * map->stripe_len; | |
5667 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5668 | ||
5669 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5670 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5671 | bbio->stripes[i].length = stripes_per_dev * | |
5672 | map->stripe_len; | |
5673 | ||
5674 | if (i / sub_stripes < remaining_stripes) | |
5675 | bbio->stripes[i].length += | |
5676 | map->stripe_len; | |
5677 | ||
5678 | /* | |
5679 | * Special for the first stripe and | |
5680 | * the last stripe: | |
5681 | * | |
5682 | * |-------|...|-------| | |
5683 | * |----------| | |
5684 | * off end_off | |
5685 | */ | |
5686 | if (i < sub_stripes) | |
5687 | bbio->stripes[i].length -= | |
5688 | stripe_offset; | |
5689 | ||
5690 | if (stripe_index >= last_stripe && | |
5691 | stripe_index <= (last_stripe + | |
5692 | sub_stripes - 1)) | |
5693 | bbio->stripes[i].length -= | |
5694 | stripe_end_offset; | |
5695 | ||
5696 | if (i == sub_stripes - 1) | |
5697 | stripe_offset = 0; | |
5698 | } else { | |
5699 | bbio->stripes[i].length = length; | |
5700 | } | |
5701 | ||
5702 | stripe_index++; | |
5703 | if (stripe_index == map->num_stripes) { | |
5704 | stripe_index = 0; | |
5705 | stripe_nr++; | |
5706 | } | |
5707 | } | |
5708 | ||
5709 | *bbio_ret = bbio; | |
5710 | bbio->map_type = map->type; | |
5711 | bbio->num_stripes = num_stripes; | |
5712 | out: | |
5713 | free_extent_map(em); | |
5714 | return ret; | |
5715 | } | |
5716 | ||
5ab56090 LB |
5717 | /* |
5718 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5719 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5720 | * left cursor can also be read from the target drive. | |
5721 | * | |
5722 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5723 | * array of stripes. | |
5724 | * For READ, it also needs to be supported using the same mirror number. | |
5725 | * | |
5726 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5727 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5728 | * case. | |
5729 | */ | |
5730 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5731 | u64 logical, u64 length, | |
5732 | u64 srcdev_devid, int *mirror_num, | |
5733 | u64 *physical) | |
5734 | { | |
5735 | struct btrfs_bio *bbio = NULL; | |
5736 | int num_stripes; | |
5737 | int index_srcdev = 0; | |
5738 | int found = 0; | |
5739 | u64 physical_of_found = 0; | |
5740 | int i; | |
5741 | int ret = 0; | |
5742 | ||
5743 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5744 | logical, &length, &bbio, 0, 0); | |
5745 | if (ret) { | |
5746 | ASSERT(bbio == NULL); | |
5747 | return ret; | |
5748 | } | |
5749 | ||
5750 | num_stripes = bbio->num_stripes; | |
5751 | if (*mirror_num > num_stripes) { | |
5752 | /* | |
5753 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5754 | * that means that the requested area is not left of the left | |
5755 | * cursor | |
5756 | */ | |
5757 | btrfs_put_bbio(bbio); | |
5758 | return -EIO; | |
5759 | } | |
5760 | ||
5761 | /* | |
5762 | * process the rest of the function using the mirror_num of the source | |
5763 | * drive. Therefore look it up first. At the end, patch the device | |
5764 | * pointer to the one of the target drive. | |
5765 | */ | |
5766 | for (i = 0; i < num_stripes; i++) { | |
5767 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5768 | continue; | |
5769 | ||
5770 | /* | |
5771 | * In case of DUP, in order to keep it simple, only add the | |
5772 | * mirror with the lowest physical address | |
5773 | */ | |
5774 | if (found && | |
5775 | physical_of_found <= bbio->stripes[i].physical) | |
5776 | continue; | |
5777 | ||
5778 | index_srcdev = i; | |
5779 | found = 1; | |
5780 | physical_of_found = bbio->stripes[i].physical; | |
5781 | } | |
5782 | ||
5783 | btrfs_put_bbio(bbio); | |
5784 | ||
5785 | ASSERT(found); | |
5786 | if (!found) | |
5787 | return -EIO; | |
5788 | ||
5789 | *mirror_num = index_srcdev + 1; | |
5790 | *physical = physical_of_found; | |
5791 | return ret; | |
5792 | } | |
5793 | ||
73c0f228 LB |
5794 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5795 | struct btrfs_bio **bbio_ret, | |
5796 | struct btrfs_dev_replace *dev_replace, | |
5797 | int *num_stripes_ret, int *max_errors_ret) | |
5798 | { | |
5799 | struct btrfs_bio *bbio = *bbio_ret; | |
5800 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5801 | int tgtdev_indexes = 0; | |
5802 | int num_stripes = *num_stripes_ret; | |
5803 | int max_errors = *max_errors_ret; | |
5804 | int i; | |
5805 | ||
5806 | if (op == BTRFS_MAP_WRITE) { | |
5807 | int index_where_to_add; | |
5808 | ||
5809 | /* | |
5810 | * duplicate the write operations while the dev replace | |
5811 | * procedure is running. Since the copying of the old disk to | |
5812 | * the new disk takes place at run time while the filesystem is | |
5813 | * mounted writable, the regular write operations to the old | |
5814 | * disk have to be duplicated to go to the new disk as well. | |
5815 | * | |
5816 | * Note that device->missing is handled by the caller, and that | |
5817 | * the write to the old disk is already set up in the stripes | |
5818 | * array. | |
5819 | */ | |
5820 | index_where_to_add = num_stripes; | |
5821 | for (i = 0; i < num_stripes; i++) { | |
5822 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5823 | /* write to new disk, too */ | |
5824 | struct btrfs_bio_stripe *new = | |
5825 | bbio->stripes + index_where_to_add; | |
5826 | struct btrfs_bio_stripe *old = | |
5827 | bbio->stripes + i; | |
5828 | ||
5829 | new->physical = old->physical; | |
5830 | new->length = old->length; | |
5831 | new->dev = dev_replace->tgtdev; | |
5832 | bbio->tgtdev_map[i] = index_where_to_add; | |
5833 | index_where_to_add++; | |
5834 | max_errors++; | |
5835 | tgtdev_indexes++; | |
5836 | } | |
5837 | } | |
5838 | num_stripes = index_where_to_add; | |
5839 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5840 | int index_srcdev = 0; | |
5841 | int found = 0; | |
5842 | u64 physical_of_found = 0; | |
5843 | ||
5844 | /* | |
5845 | * During the dev-replace procedure, the target drive can also | |
5846 | * be used to read data in case it is needed to repair a corrupt | |
5847 | * block elsewhere. This is possible if the requested area is | |
5848 | * left of the left cursor. In this area, the target drive is a | |
5849 | * full copy of the source drive. | |
5850 | */ | |
5851 | for (i = 0; i < num_stripes; i++) { | |
5852 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5853 | /* | |
5854 | * In case of DUP, in order to keep it simple, | |
5855 | * only add the mirror with the lowest physical | |
5856 | * address | |
5857 | */ | |
5858 | if (found && | |
5859 | physical_of_found <= | |
5860 | bbio->stripes[i].physical) | |
5861 | continue; | |
5862 | index_srcdev = i; | |
5863 | found = 1; | |
5864 | physical_of_found = bbio->stripes[i].physical; | |
5865 | } | |
5866 | } | |
5867 | if (found) { | |
5868 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5869 | bbio->stripes + num_stripes; | |
5870 | ||
5871 | tgtdev_stripe->physical = physical_of_found; | |
5872 | tgtdev_stripe->length = | |
5873 | bbio->stripes[index_srcdev].length; | |
5874 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5875 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5876 | ||
5877 | tgtdev_indexes++; | |
5878 | num_stripes++; | |
5879 | } | |
5880 | } | |
5881 | ||
5882 | *num_stripes_ret = num_stripes; | |
5883 | *max_errors_ret = max_errors; | |
5884 | bbio->num_tgtdevs = tgtdev_indexes; | |
5885 | *bbio_ret = bbio; | |
5886 | } | |
5887 | ||
2b19a1fe LB |
5888 | static bool need_full_stripe(enum btrfs_map_op op) |
5889 | { | |
5890 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5891 | } | |
5892 | ||
5f141126 NB |
5893 | /* |
5894 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5895 | * tuple. This information is used to calculate how big a | |
5896 | * particular bio can get before it straddles a stripe. | |
5897 | * | |
5898 | * @fs_info - the filesystem | |
5899 | * @logical - address that we want to figure out the geometry of | |
5900 | * @len - the length of IO we are going to perform, starting at @logical | |
5901 | * @op - type of operation - write or read | |
5902 | * @io_geom - pointer used to return values | |
5903 | * | |
5904 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5905 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5906 | */ | |
5907 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5908 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5909 | { |
5910 | struct extent_map *em; | |
5911 | struct map_lookup *map; | |
5912 | u64 offset; | |
5913 | u64 stripe_offset; | |
5914 | u64 stripe_nr; | |
5915 | u64 stripe_len; | |
5916 | u64 raid56_full_stripe_start = (u64)-1; | |
5917 | int data_stripes; | |
373c3b80 | 5918 | int ret = 0; |
5f141126 NB |
5919 | |
5920 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5921 | ||
5922 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5923 | if (IS_ERR(em)) | |
5924 | return PTR_ERR(em); | |
5925 | ||
5926 | map = em->map_lookup; | |
5927 | /* Offset of this logical address in the chunk */ | |
5928 | offset = logical - em->start; | |
5929 | /* Len of a stripe in a chunk */ | |
5930 | stripe_len = map->stripe_len; | |
5931 | /* Stripe wher this block falls in */ | |
5932 | stripe_nr = div64_u64(offset, stripe_len); | |
5933 | /* Offset of stripe in the chunk */ | |
5934 | stripe_offset = stripe_nr * stripe_len; | |
5935 | if (offset < stripe_offset) { | |
5936 | btrfs_crit(fs_info, | |
5937 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5938 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5939 | ret = -EINVAL; |
5940 | goto out; | |
5f141126 NB |
5941 | } |
5942 | ||
5943 | /* stripe_offset is the offset of this block in its stripe */ | |
5944 | stripe_offset = offset - stripe_offset; | |
5945 | data_stripes = nr_data_stripes(map); | |
5946 | ||
5947 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5948 | u64 max_len = stripe_len - stripe_offset; | |
5949 | ||
5950 | /* | |
5951 | * In case of raid56, we need to know the stripe aligned start | |
5952 | */ | |
5953 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5954 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5955 | raid56_full_stripe_start = offset; | |
5956 | ||
5957 | /* | |
5958 | * Allow a write of a full stripe, but make sure we | |
5959 | * don't allow straddling of stripes | |
5960 | */ | |
5961 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5962 | full_stripe_len); | |
5963 | raid56_full_stripe_start *= full_stripe_len; | |
5964 | ||
5965 | /* | |
5966 | * For writes to RAID[56], allow a full stripeset across | |
5967 | * all disks. For other RAID types and for RAID[56] | |
5968 | * reads, just allow a single stripe (on a single disk). | |
5969 | */ | |
5970 | if (op == BTRFS_MAP_WRITE) { | |
5971 | max_len = stripe_len * data_stripes - | |
5972 | (offset - raid56_full_stripe_start); | |
5973 | } | |
5974 | } | |
5975 | len = min_t(u64, em->len - offset, max_len); | |
5976 | } else { | |
5977 | len = em->len - offset; | |
5978 | } | |
5979 | ||
5980 | io_geom->len = len; | |
5981 | io_geom->offset = offset; | |
5982 | io_geom->stripe_len = stripe_len; | |
5983 | io_geom->stripe_nr = stripe_nr; | |
5984 | io_geom->stripe_offset = stripe_offset; | |
5985 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
5986 | ||
373c3b80 JT |
5987 | out: |
5988 | /* once for us */ | |
5989 | free_extent_map(em); | |
5990 | return ret; | |
5f141126 NB |
5991 | } |
5992 | ||
cf8cddd3 CH |
5993 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5994 | enum btrfs_map_op op, | |
f2d8d74d | 5995 | u64 logical, u64 *length, |
a1d3c478 | 5996 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5997 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5998 | { |
5999 | struct extent_map *em; | |
6000 | struct map_lookup *map; | |
593060d7 CM |
6001 | u64 stripe_offset; |
6002 | u64 stripe_nr; | |
53b381b3 | 6003 | u64 stripe_len; |
9d644a62 | 6004 | u32 stripe_index; |
cff82672 | 6005 | int data_stripes; |
cea9e445 | 6006 | int i; |
de11cc12 | 6007 | int ret = 0; |
f2d8d74d | 6008 | int num_stripes; |
a236aed1 | 6009 | int max_errors = 0; |
2c8cdd6e | 6010 | int tgtdev_indexes = 0; |
a1d3c478 | 6011 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6012 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6013 | int dev_replace_is_ongoing = 0; | |
6014 | int num_alloc_stripes; | |
ad6d620e SB |
6015 | int patch_the_first_stripe_for_dev_replace = 0; |
6016 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6017 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6018 | struct btrfs_io_geometry geom; |
6019 | ||
6020 | ASSERT(bbio_ret); | |
0b86a832 | 6021 | |
0b3d4cd3 LB |
6022 | if (op == BTRFS_MAP_DISCARD) |
6023 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6024 | *length, bbio_ret); | |
6025 | ||
89b798ad NB |
6026 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
6027 | if (ret < 0) | |
6028 | return ret; | |
0b86a832 | 6029 | |
89b798ad | 6030 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 6031 | ASSERT(!IS_ERR(em)); |
95617d69 | 6032 | map = em->map_lookup; |
593060d7 | 6033 | |
89b798ad | 6034 | *length = geom.len; |
89b798ad NB |
6035 | stripe_len = geom.stripe_len; |
6036 | stripe_nr = geom.stripe_nr; | |
6037 | stripe_offset = geom.stripe_offset; | |
6038 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6039 | data_stripes = nr_data_stripes(map); |
593060d7 | 6040 | |
cb5583dd | 6041 | down_read(&dev_replace->rwsem); |
472262f3 | 6042 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6043 | /* |
6044 | * Hold the semaphore for read during the whole operation, write is | |
6045 | * requested at commit time but must wait. | |
6046 | */ | |
472262f3 | 6047 | if (!dev_replace_is_ongoing) |
cb5583dd | 6048 | up_read(&dev_replace->rwsem); |
472262f3 | 6049 | |
ad6d620e | 6050 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6051 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6052 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6053 | dev_replace->srcdev->devid, | |
6054 | &mirror_num, | |
6055 | &physical_to_patch_in_first_stripe); | |
6056 | if (ret) | |
ad6d620e | 6057 | goto out; |
5ab56090 LB |
6058 | else |
6059 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6060 | } else if (mirror_num > map->num_stripes) { |
6061 | mirror_num = 0; | |
6062 | } | |
6063 | ||
f2d8d74d | 6064 | num_stripes = 1; |
cea9e445 | 6065 | stripe_index = 0; |
fce3bb9a | 6066 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6067 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6068 | &stripe_index); | |
de483734 | 6069 | if (!need_full_stripe(op)) |
28e1cc7d | 6070 | mirror_num = 1; |
c7369b3f | 6071 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6072 | if (need_full_stripe(op)) |
f2d8d74d | 6073 | num_stripes = map->num_stripes; |
2fff734f | 6074 | else if (mirror_num) |
f188591e | 6075 | stripe_index = mirror_num - 1; |
dfe25020 | 6076 | else { |
30d9861f | 6077 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6078 | dev_replace_is_ongoing); |
a1d3c478 | 6079 | mirror_num = stripe_index + 1; |
dfe25020 | 6080 | } |
2fff734f | 6081 | |
611f0e00 | 6082 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6083 | if (need_full_stripe(op)) { |
f2d8d74d | 6084 | num_stripes = map->num_stripes; |
a1d3c478 | 6085 | } else if (mirror_num) { |
f188591e | 6086 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6087 | } else { |
6088 | mirror_num = 1; | |
6089 | } | |
2fff734f | 6090 | |
321aecc6 | 6091 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6092 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6093 | |
47c5713f | 6094 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6095 | stripe_index *= map->sub_stripes; |
6096 | ||
de483734 | 6097 | if (need_full_stripe(op)) |
f2d8d74d | 6098 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6099 | else if (mirror_num) |
6100 | stripe_index += mirror_num - 1; | |
dfe25020 | 6101 | else { |
3e74317a | 6102 | int old_stripe_index = stripe_index; |
30d9861f SB |
6103 | stripe_index = find_live_mirror(fs_info, map, |
6104 | stripe_index, | |
30d9861f | 6105 | dev_replace_is_ongoing); |
3e74317a | 6106 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6107 | } |
53b381b3 | 6108 | |
ffe2d203 | 6109 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6110 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6111 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6112 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6113 | stripe_len * data_stripes); |
53b381b3 DW |
6114 | |
6115 | /* RAID[56] write or recovery. Return all stripes */ | |
6116 | num_stripes = map->num_stripes; | |
6117 | max_errors = nr_parity_stripes(map); | |
6118 | ||
53b381b3 DW |
6119 | *length = map->stripe_len; |
6120 | stripe_index = 0; | |
6121 | stripe_offset = 0; | |
6122 | } else { | |
6123 | /* | |
6124 | * Mirror #0 or #1 means the original data block. | |
6125 | * Mirror #2 is RAID5 parity block. | |
6126 | * Mirror #3 is RAID6 Q block. | |
6127 | */ | |
47c5713f | 6128 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6129 | data_stripes, &stripe_index); |
53b381b3 | 6130 | if (mirror_num > 1) |
cff82672 | 6131 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6132 | |
6133 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6134 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6135 | &stripe_index); | |
de483734 | 6136 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6137 | mirror_num = 1; |
53b381b3 | 6138 | } |
8790d502 CM |
6139 | } else { |
6140 | /* | |
47c5713f DS |
6141 | * after this, stripe_nr is the number of stripes on this |
6142 | * device we have to walk to find the data, and stripe_index is | |
6143 | * the number of our device in the stripe array | |
8790d502 | 6144 | */ |
47c5713f DS |
6145 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6146 | &stripe_index); | |
a1d3c478 | 6147 | mirror_num = stripe_index + 1; |
8790d502 | 6148 | } |
e042d1ec | 6149 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6150 | btrfs_crit(fs_info, |
6151 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6152 | stripe_index, map->num_stripes); |
6153 | ret = -EINVAL; | |
6154 | goto out; | |
6155 | } | |
cea9e445 | 6156 | |
472262f3 | 6157 | num_alloc_stripes = num_stripes; |
6fad823f | 6158 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6159 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6160 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6161 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6162 | num_alloc_stripes++; |
2c8cdd6e | 6163 | tgtdev_indexes = num_stripes; |
ad6d620e | 6164 | } |
2c8cdd6e | 6165 | |
6e9606d2 | 6166 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6167 | if (!bbio) { |
6168 | ret = -ENOMEM; | |
6169 | goto out; | |
6170 | } | |
6fad823f | 6171 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6172 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6173 | |
8e5cfb55 | 6174 | /* build raid_map */ |
2b19a1fe LB |
6175 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6176 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6177 | u64 tmp; |
9d644a62 | 6178 | unsigned rot; |
8e5cfb55 ZL |
6179 | |
6180 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6181 | sizeof(struct btrfs_bio_stripe) * | |
6182 | num_alloc_stripes + | |
6183 | sizeof(int) * tgtdev_indexes); | |
6184 | ||
6185 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6186 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6187 | |
6188 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6189 | tmp = stripe_nr * data_stripes; |
6190 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6191 | bbio->raid_map[(i+rot) % num_stripes] = |
6192 | em->start + (tmp + i) * map->stripe_len; | |
6193 | ||
6194 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6195 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6196 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6197 | RAID6_Q_STRIPE; | |
6198 | } | |
6199 | ||
b89203f7 | 6200 | |
0b3d4cd3 LB |
6201 | for (i = 0; i < num_stripes; i++) { |
6202 | bbio->stripes[i].physical = | |
6203 | map->stripes[stripe_index].physical + | |
6204 | stripe_offset + | |
6205 | stripe_nr * map->stripe_len; | |
6206 | bbio->stripes[i].dev = | |
6207 | map->stripes[stripe_index].dev; | |
6208 | stripe_index++; | |
593060d7 | 6209 | } |
de11cc12 | 6210 | |
2b19a1fe | 6211 | if (need_full_stripe(op)) |
d20983b4 | 6212 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6213 | |
8e5cfb55 ZL |
6214 | if (bbio->raid_map) |
6215 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6216 | |
73c0f228 | 6217 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6218 | need_full_stripe(op)) { |
73c0f228 LB |
6219 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6220 | &max_errors); | |
472262f3 SB |
6221 | } |
6222 | ||
de11cc12 | 6223 | *bbio_ret = bbio; |
10f11900 | 6224 | bbio->map_type = map->type; |
de11cc12 LZ |
6225 | bbio->num_stripes = num_stripes; |
6226 | bbio->max_errors = max_errors; | |
6227 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6228 | |
6229 | /* | |
6230 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6231 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6232 | * available as a mirror | |
6233 | */ | |
6234 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6235 | WARN_ON(num_stripes > 1); | |
6236 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6237 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6238 | bbio->mirror_num = map->num_stripes + 1; | |
6239 | } | |
cea9e445 | 6240 | out: |
73beece9 | 6241 | if (dev_replace_is_ongoing) { |
53176dde DS |
6242 | lockdep_assert_held(&dev_replace->rwsem); |
6243 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6244 | up_read(&dev_replace->rwsem); |
73beece9 | 6245 | } |
0b86a832 | 6246 | free_extent_map(em); |
de11cc12 | 6247 | return ret; |
0b86a832 CM |
6248 | } |
6249 | ||
cf8cddd3 | 6250 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6251 | u64 logical, u64 *length, |
a1d3c478 | 6252 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6253 | { |
b3d3fa51 | 6254 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6255 | mirror_num, 0); |
f2d8d74d CM |
6256 | } |
6257 | ||
af8e2d1d | 6258 | /* For Scrub/replace */ |
cf8cddd3 | 6259 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6260 | u64 logical, u64 *length, |
825ad4c9 | 6261 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6262 | { |
825ad4c9 | 6263 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6264 | } |
6265 | ||
63a9c7b9 NB |
6266 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6267 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6268 | { |
a512bbf8 YZ |
6269 | struct extent_map *em; |
6270 | struct map_lookup *map; | |
6271 | u64 *buf; | |
6272 | u64 bytenr; | |
6273 | u64 length; | |
6274 | u64 stripe_nr; | |
53b381b3 | 6275 | u64 rmap_len; |
a512bbf8 YZ |
6276 | int i, j, nr = 0; |
6277 | ||
60ca842e | 6278 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6279 | if (IS_ERR(em)) |
835d974f | 6280 | return -EIO; |
835d974f | 6281 | |
95617d69 | 6282 | map = em->map_lookup; |
a512bbf8 | 6283 | length = em->len; |
53b381b3 DW |
6284 | rmap_len = map->stripe_len; |
6285 | ||
a512bbf8 | 6286 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6287 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6288 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6289 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6290 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6291 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6292 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6293 | } | |
a512bbf8 | 6294 | |
31e818fe | 6295 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6296 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6297 | |
6298 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6299 | if (map->stripes[i].physical > physical || |
6300 | map->stripes[i].physical + length <= physical) | |
6301 | continue; | |
6302 | ||
6303 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6304 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6305 | |
6306 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6307 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6308 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6309 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6310 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6311 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6312 | * Alternatively, just use rmap_len below instead of | |
6313 | * map->stripe_len */ | |
6314 | ||
6315 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6316 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6317 | for (j = 0; j < nr; j++) { |
6318 | if (buf[j] == bytenr) | |
6319 | break; | |
6320 | } | |
934d375b CM |
6321 | if (j == nr) { |
6322 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6323 | buf[nr++] = bytenr; |
934d375b | 6324 | } |
a512bbf8 YZ |
6325 | } |
6326 | ||
a512bbf8 YZ |
6327 | *logical = buf; |
6328 | *naddrs = nr; | |
53b381b3 | 6329 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6330 | |
6331 | free_extent_map(em); | |
6332 | return 0; | |
f2d8d74d CM |
6333 | } |
6334 | ||
4246a0b6 | 6335 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6336 | { |
326e1dbb MS |
6337 | bio->bi_private = bbio->private; |
6338 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6339 | bio_endio(bio); |
326e1dbb | 6340 | |
6e9606d2 | 6341 | btrfs_put_bbio(bbio); |
8408c716 MX |
6342 | } |
6343 | ||
4246a0b6 | 6344 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6345 | { |
9be3395b | 6346 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6347 | int is_orig_bio = 0; |
8790d502 | 6348 | |
4e4cbee9 | 6349 | if (bio->bi_status) { |
a1d3c478 | 6350 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6351 | if (bio->bi_status == BLK_STS_IOERR || |
6352 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6353 | unsigned int stripe_index = |
9be3395b | 6354 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6355 | struct btrfs_device *dev; |
442a4f63 SB |
6356 | |
6357 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6358 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6359 | if (dev->bdev) { |
37226b21 | 6360 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6361 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6362 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6363 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6364 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6365 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6366 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6367 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6368 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6369 | } |
442a4f63 SB |
6370 | } |
6371 | } | |
8790d502 | 6372 | |
a1d3c478 | 6373 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6374 | is_orig_bio = 1; |
6375 | ||
c404e0dc MX |
6376 | btrfs_bio_counter_dec(bbio->fs_info); |
6377 | ||
a1d3c478 | 6378 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6379 | if (!is_orig_bio) { |
6380 | bio_put(bio); | |
a1d3c478 | 6381 | bio = bbio->orig_bio; |
7d2b4daa | 6382 | } |
c7b22bb1 | 6383 | |
9be3395b | 6384 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6385 | /* only send an error to the higher layers if it is |
53b381b3 | 6386 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6387 | */ |
a1d3c478 | 6388 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6389 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6390 | } else { |
1259ab75 CM |
6391 | /* |
6392 | * this bio is actually up to date, we didn't | |
6393 | * go over the max number of errors | |
6394 | */ | |
2dbe0c77 | 6395 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6396 | } |
c55f1396 | 6397 | |
4246a0b6 | 6398 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6399 | } else if (!is_orig_bio) { |
8790d502 CM |
6400 | bio_put(bio); |
6401 | } | |
8790d502 CM |
6402 | } |
6403 | ||
8b712842 CM |
6404 | /* |
6405 | * see run_scheduled_bios for a description of why bios are collected for | |
6406 | * async submit. | |
6407 | * | |
6408 | * This will add one bio to the pending list for a device and make sure | |
6409 | * the work struct is scheduled. | |
6410 | */ | |
2ff7e61e | 6411 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6412 | struct bio *bio) |
8b712842 | 6413 | { |
0b246afa | 6414 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6415 | int should_queue = 1; |
ffbd517d | 6416 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6417 | |
6418 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6419 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6420 | btrfsic_submit_bio(bio); |
143bede5 | 6421 | return; |
8b712842 CM |
6422 | } |
6423 | ||
492bb6de | 6424 | WARN_ON(bio->bi_next); |
8b712842 | 6425 | bio->bi_next = NULL; |
8b712842 CM |
6426 | |
6427 | spin_lock(&device->io_lock); | |
67f055c7 | 6428 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6429 | pending_bios = &device->pending_sync_bios; |
6430 | else | |
6431 | pending_bios = &device->pending_bios; | |
8b712842 | 6432 | |
ffbd517d CM |
6433 | if (pending_bios->tail) |
6434 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6435 | |
ffbd517d CM |
6436 | pending_bios->tail = bio; |
6437 | if (!pending_bios->head) | |
6438 | pending_bios->head = bio; | |
8b712842 CM |
6439 | if (device->running_pending) |
6440 | should_queue = 0; | |
6441 | ||
6442 | spin_unlock(&device->io_lock); | |
6443 | ||
6444 | if (should_queue) | |
0b246afa | 6445 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6446 | } |
6447 | ||
2ff7e61e JM |
6448 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6449 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6450 | { |
6451 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6452 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6453 | |
6454 | bio->bi_private = bbio; | |
9be3395b | 6455 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6456 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6457 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6458 | btrfs_debug_in_rcu(fs_info, |
6459 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6460 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6461 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6462 | bio->bi_iter.bi_size); | |
74d46992 | 6463 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6464 | |
2ff7e61e | 6465 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6466 | |
de1ee92a | 6467 | if (async) |
2ff7e61e | 6468 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6469 | else |
4e49ea4a | 6470 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6471 | } |
6472 | ||
de1ee92a JB |
6473 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6474 | { | |
6475 | atomic_inc(&bbio->error); | |
6476 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6477 | /* Should be the original bio. */ |
8408c716 MX |
6478 | WARN_ON(bio != bbio->orig_bio); |
6479 | ||
9be3395b | 6480 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6481 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6482 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6483 | bio->bi_status = BLK_STS_IOERR; | |
6484 | else | |
6485 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6486 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6487 | } |
6488 | } | |
6489 | ||
58efbc9f OS |
6490 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6491 | int mirror_num, int async_submit) | |
0b86a832 | 6492 | { |
0b86a832 | 6493 | struct btrfs_device *dev; |
8790d502 | 6494 | struct bio *first_bio = bio; |
4f024f37 | 6495 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6496 | u64 length = 0; |
6497 | u64 map_length; | |
0b86a832 | 6498 | int ret; |
08da757d ZL |
6499 | int dev_nr; |
6500 | int total_devs; | |
a1d3c478 | 6501 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6502 | |
4f024f37 | 6503 | length = bio->bi_iter.bi_size; |
0b86a832 | 6504 | map_length = length; |
cea9e445 | 6505 | |
0b246afa | 6506 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6507 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6508 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6509 | if (ret) { |
0b246afa | 6510 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6511 | return errno_to_blk_status(ret); |
c404e0dc | 6512 | } |
cea9e445 | 6513 | |
a1d3c478 | 6514 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6515 | bbio->orig_bio = first_bio; |
6516 | bbio->private = first_bio->bi_private; | |
6517 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6518 | bbio->fs_info = fs_info; |
53b381b3 DW |
6519 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6520 | ||
ad1ba2a0 | 6521 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6522 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6523 | /* In this case, map_length has been set to the length of |
6524 | a single stripe; not the whole write */ | |
37226b21 | 6525 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6526 | ret = raid56_parity_write(fs_info, bio, bbio, |
6527 | map_length); | |
53b381b3 | 6528 | } else { |
2ff7e61e JM |
6529 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6530 | map_length, mirror_num, 1); | |
53b381b3 | 6531 | } |
4245215d | 6532 | |
0b246afa | 6533 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6534 | return errno_to_blk_status(ret); |
53b381b3 DW |
6535 | } |
6536 | ||
cea9e445 | 6537 | if (map_length < length) { |
0b246afa | 6538 | btrfs_crit(fs_info, |
5d163e0e JM |
6539 | "mapping failed logical %llu bio len %llu len %llu", |
6540 | logical, length, map_length); | |
cea9e445 CM |
6541 | BUG(); |
6542 | } | |
a1d3c478 | 6543 | |
08da757d | 6544 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6545 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6546 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6547 | &dev->dev_state) || | |
ebbede42 AJ |
6548 | (bio_op(first_bio) == REQ_OP_WRITE && |
6549 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6550 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6551 | continue; |
6552 | } | |
6553 | ||
3aa8e074 | 6554 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6555 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6556 | else |
a1d3c478 | 6557 | bio = first_bio; |
de1ee92a | 6558 | |
2ff7e61e JM |
6559 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6560 | dev_nr, async_submit); | |
8790d502 | 6561 | } |
0b246afa | 6562 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6563 | return BLK_STS_OK; |
0b86a832 CM |
6564 | } |
6565 | ||
09ba3bc9 AJ |
6566 | /* |
6567 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6568 | * return NULL. | |
6569 | * | |
6570 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6571 | * only devid is used. | |
6572 | * | |
6573 | * If @seed is true, traverse through the seed devices. | |
6574 | */ | |
e4319cd9 | 6575 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6576 | u64 devid, u8 *uuid, u8 *fsid, |
6577 | bool seed) | |
0b86a832 | 6578 | { |
2b82032c | 6579 | struct btrfs_device *device; |
2b82032c | 6580 | |
e4319cd9 | 6581 | while (fs_devices) { |
2b82032c | 6582 | if (!fsid || |
e4319cd9 | 6583 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6584 | list_for_each_entry(device, &fs_devices->devices, |
6585 | dev_list) { | |
6586 | if (device->devid == devid && | |
6587 | (!uuid || memcmp(device->uuid, uuid, | |
6588 | BTRFS_UUID_SIZE) == 0)) | |
6589 | return device; | |
6590 | } | |
2b82032c | 6591 | } |
09ba3bc9 AJ |
6592 | if (seed) |
6593 | fs_devices = fs_devices->seed; | |
6594 | else | |
6595 | return NULL; | |
2b82032c YZ |
6596 | } |
6597 | return NULL; | |
0b86a832 CM |
6598 | } |
6599 | ||
2ff7e61e | 6600 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6601 | u64 devid, u8 *dev_uuid) |
6602 | { | |
6603 | struct btrfs_device *device; | |
dfe25020 | 6604 | |
12bd2fc0 ID |
6605 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6606 | if (IS_ERR(device)) | |
adfb69af | 6607 | return device; |
12bd2fc0 ID |
6608 | |
6609 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6610 | device->fs_devices = fs_devices; |
dfe25020 | 6611 | fs_devices->num_devices++; |
12bd2fc0 | 6612 | |
e6e674bd | 6613 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6614 | fs_devices->missing_devices++; |
12bd2fc0 | 6615 | |
dfe25020 CM |
6616 | return device; |
6617 | } | |
6618 | ||
12bd2fc0 ID |
6619 | /** |
6620 | * btrfs_alloc_device - allocate struct btrfs_device | |
6621 | * @fs_info: used only for generating a new devid, can be NULL if | |
6622 | * devid is provided (i.e. @devid != NULL). | |
6623 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6624 | * is generated. | |
6625 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6626 | * is generated. | |
6627 | * | |
6628 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6629 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6630 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6631 | */ |
6632 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6633 | const u64 *devid, | |
6634 | const u8 *uuid) | |
6635 | { | |
6636 | struct btrfs_device *dev; | |
6637 | u64 tmp; | |
6638 | ||
fae7f21c | 6639 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6640 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6641 | |
6642 | dev = __alloc_device(); | |
6643 | if (IS_ERR(dev)) | |
6644 | return dev; | |
6645 | ||
6646 | if (devid) | |
6647 | tmp = *devid; | |
6648 | else { | |
6649 | int ret; | |
6650 | ||
6651 | ret = find_next_devid(fs_info, &tmp); | |
6652 | if (ret) { | |
a425f9d4 | 6653 | btrfs_free_device(dev); |
12bd2fc0 ID |
6654 | return ERR_PTR(ret); |
6655 | } | |
6656 | } | |
6657 | dev->devid = tmp; | |
6658 | ||
6659 | if (uuid) | |
6660 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6661 | else | |
6662 | generate_random_uuid(dev->uuid); | |
6663 | ||
9e0af237 LB |
6664 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6665 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6666 | |
6667 | return dev; | |
6668 | } | |
6669 | ||
5a2b8e60 | 6670 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6671 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6672 | { |
2b902dfc AJ |
6673 | if (error) |
6674 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6675 | devid, uuid); | |
6676 | else | |
6677 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6678 | devid, uuid); | |
5a2b8e60 AJ |
6679 | } |
6680 | ||
39e264a4 NB |
6681 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6682 | { | |
6683 | int index = btrfs_bg_flags_to_raid_index(type); | |
6684 | int ncopies = btrfs_raid_array[index].ncopies; | |
6685 | int data_stripes; | |
6686 | ||
6687 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
6688 | case BTRFS_BLOCK_GROUP_RAID5: | |
6689 | data_stripes = num_stripes - 1; | |
6690 | break; | |
6691 | case BTRFS_BLOCK_GROUP_RAID6: | |
6692 | data_stripes = num_stripes - 2; | |
6693 | break; | |
6694 | default: | |
6695 | data_stripes = num_stripes / ncopies; | |
6696 | break; | |
6697 | } | |
6698 | return div_u64(chunk_len, data_stripes); | |
6699 | } | |
6700 | ||
9690ac09 | 6701 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6702 | struct btrfs_chunk *chunk) |
6703 | { | |
9690ac09 | 6704 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6705 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6706 | struct map_lookup *map; |
6707 | struct extent_map *em; | |
6708 | u64 logical; | |
6709 | u64 length; | |
e06cd3dd LB |
6710 | u64 devid; |
6711 | u8 uuid[BTRFS_UUID_SIZE]; | |
6712 | int num_stripes; | |
6713 | int ret; | |
6714 | int i; | |
6715 | ||
6716 | logical = key->offset; | |
6717 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6718 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6719 | ||
075cb3c7 QW |
6720 | /* |
6721 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6722 | * as chunk item in tree block is already verified by tree-checker. | |
6723 | */ | |
6724 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6725 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6726 | if (ret) |
6727 | return ret; | |
6728 | } | |
a061fc8d | 6729 | |
c8bf1b67 DS |
6730 | read_lock(&map_tree->lock); |
6731 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6732 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6733 | |
6734 | /* already mapped? */ | |
6735 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6736 | free_extent_map(em); | |
0b86a832 CM |
6737 | return 0; |
6738 | } else if (em) { | |
6739 | free_extent_map(em); | |
6740 | } | |
0b86a832 | 6741 | |
172ddd60 | 6742 | em = alloc_extent_map(); |
0b86a832 CM |
6743 | if (!em) |
6744 | return -ENOMEM; | |
593060d7 | 6745 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6746 | if (!map) { |
6747 | free_extent_map(em); | |
6748 | return -ENOMEM; | |
6749 | } | |
6750 | ||
298a8f9c | 6751 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6752 | em->map_lookup = map; |
0b86a832 CM |
6753 | em->start = logical; |
6754 | em->len = length; | |
70c8a91c | 6755 | em->orig_start = 0; |
0b86a832 | 6756 | em->block_start = 0; |
c8b97818 | 6757 | em->block_len = em->len; |
0b86a832 | 6758 | |
593060d7 CM |
6759 | map->num_stripes = num_stripes; |
6760 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6761 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6762 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6763 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6764 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6765 | map->verified_stripes = 0; |
39e264a4 NB |
6766 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6767 | map->num_stripes); | |
593060d7 CM |
6768 | for (i = 0; i < num_stripes; i++) { |
6769 | map->stripes[i].physical = | |
6770 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6771 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6772 | read_extent_buffer(leaf, uuid, (unsigned long) |
6773 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6774 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6775 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6776 | devid, uuid, NULL, true); |
3cdde224 | 6777 | if (!map->stripes[i].dev && |
0b246afa | 6778 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6779 | free_extent_map(em); |
2b902dfc | 6780 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6781 | return -ENOENT; |
593060d7 | 6782 | } |
dfe25020 CM |
6783 | if (!map->stripes[i].dev) { |
6784 | map->stripes[i].dev = | |
2ff7e61e JM |
6785 | add_missing_dev(fs_info->fs_devices, devid, |
6786 | uuid); | |
adfb69af | 6787 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6788 | free_extent_map(em); |
adfb69af AJ |
6789 | btrfs_err(fs_info, |
6790 | "failed to init missing dev %llu: %ld", | |
6791 | devid, PTR_ERR(map->stripes[i].dev)); | |
6792 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6793 | } |
2b902dfc | 6794 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6795 | } |
e12c9621 AJ |
6796 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6797 | &(map->stripes[i].dev->dev_state)); | |
6798 | ||
0b86a832 CM |
6799 | } |
6800 | ||
c8bf1b67 DS |
6801 | write_lock(&map_tree->lock); |
6802 | ret = add_extent_mapping(map_tree, em, 0); | |
6803 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6804 | if (ret < 0) { |
6805 | btrfs_err(fs_info, | |
6806 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6807 | em->start, em->len, ret); | |
6808 | } | |
0b86a832 CM |
6809 | free_extent_map(em); |
6810 | ||
64f64f43 | 6811 | return ret; |
0b86a832 CM |
6812 | } |
6813 | ||
143bede5 | 6814 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6815 | struct btrfs_dev_item *dev_item, |
6816 | struct btrfs_device *device) | |
6817 | { | |
6818 | unsigned long ptr; | |
0b86a832 CM |
6819 | |
6820 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6821 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6822 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6823 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6824 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6825 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6826 | device->type = btrfs_device_type(leaf, dev_item); |
6827 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6828 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6829 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6830 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6831 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6832 | |
410ba3a2 | 6833 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6834 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6835 | } |
6836 | ||
2ff7e61e | 6837 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6838 | u8 *fsid) |
2b82032c YZ |
6839 | { |
6840 | struct btrfs_fs_devices *fs_devices; | |
6841 | int ret; | |
6842 | ||
a32bf9a3 | 6843 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6844 | ASSERT(fsid); |
2b82032c | 6845 | |
0b246afa | 6846 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6847 | while (fs_devices) { |
44880fdc | 6848 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6849 | return fs_devices; |
6850 | ||
2b82032c YZ |
6851 | fs_devices = fs_devices->seed; |
6852 | } | |
6853 | ||
7239ff4b | 6854 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6855 | if (!fs_devices) { |
0b246afa | 6856 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6857 | return ERR_PTR(-ENOENT); |
6858 | ||
7239ff4b | 6859 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6860 | if (IS_ERR(fs_devices)) |
6861 | return fs_devices; | |
6862 | ||
6863 | fs_devices->seeding = 1; | |
6864 | fs_devices->opened = 1; | |
6865 | return fs_devices; | |
2b82032c | 6866 | } |
e4404d6e YZ |
6867 | |
6868 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6869 | if (IS_ERR(fs_devices)) |
6870 | return fs_devices; | |
2b82032c | 6871 | |
897fb573 | 6872 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6873 | if (ret) { |
6874 | free_fs_devices(fs_devices); | |
5f375835 | 6875 | fs_devices = ERR_PTR(ret); |
2b82032c | 6876 | goto out; |
48d28232 | 6877 | } |
2b82032c YZ |
6878 | |
6879 | if (!fs_devices->seeding) { | |
0226e0eb | 6880 | close_fs_devices(fs_devices); |
e4404d6e | 6881 | free_fs_devices(fs_devices); |
5f375835 | 6882 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6883 | goto out; |
6884 | } | |
6885 | ||
0b246afa JM |
6886 | fs_devices->seed = fs_info->fs_devices->seed; |
6887 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6888 | out: |
5f375835 | 6889 | return fs_devices; |
2b82032c YZ |
6890 | } |
6891 | ||
17850759 | 6892 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6893 | struct btrfs_dev_item *dev_item) |
6894 | { | |
17850759 | 6895 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6896 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6897 | struct btrfs_device *device; |
6898 | u64 devid; | |
6899 | int ret; | |
44880fdc | 6900 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6901 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6902 | ||
0b86a832 | 6903 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6904 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6905 | BTRFS_UUID_SIZE); |
1473b24e | 6906 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6907 | BTRFS_FSID_SIZE); |
2b82032c | 6908 | |
de37aa51 | 6909 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6910 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6911 | if (IS_ERR(fs_devices)) |
6912 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6913 | } |
6914 | ||
e4319cd9 | 6915 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6916 | fs_uuid, true); |
5f375835 | 6917 | if (!device) { |
c5502451 | 6918 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6919 | btrfs_report_missing_device(fs_info, devid, |
6920 | dev_uuid, true); | |
45dbdbc9 | 6921 | return -ENOENT; |
c5502451 | 6922 | } |
2b82032c | 6923 | |
2ff7e61e | 6924 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6925 | if (IS_ERR(device)) { |
6926 | btrfs_err(fs_info, | |
6927 | "failed to add missing dev %llu: %ld", | |
6928 | devid, PTR_ERR(device)); | |
6929 | return PTR_ERR(device); | |
6930 | } | |
2b902dfc | 6931 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6932 | } else { |
c5502451 | 6933 | if (!device->bdev) { |
2b902dfc AJ |
6934 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6935 | btrfs_report_missing_device(fs_info, | |
6936 | devid, dev_uuid, true); | |
45dbdbc9 | 6937 | return -ENOENT; |
2b902dfc AJ |
6938 | } |
6939 | btrfs_report_missing_device(fs_info, devid, | |
6940 | dev_uuid, false); | |
c5502451 | 6941 | } |
5f375835 | 6942 | |
e6e674bd AJ |
6943 | if (!device->bdev && |
6944 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6945 | /* |
6946 | * this happens when a device that was properly setup | |
6947 | * in the device info lists suddenly goes bad. | |
6948 | * device->bdev is NULL, and so we have to set | |
6949 | * device->missing to one here | |
6950 | */ | |
5f375835 | 6951 | device->fs_devices->missing_devices++; |
e6e674bd | 6952 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6953 | } |
5f375835 MX |
6954 | |
6955 | /* Move the device to its own fs_devices */ | |
6956 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6957 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6958 | &device->dev_state)); | |
5f375835 MX |
6959 | |
6960 | list_move(&device->dev_list, &fs_devices->devices); | |
6961 | device->fs_devices->num_devices--; | |
6962 | fs_devices->num_devices++; | |
6963 | ||
6964 | device->fs_devices->missing_devices--; | |
6965 | fs_devices->missing_devices++; | |
6966 | ||
6967 | device->fs_devices = fs_devices; | |
6968 | } | |
2b82032c YZ |
6969 | } |
6970 | ||
0b246afa | 6971 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6972 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6973 | if (device->generation != |
6974 | btrfs_device_generation(leaf, dev_item)) | |
6975 | return -EINVAL; | |
6324fbf3 | 6976 | } |
0b86a832 CM |
6977 | |
6978 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6979 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6980 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6981 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6982 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6983 | atomic64_add(device->total_bytes - device->bytes_used, |
6984 | &fs_info->free_chunk_space); | |
2bf64758 | 6985 | } |
0b86a832 | 6986 | ret = 0; |
0b86a832 CM |
6987 | return ret; |
6988 | } | |
6989 | ||
6bccf3ab | 6990 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6991 | { |
6bccf3ab | 6992 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6993 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6994 | struct extent_buffer *sb; |
0b86a832 | 6995 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6996 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6997 | u8 *array_ptr; |
6998 | unsigned long sb_array_offset; | |
84eed90f | 6999 | int ret = 0; |
0b86a832 CM |
7000 | u32 num_stripes; |
7001 | u32 array_size; | |
7002 | u32 len = 0; | |
1ffb22cf | 7003 | u32 cur_offset; |
e06cd3dd | 7004 | u64 type; |
84eed90f | 7005 | struct btrfs_key key; |
0b86a832 | 7006 | |
0b246afa | 7007 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7008 | /* |
7009 | * This will create extent buffer of nodesize, superblock size is | |
7010 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7011 | * overallocate but we can keep it as-is, only the first page is used. | |
7012 | */ | |
2ff7e61e | 7013 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7014 | if (IS_ERR(sb)) |
7015 | return PTR_ERR(sb); | |
4db8c528 | 7016 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7017 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7018 | /* |
01327610 | 7019 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7020 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7021 | * pages up-to-date when the page is larger: extent does not cover the |
7022 | * whole page and consequently check_page_uptodate does not find all | |
7023 | * the page's extents up-to-date (the hole beyond sb), | |
7024 | * write_extent_buffer then triggers a WARN_ON. | |
7025 | * | |
7026 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7027 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7028 | * to silence the warning eg. on PowerPC 64. | |
7029 | */ | |
09cbfeaf | 7030 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7031 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7032 | |
a061fc8d | 7033 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7034 | array_size = btrfs_super_sys_array_size(super_copy); |
7035 | ||
1ffb22cf DS |
7036 | array_ptr = super_copy->sys_chunk_array; |
7037 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7038 | cur_offset = 0; | |
0b86a832 | 7039 | |
1ffb22cf DS |
7040 | while (cur_offset < array_size) { |
7041 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7042 | len = sizeof(*disk_key); |
7043 | if (cur_offset + len > array_size) | |
7044 | goto out_short_read; | |
7045 | ||
0b86a832 CM |
7046 | btrfs_disk_key_to_cpu(&key, disk_key); |
7047 | ||
1ffb22cf DS |
7048 | array_ptr += len; |
7049 | sb_array_offset += len; | |
7050 | cur_offset += len; | |
0b86a832 | 7051 | |
0d81ba5d | 7052 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7053 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7054 | /* |
7055 | * At least one btrfs_chunk with one stripe must be | |
7056 | * present, exact stripe count check comes afterwards | |
7057 | */ | |
7058 | len = btrfs_chunk_item_size(1); | |
7059 | if (cur_offset + len > array_size) | |
7060 | goto out_short_read; | |
7061 | ||
7062 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7063 | if (!num_stripes) { |
ab8d0fc4 JM |
7064 | btrfs_err(fs_info, |
7065 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7066 | num_stripes, cur_offset); |
7067 | ret = -EIO; | |
7068 | break; | |
7069 | } | |
7070 | ||
e06cd3dd LB |
7071 | type = btrfs_chunk_type(sb, chunk); |
7072 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7073 | btrfs_err(fs_info, |
e06cd3dd LB |
7074 | "invalid chunk type %llu in sys_array at offset %u", |
7075 | type, cur_offset); | |
7076 | ret = -EIO; | |
7077 | break; | |
7078 | } | |
7079 | ||
e3540eab DS |
7080 | len = btrfs_chunk_item_size(num_stripes); |
7081 | if (cur_offset + len > array_size) | |
7082 | goto out_short_read; | |
7083 | ||
9690ac09 | 7084 | ret = read_one_chunk(&key, sb, chunk); |
84eed90f CM |
7085 | if (ret) |
7086 | break; | |
0b86a832 | 7087 | } else { |
ab8d0fc4 JM |
7088 | btrfs_err(fs_info, |
7089 | "unexpected item type %u in sys_array at offset %u", | |
7090 | (u32)key.type, cur_offset); | |
84eed90f CM |
7091 | ret = -EIO; |
7092 | break; | |
0b86a832 | 7093 | } |
1ffb22cf DS |
7094 | array_ptr += len; |
7095 | sb_array_offset += len; | |
7096 | cur_offset += len; | |
0b86a832 | 7097 | } |
d865177a | 7098 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7099 | free_extent_buffer_stale(sb); |
84eed90f | 7100 | return ret; |
e3540eab DS |
7101 | |
7102 | out_short_read: | |
ab8d0fc4 | 7103 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7104 | len, cur_offset); |
d865177a | 7105 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7106 | free_extent_buffer_stale(sb); |
e3540eab | 7107 | return -EIO; |
0b86a832 CM |
7108 | } |
7109 | ||
21634a19 QW |
7110 | /* |
7111 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7112 | * | |
6528b99d AJ |
7113 | * If the @failing_dev is specified, it's accounted as missing. |
7114 | * | |
21634a19 QW |
7115 | * Return true if all chunks meet the minimal RW mount requirements. |
7116 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7117 | */ | |
6528b99d AJ |
7118 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7119 | struct btrfs_device *failing_dev) | |
21634a19 | 7120 | { |
c8bf1b67 | 7121 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7122 | struct extent_map *em; |
7123 | u64 next_start = 0; | |
7124 | bool ret = true; | |
7125 | ||
c8bf1b67 DS |
7126 | read_lock(&map_tree->lock); |
7127 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7128 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7129 | /* No chunk at all? Return false anyway */ |
7130 | if (!em) { | |
7131 | ret = false; | |
7132 | goto out; | |
7133 | } | |
7134 | while (em) { | |
7135 | struct map_lookup *map; | |
7136 | int missing = 0; | |
7137 | int max_tolerated; | |
7138 | int i; | |
7139 | ||
7140 | map = em->map_lookup; | |
7141 | max_tolerated = | |
7142 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7143 | map->type); | |
7144 | for (i = 0; i < map->num_stripes; i++) { | |
7145 | struct btrfs_device *dev = map->stripes[i].dev; | |
7146 | ||
e6e674bd AJ |
7147 | if (!dev || !dev->bdev || |
7148 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7149 | dev->last_flush_error) |
7150 | missing++; | |
6528b99d AJ |
7151 | else if (failing_dev && failing_dev == dev) |
7152 | missing++; | |
21634a19 QW |
7153 | } |
7154 | if (missing > max_tolerated) { | |
6528b99d AJ |
7155 | if (!failing_dev) |
7156 | btrfs_warn(fs_info, | |
52042d8e | 7157 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7158 | em->start, missing, max_tolerated); |
7159 | free_extent_map(em); | |
7160 | ret = false; | |
7161 | goto out; | |
7162 | } | |
7163 | next_start = extent_map_end(em); | |
7164 | free_extent_map(em); | |
7165 | ||
c8bf1b67 DS |
7166 | read_lock(&map_tree->lock); |
7167 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7168 | (u64)(-1) - next_start); |
c8bf1b67 | 7169 | read_unlock(&map_tree->lock); |
21634a19 QW |
7170 | } |
7171 | out: | |
7172 | return ret; | |
7173 | } | |
7174 | ||
5b4aacef | 7175 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7176 | { |
5b4aacef | 7177 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7178 | struct btrfs_path *path; |
7179 | struct extent_buffer *leaf; | |
7180 | struct btrfs_key key; | |
7181 | struct btrfs_key found_key; | |
7182 | int ret; | |
7183 | int slot; | |
99e3ecfc | 7184 | u64 total_dev = 0; |
0b86a832 | 7185 | |
0b86a832 CM |
7186 | path = btrfs_alloc_path(); |
7187 | if (!path) | |
7188 | return -ENOMEM; | |
7189 | ||
3dd0f7a3 AJ |
7190 | /* |
7191 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7192 | * otherwise we don't need it. | |
7193 | */ | |
b367e47f | 7194 | mutex_lock(&uuid_mutex); |
34441361 | 7195 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7196 | |
395927a9 FDBM |
7197 | /* |
7198 | * Read all device items, and then all the chunk items. All | |
7199 | * device items are found before any chunk item (their object id | |
7200 | * is smaller than the lowest possible object id for a chunk | |
7201 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7202 | */ |
7203 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7204 | key.offset = 0; | |
7205 | key.type = 0; | |
0b86a832 | 7206 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7207 | if (ret < 0) |
7208 | goto error; | |
d397712b | 7209 | while (1) { |
0b86a832 CM |
7210 | leaf = path->nodes[0]; |
7211 | slot = path->slots[0]; | |
7212 | if (slot >= btrfs_header_nritems(leaf)) { | |
7213 | ret = btrfs_next_leaf(root, path); | |
7214 | if (ret == 0) | |
7215 | continue; | |
7216 | if (ret < 0) | |
7217 | goto error; | |
7218 | break; | |
7219 | } | |
7220 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7221 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7222 | struct btrfs_dev_item *dev_item; | |
7223 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7224 | struct btrfs_dev_item); |
17850759 | 7225 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7226 | if (ret) |
7227 | goto error; | |
99e3ecfc | 7228 | total_dev++; |
0b86a832 CM |
7229 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7230 | struct btrfs_chunk *chunk; | |
7231 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
9690ac09 | 7232 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7233 | if (ret) |
7234 | goto error; | |
0b86a832 CM |
7235 | } |
7236 | path->slots[0]++; | |
7237 | } | |
99e3ecfc LB |
7238 | |
7239 | /* | |
7240 | * After loading chunk tree, we've got all device information, | |
7241 | * do another round of validation checks. | |
7242 | */ | |
0b246afa JM |
7243 | if (total_dev != fs_info->fs_devices->total_devices) { |
7244 | btrfs_err(fs_info, | |
99e3ecfc | 7245 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7246 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7247 | total_dev); |
7248 | ret = -EINVAL; | |
7249 | goto error; | |
7250 | } | |
0b246afa JM |
7251 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7252 | fs_info->fs_devices->total_rw_bytes) { | |
7253 | btrfs_err(fs_info, | |
99e3ecfc | 7254 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7255 | btrfs_super_total_bytes(fs_info->super_copy), |
7256 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7257 | ret = -EINVAL; |
7258 | goto error; | |
7259 | } | |
0b86a832 CM |
7260 | ret = 0; |
7261 | error: | |
34441361 | 7262 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7263 | mutex_unlock(&uuid_mutex); |
7264 | ||
2b82032c | 7265 | btrfs_free_path(path); |
0b86a832 CM |
7266 | return ret; |
7267 | } | |
442a4f63 | 7268 | |
cb517eab MX |
7269 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7270 | { | |
7271 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7272 | struct btrfs_device *device; | |
7273 | ||
29cc83f6 LB |
7274 | while (fs_devices) { |
7275 | mutex_lock(&fs_devices->device_list_mutex); | |
7276 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7277 | device->fs_info = fs_info; |
29cc83f6 LB |
7278 | mutex_unlock(&fs_devices->device_list_mutex); |
7279 | ||
7280 | fs_devices = fs_devices->seed; | |
7281 | } | |
cb517eab MX |
7282 | } |
7283 | ||
733f4fbb SB |
7284 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7285 | { | |
7286 | struct btrfs_key key; | |
733f4fbb SB |
7287 | struct btrfs_root *dev_root = fs_info->dev_root; |
7288 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7289 | struct extent_buffer *eb; | |
7290 | int slot; | |
7291 | int ret = 0; | |
7292 | struct btrfs_device *device; | |
7293 | struct btrfs_path *path = NULL; | |
7294 | int i; | |
7295 | ||
7296 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7297 | if (!path) |
7298 | return -ENOMEM; | |
733f4fbb SB |
7299 | |
7300 | mutex_lock(&fs_devices->device_list_mutex); | |
7301 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7302 | int item_size; | |
7303 | struct btrfs_dev_stats_item *ptr; | |
7304 | ||
242e2956 DS |
7305 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7306 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7307 | key.offset = device->devid; |
7308 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7309 | if (ret) { | |
ae4b9b4c AJ |
7310 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7311 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7312 | device->dev_stats_valid = 1; |
7313 | btrfs_release_path(path); | |
7314 | continue; | |
7315 | } | |
7316 | slot = path->slots[0]; | |
7317 | eb = path->nodes[0]; | |
733f4fbb SB |
7318 | item_size = btrfs_item_size_nr(eb, slot); |
7319 | ||
7320 | ptr = btrfs_item_ptr(eb, slot, | |
7321 | struct btrfs_dev_stats_item); | |
7322 | ||
7323 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7324 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7325 | btrfs_dev_stat_set(device, i, | |
7326 | btrfs_dev_stats_value(eb, ptr, i)); | |
7327 | else | |
4e411a7d | 7328 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7329 | } |
7330 | ||
7331 | device->dev_stats_valid = 1; | |
7332 | btrfs_dev_stat_print_on_load(device); | |
7333 | btrfs_release_path(path); | |
7334 | } | |
7335 | mutex_unlock(&fs_devices->device_list_mutex); | |
7336 | ||
733f4fbb SB |
7337 | btrfs_free_path(path); |
7338 | return ret < 0 ? ret : 0; | |
7339 | } | |
7340 | ||
7341 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7342 | struct btrfs_device *device) |
7343 | { | |
5495f195 | 7344 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7345 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7346 | struct btrfs_path *path; |
7347 | struct btrfs_key key; | |
7348 | struct extent_buffer *eb; | |
7349 | struct btrfs_dev_stats_item *ptr; | |
7350 | int ret; | |
7351 | int i; | |
7352 | ||
242e2956 DS |
7353 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7354 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7355 | key.offset = device->devid; |
7356 | ||
7357 | path = btrfs_alloc_path(); | |
fa252992 DS |
7358 | if (!path) |
7359 | return -ENOMEM; | |
733f4fbb SB |
7360 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7361 | if (ret < 0) { | |
0b246afa | 7362 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7363 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7364 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7365 | goto out; |
7366 | } | |
7367 | ||
7368 | if (ret == 0 && | |
7369 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7370 | /* need to delete old one and insert a new one */ | |
7371 | ret = btrfs_del_item(trans, dev_root, path); | |
7372 | if (ret != 0) { | |
0b246afa | 7373 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7374 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7375 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7376 | goto out; |
7377 | } | |
7378 | ret = 1; | |
7379 | } | |
7380 | ||
7381 | if (ret == 1) { | |
7382 | /* need to insert a new item */ | |
7383 | btrfs_release_path(path); | |
7384 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7385 | &key, sizeof(*ptr)); | |
7386 | if (ret < 0) { | |
0b246afa | 7387 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7388 | "insert dev_stats item for device %s failed %d", |
7389 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7390 | goto out; |
7391 | } | |
7392 | } | |
7393 | ||
7394 | eb = path->nodes[0]; | |
7395 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7396 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7397 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7398 | btrfs_dev_stat_read(device, i)); | |
7399 | btrfs_mark_buffer_dirty(eb); | |
7400 | ||
7401 | out: | |
7402 | btrfs_free_path(path); | |
7403 | return ret; | |
7404 | } | |
7405 | ||
7406 | /* | |
7407 | * called from commit_transaction. Writes all changed device stats to disk. | |
7408 | */ | |
196c9d8d | 7409 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7410 | { |
196c9d8d | 7411 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7412 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7413 | struct btrfs_device *device; | |
addc3fa7 | 7414 | int stats_cnt; |
733f4fbb SB |
7415 | int ret = 0; |
7416 | ||
7417 | mutex_lock(&fs_devices->device_list_mutex); | |
7418 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7419 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7420 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7421 | continue; |
7422 | ||
9deae968 NB |
7423 | |
7424 | /* | |
7425 | * There is a LOAD-LOAD control dependency between the value of | |
7426 | * dev_stats_ccnt and updating the on-disk values which requires | |
7427 | * reading the in-memory counters. Such control dependencies | |
7428 | * require explicit read memory barriers. | |
7429 | * | |
7430 | * This memory barriers pairs with smp_mb__before_atomic in | |
7431 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7432 | * barrier implied by atomic_xchg in | |
7433 | * btrfs_dev_stats_read_and_reset | |
7434 | */ | |
7435 | smp_rmb(); | |
7436 | ||
5495f195 | 7437 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7438 | if (!ret) |
addc3fa7 | 7439 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7440 | } |
7441 | mutex_unlock(&fs_devices->device_list_mutex); | |
7442 | ||
7443 | return ret; | |
7444 | } | |
7445 | ||
442a4f63 SB |
7446 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7447 | { | |
7448 | btrfs_dev_stat_inc(dev, index); | |
7449 | btrfs_dev_stat_print_on_error(dev); | |
7450 | } | |
7451 | ||
48a3b636 | 7452 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7453 | { |
733f4fbb SB |
7454 | if (!dev->dev_stats_valid) |
7455 | return; | |
fb456252 | 7456 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7457 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7458 | rcu_str_deref(dev->name), |
442a4f63 SB |
7459 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7460 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7461 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7462 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7463 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7464 | } |
c11d2c23 | 7465 | |
733f4fbb SB |
7466 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7467 | { | |
a98cdb85 SB |
7468 | int i; |
7469 | ||
7470 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7471 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7472 | break; | |
7473 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7474 | return; /* all values == 0, suppress message */ | |
7475 | ||
fb456252 | 7476 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7477 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7478 | rcu_str_deref(dev->name), |
733f4fbb SB |
7479 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7480 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7481 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7482 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7483 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7484 | } | |
7485 | ||
2ff7e61e | 7486 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7487 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7488 | { |
7489 | struct btrfs_device *dev; | |
0b246afa | 7490 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7491 | int i; |
7492 | ||
7493 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7494 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7495 | true); | |
c11d2c23 SB |
7496 | mutex_unlock(&fs_devices->device_list_mutex); |
7497 | ||
7498 | if (!dev) { | |
0b246afa | 7499 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7500 | return -ENODEV; |
733f4fbb | 7501 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7502 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7503 | return -ENODEV; |
b27f7c0c | 7504 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7505 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7506 | if (stats->nr_items > i) | |
7507 | stats->values[i] = | |
7508 | btrfs_dev_stat_read_and_reset(dev, i); | |
7509 | else | |
4e411a7d | 7510 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 SB |
7511 | } |
7512 | } else { | |
7513 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7514 | if (stats->nr_items > i) | |
7515 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7516 | } | |
7517 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7518 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7519 | return 0; | |
7520 | } | |
a8a6dab7 | 7521 | |
da353f6b | 7522 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7523 | { |
7524 | struct buffer_head *bh; | |
7525 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7526 | int copy_num; |
a8a6dab7 | 7527 | |
12b1c263 AJ |
7528 | if (!bdev) |
7529 | return; | |
a8a6dab7 | 7530 | |
12b1c263 AJ |
7531 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7532 | copy_num++) { | |
a8a6dab7 | 7533 | |
12b1c263 AJ |
7534 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7535 | continue; | |
7536 | ||
7537 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7538 | ||
7539 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7540 | set_buffer_dirty(bh); | |
7541 | sync_dirty_buffer(bh); | |
7542 | brelse(bh); | |
7543 | } | |
7544 | ||
7545 | /* Notify udev that device has changed */ | |
7546 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7547 | ||
7548 | /* Update ctime/mtime for device path for libblkid */ | |
7549 | update_dev_time(device_path); | |
a8a6dab7 | 7550 | } |
935e5cc9 MX |
7551 | |
7552 | /* | |
bbbf7243 NB |
7553 | * Update the size and bytes used for each device where it changed. This is |
7554 | * delayed since we would otherwise get errors while writing out the | |
7555 | * superblocks. | |
7556 | * | |
7557 | * Must be invoked during transaction commit. | |
935e5cc9 | 7558 | */ |
bbbf7243 | 7559 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7560 | { |
935e5cc9 MX |
7561 | struct btrfs_device *curr, *next; |
7562 | ||
bbbf7243 | 7563 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7564 | |
bbbf7243 | 7565 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7566 | return; |
7567 | ||
bbbf7243 NB |
7568 | /* |
7569 | * We don't need the device_list_mutex here. This list is owned by the | |
7570 | * transaction and the transaction must complete before the device is | |
7571 | * released. | |
7572 | */ | |
7573 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7574 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7575 | post_commit_list) { | |
7576 | list_del_init(&curr->post_commit_list); | |
7577 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7578 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7579 | } |
bbbf7243 | 7580 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7581 | } |
5a13f430 AJ |
7582 | |
7583 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7584 | { | |
7585 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7586 | while (fs_devices) { | |
7587 | fs_devices->fs_info = fs_info; | |
7588 | fs_devices = fs_devices->seed; | |
7589 | } | |
7590 | } | |
7591 | ||
7592 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7593 | { | |
7594 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7595 | while (fs_devices) { | |
7596 | fs_devices->fs_info = NULL; | |
7597 | fs_devices = fs_devices->seed; | |
7598 | } | |
7599 | } | |
46df06b8 DS |
7600 | |
7601 | /* | |
7602 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7603 | */ | |
7604 | int btrfs_bg_type_to_factor(u64 flags) | |
7605 | { | |
44b28ada DS |
7606 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7607 | ||
7608 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7609 | } |
cf90d884 QW |
7610 | |
7611 | ||
cf90d884 QW |
7612 | |
7613 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7614 | u64 chunk_offset, u64 devid, | |
7615 | u64 physical_offset, u64 physical_len) | |
7616 | { | |
c8bf1b67 | 7617 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7618 | struct extent_map *em; |
7619 | struct map_lookup *map; | |
05a37c48 | 7620 | struct btrfs_device *dev; |
cf90d884 QW |
7621 | u64 stripe_len; |
7622 | bool found = false; | |
7623 | int ret = 0; | |
7624 | int i; | |
7625 | ||
7626 | read_lock(&em_tree->lock); | |
7627 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7628 | read_unlock(&em_tree->lock); | |
7629 | ||
7630 | if (!em) { | |
7631 | btrfs_err(fs_info, | |
7632 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7633 | physical_offset, devid); | |
7634 | ret = -EUCLEAN; | |
7635 | goto out; | |
7636 | } | |
7637 | ||
7638 | map = em->map_lookup; | |
7639 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7640 | if (physical_len != stripe_len) { | |
7641 | btrfs_err(fs_info, | |
7642 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7643 | physical_offset, devid, em->start, physical_len, | |
7644 | stripe_len); | |
7645 | ret = -EUCLEAN; | |
7646 | goto out; | |
7647 | } | |
7648 | ||
7649 | for (i = 0; i < map->num_stripes; i++) { | |
7650 | if (map->stripes[i].dev->devid == devid && | |
7651 | map->stripes[i].physical == physical_offset) { | |
7652 | found = true; | |
7653 | if (map->verified_stripes >= map->num_stripes) { | |
7654 | btrfs_err(fs_info, | |
7655 | "too many dev extents for chunk %llu found", | |
7656 | em->start); | |
7657 | ret = -EUCLEAN; | |
7658 | goto out; | |
7659 | } | |
7660 | map->verified_stripes++; | |
7661 | break; | |
7662 | } | |
7663 | } | |
7664 | if (!found) { | |
7665 | btrfs_err(fs_info, | |
7666 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7667 | physical_offset, devid); | |
7668 | ret = -EUCLEAN; | |
7669 | } | |
05a37c48 QW |
7670 | |
7671 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7672 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7673 | if (!dev) { |
7674 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7675 | ret = -EUCLEAN; | |
7676 | goto out; | |
7677 | } | |
1b3922a8 QW |
7678 | |
7679 | /* It's possible this device is a dummy for seed device */ | |
7680 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7681 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7682 | NULL, false); | |
1b3922a8 QW |
7683 | if (!dev) { |
7684 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7685 | devid); | |
7686 | ret = -EUCLEAN; | |
7687 | goto out; | |
7688 | } | |
7689 | } | |
7690 | ||
05a37c48 QW |
7691 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7692 | btrfs_err(fs_info, | |
7693 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7694 | devid, physical_offset, physical_len, | |
7695 | dev->disk_total_bytes); | |
7696 | ret = -EUCLEAN; | |
7697 | goto out; | |
7698 | } | |
cf90d884 QW |
7699 | out: |
7700 | free_extent_map(em); | |
7701 | return ret; | |
7702 | } | |
7703 | ||
7704 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7705 | { | |
c8bf1b67 | 7706 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7707 | struct extent_map *em; |
7708 | struct rb_node *node; | |
7709 | int ret = 0; | |
7710 | ||
7711 | read_lock(&em_tree->lock); | |
07e1ce09 | 7712 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7713 | em = rb_entry(node, struct extent_map, rb_node); |
7714 | if (em->map_lookup->num_stripes != | |
7715 | em->map_lookup->verified_stripes) { | |
7716 | btrfs_err(fs_info, | |
7717 | "chunk %llu has missing dev extent, have %d expect %d", | |
7718 | em->start, em->map_lookup->verified_stripes, | |
7719 | em->map_lookup->num_stripes); | |
7720 | ret = -EUCLEAN; | |
7721 | goto out; | |
7722 | } | |
7723 | } | |
7724 | out: | |
7725 | read_unlock(&em_tree->lock); | |
7726 | return ret; | |
7727 | } | |
7728 | ||
7729 | /* | |
7730 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7731 | * later chunk allocation/free would cause unexpected behavior. | |
7732 | * | |
7733 | * NOTE: This will iterate through the whole device tree, which should be of | |
7734 | * the same size level as the chunk tree. This slightly increases mount time. | |
7735 | */ | |
7736 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7737 | { | |
7738 | struct btrfs_path *path; | |
7739 | struct btrfs_root *root = fs_info->dev_root; | |
7740 | struct btrfs_key key; | |
5eb19381 QW |
7741 | u64 prev_devid = 0; |
7742 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7743 | int ret = 0; |
7744 | ||
7745 | key.objectid = 1; | |
7746 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7747 | key.offset = 0; | |
7748 | ||
7749 | path = btrfs_alloc_path(); | |
7750 | if (!path) | |
7751 | return -ENOMEM; | |
7752 | ||
7753 | path->reada = READA_FORWARD; | |
7754 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7755 | if (ret < 0) | |
7756 | goto out; | |
7757 | ||
7758 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7759 | ret = btrfs_next_item(root, path); | |
7760 | if (ret < 0) | |
7761 | goto out; | |
7762 | /* No dev extents at all? Not good */ | |
7763 | if (ret > 0) { | |
7764 | ret = -EUCLEAN; | |
7765 | goto out; | |
7766 | } | |
7767 | } | |
7768 | while (1) { | |
7769 | struct extent_buffer *leaf = path->nodes[0]; | |
7770 | struct btrfs_dev_extent *dext; | |
7771 | int slot = path->slots[0]; | |
7772 | u64 chunk_offset; | |
7773 | u64 physical_offset; | |
7774 | u64 physical_len; | |
7775 | u64 devid; | |
7776 | ||
7777 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7778 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7779 | break; | |
7780 | devid = key.objectid; | |
7781 | physical_offset = key.offset; | |
7782 | ||
7783 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7784 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7785 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7786 | ||
5eb19381 QW |
7787 | /* Check if this dev extent overlaps with the previous one */ |
7788 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7789 | btrfs_err(fs_info, | |
7790 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7791 | devid, physical_offset, prev_dev_ext_end); | |
7792 | ret = -EUCLEAN; | |
7793 | goto out; | |
7794 | } | |
7795 | ||
cf90d884 QW |
7796 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7797 | physical_offset, physical_len); | |
7798 | if (ret < 0) | |
7799 | goto out; | |
5eb19381 QW |
7800 | prev_devid = devid; |
7801 | prev_dev_ext_end = physical_offset + physical_len; | |
7802 | ||
cf90d884 QW |
7803 | ret = btrfs_next_item(root, path); |
7804 | if (ret < 0) | |
7805 | goto out; | |
7806 | if (ret > 0) { | |
7807 | ret = 0; | |
7808 | break; | |
7809 | } | |
7810 | } | |
7811 | ||
7812 | /* Ensure all chunks have corresponding dev extents */ | |
7813 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7814 | out: | |
7815 | btrfs_free_path(path); | |
7816 | return ret; | |
7817 | } | |
eede2bf3 OS |
7818 | |
7819 | /* | |
7820 | * Check whether the given block group or device is pinned by any inode being | |
7821 | * used as a swapfile. | |
7822 | */ | |
7823 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7824 | { | |
7825 | struct btrfs_swapfile_pin *sp; | |
7826 | struct rb_node *node; | |
7827 | ||
7828 | spin_lock(&fs_info->swapfile_pins_lock); | |
7829 | node = fs_info->swapfile_pins.rb_node; | |
7830 | while (node) { | |
7831 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7832 | if (ptr < sp->ptr) | |
7833 | node = node->rb_left; | |
7834 | else if (ptr > sp->ptr) | |
7835 | node = node->rb_right; | |
7836 | else | |
7837 | break; | |
7838 | } | |
7839 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7840 | return node != NULL; | |
7841 | } |