]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 CM |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
8a4b83cc | 9 | #include <linux/buffer_head.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
784352fe | 17 | #include "misc.h" |
0b86a832 CM |
18 | #include "ctree.h" |
19 | #include "extent_map.h" | |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "print-tree.h" | |
23 | #include "volumes.h" | |
53b381b3 | 24 | #include "raid56.h" |
8b712842 | 25 | #include "async-thread.h" |
21adbd5c | 26 | #include "check-integrity.h" |
606686ee | 27 | #include "rcu-string.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
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; | |
d2979aa2 | 1118 | int ret = 0; |
e4404d6e | 1119 | |
7239ff4b | 1120 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1121 | if (IS_ERR(fs_devices)) |
1122 | return fs_devices; | |
e4404d6e | 1123 | |
adbbb863 | 1124 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1125 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1126 | |
1127 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1128 | struct rcu_string *name; |
1129 | ||
12bd2fc0 ID |
1130 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1131 | orig_dev->uuid); | |
d2979aa2 AJ |
1132 | if (IS_ERR(device)) { |
1133 | ret = PTR_ERR(device); | |
e4404d6e | 1134 | goto error; |
d2979aa2 | 1135 | } |
e4404d6e | 1136 | |
606686ee JB |
1137 | /* |
1138 | * This is ok to do without rcu read locked because we hold the | |
1139 | * uuid mutex so nothing we touch in here is going to disappear. | |
1140 | */ | |
e755f780 | 1141 | if (orig_dev->name) { |
78f2c9e6 DS |
1142 | name = rcu_string_strdup(orig_dev->name->str, |
1143 | GFP_KERNEL); | |
e755f780 | 1144 | if (!name) { |
a425f9d4 | 1145 | btrfs_free_device(device); |
d2979aa2 | 1146 | ret = -ENOMEM; |
e755f780 AJ |
1147 | goto error; |
1148 | } | |
1149 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1150 | } |
e4404d6e | 1151 | |
e4404d6e YZ |
1152 | list_add(&device->dev_list, &fs_devices->devices); |
1153 | device->fs_devices = fs_devices; | |
1154 | fs_devices->num_devices++; | |
1155 | } | |
adbbb863 | 1156 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1157 | return fs_devices; |
1158 | error: | |
adbbb863 | 1159 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1160 | free_fs_devices(fs_devices); |
d2979aa2 | 1161 | return ERR_PTR(ret); |
e4404d6e YZ |
1162 | } |
1163 | ||
9b99b115 AJ |
1164 | /* |
1165 | * After we have read the system tree and know devids belonging to | |
1166 | * this filesystem, remove the device which does not belong there. | |
1167 | */ | |
1168 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1169 | { |
c6e30871 | 1170 | struct btrfs_device *device, *next; |
443f24fe | 1171 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1172 | |
dfe25020 CM |
1173 | mutex_lock(&uuid_mutex); |
1174 | again: | |
46224705 | 1175 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1176 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1177 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1178 | &device->dev_state)) { | |
401e29c1 AJ |
1179 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1180 | &device->dev_state) && | |
1181 | (!latest_dev || | |
1182 | device->generation > latest_dev->generation)) { | |
443f24fe | 1183 | latest_dev = device; |
a6b0d5c8 | 1184 | } |
2b82032c | 1185 | continue; |
a6b0d5c8 | 1186 | } |
2b82032c | 1187 | |
8dabb742 SB |
1188 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1189 | /* | |
1190 | * In the first step, keep the device which has | |
1191 | * the correct fsid and the devid that is used | |
1192 | * for the dev_replace procedure. | |
1193 | * In the second step, the dev_replace state is | |
1194 | * read from the device tree and it is known | |
1195 | * whether the procedure is really active or | |
1196 | * not, which means whether this device is | |
1197 | * used or whether it should be removed. | |
1198 | */ | |
401e29c1 AJ |
1199 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1200 | &device->dev_state)) { | |
8dabb742 SB |
1201 | continue; |
1202 | } | |
1203 | } | |
2b82032c | 1204 | if (device->bdev) { |
d4d77629 | 1205 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1206 | device->bdev = NULL; |
1207 | fs_devices->open_devices--; | |
1208 | } | |
ebbede42 | 1209 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1210 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1211 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1212 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1213 | &device->dev_state)) | |
8dabb742 | 1214 | fs_devices->rw_devices--; |
2b82032c | 1215 | } |
e4404d6e YZ |
1216 | list_del_init(&device->dev_list); |
1217 | fs_devices->num_devices--; | |
a425f9d4 | 1218 | btrfs_free_device(device); |
dfe25020 | 1219 | } |
2b82032c YZ |
1220 | |
1221 | if (fs_devices->seed) { | |
1222 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1223 | goto again; |
1224 | } | |
1225 | ||
443f24fe | 1226 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1227 | |
dfe25020 | 1228 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1229 | } |
a0af469b | 1230 | |
14238819 AJ |
1231 | static void btrfs_close_bdev(struct btrfs_device *device) |
1232 | { | |
08ffcae8 DS |
1233 | if (!device->bdev) |
1234 | return; | |
1235 | ||
ebbede42 | 1236 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1237 | sync_blockdev(device->bdev); |
1238 | invalidate_bdev(device->bdev); | |
1239 | } | |
1240 | ||
08ffcae8 | 1241 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1242 | } |
1243 | ||
959b1c04 | 1244 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1245 | { |
1246 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1247 | struct btrfs_device *new_device; | |
1248 | struct rcu_string *name; | |
1249 | ||
1250 | if (device->bdev) | |
1251 | fs_devices->open_devices--; | |
1252 | ||
ebbede42 | 1253 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1254 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1255 | list_del_init(&device->dev_alloc_list); | |
1256 | fs_devices->rw_devices--; | |
1257 | } | |
1258 | ||
e6e674bd | 1259 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1260 | fs_devices->missing_devices--; |
1261 | ||
959b1c04 NB |
1262 | btrfs_close_bdev(device); |
1263 | ||
f448341a AJ |
1264 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1265 | device->uuid); | |
1266 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1267 | ||
1268 | /* Safe because we are under uuid_mutex */ | |
1269 | if (device->name) { | |
1270 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1271 | BUG_ON(!name); /* -ENOMEM */ | |
1272 | rcu_assign_pointer(new_device->name, name); | |
1273 | } | |
1274 | ||
1275 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1276 | new_device->fs_devices = device->fs_devices; | |
959b1c04 | 1277 | |
8e75fd89 NB |
1278 | synchronize_rcu(); |
1279 | btrfs_free_device(device); | |
f448341a AJ |
1280 | } |
1281 | ||
0226e0eb | 1282 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1283 | { |
2037a093 | 1284 | struct btrfs_device *device, *tmp; |
e4404d6e | 1285 | |
2b82032c YZ |
1286 | if (--fs_devices->opened > 0) |
1287 | return 0; | |
8a4b83cc | 1288 | |
c9513edb | 1289 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1290 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1291 | btrfs_close_one_device(device); |
8a4b83cc | 1292 | } |
c9513edb XG |
1293 | mutex_unlock(&fs_devices->device_list_mutex); |
1294 | ||
e4404d6e YZ |
1295 | WARN_ON(fs_devices->open_devices); |
1296 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1297 | fs_devices->opened = 0; |
1298 | fs_devices->seeding = 0; | |
2b82032c | 1299 | |
8a4b83cc CM |
1300 | return 0; |
1301 | } | |
1302 | ||
2b82032c YZ |
1303 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1304 | { | |
e4404d6e | 1305 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1306 | int ret; |
1307 | ||
1308 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1309 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1310 | if (!fs_devices->opened) { |
1311 | seed_devices = fs_devices->seed; | |
1312 | fs_devices->seed = NULL; | |
1313 | } | |
2b82032c | 1314 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1315 | |
1316 | while (seed_devices) { | |
1317 | fs_devices = seed_devices; | |
1318 | seed_devices = fs_devices->seed; | |
0226e0eb | 1319 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1320 | free_fs_devices(fs_devices); |
1321 | } | |
2b82032c YZ |
1322 | return ret; |
1323 | } | |
1324 | ||
897fb573 | 1325 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1326 | fmode_t flags, void *holder) |
8a4b83cc | 1327 | { |
8a4b83cc | 1328 | struct btrfs_device *device; |
443f24fe | 1329 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1330 | int ret = 0; |
8a4b83cc | 1331 | |
d4d77629 TH |
1332 | flags |= FMODE_EXCL; |
1333 | ||
f117e290 | 1334 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1335 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1336 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1337 | continue; |
a0af469b | 1338 | |
9f050db4 AJ |
1339 | if (!latest_dev || |
1340 | device->generation > latest_dev->generation) | |
1341 | latest_dev = device; | |
8a4b83cc | 1342 | } |
a0af469b | 1343 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1344 | ret = -EINVAL; |
a0af469b CM |
1345 | goto out; |
1346 | } | |
2b82032c | 1347 | fs_devices->opened = 1; |
443f24fe | 1348 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1349 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1350 | out: |
2b82032c YZ |
1351 | return ret; |
1352 | } | |
1353 | ||
f8e10cd3 AJ |
1354 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1355 | { | |
1356 | struct btrfs_device *dev1, *dev2; | |
1357 | ||
1358 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1359 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1360 | ||
1361 | if (dev1->devid < dev2->devid) | |
1362 | return -1; | |
1363 | else if (dev1->devid > dev2->devid) | |
1364 | return 1; | |
1365 | return 0; | |
1366 | } | |
1367 | ||
2b82032c | 1368 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1369 | fmode_t flags, void *holder) |
2b82032c YZ |
1370 | { |
1371 | int ret; | |
1372 | ||
f5194e34 DS |
1373 | lockdep_assert_held(&uuid_mutex); |
1374 | ||
542c5908 | 1375 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1376 | if (fs_devices->opened) { |
e4404d6e YZ |
1377 | fs_devices->opened++; |
1378 | ret = 0; | |
2b82032c | 1379 | } else { |
f8e10cd3 | 1380 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1381 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1382 | } |
542c5908 AJ |
1383 | mutex_unlock(&fs_devices->device_list_mutex); |
1384 | ||
8a4b83cc CM |
1385 | return ret; |
1386 | } | |
1387 | ||
c9162bdf | 1388 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1389 | { |
1390 | kunmap(page); | |
1391 | put_page(page); | |
1392 | } | |
1393 | ||
c9162bdf OS |
1394 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1395 | struct page **page, | |
1396 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1397 | { |
1398 | void *p; | |
1399 | pgoff_t index; | |
1400 | ||
1401 | /* make sure our super fits in the device */ | |
1402 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1403 | return 1; | |
1404 | ||
1405 | /* make sure our super fits in the page */ | |
1406 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1407 | return 1; | |
1408 | ||
1409 | /* make sure our super doesn't straddle pages on disk */ | |
1410 | index = bytenr >> PAGE_SHIFT; | |
1411 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1412 | return 1; | |
1413 | ||
1414 | /* pull in the page with our super */ | |
1415 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1416 | index, GFP_KERNEL); | |
1417 | ||
1418 | if (IS_ERR_OR_NULL(*page)) | |
1419 | return 1; | |
1420 | ||
1421 | p = kmap(*page); | |
1422 | ||
1423 | /* align our pointer to the offset of the super block */ | |
7073017a | 1424 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1425 | |
1426 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1427 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1428 | btrfs_release_disk_super(*page); | |
1429 | return 1; | |
1430 | } | |
1431 | ||
1432 | if ((*disk_super)->label[0] && | |
1433 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1434 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1435 | ||
1436 | return 0; | |
1437 | } | |
1438 | ||
228a73ab AJ |
1439 | int btrfs_forget_devices(const char *path) |
1440 | { | |
1441 | int ret; | |
1442 | ||
1443 | mutex_lock(&uuid_mutex); | |
1444 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1445 | mutex_unlock(&uuid_mutex); | |
1446 | ||
1447 | return ret; | |
1448 | } | |
1449 | ||
6f60cbd3 DS |
1450 | /* |
1451 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1452 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1453 | * is read via pagecache | |
1454 | */ | |
36350e95 GJ |
1455 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1456 | void *holder) | |
8a4b83cc CM |
1457 | { |
1458 | struct btrfs_super_block *disk_super; | |
4306a974 | 1459 | bool new_device_added = false; |
36350e95 | 1460 | struct btrfs_device *device = NULL; |
8a4b83cc | 1461 | struct block_device *bdev; |
6f60cbd3 | 1462 | struct page *page; |
6f60cbd3 | 1463 | u64 bytenr; |
8a4b83cc | 1464 | |
899f9307 DS |
1465 | lockdep_assert_held(&uuid_mutex); |
1466 | ||
6f60cbd3 DS |
1467 | /* |
1468 | * we would like to check all the supers, but that would make | |
1469 | * a btrfs mount succeed after a mkfs from a different FS. | |
1470 | * So, we need to add a special mount option to scan for | |
1471 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1472 | */ | |
1473 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1474 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1475 | |
1476 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1477 | if (IS_ERR(bdev)) |
36350e95 | 1478 | return ERR_CAST(bdev); |
6f60cbd3 | 1479 | |
05a5c55d | 1480 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1481 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1482 | goto error_bdev_put; |
05a5c55d | 1483 | } |
6f60cbd3 | 1484 | |
4306a974 | 1485 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1486 | if (!IS_ERR(device)) { |
4306a974 AJ |
1487 | if (new_device_added) |
1488 | btrfs_free_stale_devices(path, device); | |
1489 | } | |
6f60cbd3 | 1490 | |
6cf86a00 | 1491 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1492 | |
1493 | error_bdev_put: | |
d4d77629 | 1494 | blkdev_put(bdev, flags); |
b6ed73bc | 1495 | |
36350e95 | 1496 | return device; |
8a4b83cc | 1497 | } |
0b86a832 | 1498 | |
1c11b63e JM |
1499 | /* |
1500 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1501 | * such is found, record the end of it in *start | |
1502 | */ | |
1c11b63e JM |
1503 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1504 | u64 len) | |
6df9a95e | 1505 | { |
1c11b63e | 1506 | u64 physical_start, physical_end; |
6df9a95e | 1507 | |
1c11b63e | 1508 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1509 | |
1c11b63e JM |
1510 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1511 | &physical_start, &physical_end, | |
1512 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1513 | |
1c11b63e JM |
1514 | if (in_range(physical_start, *start, len) || |
1515 | in_range(*start, physical_start, | |
1516 | physical_end - physical_start)) { | |
1517 | *start = physical_end + 1; | |
1518 | return true; | |
6df9a95e JB |
1519 | } |
1520 | } | |
1c11b63e | 1521 | return false; |
6df9a95e JB |
1522 | } |
1523 | ||
1524 | ||
0b86a832 | 1525 | /* |
499f377f JM |
1526 | * find_free_dev_extent_start - find free space in the specified device |
1527 | * @device: the device which we search the free space in | |
1528 | * @num_bytes: the size of the free space that we need | |
1529 | * @search_start: the position from which to begin the search | |
1530 | * @start: store the start of the free space. | |
1531 | * @len: the size of the free space. that we find, or the size | |
1532 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1533 | * |
0b86a832 CM |
1534 | * this uses a pretty simple search, the expectation is that it is |
1535 | * called very infrequently and that a given device has a small number | |
1536 | * of extents | |
7bfc837d MX |
1537 | * |
1538 | * @start is used to store the start of the free space if we find. But if we | |
1539 | * don't find suitable free space, it will be used to store the start position | |
1540 | * of the max free space. | |
1541 | * | |
1542 | * @len is used to store the size of the free space that we find. | |
1543 | * But if we don't find suitable free space, it is used to store the size of | |
1544 | * the max free space. | |
135da976 QW |
1545 | * |
1546 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1547 | * check to ensure dev extents are not double allocated. | |
1548 | * This makes the function safe to allocate dev extents but may not report | |
1549 | * correct usable device space, as device extent freed in current transaction | |
1550 | * is not reported as avaiable. | |
0b86a832 | 1551 | */ |
9e3246a5 QW |
1552 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1553 | u64 num_bytes, u64 search_start, u64 *start, | |
1554 | u64 *len) | |
0b86a832 | 1555 | { |
0b246afa JM |
1556 | struct btrfs_fs_info *fs_info = device->fs_info; |
1557 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1558 | struct btrfs_key key; |
7bfc837d | 1559 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1560 | struct btrfs_path *path; |
7bfc837d MX |
1561 | u64 hole_size; |
1562 | u64 max_hole_start; | |
1563 | u64 max_hole_size; | |
1564 | u64 extent_end; | |
0b86a832 CM |
1565 | u64 search_end = device->total_bytes; |
1566 | int ret; | |
7bfc837d | 1567 | int slot; |
0b86a832 | 1568 | struct extent_buffer *l; |
8cdc7c5b FM |
1569 | |
1570 | /* | |
1571 | * We don't want to overwrite the superblock on the drive nor any area | |
1572 | * used by the boot loader (grub for example), so we make sure to start | |
1573 | * at an offset of at least 1MB. | |
1574 | */ | |
0d0c71b3 | 1575 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1576 | |
6df9a95e JB |
1577 | path = btrfs_alloc_path(); |
1578 | if (!path) | |
1579 | return -ENOMEM; | |
f2ab7618 | 1580 | |
7bfc837d MX |
1581 | max_hole_start = search_start; |
1582 | max_hole_size = 0; | |
1583 | ||
f2ab7618 | 1584 | again: |
401e29c1 AJ |
1585 | if (search_start >= search_end || |
1586 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1587 | ret = -ENOSPC; |
6df9a95e | 1588 | goto out; |
7bfc837d MX |
1589 | } |
1590 | ||
e4058b54 | 1591 | path->reada = READA_FORWARD; |
6df9a95e JB |
1592 | path->search_commit_root = 1; |
1593 | path->skip_locking = 1; | |
7bfc837d | 1594 | |
0b86a832 CM |
1595 | key.objectid = device->devid; |
1596 | key.offset = search_start; | |
1597 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1598 | |
125ccb0a | 1599 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1600 | if (ret < 0) |
7bfc837d | 1601 | goto out; |
1fcbac58 YZ |
1602 | if (ret > 0) { |
1603 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1604 | if (ret < 0) | |
7bfc837d | 1605 | goto out; |
1fcbac58 | 1606 | } |
7bfc837d | 1607 | |
0b86a832 CM |
1608 | while (1) { |
1609 | l = path->nodes[0]; | |
1610 | slot = path->slots[0]; | |
1611 | if (slot >= btrfs_header_nritems(l)) { | |
1612 | ret = btrfs_next_leaf(root, path); | |
1613 | if (ret == 0) | |
1614 | continue; | |
1615 | if (ret < 0) | |
7bfc837d MX |
1616 | goto out; |
1617 | ||
1618 | break; | |
0b86a832 CM |
1619 | } |
1620 | btrfs_item_key_to_cpu(l, &key, slot); | |
1621 | ||
1622 | if (key.objectid < device->devid) | |
1623 | goto next; | |
1624 | ||
1625 | if (key.objectid > device->devid) | |
7bfc837d | 1626 | break; |
0b86a832 | 1627 | |
962a298f | 1628 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1629 | goto next; |
9779b72f | 1630 | |
7bfc837d MX |
1631 | if (key.offset > search_start) { |
1632 | hole_size = key.offset - search_start; | |
9779b72f | 1633 | |
6df9a95e JB |
1634 | /* |
1635 | * Have to check before we set max_hole_start, otherwise | |
1636 | * we could end up sending back this offset anyway. | |
1637 | */ | |
1c11b63e | 1638 | if (contains_pending_extent(device, &search_start, |
1b984508 | 1639 | hole_size)) { |
1c11b63e | 1640 | if (key.offset >= search_start) |
1b984508 | 1641 | hole_size = key.offset - search_start; |
1c11b63e | 1642 | else |
1b984508 | 1643 | hole_size = 0; |
1b984508 | 1644 | } |
6df9a95e | 1645 | |
7bfc837d MX |
1646 | if (hole_size > max_hole_size) { |
1647 | max_hole_start = search_start; | |
1648 | max_hole_size = hole_size; | |
1649 | } | |
9779b72f | 1650 | |
7bfc837d MX |
1651 | /* |
1652 | * If this free space is greater than which we need, | |
1653 | * it must be the max free space that we have found | |
1654 | * until now, so max_hole_start must point to the start | |
1655 | * of this free space and the length of this free space | |
1656 | * is stored in max_hole_size. Thus, we return | |
1657 | * max_hole_start and max_hole_size and go back to the | |
1658 | * caller. | |
1659 | */ | |
1660 | if (hole_size >= num_bytes) { | |
1661 | ret = 0; | |
1662 | goto out; | |
0b86a832 CM |
1663 | } |
1664 | } | |
0b86a832 | 1665 | |
0b86a832 | 1666 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1667 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1668 | dev_extent); | |
1669 | if (extent_end > search_start) | |
1670 | search_start = extent_end; | |
0b86a832 CM |
1671 | next: |
1672 | path->slots[0]++; | |
1673 | cond_resched(); | |
1674 | } | |
0b86a832 | 1675 | |
38c01b96 | 1676 | /* |
1677 | * At this point, search_start should be the end of | |
1678 | * allocated dev extents, and when shrinking the device, | |
1679 | * search_end may be smaller than search_start. | |
1680 | */ | |
f2ab7618 | 1681 | if (search_end > search_start) { |
38c01b96 | 1682 | hole_size = search_end - search_start; |
1683 | ||
1c11b63e | 1684 | if (contains_pending_extent(device, &search_start, hole_size)) { |
f2ab7618 ZL |
1685 | btrfs_release_path(path); |
1686 | goto again; | |
1687 | } | |
0b86a832 | 1688 | |
f2ab7618 ZL |
1689 | if (hole_size > max_hole_size) { |
1690 | max_hole_start = search_start; | |
1691 | max_hole_size = hole_size; | |
1692 | } | |
6df9a95e JB |
1693 | } |
1694 | ||
7bfc837d | 1695 | /* See above. */ |
f2ab7618 | 1696 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1697 | ret = -ENOSPC; |
1698 | else | |
1699 | ret = 0; | |
1700 | ||
1701 | out: | |
2b82032c | 1702 | btrfs_free_path(path); |
7bfc837d | 1703 | *start = max_hole_start; |
b2117a39 | 1704 | if (len) |
7bfc837d | 1705 | *len = max_hole_size; |
0b86a832 CM |
1706 | return ret; |
1707 | } | |
1708 | ||
60dfdf25 | 1709 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1710 | u64 *start, u64 *len) |
1711 | { | |
499f377f | 1712 | /* FIXME use last free of some kind */ |
60dfdf25 | 1713 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1714 | } |
1715 | ||
b2950863 | 1716 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1717 | struct btrfs_device *device, |
2196d6e8 | 1718 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1719 | { |
0b246afa JM |
1720 | struct btrfs_fs_info *fs_info = device->fs_info; |
1721 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1722 | int ret; |
1723 | struct btrfs_path *path; | |
8f18cf13 | 1724 | struct btrfs_key key; |
a061fc8d CM |
1725 | struct btrfs_key found_key; |
1726 | struct extent_buffer *leaf = NULL; | |
1727 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1728 | |
1729 | path = btrfs_alloc_path(); | |
1730 | if (!path) | |
1731 | return -ENOMEM; | |
1732 | ||
1733 | key.objectid = device->devid; | |
1734 | key.offset = start; | |
1735 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1736 | again: |
8f18cf13 | 1737 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1738 | if (ret > 0) { |
1739 | ret = btrfs_previous_item(root, path, key.objectid, | |
1740 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1741 | if (ret) |
1742 | goto out; | |
a061fc8d CM |
1743 | leaf = path->nodes[0]; |
1744 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1745 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1746 | struct btrfs_dev_extent); | |
1747 | BUG_ON(found_key.offset > start || found_key.offset + | |
1748 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1749 | key = found_key; |
1750 | btrfs_release_path(path); | |
1751 | goto again; | |
a061fc8d CM |
1752 | } else if (ret == 0) { |
1753 | leaf = path->nodes[0]; | |
1754 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1755 | struct btrfs_dev_extent); | |
79787eaa | 1756 | } else { |
0b246afa | 1757 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1758 | goto out; |
a061fc8d | 1759 | } |
8f18cf13 | 1760 | |
2196d6e8 MX |
1761 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1762 | ||
8f18cf13 | 1763 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1764 | if (ret) { |
0b246afa JM |
1765 | btrfs_handle_fs_error(fs_info, ret, |
1766 | "Failed to remove dev extent item"); | |
13212b54 | 1767 | } else { |
3204d33c | 1768 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1769 | } |
b0b802d7 | 1770 | out: |
8f18cf13 CM |
1771 | btrfs_free_path(path); |
1772 | return ret; | |
1773 | } | |
1774 | ||
48a3b636 ES |
1775 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1776 | struct btrfs_device *device, | |
48a3b636 | 1777 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1778 | { |
1779 | int ret; | |
1780 | struct btrfs_path *path; | |
0b246afa JM |
1781 | struct btrfs_fs_info *fs_info = device->fs_info; |
1782 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1783 | struct btrfs_dev_extent *extent; |
1784 | struct extent_buffer *leaf; | |
1785 | struct btrfs_key key; | |
1786 | ||
e12c9621 | 1787 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1788 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1789 | path = btrfs_alloc_path(); |
1790 | if (!path) | |
1791 | return -ENOMEM; | |
1792 | ||
0b86a832 | 1793 | key.objectid = device->devid; |
2b82032c | 1794 | key.offset = start; |
0b86a832 CM |
1795 | key.type = BTRFS_DEV_EXTENT_KEY; |
1796 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1797 | sizeof(*extent)); | |
2cdcecbc MF |
1798 | if (ret) |
1799 | goto out; | |
0b86a832 CM |
1800 | |
1801 | leaf = path->nodes[0]; | |
1802 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1803 | struct btrfs_dev_extent); | |
b5d9071c NB |
1804 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1805 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1806 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1807 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1808 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1809 | ||
0b86a832 CM |
1810 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1811 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1812 | out: |
0b86a832 CM |
1813 | btrfs_free_path(path); |
1814 | return ret; | |
1815 | } | |
1816 | ||
6df9a95e | 1817 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1818 | { |
6df9a95e JB |
1819 | struct extent_map_tree *em_tree; |
1820 | struct extent_map *em; | |
1821 | struct rb_node *n; | |
1822 | u64 ret = 0; | |
0b86a832 | 1823 | |
c8bf1b67 | 1824 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1825 | read_lock(&em_tree->lock); |
07e1ce09 | 1826 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1827 | if (n) { |
1828 | em = rb_entry(n, struct extent_map, rb_node); | |
1829 | ret = em->start + em->len; | |
0b86a832 | 1830 | } |
6df9a95e JB |
1831 | read_unlock(&em_tree->lock); |
1832 | ||
0b86a832 CM |
1833 | return ret; |
1834 | } | |
1835 | ||
53f10659 ID |
1836 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1837 | u64 *devid_ret) | |
0b86a832 CM |
1838 | { |
1839 | int ret; | |
1840 | struct btrfs_key key; | |
1841 | struct btrfs_key found_key; | |
2b82032c YZ |
1842 | struct btrfs_path *path; |
1843 | ||
2b82032c YZ |
1844 | path = btrfs_alloc_path(); |
1845 | if (!path) | |
1846 | return -ENOMEM; | |
0b86a832 CM |
1847 | |
1848 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1849 | key.type = BTRFS_DEV_ITEM_KEY; | |
1850 | key.offset = (u64)-1; | |
1851 | ||
53f10659 | 1852 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1853 | if (ret < 0) |
1854 | goto error; | |
1855 | ||
a06dee4d AJ |
1856 | if (ret == 0) { |
1857 | /* Corruption */ | |
1858 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1859 | ret = -EUCLEAN; | |
1860 | goto error; | |
1861 | } | |
0b86a832 | 1862 | |
53f10659 ID |
1863 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1864 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1865 | BTRFS_DEV_ITEM_KEY); |
1866 | if (ret) { | |
53f10659 | 1867 | *devid_ret = 1; |
0b86a832 CM |
1868 | } else { |
1869 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1870 | path->slots[0]); | |
53f10659 | 1871 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1872 | } |
1873 | ret = 0; | |
1874 | error: | |
2b82032c | 1875 | btrfs_free_path(path); |
0b86a832 CM |
1876 | return ret; |
1877 | } | |
1878 | ||
1879 | /* | |
1880 | * the device information is stored in the chunk root | |
1881 | * the btrfs_device struct should be fully filled in | |
1882 | */ | |
c74a0b02 | 1883 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1884 | struct btrfs_device *device) |
0b86a832 CM |
1885 | { |
1886 | int ret; | |
1887 | struct btrfs_path *path; | |
1888 | struct btrfs_dev_item *dev_item; | |
1889 | struct extent_buffer *leaf; | |
1890 | struct btrfs_key key; | |
1891 | unsigned long ptr; | |
0b86a832 | 1892 | |
0b86a832 CM |
1893 | path = btrfs_alloc_path(); |
1894 | if (!path) | |
1895 | return -ENOMEM; | |
1896 | ||
0b86a832 CM |
1897 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1898 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1899 | key.offset = device->devid; |
0b86a832 | 1900 | |
8e87e856 NB |
1901 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1902 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1903 | if (ret) |
1904 | goto out; | |
1905 | ||
1906 | leaf = path->nodes[0]; | |
1907 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1908 | ||
1909 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1910 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1911 | btrfs_set_device_type(leaf, dev_item, device->type); |
1912 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1913 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1914 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1915 | btrfs_set_device_total_bytes(leaf, dev_item, |
1916 | btrfs_device_get_disk_total_bytes(device)); | |
1917 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1918 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1919 | btrfs_set_device_group(leaf, dev_item, 0); |
1920 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1921 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1922 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1923 | |
410ba3a2 | 1924 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1925 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1926 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1927 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1928 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1929 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1930 | |
2b82032c | 1931 | ret = 0; |
0b86a832 CM |
1932 | out: |
1933 | btrfs_free_path(path); | |
1934 | return ret; | |
1935 | } | |
8f18cf13 | 1936 | |
5a1972bd QW |
1937 | /* |
1938 | * Function to update ctime/mtime for a given device path. | |
1939 | * Mainly used for ctime/mtime based probe like libblkid. | |
1940 | */ | |
da353f6b | 1941 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1942 | { |
1943 | struct file *filp; | |
1944 | ||
1945 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1946 | if (IS_ERR(filp)) |
5a1972bd QW |
1947 | return; |
1948 | file_update_time(filp); | |
1949 | filp_close(filp, NULL); | |
5a1972bd QW |
1950 | } |
1951 | ||
f331a952 | 1952 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1953 | { |
f331a952 | 1954 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1955 | int ret; |
1956 | struct btrfs_path *path; | |
a061fc8d | 1957 | struct btrfs_key key; |
a061fc8d CM |
1958 | struct btrfs_trans_handle *trans; |
1959 | ||
a061fc8d CM |
1960 | path = btrfs_alloc_path(); |
1961 | if (!path) | |
1962 | return -ENOMEM; | |
1963 | ||
a22285a6 | 1964 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1965 | if (IS_ERR(trans)) { |
1966 | btrfs_free_path(path); | |
1967 | return PTR_ERR(trans); | |
1968 | } | |
a061fc8d CM |
1969 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1970 | key.type = BTRFS_DEV_ITEM_KEY; | |
1971 | key.offset = device->devid; | |
1972 | ||
1973 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1974 | if (ret) { |
1975 | if (ret > 0) | |
1976 | ret = -ENOENT; | |
1977 | btrfs_abort_transaction(trans, ret); | |
1978 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1979 | goto out; |
1980 | } | |
1981 | ||
1982 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1983 | if (ret) { |
1984 | btrfs_abort_transaction(trans, ret); | |
1985 | btrfs_end_transaction(trans); | |
1986 | } | |
1987 | ||
a061fc8d CM |
1988 | out: |
1989 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1990 | if (!ret) |
1991 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1992 | return ret; |
1993 | } | |
1994 | ||
3cc31a0d DS |
1995 | /* |
1996 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1997 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1998 | * replace. | |
1999 | */ | |
2000 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
2001 | u64 num_devices) | |
a061fc8d | 2002 | { |
a061fc8d | 2003 | u64 all_avail; |
de98ced9 | 2004 | unsigned seq; |
418775a2 | 2005 | int i; |
a061fc8d | 2006 | |
de98ced9 | 2007 | do { |
bd45ffbc | 2008 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2009 | |
bd45ffbc AJ |
2010 | all_avail = fs_info->avail_data_alloc_bits | |
2011 | fs_info->avail_system_alloc_bits | | |
2012 | fs_info->avail_metadata_alloc_bits; | |
2013 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2014 | |
418775a2 | 2015 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2016 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2017 | continue; |
a061fc8d | 2018 | |
418775a2 | 2019 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2020 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2021 | |
418775a2 DS |
2022 | if (ret) |
2023 | return ret; | |
2024 | } | |
53b381b3 DW |
2025 | } |
2026 | ||
bd45ffbc | 2027 | return 0; |
f1fa7f26 AJ |
2028 | } |
2029 | ||
c9162bdf OS |
2030 | static struct btrfs_device * btrfs_find_next_active_device( |
2031 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2032 | { |
2b82032c | 2033 | struct btrfs_device *next_device; |
88acff64 AJ |
2034 | |
2035 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2036 | if (next_device != device && | |
e6e674bd AJ |
2037 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2038 | && next_device->bdev) | |
88acff64 AJ |
2039 | return next_device; |
2040 | } | |
2041 | ||
2042 | return NULL; | |
2043 | } | |
2044 | ||
2045 | /* | |
2046 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2047 | * and replace it with the provided or the next active device, in the context | |
2048 | * where this function called, there should be always be another device (or | |
2049 | * this_dev) which is active. | |
2050 | */ | |
d6507cf1 NB |
2051 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2052 | struct btrfs_device *this_dev) | |
88acff64 | 2053 | { |
d6507cf1 | 2054 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2055 | struct btrfs_device *next_device; |
2056 | ||
2057 | if (this_dev) | |
2058 | next_device = this_dev; | |
2059 | else | |
2060 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2061 | device); | |
2062 | ASSERT(next_device); | |
2063 | ||
2064 | if (fs_info->sb->s_bdev && | |
2065 | (fs_info->sb->s_bdev == device->bdev)) | |
2066 | fs_info->sb->s_bdev = next_device->bdev; | |
2067 | ||
2068 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2069 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2070 | } | |
2071 | ||
1da73967 AJ |
2072 | /* |
2073 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2074 | * currently replaced. | |
2075 | */ | |
2076 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2077 | { | |
2078 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2079 | ||
cb5583dd | 2080 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2081 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2082 | ASSERT(num_devices > 1); | |
2083 | num_devices--; | |
2084 | } | |
cb5583dd | 2085 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2086 | |
2087 | return num_devices; | |
2088 | } | |
2089 | ||
da353f6b DS |
2090 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2091 | u64 devid) | |
f1fa7f26 AJ |
2092 | { |
2093 | struct btrfs_device *device; | |
1f78160c | 2094 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2095 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2096 | u64 num_devices; |
a061fc8d CM |
2097 | int ret = 0; |
2098 | ||
a061fc8d CM |
2099 | mutex_lock(&uuid_mutex); |
2100 | ||
1da73967 | 2101 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2102 | |
0b246afa | 2103 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2104 | if (ret) |
a061fc8d | 2105 | goto out; |
a061fc8d | 2106 | |
a27a94c2 NB |
2107 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2108 | ||
2109 | if (IS_ERR(device)) { | |
2110 | if (PTR_ERR(device) == -ENOENT && | |
2111 | strcmp(device_path, "missing") == 0) | |
2112 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2113 | else | |
2114 | ret = PTR_ERR(device); | |
53b381b3 | 2115 | goto out; |
a27a94c2 | 2116 | } |
dfe25020 | 2117 | |
eede2bf3 OS |
2118 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2119 | btrfs_warn_in_rcu(fs_info, | |
2120 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2121 | rcu_str_deref(device->name), device->devid); | |
2122 | ret = -ETXTBSY; | |
2123 | goto out; | |
2124 | } | |
2125 | ||
401e29c1 | 2126 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2127 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2128 | goto out; |
63a212ab SB |
2129 | } |
2130 | ||
ebbede42 AJ |
2131 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2132 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2133 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2134 | goto out; |
2b82032c YZ |
2135 | } |
2136 | ||
ebbede42 | 2137 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2138 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2139 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2140 | device->fs_devices->rw_devices--; |
34441361 | 2141 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2142 | } |
a061fc8d | 2143 | |
d7901554 | 2144 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2145 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2146 | mutex_lock(&uuid_mutex); |
a061fc8d | 2147 | if (ret) |
9b3517e9 | 2148 | goto error_undo; |
a061fc8d | 2149 | |
63a212ab SB |
2150 | /* |
2151 | * TODO: the superblock still includes this device in its num_devices | |
2152 | * counter although write_all_supers() is not locked out. This | |
2153 | * could give a filesystem state which requires a degraded mount. | |
2154 | */ | |
f331a952 | 2155 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2156 | if (ret) |
9b3517e9 | 2157 | goto error_undo; |
a061fc8d | 2158 | |
e12c9621 | 2159 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2160 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2161 | |
2162 | /* | |
2163 | * the device list mutex makes sure that we don't change | |
2164 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2165 | * the device supers. Whoever is writing all supers, should |
2166 | * lock the device list mutex before getting the number of | |
2167 | * devices in the super block (super_copy). Conversely, | |
2168 | * whoever updates the number of devices in the super block | |
2169 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2170 | */ |
1f78160c | 2171 | |
41a52a0f AJ |
2172 | /* |
2173 | * In normal cases the cur_devices == fs_devices. But in case | |
2174 | * of deleting a seed device, the cur_devices should point to | |
2175 | * its own fs_devices listed under the fs_devices->seed. | |
2176 | */ | |
1f78160c | 2177 | cur_devices = device->fs_devices; |
b5185197 | 2178 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2179 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2180 | |
41a52a0f AJ |
2181 | cur_devices->num_devices--; |
2182 | cur_devices->total_devices--; | |
b4993e64 AJ |
2183 | /* Update total_devices of the parent fs_devices if it's seed */ |
2184 | if (cur_devices != fs_devices) | |
2185 | fs_devices->total_devices--; | |
2b82032c | 2186 | |
e6e674bd | 2187 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2188 | cur_devices->missing_devices--; |
cd02dca5 | 2189 | |
d6507cf1 | 2190 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2191 | |
0bfaa9c5 | 2192 | if (device->bdev) { |
41a52a0f | 2193 | cur_devices->open_devices--; |
0bfaa9c5 | 2194 | /* remove sysfs entry */ |
b5185197 | 2195 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2196 | } |
99994cde | 2197 | |
0b246afa JM |
2198 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2199 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2200 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2201 | |
cea67ab9 JM |
2202 | /* |
2203 | * at this point, the device is zero sized and detached from | |
2204 | * the devices list. All that's left is to zero out the old | |
2205 | * supers and free the device. | |
2206 | */ | |
ebbede42 | 2207 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2208 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2209 | ||
2210 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2211 | synchronize_rcu(); |
2212 | btrfs_free_device(device); | |
cea67ab9 | 2213 | |
1f78160c | 2214 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2215 | while (fs_devices) { |
8321cf25 RS |
2216 | if (fs_devices->seed == cur_devices) { |
2217 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2218 | break; |
8321cf25 | 2219 | } |
e4404d6e | 2220 | fs_devices = fs_devices->seed; |
2b82032c | 2221 | } |
1f78160c | 2222 | cur_devices->seed = NULL; |
0226e0eb | 2223 | close_fs_devices(cur_devices); |
1f78160c | 2224 | free_fs_devices(cur_devices); |
2b82032c YZ |
2225 | } |
2226 | ||
a061fc8d CM |
2227 | out: |
2228 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2229 | return ret; |
24fc572f | 2230 | |
9b3517e9 | 2231 | error_undo: |
ebbede42 | 2232 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2233 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2234 | list_add(&device->dev_alloc_list, |
b5185197 | 2235 | &fs_devices->alloc_list); |
c3929c36 | 2236 | device->fs_devices->rw_devices++; |
34441361 | 2237 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2238 | } |
24fc572f | 2239 | goto out; |
a061fc8d CM |
2240 | } |
2241 | ||
68a9db5f | 2242 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2243 | { |
d51908ce AJ |
2244 | struct btrfs_fs_devices *fs_devices; |
2245 | ||
68a9db5f | 2246 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2247 | |
25e8e911 AJ |
2248 | /* |
2249 | * in case of fs with no seed, srcdev->fs_devices will point | |
2250 | * to fs_devices of fs_info. However when the dev being replaced is | |
2251 | * a seed dev it will point to the seed's local fs_devices. In short | |
2252 | * srcdev will have its correct fs_devices in both the cases. | |
2253 | */ | |
2254 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2255 | |
e93c89c1 | 2256 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2257 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2258 | fs_devices->num_devices--; |
e6e674bd | 2259 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2260 | fs_devices->missing_devices--; |
e93c89c1 | 2261 | |
ebbede42 | 2262 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2263 | fs_devices->rw_devices--; |
1357272f | 2264 | |
82372bc8 | 2265 | if (srcdev->bdev) |
d51908ce | 2266 | fs_devices->open_devices--; |
084b6e7c QW |
2267 | } |
2268 | ||
65237ee3 | 2269 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c | 2270 | { |
65237ee3 | 2271 | struct btrfs_fs_info *fs_info = srcdev->fs_info; |
084b6e7c | 2272 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
e93c89c1 | 2273 | |
ebbede42 | 2274 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2275 | /* zero out the old super if it is writable */ |
2276 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2277 | } | |
14238819 AJ |
2278 | |
2279 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2280 | synchronize_rcu(); |
2281 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2282 | |
94d5f0c2 AJ |
2283 | /* if this is no devs we rather delete the fs_devices */ |
2284 | if (!fs_devices->num_devices) { | |
2285 | struct btrfs_fs_devices *tmp_fs_devices; | |
2286 | ||
6dd38f81 AJ |
2287 | /* |
2288 | * On a mounted FS, num_devices can't be zero unless it's a | |
2289 | * seed. In case of a seed device being replaced, the replace | |
2290 | * target added to the sprout FS, so there will be no more | |
2291 | * device left under the seed FS. | |
2292 | */ | |
2293 | ASSERT(fs_devices->seeding); | |
2294 | ||
94d5f0c2 AJ |
2295 | tmp_fs_devices = fs_info->fs_devices; |
2296 | while (tmp_fs_devices) { | |
2297 | if (tmp_fs_devices->seed == fs_devices) { | |
2298 | tmp_fs_devices->seed = fs_devices->seed; | |
2299 | break; | |
2300 | } | |
2301 | tmp_fs_devices = tmp_fs_devices->seed; | |
2302 | } | |
2303 | fs_devices->seed = NULL; | |
0226e0eb | 2304 | close_fs_devices(fs_devices); |
8bef8401 | 2305 | free_fs_devices(fs_devices); |
94d5f0c2 | 2306 | } |
e93c89c1 SB |
2307 | } |
2308 | ||
4f5ad7bd | 2309 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2310 | { |
4f5ad7bd | 2311 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2312 | |
e93c89c1 | 2313 | WARN_ON(!tgtdev); |
d9a071f0 | 2314 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2315 | |
d9a071f0 | 2316 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2317 | |
779bf3fe | 2318 | if (tgtdev->bdev) |
d9a071f0 | 2319 | fs_devices->open_devices--; |
779bf3fe | 2320 | |
d9a071f0 | 2321 | fs_devices->num_devices--; |
e93c89c1 | 2322 | |
d6507cf1 | 2323 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2324 | |
e93c89c1 | 2325 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2326 | |
d9a071f0 | 2327 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2328 | |
2329 | /* | |
2330 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2331 | * may lead to a call to btrfs_show_devname() which will try | |
2332 | * to hold device_list_mutex. And here this device | |
2333 | * is already out of device list, so we don't have to hold | |
2334 | * the device_list_mutex lock. | |
2335 | */ | |
2336 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2337 | |
2338 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2339 | synchronize_rcu(); |
2340 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2341 | } |
2342 | ||
b444ad46 NB |
2343 | static struct btrfs_device *btrfs_find_device_by_path( |
2344 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2345 | { |
2346 | int ret = 0; | |
2347 | struct btrfs_super_block *disk_super; | |
2348 | u64 devid; | |
2349 | u8 *dev_uuid; | |
2350 | struct block_device *bdev; | |
2351 | struct buffer_head *bh; | |
b444ad46 | 2352 | struct btrfs_device *device; |
7ba15b7d | 2353 | |
7ba15b7d | 2354 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2355 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2356 | if (ret) |
b444ad46 | 2357 | return ERR_PTR(ret); |
7ba15b7d SB |
2358 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2359 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2360 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2361 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2362 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2363 | disk_super->metadata_uuid, true); |
7239ff4b | 2364 | else |
e4319cd9 | 2365 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2366 | disk_super->fsid, true); |
7239ff4b | 2367 | |
7ba15b7d | 2368 | brelse(bh); |
b444ad46 NB |
2369 | if (!device) |
2370 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2371 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2372 | return device; |
7ba15b7d SB |
2373 | } |
2374 | ||
5c5c0df0 DS |
2375 | /* |
2376 | * Lookup a device given by device id, or the path if the id is 0. | |
2377 | */ | |
a27a94c2 | 2378 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2379 | struct btrfs_fs_info *fs_info, u64 devid, |
2380 | const char *device_path) | |
24e0474b | 2381 | { |
a27a94c2 | 2382 | struct btrfs_device *device; |
24e0474b | 2383 | |
5c5c0df0 | 2384 | if (devid) { |
e4319cd9 | 2385 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2386 | NULL, true); |
a27a94c2 NB |
2387 | if (!device) |
2388 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2389 | return device; |
2390 | } | |
2391 | ||
2392 | if (!device_path || !device_path[0]) | |
2393 | return ERR_PTR(-EINVAL); | |
2394 | ||
2395 | if (strcmp(device_path, "missing") == 0) { | |
2396 | /* Find first missing device */ | |
2397 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2398 | dev_list) { | |
2399 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2400 | &device->dev_state) && !device->bdev) | |
2401 | return device; | |
d95a830c | 2402 | } |
6e927ceb | 2403 | return ERR_PTR(-ENOENT); |
24e0474b | 2404 | } |
6e927ceb AJ |
2405 | |
2406 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2407 | } |
2408 | ||
2b82032c YZ |
2409 | /* |
2410 | * does all the dirty work required for changing file system's UUID. | |
2411 | */ | |
2ff7e61e | 2412 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2413 | { |
0b246afa | 2414 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2415 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2416 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2417 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2418 | struct btrfs_device *device; |
2419 | u64 super_flags; | |
2420 | ||
a32bf9a3 | 2421 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2422 | if (!fs_devices->seeding) |
2b82032c YZ |
2423 | return -EINVAL; |
2424 | ||
7239ff4b | 2425 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2426 | if (IS_ERR(seed_devices)) |
2427 | return PTR_ERR(seed_devices); | |
2b82032c | 2428 | |
e4404d6e YZ |
2429 | old_devices = clone_fs_devices(fs_devices); |
2430 | if (IS_ERR(old_devices)) { | |
2431 | kfree(seed_devices); | |
2432 | return PTR_ERR(old_devices); | |
2b82032c | 2433 | } |
e4404d6e | 2434 | |
c4babc5e | 2435 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2436 | |
e4404d6e YZ |
2437 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2438 | seed_devices->opened = 1; | |
2439 | INIT_LIST_HEAD(&seed_devices->devices); | |
2440 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2441 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2442 | |
321a4bf7 | 2443 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2444 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2445 | synchronize_rcu); | |
2196d6e8 MX |
2446 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2447 | device->fs_devices = seed_devices; | |
c9513edb | 2448 | |
34441361 | 2449 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2450 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2451 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2452 | |
2b82032c YZ |
2453 | fs_devices->seeding = 0; |
2454 | fs_devices->num_devices = 0; | |
2455 | fs_devices->open_devices = 0; | |
69611ac8 | 2456 | fs_devices->missing_devices = 0; |
69611ac8 | 2457 | fs_devices->rotating = 0; |
e4404d6e | 2458 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2459 | |
2460 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2461 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2462 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2463 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2464 | |
2b82032c YZ |
2465 | super_flags = btrfs_super_flags(disk_super) & |
2466 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2467 | btrfs_set_super_flags(disk_super, super_flags); | |
2468 | ||
2469 | return 0; | |
2470 | } | |
2471 | ||
2472 | /* | |
01327610 | 2473 | * Store the expected generation for seed devices in device items. |
2b82032c | 2474 | */ |
5c466629 | 2475 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2476 | { |
5c466629 | 2477 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2478 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2479 | struct btrfs_path *path; |
2480 | struct extent_buffer *leaf; | |
2481 | struct btrfs_dev_item *dev_item; | |
2482 | struct btrfs_device *device; | |
2483 | struct btrfs_key key; | |
44880fdc | 2484 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2485 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2486 | u64 devid; | |
2487 | int ret; | |
2488 | ||
2489 | path = btrfs_alloc_path(); | |
2490 | if (!path) | |
2491 | return -ENOMEM; | |
2492 | ||
2b82032c YZ |
2493 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2494 | key.offset = 0; | |
2495 | key.type = BTRFS_DEV_ITEM_KEY; | |
2496 | ||
2497 | while (1) { | |
2498 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2499 | if (ret < 0) | |
2500 | goto error; | |
2501 | ||
2502 | leaf = path->nodes[0]; | |
2503 | next_slot: | |
2504 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2505 | ret = btrfs_next_leaf(root, path); | |
2506 | if (ret > 0) | |
2507 | break; | |
2508 | if (ret < 0) | |
2509 | goto error; | |
2510 | leaf = path->nodes[0]; | |
2511 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2512 | btrfs_release_path(path); |
2b82032c YZ |
2513 | continue; |
2514 | } | |
2515 | ||
2516 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2517 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2518 | key.type != BTRFS_DEV_ITEM_KEY) | |
2519 | break; | |
2520 | ||
2521 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2522 | struct btrfs_dev_item); | |
2523 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2524 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2525 | BTRFS_UUID_SIZE); |
1473b24e | 2526 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2527 | BTRFS_FSID_SIZE); |
e4319cd9 | 2528 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2529 | fs_uuid, true); |
79787eaa | 2530 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2531 | |
2532 | if (device->fs_devices->seeding) { | |
2533 | btrfs_set_device_generation(leaf, dev_item, | |
2534 | device->generation); | |
2535 | btrfs_mark_buffer_dirty(leaf); | |
2536 | } | |
2537 | ||
2538 | path->slots[0]++; | |
2539 | goto next_slot; | |
2540 | } | |
2541 | ret = 0; | |
2542 | error: | |
2543 | btrfs_free_path(path); | |
2544 | return ret; | |
2545 | } | |
2546 | ||
da353f6b | 2547 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2548 | { |
5112febb | 2549 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2550 | struct request_queue *q; |
788f20eb CM |
2551 | struct btrfs_trans_handle *trans; |
2552 | struct btrfs_device *device; | |
2553 | struct block_device *bdev; | |
0b246afa | 2554 | struct super_block *sb = fs_info->sb; |
606686ee | 2555 | struct rcu_string *name; |
5da54bc1 | 2556 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2557 | u64 orig_super_total_bytes; |
2558 | u64 orig_super_num_devices; | |
2b82032c | 2559 | int seeding_dev = 0; |
788f20eb | 2560 | int ret = 0; |
7132a262 | 2561 | bool unlocked = false; |
788f20eb | 2562 | |
5da54bc1 | 2563 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2564 | return -EROFS; |
788f20eb | 2565 | |
a5d16333 | 2566 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2567 | fs_info->bdev_holder); |
7f59203a JB |
2568 | if (IS_ERR(bdev)) |
2569 | return PTR_ERR(bdev); | |
a2135011 | 2570 | |
5da54bc1 | 2571 | if (fs_devices->seeding) { |
2b82032c YZ |
2572 | seeding_dev = 1; |
2573 | down_write(&sb->s_umount); | |
2574 | mutex_lock(&uuid_mutex); | |
2575 | } | |
2576 | ||
8c8bee1d | 2577 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2578 | |
5da54bc1 | 2579 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2580 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2581 | if (device->bdev == bdev) { |
2582 | ret = -EEXIST; | |
d25628bd | 2583 | mutex_unlock( |
5da54bc1 | 2584 | &fs_devices->device_list_mutex); |
2b82032c | 2585 | goto error; |
788f20eb CM |
2586 | } |
2587 | } | |
5da54bc1 | 2588 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2589 | |
0b246afa | 2590 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2591 | if (IS_ERR(device)) { |
788f20eb | 2592 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2593 | ret = PTR_ERR(device); |
2b82032c | 2594 | goto error; |
788f20eb CM |
2595 | } |
2596 | ||
78f2c9e6 | 2597 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2598 | if (!name) { |
2b82032c | 2599 | ret = -ENOMEM; |
5c4cf6c9 | 2600 | goto error_free_device; |
788f20eb | 2601 | } |
606686ee | 2602 | rcu_assign_pointer(device->name, name); |
2b82032c | 2603 | |
a22285a6 | 2604 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2605 | if (IS_ERR(trans)) { |
98d5dc13 | 2606 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2607 | goto error_free_device; |
98d5dc13 TI |
2608 | } |
2609 | ||
d5e2003c | 2610 | q = bdev_get_queue(bdev); |
ebbede42 | 2611 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2612 | device->generation = trans->transid; |
0b246afa JM |
2613 | device->io_width = fs_info->sectorsize; |
2614 | device->io_align = fs_info->sectorsize; | |
2615 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2616 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2617 | fs_info->sectorsize); | |
2cc3c559 | 2618 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2619 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2620 | device->fs_info = fs_info; |
788f20eb | 2621 | device->bdev = bdev; |
e12c9621 | 2622 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2623 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2624 | device->mode = FMODE_EXCL; |
27087f37 | 2625 | device->dev_stats_valid = 1; |
9f6d2510 | 2626 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2627 | |
2b82032c | 2628 | if (seeding_dev) { |
1751e8a6 | 2629 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2630 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2631 | if (ret) { |
2632 | btrfs_abort_transaction(trans, ret); | |
2633 | goto error_trans; | |
2634 | } | |
2b82032c | 2635 | } |
788f20eb | 2636 | |
5da54bc1 | 2637 | device->fs_devices = fs_devices; |
e5e9a520 | 2638 | |
5da54bc1 | 2639 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2640 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2641 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2642 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2643 | fs_devices->num_devices++; | |
2644 | fs_devices->open_devices++; | |
2645 | fs_devices->rw_devices++; | |
2646 | fs_devices->total_devices++; | |
2647 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2648 | |
a5ed45f8 | 2649 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2650 | |
e884f4f0 | 2651 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2652 | fs_devices->rotating = 1; |
c289811c | 2653 | |
39379faa | 2654 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2655 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2656 | round_down(orig_super_total_bytes + device->total_bytes, |
2657 | fs_info->sectorsize)); | |
788f20eb | 2658 | |
39379faa NA |
2659 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2660 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2661 | orig_super_num_devices + 1); | |
0d39376a AJ |
2662 | |
2663 | /* add sysfs device entry */ | |
5da54bc1 | 2664 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2665 | |
2196d6e8 MX |
2666 | /* |
2667 | * we've got more storage, clear any full flags on the space | |
2668 | * infos | |
2669 | */ | |
0b246afa | 2670 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2671 | |
34441361 | 2672 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2673 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2674 | |
2b82032c | 2675 | if (seeding_dev) { |
34441361 | 2676 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2677 | ret = init_first_rw_device(trans); |
34441361 | 2678 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2679 | if (ret) { |
66642832 | 2680 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2681 | goto error_sysfs; |
005d6427 | 2682 | } |
2196d6e8 MX |
2683 | } |
2684 | ||
8e87e856 | 2685 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2686 | if (ret) { |
66642832 | 2687 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2688 | goto error_sysfs; |
2196d6e8 MX |
2689 | } |
2690 | ||
2691 | if (seeding_dev) { | |
5c466629 | 2692 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2693 | if (ret) { |
66642832 | 2694 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2695 | goto error_sysfs; |
005d6427 | 2696 | } |
b2373f25 | 2697 | |
f93c3997 DS |
2698 | btrfs_sysfs_update_sprout_fsid(fs_devices, |
2699 | fs_info->fs_devices->fsid); | |
2b82032c YZ |
2700 | } |
2701 | ||
3a45bb20 | 2702 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2703 | |
2b82032c YZ |
2704 | if (seeding_dev) { |
2705 | mutex_unlock(&uuid_mutex); | |
2706 | up_write(&sb->s_umount); | |
7132a262 | 2707 | unlocked = true; |
788f20eb | 2708 | |
79787eaa JM |
2709 | if (ret) /* transaction commit */ |
2710 | return ret; | |
2711 | ||
2ff7e61e | 2712 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2713 | if (ret < 0) |
0b246afa | 2714 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2715 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2716 | trans = btrfs_attach_transaction(root); |
2717 | if (IS_ERR(trans)) { | |
2718 | if (PTR_ERR(trans) == -ENOENT) | |
2719 | return 0; | |
7132a262 AJ |
2720 | ret = PTR_ERR(trans); |
2721 | trans = NULL; | |
2722 | goto error_sysfs; | |
671415b7 | 2723 | } |
3a45bb20 | 2724 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2725 | } |
c9e9f97b | 2726 | |
5a1972bd QW |
2727 | /* Update ctime/mtime for libblkid */ |
2728 | update_dev_time(device_path); | |
2b82032c | 2729 | return ret; |
79787eaa | 2730 | |
d31c32f6 | 2731 | error_sysfs: |
5da54bc1 | 2732 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2733 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2734 | mutex_lock(&fs_info->chunk_mutex); | |
2735 | list_del_rcu(&device->dev_list); | |
2736 | list_del(&device->dev_alloc_list); | |
2737 | fs_info->fs_devices->num_devices--; | |
2738 | fs_info->fs_devices->open_devices--; | |
2739 | fs_info->fs_devices->rw_devices--; | |
2740 | fs_info->fs_devices->total_devices--; | |
2741 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2742 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2743 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2744 | orig_super_total_bytes); | |
2745 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2746 | orig_super_num_devices); | |
2747 | mutex_unlock(&fs_info->chunk_mutex); | |
2748 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2749 | error_trans: |
0af2c4bf | 2750 | if (seeding_dev) |
1751e8a6 | 2751 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2752 | if (trans) |
2753 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2754 | error_free_device: |
a425f9d4 | 2755 | btrfs_free_device(device); |
2b82032c | 2756 | error: |
e525fd89 | 2757 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2758 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2759 | mutex_unlock(&uuid_mutex); |
2760 | up_write(&sb->s_umount); | |
2761 | } | |
c9e9f97b | 2762 | return ret; |
788f20eb CM |
2763 | } |
2764 | ||
d397712b CM |
2765 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2766 | struct btrfs_device *device) | |
0b86a832 CM |
2767 | { |
2768 | int ret; | |
2769 | struct btrfs_path *path; | |
0b246afa | 2770 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2771 | struct btrfs_dev_item *dev_item; |
2772 | struct extent_buffer *leaf; | |
2773 | struct btrfs_key key; | |
2774 | ||
0b86a832 CM |
2775 | path = btrfs_alloc_path(); |
2776 | if (!path) | |
2777 | return -ENOMEM; | |
2778 | ||
2779 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2780 | key.type = BTRFS_DEV_ITEM_KEY; | |
2781 | key.offset = device->devid; | |
2782 | ||
2783 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2784 | if (ret < 0) | |
2785 | goto out; | |
2786 | ||
2787 | if (ret > 0) { | |
2788 | ret = -ENOENT; | |
2789 | goto out; | |
2790 | } | |
2791 | ||
2792 | leaf = path->nodes[0]; | |
2793 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2794 | ||
2795 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2796 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2797 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2798 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2799 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2800 | btrfs_set_device_total_bytes(leaf, dev_item, |
2801 | btrfs_device_get_disk_total_bytes(device)); | |
2802 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2803 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2804 | btrfs_mark_buffer_dirty(leaf); |
2805 | ||
2806 | out: | |
2807 | btrfs_free_path(path); | |
2808 | return ret; | |
2809 | } | |
2810 | ||
2196d6e8 | 2811 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2812 | struct btrfs_device *device, u64 new_size) |
2813 | { | |
0b246afa JM |
2814 | struct btrfs_fs_info *fs_info = device->fs_info; |
2815 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2816 | u64 old_total; |
2817 | u64 diff; | |
8f18cf13 | 2818 | |
ebbede42 | 2819 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2820 | return -EACCES; |
2196d6e8 | 2821 | |
7dfb8be1 NB |
2822 | new_size = round_down(new_size, fs_info->sectorsize); |
2823 | ||
34441361 | 2824 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2825 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2826 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2827 | |
63a212ab | 2828 | if (new_size <= device->total_bytes || |
401e29c1 | 2829 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2830 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2831 | return -EINVAL; |
2196d6e8 | 2832 | } |
2b82032c | 2833 | |
7dfb8be1 NB |
2834 | btrfs_set_super_total_bytes(super_copy, |
2835 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2836 | device->fs_devices->total_rw_bytes += diff; |
2837 | ||
7cc8e58d MX |
2838 | btrfs_device_set_total_bytes(device, new_size); |
2839 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2840 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2841 | if (list_empty(&device->post_commit_list)) |
2842 | list_add_tail(&device->post_commit_list, | |
2843 | &trans->transaction->dev_update_list); | |
34441361 | 2844 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2845 | |
8f18cf13 CM |
2846 | return btrfs_update_device(trans, device); |
2847 | } | |
2848 | ||
f4208794 | 2849 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2850 | { |
f4208794 | 2851 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2852 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2853 | int ret; |
2854 | struct btrfs_path *path; | |
2855 | struct btrfs_key key; | |
2856 | ||
8f18cf13 CM |
2857 | path = btrfs_alloc_path(); |
2858 | if (!path) | |
2859 | return -ENOMEM; | |
2860 | ||
408fbf19 | 2861 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2862 | key.offset = chunk_offset; |
2863 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2864 | ||
2865 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2866 | if (ret < 0) |
2867 | goto out; | |
2868 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2869 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2870 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2871 | ret = -ENOENT; |
2872 | goto out; | |
2873 | } | |
8f18cf13 CM |
2874 | |
2875 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2876 | if (ret < 0) |
0b246afa JM |
2877 | btrfs_handle_fs_error(fs_info, ret, |
2878 | "Failed to delete chunk item."); | |
79787eaa | 2879 | out: |
8f18cf13 | 2880 | btrfs_free_path(path); |
65a246c5 | 2881 | return ret; |
8f18cf13 CM |
2882 | } |
2883 | ||
408fbf19 | 2884 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2885 | { |
0b246afa | 2886 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2887 | struct btrfs_disk_key *disk_key; |
2888 | struct btrfs_chunk *chunk; | |
2889 | u8 *ptr; | |
2890 | int ret = 0; | |
2891 | u32 num_stripes; | |
2892 | u32 array_size; | |
2893 | u32 len = 0; | |
2894 | u32 cur; | |
2895 | struct btrfs_key key; | |
2896 | ||
34441361 | 2897 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2898 | array_size = btrfs_super_sys_array_size(super_copy); |
2899 | ||
2900 | ptr = super_copy->sys_chunk_array; | |
2901 | cur = 0; | |
2902 | ||
2903 | while (cur < array_size) { | |
2904 | disk_key = (struct btrfs_disk_key *)ptr; | |
2905 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2906 | ||
2907 | len = sizeof(*disk_key); | |
2908 | ||
2909 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2910 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2911 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2912 | len += btrfs_chunk_item_size(num_stripes); | |
2913 | } else { | |
2914 | ret = -EIO; | |
2915 | break; | |
2916 | } | |
408fbf19 | 2917 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2918 | key.offset == chunk_offset) { |
2919 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2920 | array_size -= len; | |
2921 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2922 | } else { | |
2923 | ptr += len; | |
2924 | cur += len; | |
2925 | } | |
2926 | } | |
34441361 | 2927 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2928 | return ret; |
2929 | } | |
2930 | ||
60ca842e OS |
2931 | /* |
2932 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2933 | * @logical: Logical block offset in bytes. | |
2934 | * @length: Length of extent in bytes. | |
2935 | * | |
2936 | * Return: Chunk mapping or ERR_PTR. | |
2937 | */ | |
2938 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2939 | u64 logical, u64 length) | |
592d92ee LB |
2940 | { |
2941 | struct extent_map_tree *em_tree; | |
2942 | struct extent_map *em; | |
2943 | ||
c8bf1b67 | 2944 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2945 | read_lock(&em_tree->lock); |
2946 | em = lookup_extent_mapping(em_tree, logical, length); | |
2947 | read_unlock(&em_tree->lock); | |
2948 | ||
2949 | if (!em) { | |
2950 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2951 | logical, length); | |
2952 | return ERR_PTR(-EINVAL); | |
2953 | } | |
2954 | ||
2955 | if (em->start > logical || em->start + em->len < logical) { | |
2956 | btrfs_crit(fs_info, | |
2957 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2958 | logical, length, em->start, em->start + em->len); | |
2959 | free_extent_map(em); | |
2960 | return ERR_PTR(-EINVAL); | |
2961 | } | |
2962 | ||
2963 | /* callers are responsible for dropping em's ref. */ | |
2964 | return em; | |
2965 | } | |
2966 | ||
97aff912 | 2967 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2968 | { |
97aff912 | 2969 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2970 | struct extent_map *em; |
2971 | struct map_lookup *map; | |
2196d6e8 | 2972 | u64 dev_extent_len = 0; |
47ab2a6c | 2973 | int i, ret = 0; |
0b246afa | 2974 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2975 | |
60ca842e | 2976 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2977 | if (IS_ERR(em)) { |
47ab2a6c JB |
2978 | /* |
2979 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2980 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2981 | * do anything we still error out. |
2982 | */ | |
2983 | ASSERT(0); | |
592d92ee | 2984 | return PTR_ERR(em); |
47ab2a6c | 2985 | } |
95617d69 | 2986 | map = em->map_lookup; |
34441361 | 2987 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2988 | check_system_chunk(trans, map->type); |
34441361 | 2989 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2990 | |
57ba4cb8 FM |
2991 | /* |
2992 | * Take the device list mutex to prevent races with the final phase of | |
2993 | * a device replace operation that replaces the device object associated | |
2994 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2995 | */ | |
2996 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2997 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2998 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2999 | ret = btrfs_free_dev_extent(trans, device, |
3000 | map->stripes[i].physical, | |
3001 | &dev_extent_len); | |
47ab2a6c | 3002 | if (ret) { |
57ba4cb8 | 3003 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3004 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3005 | goto out; |
3006 | } | |
a061fc8d | 3007 | |
2196d6e8 | 3008 | if (device->bytes_used > 0) { |
34441361 | 3009 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3010 | btrfs_device_set_bytes_used(device, |
3011 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3012 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3013 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3014 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3015 | } |
a061fc8d | 3016 | |
64bc6c2a NB |
3017 | ret = btrfs_update_device(trans, device); |
3018 | if (ret) { | |
3019 | mutex_unlock(&fs_devices->device_list_mutex); | |
3020 | btrfs_abort_transaction(trans, ret); | |
3021 | goto out; | |
dfe25020 | 3022 | } |
8f18cf13 | 3023 | } |
57ba4cb8 FM |
3024 | mutex_unlock(&fs_devices->device_list_mutex); |
3025 | ||
f4208794 | 3026 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3027 | if (ret) { |
66642832 | 3028 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3029 | goto out; |
3030 | } | |
8f18cf13 | 3031 | |
6bccf3ab | 3032 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3033 | |
8f18cf13 | 3034 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3035 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3036 | if (ret) { |
66642832 | 3037 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3038 | goto out; |
3039 | } | |
8f18cf13 CM |
3040 | } |
3041 | ||
5a98ec01 | 3042 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3043 | if (ret) { |
66642832 | 3044 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3045 | goto out; |
3046 | } | |
2b82032c | 3047 | |
47ab2a6c | 3048 | out: |
2b82032c YZ |
3049 | /* once for us */ |
3050 | free_extent_map(em); | |
47ab2a6c JB |
3051 | return ret; |
3052 | } | |
2b82032c | 3053 | |
5b4aacef | 3054 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3055 | { |
5b4aacef | 3056 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3057 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3058 | int ret; |
2b82032c | 3059 | |
67c5e7d4 FM |
3060 | /* |
3061 | * Prevent races with automatic removal of unused block groups. | |
3062 | * After we relocate and before we remove the chunk with offset | |
3063 | * chunk_offset, automatic removal of the block group can kick in, | |
3064 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3065 | * | |
3066 | * Make sure to acquire this mutex before doing a tree search (dev | |
3067 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3068 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3069 | * we release the path used to search the chunk/dev tree and before | |
3070 | * the current task acquires this mutex and calls us. | |
3071 | */ | |
a32bf9a3 | 3072 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3073 | |
47ab2a6c | 3074 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3075 | btrfs_scrub_pause(fs_info); |
0b246afa | 3076 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3077 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3078 | if (ret) |
3079 | return ret; | |
3080 | ||
19c4d2f9 CM |
3081 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3082 | chunk_offset); | |
3083 | if (IS_ERR(trans)) { | |
3084 | ret = PTR_ERR(trans); | |
3085 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3086 | return ret; | |
3087 | } | |
3088 | ||
47ab2a6c | 3089 | /* |
19c4d2f9 CM |
3090 | * step two, delete the device extents and the |
3091 | * chunk tree entries | |
47ab2a6c | 3092 | */ |
97aff912 | 3093 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3094 | btrfs_end_transaction(trans); |
19c4d2f9 | 3095 | return ret; |
2b82032c YZ |
3096 | } |
3097 | ||
2ff7e61e | 3098 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3099 | { |
0b246afa | 3100 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3101 | struct btrfs_path *path; |
3102 | struct extent_buffer *leaf; | |
3103 | struct btrfs_chunk *chunk; | |
3104 | struct btrfs_key key; | |
3105 | struct btrfs_key found_key; | |
2b82032c | 3106 | u64 chunk_type; |
ba1bf481 JB |
3107 | bool retried = false; |
3108 | int failed = 0; | |
2b82032c YZ |
3109 | int ret; |
3110 | ||
3111 | path = btrfs_alloc_path(); | |
3112 | if (!path) | |
3113 | return -ENOMEM; | |
3114 | ||
ba1bf481 | 3115 | again: |
2b82032c YZ |
3116 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3117 | key.offset = (u64)-1; | |
3118 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3119 | ||
3120 | while (1) { | |
0b246afa | 3121 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3122 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3123 | if (ret < 0) { |
0b246afa | 3124 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3125 | goto error; |
67c5e7d4 | 3126 | } |
79787eaa | 3127 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3128 | |
3129 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3130 | key.type); | |
67c5e7d4 | 3131 | if (ret) |
0b246afa | 3132 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3133 | if (ret < 0) |
3134 | goto error; | |
3135 | if (ret > 0) | |
3136 | break; | |
1a40e23b | 3137 | |
2b82032c YZ |
3138 | leaf = path->nodes[0]; |
3139 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3140 | |
2b82032c YZ |
3141 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3142 | struct btrfs_chunk); | |
3143 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3144 | btrfs_release_path(path); |
8f18cf13 | 3145 | |
2b82032c | 3146 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3147 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3148 | if (ret == -ENOSPC) |
3149 | failed++; | |
14586651 HS |
3150 | else |
3151 | BUG_ON(ret); | |
2b82032c | 3152 | } |
0b246afa | 3153 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3154 | |
2b82032c YZ |
3155 | if (found_key.offset == 0) |
3156 | break; | |
3157 | key.offset = found_key.offset - 1; | |
3158 | } | |
3159 | ret = 0; | |
ba1bf481 JB |
3160 | if (failed && !retried) { |
3161 | failed = 0; | |
3162 | retried = true; | |
3163 | goto again; | |
fae7f21c | 3164 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3165 | ret = -ENOSPC; |
3166 | } | |
2b82032c YZ |
3167 | error: |
3168 | btrfs_free_path(path); | |
3169 | return ret; | |
8f18cf13 CM |
3170 | } |
3171 | ||
a6f93c71 LB |
3172 | /* |
3173 | * return 1 : allocate a data chunk successfully, | |
3174 | * return <0: errors during allocating a data chunk, | |
3175 | * return 0 : no need to allocate a data chunk. | |
3176 | */ | |
3177 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3178 | u64 chunk_offset) | |
3179 | { | |
3180 | struct btrfs_block_group_cache *cache; | |
3181 | u64 bytes_used; | |
3182 | u64 chunk_type; | |
3183 | ||
3184 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3185 | ASSERT(cache); | |
3186 | chunk_type = cache->flags; | |
3187 | btrfs_put_block_group(cache); | |
3188 | ||
3189 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3190 | spin_lock(&fs_info->data_sinfo->lock); | |
3191 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3192 | spin_unlock(&fs_info->data_sinfo->lock); | |
3193 | ||
3194 | if (!bytes_used) { | |
3195 | struct btrfs_trans_handle *trans; | |
3196 | int ret; | |
3197 | ||
3198 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3199 | if (IS_ERR(trans)) | |
3200 | return PTR_ERR(trans); | |
3201 | ||
43a7e99d | 3202 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3203 | BTRFS_BLOCK_GROUP_DATA); |
3204 | btrfs_end_transaction(trans); | |
3205 | if (ret < 0) | |
3206 | return ret; | |
a6f93c71 LB |
3207 | return 1; |
3208 | } | |
3209 | } | |
3210 | return 0; | |
3211 | } | |
3212 | ||
6bccf3ab | 3213 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3214 | struct btrfs_balance_control *bctl) |
3215 | { | |
6bccf3ab | 3216 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3217 | struct btrfs_trans_handle *trans; |
3218 | struct btrfs_balance_item *item; | |
3219 | struct btrfs_disk_balance_args disk_bargs; | |
3220 | struct btrfs_path *path; | |
3221 | struct extent_buffer *leaf; | |
3222 | struct btrfs_key key; | |
3223 | int ret, err; | |
3224 | ||
3225 | path = btrfs_alloc_path(); | |
3226 | if (!path) | |
3227 | return -ENOMEM; | |
3228 | ||
3229 | trans = btrfs_start_transaction(root, 0); | |
3230 | if (IS_ERR(trans)) { | |
3231 | btrfs_free_path(path); | |
3232 | return PTR_ERR(trans); | |
3233 | } | |
3234 | ||
3235 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3236 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3237 | key.offset = 0; |
3238 | ||
3239 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3240 | sizeof(*item)); | |
3241 | if (ret) | |
3242 | goto out; | |
3243 | ||
3244 | leaf = path->nodes[0]; | |
3245 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3246 | ||
b159fa28 | 3247 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3248 | |
3249 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3250 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3251 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3252 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3253 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3254 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3255 | ||
3256 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3257 | ||
3258 | btrfs_mark_buffer_dirty(leaf); | |
3259 | out: | |
3260 | btrfs_free_path(path); | |
3a45bb20 | 3261 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3262 | if (err && !ret) |
3263 | ret = err; | |
3264 | return ret; | |
3265 | } | |
3266 | ||
6bccf3ab | 3267 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3268 | { |
6bccf3ab | 3269 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3270 | struct btrfs_trans_handle *trans; |
3271 | struct btrfs_path *path; | |
3272 | struct btrfs_key key; | |
3273 | int ret, err; | |
3274 | ||
3275 | path = btrfs_alloc_path(); | |
3276 | if (!path) | |
3277 | return -ENOMEM; | |
3278 | ||
3279 | trans = btrfs_start_transaction(root, 0); | |
3280 | if (IS_ERR(trans)) { | |
3281 | btrfs_free_path(path); | |
3282 | return PTR_ERR(trans); | |
3283 | } | |
3284 | ||
3285 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3286 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3287 | key.offset = 0; |
3288 | ||
3289 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3290 | if (ret < 0) | |
3291 | goto out; | |
3292 | if (ret > 0) { | |
3293 | ret = -ENOENT; | |
3294 | goto out; | |
3295 | } | |
3296 | ||
3297 | ret = btrfs_del_item(trans, root, path); | |
3298 | out: | |
3299 | btrfs_free_path(path); | |
3a45bb20 | 3300 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3301 | if (err && !ret) |
3302 | ret = err; | |
3303 | return ret; | |
3304 | } | |
3305 | ||
59641015 ID |
3306 | /* |
3307 | * This is a heuristic used to reduce the number of chunks balanced on | |
3308 | * resume after balance was interrupted. | |
3309 | */ | |
3310 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3311 | { | |
3312 | /* | |
3313 | * Turn on soft mode for chunk types that were being converted. | |
3314 | */ | |
3315 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3316 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3317 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3318 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3319 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3320 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3321 | ||
3322 | /* | |
3323 | * Turn on usage filter if is not already used. The idea is | |
3324 | * that chunks that we have already balanced should be | |
3325 | * reasonably full. Don't do it for chunks that are being | |
3326 | * converted - that will keep us from relocating unconverted | |
3327 | * (albeit full) chunks. | |
3328 | */ | |
3329 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3330 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3331 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3332 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3333 | bctl->data.usage = 90; | |
3334 | } | |
3335 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3336 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3337 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3338 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3339 | bctl->sys.usage = 90; | |
3340 | } | |
3341 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3342 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3343 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3344 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3345 | bctl->meta.usage = 90; | |
3346 | } | |
3347 | } | |
3348 | ||
149196a2 DS |
3349 | /* |
3350 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3351 | */ | |
3352 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3353 | { |
3354 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3355 | int ret; |
c9e9f97b ID |
3356 | |
3357 | BUG_ON(!fs_info->balance_ctl); | |
3358 | ||
3359 | spin_lock(&fs_info->balance_lock); | |
3360 | fs_info->balance_ctl = NULL; | |
3361 | spin_unlock(&fs_info->balance_lock); | |
3362 | ||
3363 | kfree(bctl); | |
149196a2 DS |
3364 | ret = del_balance_item(fs_info); |
3365 | if (ret) | |
3366 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3367 | } |
3368 | ||
ed25e9b2 ID |
3369 | /* |
3370 | * Balance filters. Return 1 if chunk should be filtered out | |
3371 | * (should not be balanced). | |
3372 | */ | |
899c81ea | 3373 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3374 | struct btrfs_balance_args *bargs) |
3375 | { | |
899c81ea ID |
3376 | chunk_type = chunk_to_extended(chunk_type) & |
3377 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3378 | |
899c81ea | 3379 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3380 | return 0; |
3381 | ||
3382 | return 1; | |
3383 | } | |
3384 | ||
dba72cb3 | 3385 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3386 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3387 | { |
3388 | struct btrfs_block_group_cache *cache; | |
3389 | u64 chunk_used; | |
3390 | u64 user_thresh_min; | |
3391 | u64 user_thresh_max; | |
3392 | int ret = 1; | |
3393 | ||
3394 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3395 | chunk_used = btrfs_block_group_used(&cache->item); | |
3396 | ||
3397 | if (bargs->usage_min == 0) | |
3398 | user_thresh_min = 0; | |
3399 | else | |
3400 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3401 | bargs->usage_min); | |
3402 | ||
3403 | if (bargs->usage_max == 0) | |
3404 | user_thresh_max = 1; | |
3405 | else if (bargs->usage_max > 100) | |
3406 | user_thresh_max = cache->key.offset; | |
3407 | else | |
3408 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3409 | bargs->usage_max); | |
3410 | ||
3411 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3412 | ret = 0; | |
3413 | ||
3414 | btrfs_put_block_group(cache); | |
3415 | return ret; | |
3416 | } | |
3417 | ||
dba72cb3 | 3418 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3419 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3420 | { |
3421 | struct btrfs_block_group_cache *cache; | |
3422 | u64 chunk_used, user_thresh; | |
3423 | int ret = 1; | |
3424 | ||
3425 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3426 | chunk_used = btrfs_block_group_used(&cache->item); | |
3427 | ||
bc309467 | 3428 | if (bargs->usage_min == 0) |
3e39cea6 | 3429 | user_thresh = 1; |
a105bb88 ID |
3430 | else if (bargs->usage > 100) |
3431 | user_thresh = cache->key.offset; | |
3432 | else | |
3433 | user_thresh = div_factor_fine(cache->key.offset, | |
3434 | bargs->usage); | |
3435 | ||
5ce5b3c0 ID |
3436 | if (chunk_used < user_thresh) |
3437 | ret = 0; | |
3438 | ||
3439 | btrfs_put_block_group(cache); | |
3440 | return ret; | |
3441 | } | |
3442 | ||
409d404b ID |
3443 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3444 | struct btrfs_chunk *chunk, | |
3445 | struct btrfs_balance_args *bargs) | |
3446 | { | |
3447 | struct btrfs_stripe *stripe; | |
3448 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3449 | int i; | |
3450 | ||
3451 | for (i = 0; i < num_stripes; i++) { | |
3452 | stripe = btrfs_stripe_nr(chunk, i); | |
3453 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3454 | return 0; | |
3455 | } | |
3456 | ||
3457 | return 1; | |
3458 | } | |
3459 | ||
946c9256 DS |
3460 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3461 | { | |
3462 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3463 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3464 | const int nparity = btrfs_raid_array[index].nparity; | |
3465 | ||
3466 | if (nparity) | |
3467 | return num_stripes - nparity; | |
3468 | else | |
3469 | return num_stripes / ncopies; | |
3470 | } | |
3471 | ||
94e60d5a ID |
3472 | /* [pstart, pend) */ |
3473 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3474 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3475 | struct btrfs_balance_args *bargs) |
3476 | { | |
3477 | struct btrfs_stripe *stripe; | |
3478 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3479 | u64 stripe_offset; | |
3480 | u64 stripe_length; | |
946c9256 | 3481 | u64 type; |
94e60d5a ID |
3482 | int factor; |
3483 | int i; | |
3484 | ||
3485 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3486 | return 0; | |
3487 | ||
946c9256 DS |
3488 | type = btrfs_chunk_type(leaf, chunk); |
3489 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3490 | |
3491 | for (i = 0; i < num_stripes; i++) { | |
3492 | stripe = btrfs_stripe_nr(chunk, i); | |
3493 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3494 | continue; | |
3495 | ||
3496 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3497 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3498 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3499 | |
3500 | if (stripe_offset < bargs->pend && | |
3501 | stripe_offset + stripe_length > bargs->pstart) | |
3502 | return 0; | |
3503 | } | |
3504 | ||
3505 | return 1; | |
3506 | } | |
3507 | ||
ea67176a ID |
3508 | /* [vstart, vend) */ |
3509 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3510 | struct btrfs_chunk *chunk, | |
3511 | u64 chunk_offset, | |
3512 | struct btrfs_balance_args *bargs) | |
3513 | { | |
3514 | if (chunk_offset < bargs->vend && | |
3515 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3516 | /* at least part of the chunk is inside this vrange */ | |
3517 | return 0; | |
3518 | ||
3519 | return 1; | |
3520 | } | |
3521 | ||
dee32d0a GAP |
3522 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3523 | struct btrfs_chunk *chunk, | |
3524 | struct btrfs_balance_args *bargs) | |
3525 | { | |
3526 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3527 | ||
3528 | if (bargs->stripes_min <= num_stripes | |
3529 | && num_stripes <= bargs->stripes_max) | |
3530 | return 0; | |
3531 | ||
3532 | return 1; | |
3533 | } | |
3534 | ||
899c81ea | 3535 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3536 | struct btrfs_balance_args *bargs) |
3537 | { | |
3538 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3539 | return 0; | |
3540 | ||
899c81ea ID |
3541 | chunk_type = chunk_to_extended(chunk_type) & |
3542 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3543 | |
899c81ea | 3544 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3545 | return 1; |
3546 | ||
3547 | return 0; | |
3548 | } | |
3549 | ||
6ec0896c | 3550 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3551 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3552 | { | |
6ec0896c | 3553 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3554 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3555 | struct btrfs_balance_args *bargs = NULL; |
3556 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3557 | ||
3558 | /* type filter */ | |
3559 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3560 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3561 | return 0; | |
3562 | } | |
3563 | ||
3564 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3565 | bargs = &bctl->data; | |
3566 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3567 | bargs = &bctl->sys; | |
3568 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3569 | bargs = &bctl->meta; | |
3570 | ||
ed25e9b2 ID |
3571 | /* profiles filter */ |
3572 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3573 | chunk_profiles_filter(chunk_type, bargs)) { | |
3574 | return 0; | |
5ce5b3c0 ID |
3575 | } |
3576 | ||
3577 | /* usage filter */ | |
3578 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3579 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3580 | return 0; |
bc309467 | 3581 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3582 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3583 | return 0; |
409d404b ID |
3584 | } |
3585 | ||
3586 | /* devid filter */ | |
3587 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3588 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3589 | return 0; | |
94e60d5a ID |
3590 | } |
3591 | ||
3592 | /* drange filter, makes sense only with devid filter */ | |
3593 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3594 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3595 | return 0; |
ea67176a ID |
3596 | } |
3597 | ||
3598 | /* vrange filter */ | |
3599 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3600 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3601 | return 0; | |
ed25e9b2 ID |
3602 | } |
3603 | ||
dee32d0a GAP |
3604 | /* stripes filter */ |
3605 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3606 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3607 | return 0; | |
3608 | } | |
3609 | ||
cfa4c961 ID |
3610 | /* soft profile changing mode */ |
3611 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3612 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3613 | return 0; | |
3614 | } | |
3615 | ||
7d824b6f DS |
3616 | /* |
3617 | * limited by count, must be the last filter | |
3618 | */ | |
3619 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3620 | if (bargs->limit == 0) | |
3621 | return 0; | |
3622 | else | |
3623 | bargs->limit--; | |
12907fc7 DS |
3624 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3625 | /* | |
3626 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3627 | * determined here because we do not have the global information |
12907fc7 DS |
3628 | * about the count of all chunks that satisfy the filters. |
3629 | */ | |
3630 | if (bargs->limit_max == 0) | |
3631 | return 0; | |
3632 | else | |
3633 | bargs->limit_max--; | |
7d824b6f DS |
3634 | } |
3635 | ||
f43ffb60 ID |
3636 | return 1; |
3637 | } | |
3638 | ||
c9e9f97b | 3639 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3640 | { |
19a39dce | 3641 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3642 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3643 | u64 chunk_type; |
f43ffb60 | 3644 | struct btrfs_chunk *chunk; |
5a488b9d | 3645 | struct btrfs_path *path = NULL; |
ec44a35c | 3646 | struct btrfs_key key; |
ec44a35c | 3647 | struct btrfs_key found_key; |
f43ffb60 ID |
3648 | struct extent_buffer *leaf; |
3649 | int slot; | |
c9e9f97b ID |
3650 | int ret; |
3651 | int enospc_errors = 0; | |
19a39dce | 3652 | bool counting = true; |
12907fc7 | 3653 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3654 | u64 limit_data = bctl->data.limit; |
3655 | u64 limit_meta = bctl->meta.limit; | |
3656 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3657 | u32 count_data = 0; |
3658 | u32 count_meta = 0; | |
3659 | u32 count_sys = 0; | |
2c9fe835 | 3660 | int chunk_reserved = 0; |
ec44a35c | 3661 | |
ec44a35c | 3662 | path = btrfs_alloc_path(); |
17e9f796 MF |
3663 | if (!path) { |
3664 | ret = -ENOMEM; | |
3665 | goto error; | |
3666 | } | |
19a39dce ID |
3667 | |
3668 | /* zero out stat counters */ | |
3669 | spin_lock(&fs_info->balance_lock); | |
3670 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3671 | spin_unlock(&fs_info->balance_lock); | |
3672 | again: | |
7d824b6f | 3673 | if (!counting) { |
12907fc7 DS |
3674 | /* |
3675 | * The single value limit and min/max limits use the same bytes | |
3676 | * in the | |
3677 | */ | |
7d824b6f DS |
3678 | bctl->data.limit = limit_data; |
3679 | bctl->meta.limit = limit_meta; | |
3680 | bctl->sys.limit = limit_sys; | |
3681 | } | |
ec44a35c CM |
3682 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3683 | key.offset = (u64)-1; | |
3684 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3685 | ||
d397712b | 3686 | while (1) { |
19a39dce | 3687 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3688 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3689 | ret = -ECANCELED; |
3690 | goto error; | |
3691 | } | |
3692 | ||
67c5e7d4 | 3693 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3694 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3695 | if (ret < 0) { |
3696 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3697 | goto error; |
67c5e7d4 | 3698 | } |
ec44a35c CM |
3699 | |
3700 | /* | |
3701 | * this shouldn't happen, it means the last relocate | |
3702 | * failed | |
3703 | */ | |
3704 | if (ret == 0) | |
c9e9f97b | 3705 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3706 | |
3707 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3708 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3709 | if (ret) { |
67c5e7d4 | 3710 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3711 | ret = 0; |
ec44a35c | 3712 | break; |
c9e9f97b | 3713 | } |
7d9eb12c | 3714 | |
f43ffb60 ID |
3715 | leaf = path->nodes[0]; |
3716 | slot = path->slots[0]; | |
3717 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3718 | |
67c5e7d4 FM |
3719 | if (found_key.objectid != key.objectid) { |
3720 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3721 | break; |
67c5e7d4 | 3722 | } |
7d9eb12c | 3723 | |
f43ffb60 | 3724 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3725 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3726 | |
19a39dce ID |
3727 | if (!counting) { |
3728 | spin_lock(&fs_info->balance_lock); | |
3729 | bctl->stat.considered++; | |
3730 | spin_unlock(&fs_info->balance_lock); | |
3731 | } | |
3732 | ||
6ec0896c | 3733 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3734 | |
b3b4aa74 | 3735 | btrfs_release_path(path); |
67c5e7d4 FM |
3736 | if (!ret) { |
3737 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3738 | goto loop; |
67c5e7d4 | 3739 | } |
f43ffb60 | 3740 | |
19a39dce | 3741 | if (counting) { |
67c5e7d4 | 3742 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3743 | spin_lock(&fs_info->balance_lock); |
3744 | bctl->stat.expected++; | |
3745 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3746 | |
3747 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3748 | count_data++; | |
3749 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3750 | count_sys++; | |
3751 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3752 | count_meta++; | |
3753 | ||
3754 | goto loop; | |
3755 | } | |
3756 | ||
3757 | /* | |
3758 | * Apply limit_min filter, no need to check if the LIMITS | |
3759 | * filter is used, limit_min is 0 by default | |
3760 | */ | |
3761 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3762 | count_data < bctl->data.limit_min) | |
3763 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3764 | count_meta < bctl->meta.limit_min) | |
3765 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3766 | count_sys < bctl->sys.limit_min)) { | |
3767 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3768 | goto loop; |
3769 | } | |
3770 | ||
a6f93c71 LB |
3771 | if (!chunk_reserved) { |
3772 | /* | |
3773 | * We may be relocating the only data chunk we have, | |
3774 | * which could potentially end up with losing data's | |
3775 | * raid profile, so lets allocate an empty one in | |
3776 | * advance. | |
3777 | */ | |
3778 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3779 | found_key.offset); | |
2c9fe835 ZL |
3780 | if (ret < 0) { |
3781 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3782 | goto error; | |
a6f93c71 LB |
3783 | } else if (ret == 1) { |
3784 | chunk_reserved = 1; | |
2c9fe835 | 3785 | } |
2c9fe835 ZL |
3786 | } |
3787 | ||
5b4aacef | 3788 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3789 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3790 | if (ret == -ENOSPC) { |
c9e9f97b | 3791 | enospc_errors++; |
eede2bf3 OS |
3792 | } else if (ret == -ETXTBSY) { |
3793 | btrfs_info(fs_info, | |
3794 | "skipping relocation of block group %llu due to active swapfile", | |
3795 | found_key.offset); | |
3796 | ret = 0; | |
3797 | } else if (ret) { | |
3798 | goto error; | |
19a39dce ID |
3799 | } else { |
3800 | spin_lock(&fs_info->balance_lock); | |
3801 | bctl->stat.completed++; | |
3802 | spin_unlock(&fs_info->balance_lock); | |
3803 | } | |
f43ffb60 | 3804 | loop: |
795a3321 ID |
3805 | if (found_key.offset == 0) |
3806 | break; | |
ba1bf481 | 3807 | key.offset = found_key.offset - 1; |
ec44a35c | 3808 | } |
c9e9f97b | 3809 | |
19a39dce ID |
3810 | if (counting) { |
3811 | btrfs_release_path(path); | |
3812 | counting = false; | |
3813 | goto again; | |
3814 | } | |
ec44a35c CM |
3815 | error: |
3816 | btrfs_free_path(path); | |
c9e9f97b | 3817 | if (enospc_errors) { |
efe120a0 | 3818 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3819 | enospc_errors); |
c9e9f97b ID |
3820 | if (!ret) |
3821 | ret = -ENOSPC; | |
3822 | } | |
3823 | ||
ec44a35c CM |
3824 | return ret; |
3825 | } | |
3826 | ||
0c460c0d ID |
3827 | /** |
3828 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3829 | * @flags: profile to validate | |
3830 | * @extended: if true @flags is treated as an extended profile | |
3831 | */ | |
3832 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3833 | { | |
3834 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3835 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3836 | ||
3837 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3838 | ||
3839 | /* 1) check that all other bits are zeroed */ | |
3840 | if (flags & ~mask) | |
3841 | return 0; | |
3842 | ||
3843 | /* 2) see if profile is reduced */ | |
3844 | if (flags == 0) | |
3845 | return !extended; /* "0" is valid for usual profiles */ | |
3846 | ||
3847 | /* true if exactly one bit set */ | |
818255fe | 3848 | return is_power_of_2(flags); |
0c460c0d ID |
3849 | } |
3850 | ||
837d5b6e ID |
3851 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3852 | { | |
a7e99c69 ID |
3853 | /* cancel requested || normal exit path */ |
3854 | return atomic_read(&fs_info->balance_cancel_req) || | |
3855 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3856 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3857 | } |
3858 | ||
bdcd3c97 AM |
3859 | /* Non-zero return value signifies invalidity */ |
3860 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3861 | u64 allowed) | |
3862 | { | |
3863 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3864 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3865 | (bctl_arg->target & ~allowed))); | |
3866 | } | |
3867 | ||
56fc37d9 AJ |
3868 | /* |
3869 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3870 | * @size_buf including the terminating null. The output may be trimmed if it | |
3871 | * does not fit into the provided buffer. | |
3872 | */ | |
3873 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3874 | u32 size_buf) | |
3875 | { | |
3876 | int ret; | |
3877 | u32 size_bp = size_buf; | |
3878 | char *bp = buf; | |
3879 | u64 flags = bargs->flags; | |
3880 | char tmp_buf[128] = {'\0'}; | |
3881 | ||
3882 | if (!flags) | |
3883 | return; | |
3884 | ||
3885 | #define CHECK_APPEND_NOARG(a) \ | |
3886 | do { \ | |
3887 | ret = snprintf(bp, size_bp, (a)); \ | |
3888 | if (ret < 0 || ret >= size_bp) \ | |
3889 | goto out_overflow; \ | |
3890 | size_bp -= ret; \ | |
3891 | bp += ret; \ | |
3892 | } while (0) | |
3893 | ||
3894 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3895 | do { \ | |
3896 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3897 | if (ret < 0 || ret >= size_bp) \ | |
3898 | goto out_overflow; \ | |
3899 | size_bp -= ret; \ | |
3900 | bp += ret; \ | |
3901 | } while (0) | |
3902 | ||
3903 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3904 | do { \ | |
3905 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3906 | if (ret < 0 || ret >= size_bp) \ | |
3907 | goto out_overflow; \ | |
3908 | size_bp -= ret; \ | |
3909 | bp += ret; \ | |
3910 | } while (0) | |
3911 | ||
158da513 DS |
3912 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
3913 | CHECK_APPEND_1ARG("convert=%s,", | |
3914 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
3915 | |
3916 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3917 | CHECK_APPEND_NOARG("soft,"); | |
3918 | ||
3919 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3920 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3921 | sizeof(tmp_buf)); | |
3922 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3923 | } | |
3924 | ||
3925 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3926 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3927 | ||
3928 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3929 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3930 | bargs->usage_min, bargs->usage_max); | |
3931 | ||
3932 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3933 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3934 | ||
3935 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3936 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3937 | bargs->pstart, bargs->pend); | |
3938 | ||
3939 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3940 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3941 | bargs->vstart, bargs->vend); | |
3942 | ||
3943 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3944 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3945 | ||
3946 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3947 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3948 | bargs->limit_min, bargs->limit_max); | |
3949 | ||
3950 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3951 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3952 | bargs->stripes_min, bargs->stripes_max); | |
3953 | ||
3954 | #undef CHECK_APPEND_2ARG | |
3955 | #undef CHECK_APPEND_1ARG | |
3956 | #undef CHECK_APPEND_NOARG | |
3957 | ||
3958 | out_overflow: | |
3959 | ||
3960 | if (size_bp < size_buf) | |
3961 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3962 | else | |
3963 | buf[0] = '\0'; | |
3964 | } | |
3965 | ||
3966 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3967 | { | |
3968 | u32 size_buf = 1024; | |
3969 | char tmp_buf[192] = {'\0'}; | |
3970 | char *buf; | |
3971 | char *bp; | |
3972 | u32 size_bp = size_buf; | |
3973 | int ret; | |
3974 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3975 | ||
3976 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3977 | if (!buf) | |
3978 | return; | |
3979 | ||
3980 | bp = buf; | |
3981 | ||
3982 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3983 | do { \ | |
3984 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3985 | if (ret < 0 || ret >= size_bp) \ | |
3986 | goto out_overflow; \ | |
3987 | size_bp -= ret; \ | |
3988 | bp += ret; \ | |
3989 | } while (0) | |
3990 | ||
3991 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
3992 | CHECK_APPEND_1ARG("%s", "-f "); | |
3993 | ||
3994 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
3995 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
3996 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
3997 | } | |
3998 | ||
3999 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4000 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4001 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4002 | } | |
4003 | ||
4004 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4005 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4006 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4007 | } | |
4008 | ||
4009 | #undef CHECK_APPEND_1ARG | |
4010 | ||
4011 | out_overflow: | |
4012 | ||
4013 | if (size_bp < size_buf) | |
4014 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4015 | btrfs_info(fs_info, "balance: %s %s", | |
4016 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4017 | "resume" : "start", buf); | |
4018 | ||
4019 | kfree(buf); | |
4020 | } | |
4021 | ||
c9e9f97b | 4022 | /* |
dccdb07b | 4023 | * Should be called with balance mutexe held |
c9e9f97b | 4024 | */ |
6fcf6e2b DS |
4025 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4026 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4027 | struct btrfs_ioctl_balance_args *bargs) |
4028 | { | |
14506127 | 4029 | u64 meta_target, data_target; |
f43ffb60 | 4030 | u64 allowed; |
e4837f8f | 4031 | int mixed = 0; |
c9e9f97b | 4032 | int ret; |
8dabb742 | 4033 | u64 num_devices; |
de98ced9 | 4034 | unsigned seq; |
5a8067c0 | 4035 | bool reducing_integrity; |
081db89b | 4036 | int i; |
c9e9f97b | 4037 | |
837d5b6e | 4038 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4039 | atomic_read(&fs_info->balance_pause_req) || |
4040 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4041 | ret = -EINVAL; |
4042 | goto out; | |
4043 | } | |
4044 | ||
e4837f8f ID |
4045 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4046 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4047 | mixed = 1; | |
4048 | ||
f43ffb60 ID |
4049 | /* |
4050 | * In case of mixed groups both data and meta should be picked, | |
4051 | * and identical options should be given for both of them. | |
4052 | */ | |
e4837f8f ID |
4053 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4054 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4055 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4056 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4057 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4058 | btrfs_err(fs_info, |
6dac13f8 | 4059 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4060 | ret = -EINVAL; |
4061 | goto out; | |
4062 | } | |
4063 | } | |
4064 | ||
1da73967 | 4065 | num_devices = btrfs_num_devices(fs_info); |
081db89b DS |
4066 | allowed = 0; |
4067 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) | |
4068 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4069 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4070 | |
bdcd3c97 | 4071 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e | 4072 | btrfs_err(fs_info, |
6dac13f8 | 4073 | "balance: invalid convert data profile %s", |
158da513 | 4074 | btrfs_bg_type_to_raid_name(bctl->data.target)); |
e4d8ec0f ID |
4075 | ret = -EINVAL; |
4076 | goto out; | |
4077 | } | |
bdcd3c97 | 4078 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 4079 | btrfs_err(fs_info, |
6dac13f8 | 4080 | "balance: invalid convert metadata profile %s", |
158da513 | 4081 | btrfs_bg_type_to_raid_name(bctl->meta.target)); |
e4d8ec0f ID |
4082 | ret = -EINVAL; |
4083 | goto out; | |
4084 | } | |
bdcd3c97 | 4085 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 4086 | btrfs_err(fs_info, |
6dac13f8 | 4087 | "balance: invalid convert system profile %s", |
158da513 | 4088 | btrfs_bg_type_to_raid_name(bctl->sys.target)); |
e4d8ec0f ID |
4089 | ret = -EINVAL; |
4090 | goto out; | |
4091 | } | |
4092 | ||
6079e12c DS |
4093 | /* |
4094 | * Allow to reduce metadata or system integrity only if force set for | |
4095 | * profiles with redundancy (copies, parity) | |
4096 | */ | |
4097 | allowed = 0; | |
4098 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4099 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4100 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4101 | allowed |= btrfs_raid_array[i].bg_flag; | |
4102 | } | |
de98ced9 MX |
4103 | do { |
4104 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4105 | ||
4106 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4107 | (fs_info->avail_system_alloc_bits & allowed) && | |
4108 | !(bctl->sys.target & allowed)) || | |
4109 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4110 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4111 | !(bctl->meta.target & allowed))) |
4112 | reducing_integrity = true; | |
4113 | else | |
4114 | reducing_integrity = false; | |
4115 | ||
4116 | /* if we're not converting, the target field is uninitialized */ | |
4117 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4118 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4119 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4120 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4121 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4122 | |
5a8067c0 FM |
4123 | if (reducing_integrity) { |
4124 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4125 | btrfs_info(fs_info, | |
4126 | "balance: force reducing metadata integrity"); | |
4127 | } else { | |
4128 | btrfs_err(fs_info, | |
4129 | "balance: reduces metadata integrity, use --force if you want this"); | |
4130 | ret = -EINVAL; | |
4131 | goto out; | |
4132 | } | |
4133 | } | |
4134 | ||
14506127 AB |
4135 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4136 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4137 | btrfs_warn(fs_info, |
6dac13f8 | 4138 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4139 | btrfs_bg_type_to_raid_name(meta_target), |
4140 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4141 | } |
4142 | ||
9e967495 FM |
4143 | if (fs_info->send_in_progress) { |
4144 | btrfs_warn_rl(fs_info, | |
4145 | "cannot run balance while send operations are in progress (%d in progress)", | |
4146 | fs_info->send_in_progress); | |
4147 | ret = -EAGAIN; | |
4148 | goto out; | |
4149 | } | |
4150 | ||
6bccf3ab | 4151 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4152 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4153 | goto out; |
4154 | ||
59641015 ID |
4155 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4156 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4157 | BUG_ON(fs_info->balance_ctl); |
4158 | spin_lock(&fs_info->balance_lock); | |
4159 | fs_info->balance_ctl = bctl; | |
4160 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4161 | } else { |
4162 | BUG_ON(ret != -EEXIST); | |
4163 | spin_lock(&fs_info->balance_lock); | |
4164 | update_balance_args(bctl); | |
4165 | spin_unlock(&fs_info->balance_lock); | |
4166 | } | |
c9e9f97b | 4167 | |
3009a62f DS |
4168 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4169 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4170 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4171 | mutex_unlock(&fs_info->balance_mutex); |
4172 | ||
4173 | ret = __btrfs_balance(fs_info); | |
4174 | ||
4175 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4176 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4177 | btrfs_info(fs_info, "balance: paused"); | |
4178 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4179 | btrfs_info(fs_info, "balance: canceled"); | |
4180 | else | |
4181 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4182 | ||
3009a62f | 4183 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4184 | |
4185 | if (bargs) { | |
4186 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4187 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4188 | } |
4189 | ||
3a01aa7a ID |
4190 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4191 | balance_need_close(fs_info)) { | |
149196a2 | 4192 | reset_balance_state(fs_info); |
a17c95df | 4193 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4194 | } |
4195 | ||
837d5b6e | 4196 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4197 | |
4198 | return ret; | |
4199 | out: | |
59641015 | 4200 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4201 | reset_balance_state(fs_info); |
a17c95df | 4202 | else |
59641015 | 4203 | kfree(bctl); |
a17c95df DS |
4204 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4205 | ||
59641015 ID |
4206 | return ret; |
4207 | } | |
4208 | ||
4209 | static int balance_kthread(void *data) | |
4210 | { | |
2b6ba629 | 4211 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4212 | int ret = 0; |
59641015 | 4213 | |
59641015 | 4214 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4215 | if (fs_info->balance_ctl) |
6fcf6e2b | 4216 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4217 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4218 | |
59641015 ID |
4219 | return ret; |
4220 | } | |
4221 | ||
2b6ba629 ID |
4222 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4223 | { | |
4224 | struct task_struct *tsk; | |
4225 | ||
1354e1a1 | 4226 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4227 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4228 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4229 | return 0; |
4230 | } | |
1354e1a1 | 4231 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4232 | |
3cdde224 | 4233 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4234 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4235 | return 0; |
4236 | } | |
4237 | ||
02ee654d AJ |
4238 | /* |
4239 | * A ro->rw remount sequence should continue with the paused balance | |
4240 | * regardless of who pauses it, system or the user as of now, so set | |
4241 | * the resume flag. | |
4242 | */ | |
4243 | spin_lock(&fs_info->balance_lock); | |
4244 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4245 | spin_unlock(&fs_info->balance_lock); | |
4246 | ||
2b6ba629 | 4247 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4248 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4249 | } |
4250 | ||
68310a5e | 4251 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4252 | { |
59641015 ID |
4253 | struct btrfs_balance_control *bctl; |
4254 | struct btrfs_balance_item *item; | |
4255 | struct btrfs_disk_balance_args disk_bargs; | |
4256 | struct btrfs_path *path; | |
4257 | struct extent_buffer *leaf; | |
4258 | struct btrfs_key key; | |
4259 | int ret; | |
4260 | ||
4261 | path = btrfs_alloc_path(); | |
4262 | if (!path) | |
4263 | return -ENOMEM; | |
4264 | ||
59641015 | 4265 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4266 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4267 | key.offset = 0; |
4268 | ||
68310a5e | 4269 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4270 | if (ret < 0) |
68310a5e | 4271 | goto out; |
59641015 ID |
4272 | if (ret > 0) { /* ret = -ENOENT; */ |
4273 | ret = 0; | |
68310a5e ID |
4274 | goto out; |
4275 | } | |
4276 | ||
4277 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4278 | if (!bctl) { | |
4279 | ret = -ENOMEM; | |
4280 | goto out; | |
59641015 ID |
4281 | } |
4282 | ||
4283 | leaf = path->nodes[0]; | |
4284 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4285 | ||
68310a5e ID |
4286 | bctl->flags = btrfs_balance_flags(leaf, item); |
4287 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4288 | |
4289 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4290 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4291 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4292 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4293 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4294 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4295 | ||
eee95e3f DS |
4296 | /* |
4297 | * This should never happen, as the paused balance state is recovered | |
4298 | * during mount without any chance of other exclusive ops to collide. | |
4299 | * | |
4300 | * This gives the exclusive op status to balance and keeps in paused | |
4301 | * state until user intervention (cancel or umount). If the ownership | |
4302 | * cannot be assigned, show a message but do not fail. The balance | |
4303 | * is in a paused state and must have fs_info::balance_ctl properly | |
4304 | * set up. | |
4305 | */ | |
4306 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4307 | btrfs_warn(fs_info, | |
6dac13f8 | 4308 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4309 | |
68310a5e | 4310 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4311 | BUG_ON(fs_info->balance_ctl); |
4312 | spin_lock(&fs_info->balance_lock); | |
4313 | fs_info->balance_ctl = bctl; | |
4314 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4315 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4316 | out: |
4317 | btrfs_free_path(path); | |
ec44a35c CM |
4318 | return ret; |
4319 | } | |
4320 | ||
837d5b6e ID |
4321 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4322 | { | |
4323 | int ret = 0; | |
4324 | ||
4325 | mutex_lock(&fs_info->balance_mutex); | |
4326 | if (!fs_info->balance_ctl) { | |
4327 | mutex_unlock(&fs_info->balance_mutex); | |
4328 | return -ENOTCONN; | |
4329 | } | |
4330 | ||
3009a62f | 4331 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4332 | atomic_inc(&fs_info->balance_pause_req); |
4333 | mutex_unlock(&fs_info->balance_mutex); | |
4334 | ||
4335 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4336 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4337 | |
4338 | mutex_lock(&fs_info->balance_mutex); | |
4339 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4340 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4341 | atomic_dec(&fs_info->balance_pause_req); |
4342 | } else { | |
4343 | ret = -ENOTCONN; | |
4344 | } | |
4345 | ||
4346 | mutex_unlock(&fs_info->balance_mutex); | |
4347 | return ret; | |
4348 | } | |
4349 | ||
a7e99c69 ID |
4350 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4351 | { | |
4352 | mutex_lock(&fs_info->balance_mutex); | |
4353 | if (!fs_info->balance_ctl) { | |
4354 | mutex_unlock(&fs_info->balance_mutex); | |
4355 | return -ENOTCONN; | |
4356 | } | |
4357 | ||
cf7d20f4 DS |
4358 | /* |
4359 | * A paused balance with the item stored on disk can be resumed at | |
4360 | * mount time if the mount is read-write. Otherwise it's still paused | |
4361 | * and we must not allow cancelling as it deletes the item. | |
4362 | */ | |
4363 | if (sb_rdonly(fs_info->sb)) { | |
4364 | mutex_unlock(&fs_info->balance_mutex); | |
4365 | return -EROFS; | |
4366 | } | |
4367 | ||
a7e99c69 ID |
4368 | atomic_inc(&fs_info->balance_cancel_req); |
4369 | /* | |
4370 | * if we are running just wait and return, balance item is | |
4371 | * deleted in btrfs_balance in this case | |
4372 | */ | |
3009a62f | 4373 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4374 | mutex_unlock(&fs_info->balance_mutex); |
4375 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4376 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4377 | mutex_lock(&fs_info->balance_mutex); |
4378 | } else { | |
a7e99c69 | 4379 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4380 | /* |
4381 | * Lock released to allow other waiters to continue, we'll | |
4382 | * reexamine the status again. | |
4383 | */ | |
a7e99c69 ID |
4384 | mutex_lock(&fs_info->balance_mutex); |
4385 | ||
a17c95df | 4386 | if (fs_info->balance_ctl) { |
149196a2 | 4387 | reset_balance_state(fs_info); |
a17c95df | 4388 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4389 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4390 | } |
a7e99c69 ID |
4391 | } |
4392 | ||
3009a62f DS |
4393 | BUG_ON(fs_info->balance_ctl || |
4394 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4395 | atomic_dec(&fs_info->balance_cancel_req); |
4396 | mutex_unlock(&fs_info->balance_mutex); | |
4397 | return 0; | |
4398 | } | |
4399 | ||
803b2f54 SB |
4400 | static int btrfs_uuid_scan_kthread(void *data) |
4401 | { | |
4402 | struct btrfs_fs_info *fs_info = data; | |
4403 | struct btrfs_root *root = fs_info->tree_root; | |
4404 | struct btrfs_key key; | |
803b2f54 SB |
4405 | struct btrfs_path *path = NULL; |
4406 | int ret = 0; | |
4407 | struct extent_buffer *eb; | |
4408 | int slot; | |
4409 | struct btrfs_root_item root_item; | |
4410 | u32 item_size; | |
f45388f3 | 4411 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4412 | |
4413 | path = btrfs_alloc_path(); | |
4414 | if (!path) { | |
4415 | ret = -ENOMEM; | |
4416 | goto out; | |
4417 | } | |
4418 | ||
4419 | key.objectid = 0; | |
4420 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4421 | key.offset = 0; | |
4422 | ||
803b2f54 | 4423 | while (1) { |
7c829b72 AJ |
4424 | ret = btrfs_search_forward(root, &key, path, |
4425 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4426 | if (ret) { |
4427 | if (ret > 0) | |
4428 | ret = 0; | |
4429 | break; | |
4430 | } | |
4431 | ||
4432 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4433 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4434 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4435 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4436 | goto skip; | |
4437 | ||
4438 | eb = path->nodes[0]; | |
4439 | slot = path->slots[0]; | |
4440 | item_size = btrfs_item_size_nr(eb, slot); | |
4441 | if (item_size < sizeof(root_item)) | |
4442 | goto skip; | |
4443 | ||
803b2f54 SB |
4444 | read_extent_buffer(eb, &root_item, |
4445 | btrfs_item_ptr_offset(eb, slot), | |
4446 | (int)sizeof(root_item)); | |
4447 | if (btrfs_root_refs(&root_item) == 0) | |
4448 | goto skip; | |
f45388f3 FDBM |
4449 | |
4450 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4451 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4452 | if (trans) | |
4453 | goto update_tree; | |
4454 | ||
4455 | btrfs_release_path(path); | |
803b2f54 SB |
4456 | /* |
4457 | * 1 - subvol uuid item | |
4458 | * 1 - received_subvol uuid item | |
4459 | */ | |
4460 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4461 | if (IS_ERR(trans)) { | |
4462 | ret = PTR_ERR(trans); | |
4463 | break; | |
4464 | } | |
f45388f3 FDBM |
4465 | continue; |
4466 | } else { | |
4467 | goto skip; | |
4468 | } | |
4469 | update_tree: | |
4470 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4471 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4472 | BTRFS_UUID_KEY_SUBVOL, |
4473 | key.objectid); | |
4474 | if (ret < 0) { | |
efe120a0 | 4475 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4476 | ret); |
803b2f54 SB |
4477 | break; |
4478 | } | |
4479 | } | |
4480 | ||
4481 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4482 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4483 | root_item.received_uuid, |
4484 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4485 | key.objectid); | |
4486 | if (ret < 0) { | |
efe120a0 | 4487 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4488 | ret); |
803b2f54 SB |
4489 | break; |
4490 | } | |
4491 | } | |
4492 | ||
f45388f3 | 4493 | skip: |
803b2f54 | 4494 | if (trans) { |
3a45bb20 | 4495 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4496 | trans = NULL; |
803b2f54 SB |
4497 | if (ret) |
4498 | break; | |
4499 | } | |
4500 | ||
803b2f54 SB |
4501 | btrfs_release_path(path); |
4502 | if (key.offset < (u64)-1) { | |
4503 | key.offset++; | |
4504 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4505 | key.offset = 0; | |
4506 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4507 | } else if (key.objectid < (u64)-1) { | |
4508 | key.offset = 0; | |
4509 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4510 | key.objectid++; | |
4511 | } else { | |
4512 | break; | |
4513 | } | |
4514 | cond_resched(); | |
4515 | } | |
4516 | ||
4517 | out: | |
4518 | btrfs_free_path(path); | |
f45388f3 | 4519 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4520 | btrfs_end_transaction(trans); |
803b2f54 | 4521 | if (ret) |
efe120a0 | 4522 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4523 | else |
afcdd129 | 4524 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4525 | up(&fs_info->uuid_tree_rescan_sem); |
4526 | return 0; | |
4527 | } | |
4528 | ||
70f80175 SB |
4529 | /* |
4530 | * Callback for btrfs_uuid_tree_iterate(). | |
4531 | * returns: | |
4532 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4533 | * < 0 if an error occurred. |
70f80175 SB |
4534 | * > 0 if the check failed, which means the caller shall remove the entry. |
4535 | */ | |
4536 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4537 | u8 *uuid, u8 type, u64 subid) | |
4538 | { | |
4539 | struct btrfs_key key; | |
4540 | int ret = 0; | |
4541 | struct btrfs_root *subvol_root; | |
4542 | ||
4543 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4544 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4545 | goto out; | |
4546 | ||
4547 | key.objectid = subid; | |
4548 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4549 | key.offset = (u64)-1; | |
4550 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4551 | if (IS_ERR(subvol_root)) { | |
4552 | ret = PTR_ERR(subvol_root); | |
4553 | if (ret == -ENOENT) | |
4554 | ret = 1; | |
4555 | goto out; | |
4556 | } | |
4557 | ||
4558 | switch (type) { | |
4559 | case BTRFS_UUID_KEY_SUBVOL: | |
4560 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4561 | ret = 1; | |
4562 | break; | |
4563 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4564 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4565 | BTRFS_UUID_SIZE)) | |
4566 | ret = 1; | |
4567 | break; | |
4568 | } | |
4569 | ||
4570 | out: | |
4571 | return ret; | |
4572 | } | |
4573 | ||
4574 | static int btrfs_uuid_rescan_kthread(void *data) | |
4575 | { | |
4576 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4577 | int ret; | |
4578 | ||
4579 | /* | |
4580 | * 1st step is to iterate through the existing UUID tree and | |
4581 | * to delete all entries that contain outdated data. | |
4582 | * 2nd step is to add all missing entries to the UUID tree. | |
4583 | */ | |
4584 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4585 | if (ret < 0) { | |
efe120a0 | 4586 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4587 | up(&fs_info->uuid_tree_rescan_sem); |
4588 | return ret; | |
4589 | } | |
4590 | return btrfs_uuid_scan_kthread(data); | |
4591 | } | |
4592 | ||
f7a81ea4 SB |
4593 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4594 | { | |
4595 | struct btrfs_trans_handle *trans; | |
4596 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4597 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4598 | struct task_struct *task; |
4599 | int ret; | |
f7a81ea4 SB |
4600 | |
4601 | /* | |
4602 | * 1 - root node | |
4603 | * 1 - root item | |
4604 | */ | |
4605 | trans = btrfs_start_transaction(tree_root, 2); | |
4606 | if (IS_ERR(trans)) | |
4607 | return PTR_ERR(trans); | |
4608 | ||
9b7a2440 | 4609 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4610 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4611 | ret = PTR_ERR(uuid_root); |
66642832 | 4612 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4613 | btrfs_end_transaction(trans); |
6d13f549 | 4614 | return ret; |
f7a81ea4 SB |
4615 | } |
4616 | ||
4617 | fs_info->uuid_root = uuid_root; | |
4618 | ||
3a45bb20 | 4619 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4620 | if (ret) |
4621 | return ret; | |
4622 | ||
4623 | down(&fs_info->uuid_tree_rescan_sem); | |
4624 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4625 | if (IS_ERR(task)) { | |
70f80175 | 4626 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4627 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4628 | up(&fs_info->uuid_tree_rescan_sem); |
4629 | return PTR_ERR(task); | |
4630 | } | |
4631 | ||
4632 | return 0; | |
f7a81ea4 | 4633 | } |
803b2f54 | 4634 | |
70f80175 SB |
4635 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4636 | { | |
4637 | struct task_struct *task; | |
4638 | ||
4639 | down(&fs_info->uuid_tree_rescan_sem); | |
4640 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4641 | if (IS_ERR(task)) { | |
4642 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4643 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4644 | up(&fs_info->uuid_tree_rescan_sem); |
4645 | return PTR_ERR(task); | |
4646 | } | |
4647 | ||
4648 | return 0; | |
4649 | } | |
4650 | ||
8f18cf13 CM |
4651 | /* |
4652 | * shrinking a device means finding all of the device extents past | |
4653 | * the new size, and then following the back refs to the chunks. | |
4654 | * The chunk relocation code actually frees the device extent | |
4655 | */ | |
4656 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4657 | { | |
0b246afa JM |
4658 | struct btrfs_fs_info *fs_info = device->fs_info; |
4659 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4660 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4661 | struct btrfs_dev_extent *dev_extent = NULL; |
4662 | struct btrfs_path *path; | |
4663 | u64 length; | |
8f18cf13 CM |
4664 | u64 chunk_offset; |
4665 | int ret; | |
4666 | int slot; | |
ba1bf481 JB |
4667 | int failed = 0; |
4668 | bool retried = false; | |
8f18cf13 CM |
4669 | struct extent_buffer *l; |
4670 | struct btrfs_key key; | |
0b246afa | 4671 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4672 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4673 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4674 | u64 diff; |
61d0d0d2 | 4675 | u64 start; |
7dfb8be1 NB |
4676 | |
4677 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4678 | start = new_size; |
0e4324a4 | 4679 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4680 | |
401e29c1 | 4681 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4682 | return -EINVAL; |
4683 | ||
8f18cf13 CM |
4684 | path = btrfs_alloc_path(); |
4685 | if (!path) | |
4686 | return -ENOMEM; | |
4687 | ||
0338dff6 | 4688 | path->reada = READA_BACK; |
8f18cf13 | 4689 | |
61d0d0d2 NB |
4690 | trans = btrfs_start_transaction(root, 0); |
4691 | if (IS_ERR(trans)) { | |
4692 | btrfs_free_path(path); | |
4693 | return PTR_ERR(trans); | |
4694 | } | |
4695 | ||
34441361 | 4696 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4697 | |
7cc8e58d | 4698 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4699 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4700 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4701 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4702 | } |
61d0d0d2 NB |
4703 | |
4704 | /* | |
4705 | * Once the device's size has been set to the new size, ensure all | |
4706 | * in-memory chunks are synced to disk so that the loop below sees them | |
4707 | * and relocates them accordingly. | |
4708 | */ | |
1c11b63e | 4709 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4710 | mutex_unlock(&fs_info->chunk_mutex); |
4711 | ret = btrfs_commit_transaction(trans); | |
4712 | if (ret) | |
4713 | goto done; | |
4714 | } else { | |
4715 | mutex_unlock(&fs_info->chunk_mutex); | |
4716 | btrfs_end_transaction(trans); | |
4717 | } | |
8f18cf13 | 4718 | |
ba1bf481 | 4719 | again: |
8f18cf13 CM |
4720 | key.objectid = device->devid; |
4721 | key.offset = (u64)-1; | |
4722 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4723 | ||
213e64da | 4724 | do { |
0b246afa | 4725 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4726 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4727 | if (ret < 0) { |
0b246afa | 4728 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4729 | goto done; |
67c5e7d4 | 4730 | } |
8f18cf13 CM |
4731 | |
4732 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4733 | if (ret) |
0b246afa | 4734 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4735 | if (ret < 0) |
4736 | goto done; | |
4737 | if (ret) { | |
4738 | ret = 0; | |
b3b4aa74 | 4739 | btrfs_release_path(path); |
bf1fb512 | 4740 | break; |
8f18cf13 CM |
4741 | } |
4742 | ||
4743 | l = path->nodes[0]; | |
4744 | slot = path->slots[0]; | |
4745 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4746 | ||
ba1bf481 | 4747 | if (key.objectid != device->devid) { |
0b246afa | 4748 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4749 | btrfs_release_path(path); |
bf1fb512 | 4750 | break; |
ba1bf481 | 4751 | } |
8f18cf13 CM |
4752 | |
4753 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4754 | length = btrfs_dev_extent_length(l, dev_extent); | |
4755 | ||
ba1bf481 | 4756 | if (key.offset + length <= new_size) { |
0b246afa | 4757 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4758 | btrfs_release_path(path); |
d6397bae | 4759 | break; |
ba1bf481 | 4760 | } |
8f18cf13 | 4761 | |
8f18cf13 | 4762 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4763 | btrfs_release_path(path); |
8f18cf13 | 4764 | |
a6f93c71 LB |
4765 | /* |
4766 | * We may be relocating the only data chunk we have, | |
4767 | * which could potentially end up with losing data's | |
4768 | * raid profile, so lets allocate an empty one in | |
4769 | * advance. | |
4770 | */ | |
4771 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4772 | if (ret < 0) { | |
4773 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4774 | goto done; | |
4775 | } | |
4776 | ||
0b246afa JM |
4777 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4778 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4779 | if (ret == -ENOSPC) { |
ba1bf481 | 4780 | failed++; |
eede2bf3 OS |
4781 | } else if (ret) { |
4782 | if (ret == -ETXTBSY) { | |
4783 | btrfs_warn(fs_info, | |
4784 | "could not shrink block group %llu due to active swapfile", | |
4785 | chunk_offset); | |
4786 | } | |
4787 | goto done; | |
4788 | } | |
213e64da | 4789 | } while (key.offset-- > 0); |
ba1bf481 JB |
4790 | |
4791 | if (failed && !retried) { | |
4792 | failed = 0; | |
4793 | retried = true; | |
4794 | goto again; | |
4795 | } else if (failed && retried) { | |
4796 | ret = -ENOSPC; | |
ba1bf481 | 4797 | goto done; |
8f18cf13 CM |
4798 | } |
4799 | ||
d6397bae | 4800 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4801 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4802 | if (IS_ERR(trans)) { |
4803 | ret = PTR_ERR(trans); | |
4804 | goto done; | |
4805 | } | |
4806 | ||
34441361 | 4807 | mutex_lock(&fs_info->chunk_mutex); |
7cc8e58d | 4808 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4809 | if (list_empty(&device->post_commit_list)) |
4810 | list_add_tail(&device->post_commit_list, | |
4811 | &trans->transaction->dev_update_list); | |
d6397bae | 4812 | |
d6397bae | 4813 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4814 | btrfs_set_super_total_bytes(super_copy, |
4815 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4816 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4817 | |
4818 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4819 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4820 | if (ret < 0) { |
4821 | btrfs_abort_transaction(trans, ret); | |
4822 | btrfs_end_transaction(trans); | |
4823 | } else { | |
4824 | ret = btrfs_commit_transaction(trans); | |
4825 | } | |
8f18cf13 CM |
4826 | done: |
4827 | btrfs_free_path(path); | |
53e489bc | 4828 | if (ret) { |
34441361 | 4829 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4830 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4831 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4832 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4833 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4834 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4835 | } |
8f18cf13 CM |
4836 | return ret; |
4837 | } | |
4838 | ||
2ff7e61e | 4839 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4840 | struct btrfs_key *key, |
4841 | struct btrfs_chunk *chunk, int item_size) | |
4842 | { | |
0b246afa | 4843 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4844 | struct btrfs_disk_key disk_key; |
4845 | u32 array_size; | |
4846 | u8 *ptr; | |
4847 | ||
34441361 | 4848 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4849 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4850 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4851 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4852 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4853 | return -EFBIG; |
fe48a5c0 | 4854 | } |
0b86a832 CM |
4855 | |
4856 | ptr = super_copy->sys_chunk_array + array_size; | |
4857 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4858 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4859 | ptr += sizeof(disk_key); | |
4860 | memcpy(ptr, chunk, item_size); | |
4861 | item_size += sizeof(disk_key); | |
4862 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4863 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4864 | |
0b86a832 CM |
4865 | return 0; |
4866 | } | |
4867 | ||
73c5de00 AJ |
4868 | /* |
4869 | * sort the devices in descending order by max_avail, total_avail | |
4870 | */ | |
4871 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4872 | { |
73c5de00 AJ |
4873 | const struct btrfs_device_info *di_a = a; |
4874 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4875 | |
73c5de00 | 4876 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4877 | return -1; |
73c5de00 | 4878 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4879 | return 1; |
73c5de00 AJ |
4880 | if (di_a->total_avail > di_b->total_avail) |
4881 | return -1; | |
4882 | if (di_a->total_avail < di_b->total_avail) | |
4883 | return 1; | |
4884 | return 0; | |
b2117a39 | 4885 | } |
0b86a832 | 4886 | |
53b381b3 DW |
4887 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4888 | { | |
ffe2d203 | 4889 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4890 | return; |
4891 | ||
ceda0864 | 4892 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4893 | } |
4894 | ||
73c5de00 | 4895 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4896 | u64 start, u64 type) |
b2117a39 | 4897 | { |
2ff7e61e | 4898 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4899 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4900 | struct btrfs_device *device; |
73c5de00 AJ |
4901 | struct map_lookup *map = NULL; |
4902 | struct extent_map_tree *em_tree; | |
4903 | struct extent_map *em; | |
4904 | struct btrfs_device_info *devices_info = NULL; | |
4905 | u64 total_avail; | |
4906 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4907 | int data_stripes; /* number of stripes that count for |
4908 | block group size */ | |
73c5de00 AJ |
4909 | int sub_stripes; /* sub_stripes info for map */ |
4910 | int dev_stripes; /* stripes per dev */ | |
4911 | int devs_max; /* max devs to use */ | |
4912 | int devs_min; /* min devs needed */ | |
4913 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4914 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4915 | int nparity; /* number of stripes worth of bytes to |
4916 | store parity information */ | |
73c5de00 AJ |
4917 | int ret; |
4918 | u64 max_stripe_size; | |
4919 | u64 max_chunk_size; | |
4920 | u64 stripe_size; | |
23f0ff1e | 4921 | u64 chunk_size; |
73c5de00 AJ |
4922 | int ndevs; |
4923 | int i; | |
4924 | int j; | |
31e50229 | 4925 | int index; |
593060d7 | 4926 | |
0c460c0d | 4927 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4928 | |
4117f207 QW |
4929 | if (list_empty(&fs_devices->alloc_list)) { |
4930 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4931 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4932 | return -ENOSPC; |
4117f207 | 4933 | } |
b2117a39 | 4934 | |
3e72ee88 | 4935 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4936 | |
31e50229 LB |
4937 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4938 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4939 | devs_max = btrfs_raid_array[index].devs_max; | |
e3ecdb3f DS |
4940 | if (!devs_max) |
4941 | devs_max = BTRFS_MAX_DEVS(info); | |
31e50229 LB |
4942 | devs_min = btrfs_raid_array[index].devs_min; |
4943 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4944 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4945 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4946 | |
9b3f68b9 | 4947 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4948 | max_stripe_size = SZ_1G; |
fce466ea | 4949 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
9b3f68b9 | 4950 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4951 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4952 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4953 | max_stripe_size = SZ_1G; | |
1100373f | 4954 | else |
ee22184b | 4955 | max_stripe_size = SZ_256M; |
73c5de00 | 4956 | max_chunk_size = max_stripe_size; |
a40a90a0 | 4957 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4958 | max_stripe_size = SZ_32M; |
73c5de00 AJ |
4959 | max_chunk_size = 2 * max_stripe_size; |
4960 | } else { | |
351fd353 | 4961 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 4962 | type); |
290342f6 | 4963 | BUG(); |
9b3f68b9 CM |
4964 | } |
4965 | ||
52042d8e | 4966 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
4967 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
4968 | max_chunk_size); | |
9b3f68b9 | 4969 | |
31e818fe | 4970 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4971 | GFP_NOFS); |
4972 | if (!devices_info) | |
4973 | return -ENOMEM; | |
0cad8a11 | 4974 | |
9f680ce0 | 4975 | /* |
73c5de00 AJ |
4976 | * in the first pass through the devices list, we gather information |
4977 | * about the available holes on each device. | |
9f680ce0 | 4978 | */ |
73c5de00 | 4979 | ndevs = 0; |
ebcc9301 | 4980 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4981 | u64 max_avail; |
4982 | u64 dev_offset; | |
b2117a39 | 4983 | |
ebbede42 | 4984 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 4985 | WARN(1, KERN_ERR |
efe120a0 | 4986 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4987 | continue; |
4988 | } | |
b2117a39 | 4989 | |
e12c9621 AJ |
4990 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
4991 | &device->dev_state) || | |
401e29c1 | 4992 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 4993 | continue; |
b2117a39 | 4994 | |
73c5de00 AJ |
4995 | if (device->total_bytes > device->bytes_used) |
4996 | total_avail = device->total_bytes - device->bytes_used; | |
4997 | else | |
4998 | total_avail = 0; | |
38c01b96 | 4999 | |
5000 | /* If there is no space on this device, skip it. */ | |
5001 | if (total_avail == 0) | |
5002 | continue; | |
b2117a39 | 5003 | |
60dfdf25 | 5004 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
5005 | max_stripe_size * dev_stripes, |
5006 | &dev_offset, &max_avail); | |
5007 | if (ret && ret != -ENOSPC) | |
5008 | goto error; | |
b2117a39 | 5009 | |
73c5de00 AJ |
5010 | if (ret == 0) |
5011 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5012 | |
4117f207 QW |
5013 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5014 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5015 | btrfs_debug(info, | |
5016 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5017 | __func__, device->devid, max_avail, | |
5018 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5019 | continue; |
4117f207 | 5020 | } |
b2117a39 | 5021 | |
063d006f ES |
5022 | if (ndevs == fs_devices->rw_devices) { |
5023 | WARN(1, "%s: found more than %llu devices\n", | |
5024 | __func__, fs_devices->rw_devices); | |
5025 | break; | |
5026 | } | |
73c5de00 AJ |
5027 | devices_info[ndevs].dev_offset = dev_offset; |
5028 | devices_info[ndevs].max_avail = max_avail; | |
5029 | devices_info[ndevs].total_avail = total_avail; | |
5030 | devices_info[ndevs].dev = device; | |
5031 | ++ndevs; | |
5032 | } | |
b2117a39 | 5033 | |
73c5de00 AJ |
5034 | /* |
5035 | * now sort the devices by hole size / available space | |
5036 | */ | |
5037 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5038 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5039 | |
73c5de00 | 5040 | /* round down to number of usable stripes */ |
e5600fd6 | 5041 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5042 | |
ba89b802 | 5043 | if (ndevs < devs_min) { |
73c5de00 | 5044 | ret = -ENOSPC; |
4117f207 QW |
5045 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5046 | btrfs_debug(info, | |
5047 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5048 | __func__, ndevs, devs_min); |
4117f207 | 5049 | } |
73c5de00 | 5050 | goto error; |
b2117a39 | 5051 | } |
9f680ce0 | 5052 | |
f148ef4d NB |
5053 | ndevs = min(ndevs, devs_max); |
5054 | ||
73c5de00 | 5055 | /* |
92e222df HK |
5056 | * The primary goal is to maximize the number of stripes, so use as |
5057 | * many devices as possible, even if the stripes are not maximum sized. | |
5058 | * | |
5059 | * The DUP profile stores more than one stripe per device, the | |
5060 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5061 | */ |
92e222df | 5062 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5063 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5064 | |
53b381b3 DW |
5065 | /* |
5066 | * this will have to be fixed for RAID1 and RAID10 over | |
5067 | * more drives | |
5068 | */ | |
b50836ed | 5069 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5070 | |
5071 | /* | |
5072 | * Use the number of data stripes to figure out how big this chunk | |
5073 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5074 | * and compare that answer with the max chunk size. If it's higher, |
5075 | * we try to reduce stripe_size. | |
86db2578 CM |
5076 | */ |
5077 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5078 | /* |
baf92114 HK |
5079 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5080 | * then use it, unless it ends up being even bigger than the | |
5081 | * previous value we had already. | |
86db2578 | 5082 | */ |
baf92114 HK |
5083 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5084 | data_stripes), SZ_16M), | |
793ff2c8 | 5085 | stripe_size); |
86db2578 CM |
5086 | } |
5087 | ||
37db63a4 | 5088 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5089 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5090 | |
5091 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5092 | if (!map) { | |
5093 | ret = -ENOMEM; | |
5094 | goto error; | |
5095 | } | |
5096 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5097 | |
73c5de00 AJ |
5098 | for (i = 0; i < ndevs; ++i) { |
5099 | for (j = 0; j < dev_stripes; ++j) { | |
5100 | int s = i * dev_stripes + j; | |
5101 | map->stripes[s].dev = devices_info[i].dev; | |
5102 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5103 | j * stripe_size; | |
6324fbf3 | 5104 | } |
6324fbf3 | 5105 | } |
500ceed8 NB |
5106 | map->stripe_len = BTRFS_STRIPE_LEN; |
5107 | map->io_align = BTRFS_STRIPE_LEN; | |
5108 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5109 | map->type = type; |
2b82032c | 5110 | map->sub_stripes = sub_stripes; |
0b86a832 | 5111 | |
23f0ff1e | 5112 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5113 | |
23f0ff1e | 5114 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5115 | |
172ddd60 | 5116 | em = alloc_extent_map(); |
2b82032c | 5117 | if (!em) { |
298a8f9c | 5118 | kfree(map); |
b2117a39 MX |
5119 | ret = -ENOMEM; |
5120 | goto error; | |
593060d7 | 5121 | } |
298a8f9c | 5122 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5123 | em->map_lookup = map; |
2b82032c | 5124 | em->start = start; |
23f0ff1e | 5125 | em->len = chunk_size; |
2b82032c YZ |
5126 | em->block_start = 0; |
5127 | em->block_len = em->len; | |
6df9a95e | 5128 | em->orig_block_len = stripe_size; |
593060d7 | 5129 | |
c8bf1b67 | 5130 | em_tree = &info->mapping_tree; |
890871be | 5131 | write_lock(&em_tree->lock); |
09a2a8f9 | 5132 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5133 | if (ret) { |
1efb72a3 | 5134 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5135 | free_extent_map(em); |
1dd4602f | 5136 | goto error; |
0f5d42b2 | 5137 | } |
1efb72a3 NB |
5138 | write_unlock(&em_tree->lock); |
5139 | ||
23f0ff1e | 5140 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5141 | if (ret) |
5142 | goto error_del_extent; | |
2b82032c | 5143 | |
bbbf7243 NB |
5144 | for (i = 0; i < map->num_stripes; i++) { |
5145 | struct btrfs_device *dev = map->stripes[i].dev; | |
5146 | ||
5147 | btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size); | |
5148 | if (list_empty(&dev->post_commit_list)) | |
5149 | list_add_tail(&dev->post_commit_list, | |
5150 | &trans->transaction->dev_update_list); | |
5151 | } | |
43530c46 | 5152 | |
a5ed45f8 | 5153 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5154 | |
0f5d42b2 | 5155 | free_extent_map(em); |
0b246afa | 5156 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5157 | |
b2117a39 | 5158 | kfree(devices_info); |
2b82032c | 5159 | return 0; |
b2117a39 | 5160 | |
6df9a95e | 5161 | error_del_extent: |
0f5d42b2 JB |
5162 | write_lock(&em_tree->lock); |
5163 | remove_extent_mapping(em_tree, em); | |
5164 | write_unlock(&em_tree->lock); | |
5165 | ||
5166 | /* One for our allocation */ | |
5167 | free_extent_map(em); | |
5168 | /* One for the tree reference */ | |
5169 | free_extent_map(em); | |
b2117a39 | 5170 | error: |
b2117a39 MX |
5171 | kfree(devices_info); |
5172 | return ret; | |
2b82032c YZ |
5173 | } |
5174 | ||
6df9a95e | 5175 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5176 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5177 | { |
97aff912 | 5178 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5179 | struct btrfs_root *extent_root = fs_info->extent_root; |
5180 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5181 | struct btrfs_key key; |
2b82032c YZ |
5182 | struct btrfs_device *device; |
5183 | struct btrfs_chunk *chunk; | |
5184 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5185 | struct extent_map *em; |
5186 | struct map_lookup *map; | |
5187 | size_t item_size; | |
5188 | u64 dev_offset; | |
5189 | u64 stripe_size; | |
5190 | int i = 0; | |
140e639f | 5191 | int ret = 0; |
2b82032c | 5192 | |
60ca842e | 5193 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5194 | if (IS_ERR(em)) |
5195 | return PTR_ERR(em); | |
6df9a95e | 5196 | |
95617d69 | 5197 | map = em->map_lookup; |
6df9a95e JB |
5198 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5199 | stripe_size = em->orig_block_len; | |
5200 | ||
2b82032c | 5201 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5202 | if (!chunk) { |
5203 | ret = -ENOMEM; | |
5204 | goto out; | |
5205 | } | |
5206 | ||
50460e37 FM |
5207 | /* |
5208 | * Take the device list mutex to prevent races with the final phase of | |
5209 | * a device replace operation that replaces the device object associated | |
5210 | * with the map's stripes, because the device object's id can change | |
5211 | * at any time during that final phase of the device replace operation | |
5212 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5213 | */ | |
0b246afa | 5214 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5215 | for (i = 0; i < map->num_stripes; i++) { |
5216 | device = map->stripes[i].dev; | |
5217 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5218 | |
0b86a832 | 5219 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5220 | if (ret) |
50460e37 | 5221 | break; |
b5d9071c NB |
5222 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5223 | dev_offset, stripe_size); | |
6df9a95e | 5224 | if (ret) |
50460e37 FM |
5225 | break; |
5226 | } | |
5227 | if (ret) { | |
0b246afa | 5228 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5229 | goto out; |
2b82032c YZ |
5230 | } |
5231 | ||
2b82032c | 5232 | stripe = &chunk->stripe; |
6df9a95e JB |
5233 | for (i = 0; i < map->num_stripes; i++) { |
5234 | device = map->stripes[i].dev; | |
5235 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5236 | |
e17cade2 CM |
5237 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5238 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5239 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5240 | stripe++; |
0b86a832 | 5241 | } |
0b246afa | 5242 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5243 | |
2b82032c | 5244 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5245 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5246 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5247 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5248 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5249 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5250 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5251 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5252 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5253 | |
2b82032c YZ |
5254 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5255 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5256 | key.offset = chunk_offset; | |
0b86a832 | 5257 | |
2b82032c | 5258 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5259 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5260 | /* | |
5261 | * TODO: Cleanup of inserted chunk root in case of | |
5262 | * failure. | |
5263 | */ | |
2ff7e61e | 5264 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5265 | } |
1abe9b8a | 5266 | |
6df9a95e | 5267 | out: |
0b86a832 | 5268 | kfree(chunk); |
6df9a95e | 5269 | free_extent_map(em); |
4ed1d16e | 5270 | return ret; |
2b82032c | 5271 | } |
0b86a832 | 5272 | |
2b82032c | 5273 | /* |
52042d8e AG |
5274 | * Chunk allocation falls into two parts. The first part does work |
5275 | * that makes the new allocated chunk usable, but does not do any operation | |
5276 | * that modifies the chunk tree. The second part does the work that | |
5277 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5278 | * bootstrap process of adding storage to a seed btrfs. |
5279 | */ | |
c216b203 | 5280 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5281 | { |
5282 | u64 chunk_offset; | |
2b82032c | 5283 | |
c216b203 NB |
5284 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5285 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5286 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5287 | } |
5288 | ||
6f8e0fc7 | 5289 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5290 | { |
6f8e0fc7 | 5291 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c YZ |
5292 | u64 chunk_offset; |
5293 | u64 sys_chunk_offset; | |
2b82032c | 5294 | u64 alloc_profile; |
2b82032c YZ |
5295 | int ret; |
5296 | ||
6df9a95e | 5297 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5298 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5299 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5300 | if (ret) |
5301 | return ret; | |
2b82032c | 5302 | |
0b246afa | 5303 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5304 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5305 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5306 | return ret; |
2b82032c YZ |
5307 | } |
5308 | ||
d20983b4 MX |
5309 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5310 | { | |
fc9a2ac7 | 5311 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5312 | |
fc9a2ac7 | 5313 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5314 | } |
5315 | ||
2ff7e61e | 5316 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5317 | { |
5318 | struct extent_map *em; | |
5319 | struct map_lookup *map; | |
2b82032c | 5320 | int readonly = 0; |
d20983b4 | 5321 | int miss_ndevs = 0; |
2b82032c YZ |
5322 | int i; |
5323 | ||
60ca842e | 5324 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5325 | if (IS_ERR(em)) |
2b82032c YZ |
5326 | return 1; |
5327 | ||
95617d69 | 5328 | map = em->map_lookup; |
2b82032c | 5329 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5330 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5331 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5332 | miss_ndevs++; |
5333 | continue; | |
5334 | } | |
ebbede42 AJ |
5335 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5336 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5337 | readonly = 1; |
d20983b4 | 5338 | goto end; |
2b82032c YZ |
5339 | } |
5340 | } | |
d20983b4 MX |
5341 | |
5342 | /* | |
5343 | * If the number of missing devices is larger than max errors, | |
5344 | * we can not write the data into that chunk successfully, so | |
5345 | * set it readonly. | |
5346 | */ | |
5347 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5348 | readonly = 1; | |
5349 | end: | |
0b86a832 | 5350 | free_extent_map(em); |
2b82032c | 5351 | return readonly; |
0b86a832 CM |
5352 | } |
5353 | ||
c8bf1b67 | 5354 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5355 | { |
5356 | struct extent_map *em; | |
5357 | ||
d397712b | 5358 | while (1) { |
c8bf1b67 DS |
5359 | write_lock(&tree->lock); |
5360 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5361 | if (em) |
c8bf1b67 DS |
5362 | remove_extent_mapping(tree, em); |
5363 | write_unlock(&tree->lock); | |
0b86a832 CM |
5364 | if (!em) |
5365 | break; | |
0b86a832 CM |
5366 | /* once for us */ |
5367 | free_extent_map(em); | |
5368 | /* once for the tree */ | |
5369 | free_extent_map(em); | |
5370 | } | |
5371 | } | |
5372 | ||
5d964051 | 5373 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5374 | { |
5375 | struct extent_map *em; | |
5376 | struct map_lookup *map; | |
f188591e CM |
5377 | int ret; |
5378 | ||
60ca842e | 5379 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5380 | if (IS_ERR(em)) |
5381 | /* | |
5382 | * We could return errors for these cases, but that could get | |
5383 | * ugly and we'd probably do the same thing which is just not do | |
5384 | * anything else and exit, so return 1 so the callers don't try | |
5385 | * to use other copies. | |
5386 | */ | |
fb7669b5 | 5387 | return 1; |
fb7669b5 | 5388 | |
95617d69 | 5389 | map = em->map_lookup; |
c7369b3f | 5390 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5391 | ret = map->num_stripes; |
321aecc6 CM |
5392 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5393 | ret = map->sub_stripes; | |
53b381b3 DW |
5394 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5395 | ret = 2; | |
5396 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5397 | /* |
5398 | * There could be two corrupted data stripes, we need | |
5399 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5400 | * |
8810f751 LB |
5401 | * Fail a stripe at a time on every retry except the |
5402 | * stripe under reconstruction. | |
5403 | */ | |
5404 | ret = map->num_stripes; | |
f188591e CM |
5405 | else |
5406 | ret = 1; | |
5407 | free_extent_map(em); | |
ad6d620e | 5408 | |
cb5583dd | 5409 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5410 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5411 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5412 | ret++; |
cb5583dd | 5413 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5414 | |
f188591e CM |
5415 | return ret; |
5416 | } | |
5417 | ||
2ff7e61e | 5418 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5419 | u64 logical) |
5420 | { | |
5421 | struct extent_map *em; | |
5422 | struct map_lookup *map; | |
0b246afa | 5423 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5424 | |
60ca842e | 5425 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5426 | |
69f03f13 NB |
5427 | if (!WARN_ON(IS_ERR(em))) { |
5428 | map = em->map_lookup; | |
5429 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5430 | len = map->stripe_len * nr_data_stripes(map); | |
5431 | free_extent_map(em); | |
5432 | } | |
53b381b3 DW |
5433 | return len; |
5434 | } | |
5435 | ||
e4ff5fb5 | 5436 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5437 | { |
5438 | struct extent_map *em; | |
5439 | struct map_lookup *map; | |
53b381b3 DW |
5440 | int ret = 0; |
5441 | ||
60ca842e | 5442 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5443 | |
69f03f13 NB |
5444 | if(!WARN_ON(IS_ERR(em))) { |
5445 | map = em->map_lookup; | |
5446 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5447 | ret = 1; | |
5448 | free_extent_map(em); | |
5449 | } | |
53b381b3 DW |
5450 | return ret; |
5451 | } | |
5452 | ||
30d9861f | 5453 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5454 | struct map_lookup *map, int first, |
8ba0ae78 | 5455 | int dev_replace_is_ongoing) |
dfe25020 CM |
5456 | { |
5457 | int i; | |
99f92a7c | 5458 | int num_stripes; |
8ba0ae78 | 5459 | int preferred_mirror; |
30d9861f SB |
5460 | int tolerance; |
5461 | struct btrfs_device *srcdev; | |
5462 | ||
99f92a7c | 5463 | ASSERT((map->type & |
c7369b3f | 5464 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5465 | |
5466 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5467 | num_stripes = map->sub_stripes; | |
5468 | else | |
5469 | num_stripes = map->num_stripes; | |
5470 | ||
8ba0ae78 AJ |
5471 | preferred_mirror = first + current->pid % num_stripes; |
5472 | ||
30d9861f SB |
5473 | if (dev_replace_is_ongoing && |
5474 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5475 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5476 | srcdev = fs_info->dev_replace.srcdev; | |
5477 | else | |
5478 | srcdev = NULL; | |
5479 | ||
5480 | /* | |
5481 | * try to avoid the drive that is the source drive for a | |
5482 | * dev-replace procedure, only choose it if no other non-missing | |
5483 | * mirror is available | |
5484 | */ | |
5485 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5486 | if (map->stripes[preferred_mirror].dev->bdev && |
5487 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5488 | return preferred_mirror; | |
99f92a7c | 5489 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5490 | if (map->stripes[i].dev->bdev && |
5491 | (tolerance || map->stripes[i].dev != srcdev)) | |
5492 | return i; | |
5493 | } | |
dfe25020 | 5494 | } |
30d9861f | 5495 | |
dfe25020 CM |
5496 | /* we couldn't find one that doesn't fail. Just return something |
5497 | * and the io error handling code will clean up eventually | |
5498 | */ | |
8ba0ae78 | 5499 | return preferred_mirror; |
dfe25020 CM |
5500 | } |
5501 | ||
53b381b3 DW |
5502 | static inline int parity_smaller(u64 a, u64 b) |
5503 | { | |
5504 | return a > b; | |
5505 | } | |
5506 | ||
5507 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5508 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5509 | { |
5510 | struct btrfs_bio_stripe s; | |
5511 | int i; | |
5512 | u64 l; | |
5513 | int again = 1; | |
5514 | ||
5515 | while (again) { | |
5516 | again = 0; | |
cc7539ed | 5517 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5518 | if (parity_smaller(bbio->raid_map[i], |
5519 | bbio->raid_map[i+1])) { | |
53b381b3 | 5520 | s = bbio->stripes[i]; |
8e5cfb55 | 5521 | l = bbio->raid_map[i]; |
53b381b3 | 5522 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5523 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5524 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5525 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5526 | |
53b381b3 DW |
5527 | again = 1; |
5528 | } | |
5529 | } | |
5530 | } | |
5531 | } | |
5532 | ||
6e9606d2 ZL |
5533 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5534 | { | |
5535 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5536 | /* the size of the btrfs_bio */ |
6e9606d2 | 5537 | sizeof(struct btrfs_bio) + |
e57cf21e | 5538 | /* plus the variable array for the stripes */ |
6e9606d2 | 5539 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5540 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5541 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5542 | /* |
5543 | * plus the raid_map, which includes both the tgt dev | |
5544 | * and the stripes | |
5545 | */ | |
5546 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5547 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5548 | |
5549 | atomic_set(&bbio->error, 0); | |
140475ae | 5550 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5551 | |
5552 | return bbio; | |
5553 | } | |
5554 | ||
5555 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5556 | { | |
140475ae ER |
5557 | WARN_ON(!refcount_read(&bbio->refs)); |
5558 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5559 | } |
5560 | ||
5561 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5562 | { | |
5563 | if (!bbio) | |
5564 | return; | |
140475ae | 5565 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5566 | kfree(bbio); |
5567 | } | |
5568 | ||
0b3d4cd3 LB |
5569 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5570 | /* | |
5571 | * Please note that, discard won't be sent to target device of device | |
5572 | * replace. | |
5573 | */ | |
5574 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5575 | u64 logical, u64 length, | |
5576 | struct btrfs_bio **bbio_ret) | |
5577 | { | |
5578 | struct extent_map *em; | |
5579 | struct map_lookup *map; | |
5580 | struct btrfs_bio *bbio; | |
5581 | u64 offset; | |
5582 | u64 stripe_nr; | |
5583 | u64 stripe_nr_end; | |
5584 | u64 stripe_end_offset; | |
5585 | u64 stripe_cnt; | |
5586 | u64 stripe_len; | |
5587 | u64 stripe_offset; | |
5588 | u64 num_stripes; | |
5589 | u32 stripe_index; | |
5590 | u32 factor = 0; | |
5591 | u32 sub_stripes = 0; | |
5592 | u64 stripes_per_dev = 0; | |
5593 | u32 remaining_stripes = 0; | |
5594 | u32 last_stripe = 0; | |
5595 | int ret = 0; | |
5596 | int i; | |
5597 | ||
5598 | /* discard always return a bbio */ | |
5599 | ASSERT(bbio_ret); | |
5600 | ||
60ca842e | 5601 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5602 | if (IS_ERR(em)) |
5603 | return PTR_ERR(em); | |
5604 | ||
5605 | map = em->map_lookup; | |
5606 | /* we don't discard raid56 yet */ | |
5607 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5608 | ret = -EOPNOTSUPP; | |
5609 | goto out; | |
5610 | } | |
5611 | ||
5612 | offset = logical - em->start; | |
5613 | length = min_t(u64, em->len - offset, length); | |
5614 | ||
5615 | stripe_len = map->stripe_len; | |
5616 | /* | |
5617 | * stripe_nr counts the total number of stripes we have to stride | |
5618 | * to get to this block | |
5619 | */ | |
5620 | stripe_nr = div64_u64(offset, stripe_len); | |
5621 | ||
5622 | /* stripe_offset is the offset of this block in its stripe */ | |
5623 | stripe_offset = offset - stripe_nr * stripe_len; | |
5624 | ||
5625 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5626 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5627 | stripe_cnt = stripe_nr_end - stripe_nr; |
5628 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5629 | (offset + length); | |
5630 | /* | |
5631 | * after this, stripe_nr is the number of stripes on this | |
5632 | * device we have to walk to find the data, and stripe_index is | |
5633 | * the number of our device in the stripe array | |
5634 | */ | |
5635 | num_stripes = 1; | |
5636 | stripe_index = 0; | |
5637 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5638 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5639 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5640 | sub_stripes = 1; | |
5641 | else | |
5642 | sub_stripes = map->sub_stripes; | |
5643 | ||
5644 | factor = map->num_stripes / sub_stripes; | |
5645 | num_stripes = min_t(u64, map->num_stripes, | |
5646 | sub_stripes * stripe_cnt); | |
5647 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5648 | stripe_index *= sub_stripes; | |
5649 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5650 | &remaining_stripes); | |
5651 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5652 | last_stripe *= sub_stripes; | |
c7369b3f | 5653 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5654 | BTRFS_BLOCK_GROUP_DUP)) { |
5655 | num_stripes = map->num_stripes; | |
5656 | } else { | |
5657 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5658 | &stripe_index); | |
5659 | } | |
5660 | ||
5661 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5662 | if (!bbio) { | |
5663 | ret = -ENOMEM; | |
5664 | goto out; | |
5665 | } | |
5666 | ||
5667 | for (i = 0; i < num_stripes; i++) { | |
5668 | bbio->stripes[i].physical = | |
5669 | map->stripes[stripe_index].physical + | |
5670 | stripe_offset + stripe_nr * map->stripe_len; | |
5671 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5672 | ||
5673 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5674 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5675 | bbio->stripes[i].length = stripes_per_dev * | |
5676 | map->stripe_len; | |
5677 | ||
5678 | if (i / sub_stripes < remaining_stripes) | |
5679 | bbio->stripes[i].length += | |
5680 | map->stripe_len; | |
5681 | ||
5682 | /* | |
5683 | * Special for the first stripe and | |
5684 | * the last stripe: | |
5685 | * | |
5686 | * |-------|...|-------| | |
5687 | * |----------| | |
5688 | * off end_off | |
5689 | */ | |
5690 | if (i < sub_stripes) | |
5691 | bbio->stripes[i].length -= | |
5692 | stripe_offset; | |
5693 | ||
5694 | if (stripe_index >= last_stripe && | |
5695 | stripe_index <= (last_stripe + | |
5696 | sub_stripes - 1)) | |
5697 | bbio->stripes[i].length -= | |
5698 | stripe_end_offset; | |
5699 | ||
5700 | if (i == sub_stripes - 1) | |
5701 | stripe_offset = 0; | |
5702 | } else { | |
5703 | bbio->stripes[i].length = length; | |
5704 | } | |
5705 | ||
5706 | stripe_index++; | |
5707 | if (stripe_index == map->num_stripes) { | |
5708 | stripe_index = 0; | |
5709 | stripe_nr++; | |
5710 | } | |
5711 | } | |
5712 | ||
5713 | *bbio_ret = bbio; | |
5714 | bbio->map_type = map->type; | |
5715 | bbio->num_stripes = num_stripes; | |
5716 | out: | |
5717 | free_extent_map(em); | |
5718 | return ret; | |
5719 | } | |
5720 | ||
5ab56090 LB |
5721 | /* |
5722 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5723 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5724 | * left cursor can also be read from the target drive. | |
5725 | * | |
5726 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5727 | * array of stripes. | |
5728 | * For READ, it also needs to be supported using the same mirror number. | |
5729 | * | |
5730 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5731 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5732 | * case. | |
5733 | */ | |
5734 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5735 | u64 logical, u64 length, | |
5736 | u64 srcdev_devid, int *mirror_num, | |
5737 | u64 *physical) | |
5738 | { | |
5739 | struct btrfs_bio *bbio = NULL; | |
5740 | int num_stripes; | |
5741 | int index_srcdev = 0; | |
5742 | int found = 0; | |
5743 | u64 physical_of_found = 0; | |
5744 | int i; | |
5745 | int ret = 0; | |
5746 | ||
5747 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5748 | logical, &length, &bbio, 0, 0); | |
5749 | if (ret) { | |
5750 | ASSERT(bbio == NULL); | |
5751 | return ret; | |
5752 | } | |
5753 | ||
5754 | num_stripes = bbio->num_stripes; | |
5755 | if (*mirror_num > num_stripes) { | |
5756 | /* | |
5757 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5758 | * that means that the requested area is not left of the left | |
5759 | * cursor | |
5760 | */ | |
5761 | btrfs_put_bbio(bbio); | |
5762 | return -EIO; | |
5763 | } | |
5764 | ||
5765 | /* | |
5766 | * process the rest of the function using the mirror_num of the source | |
5767 | * drive. Therefore look it up first. At the end, patch the device | |
5768 | * pointer to the one of the target drive. | |
5769 | */ | |
5770 | for (i = 0; i < num_stripes; i++) { | |
5771 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5772 | continue; | |
5773 | ||
5774 | /* | |
5775 | * In case of DUP, in order to keep it simple, only add the | |
5776 | * mirror with the lowest physical address | |
5777 | */ | |
5778 | if (found && | |
5779 | physical_of_found <= bbio->stripes[i].physical) | |
5780 | continue; | |
5781 | ||
5782 | index_srcdev = i; | |
5783 | found = 1; | |
5784 | physical_of_found = bbio->stripes[i].physical; | |
5785 | } | |
5786 | ||
5787 | btrfs_put_bbio(bbio); | |
5788 | ||
5789 | ASSERT(found); | |
5790 | if (!found) | |
5791 | return -EIO; | |
5792 | ||
5793 | *mirror_num = index_srcdev + 1; | |
5794 | *physical = physical_of_found; | |
5795 | return ret; | |
5796 | } | |
5797 | ||
73c0f228 LB |
5798 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5799 | struct btrfs_bio **bbio_ret, | |
5800 | struct btrfs_dev_replace *dev_replace, | |
5801 | int *num_stripes_ret, int *max_errors_ret) | |
5802 | { | |
5803 | struct btrfs_bio *bbio = *bbio_ret; | |
5804 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5805 | int tgtdev_indexes = 0; | |
5806 | int num_stripes = *num_stripes_ret; | |
5807 | int max_errors = *max_errors_ret; | |
5808 | int i; | |
5809 | ||
5810 | if (op == BTRFS_MAP_WRITE) { | |
5811 | int index_where_to_add; | |
5812 | ||
5813 | /* | |
5814 | * duplicate the write operations while the dev replace | |
5815 | * procedure is running. Since the copying of the old disk to | |
5816 | * the new disk takes place at run time while the filesystem is | |
5817 | * mounted writable, the regular write operations to the old | |
5818 | * disk have to be duplicated to go to the new disk as well. | |
5819 | * | |
5820 | * Note that device->missing is handled by the caller, and that | |
5821 | * the write to the old disk is already set up in the stripes | |
5822 | * array. | |
5823 | */ | |
5824 | index_where_to_add = num_stripes; | |
5825 | for (i = 0; i < num_stripes; i++) { | |
5826 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5827 | /* write to new disk, too */ | |
5828 | struct btrfs_bio_stripe *new = | |
5829 | bbio->stripes + index_where_to_add; | |
5830 | struct btrfs_bio_stripe *old = | |
5831 | bbio->stripes + i; | |
5832 | ||
5833 | new->physical = old->physical; | |
5834 | new->length = old->length; | |
5835 | new->dev = dev_replace->tgtdev; | |
5836 | bbio->tgtdev_map[i] = index_where_to_add; | |
5837 | index_where_to_add++; | |
5838 | max_errors++; | |
5839 | tgtdev_indexes++; | |
5840 | } | |
5841 | } | |
5842 | num_stripes = index_where_to_add; | |
5843 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5844 | int index_srcdev = 0; | |
5845 | int found = 0; | |
5846 | u64 physical_of_found = 0; | |
5847 | ||
5848 | /* | |
5849 | * During the dev-replace procedure, the target drive can also | |
5850 | * be used to read data in case it is needed to repair a corrupt | |
5851 | * block elsewhere. This is possible if the requested area is | |
5852 | * left of the left cursor. In this area, the target drive is a | |
5853 | * full copy of the source drive. | |
5854 | */ | |
5855 | for (i = 0; i < num_stripes; i++) { | |
5856 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5857 | /* | |
5858 | * In case of DUP, in order to keep it simple, | |
5859 | * only add the mirror with the lowest physical | |
5860 | * address | |
5861 | */ | |
5862 | if (found && | |
5863 | physical_of_found <= | |
5864 | bbio->stripes[i].physical) | |
5865 | continue; | |
5866 | index_srcdev = i; | |
5867 | found = 1; | |
5868 | physical_of_found = bbio->stripes[i].physical; | |
5869 | } | |
5870 | } | |
5871 | if (found) { | |
5872 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5873 | bbio->stripes + num_stripes; | |
5874 | ||
5875 | tgtdev_stripe->physical = physical_of_found; | |
5876 | tgtdev_stripe->length = | |
5877 | bbio->stripes[index_srcdev].length; | |
5878 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5879 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5880 | ||
5881 | tgtdev_indexes++; | |
5882 | num_stripes++; | |
5883 | } | |
5884 | } | |
5885 | ||
5886 | *num_stripes_ret = num_stripes; | |
5887 | *max_errors_ret = max_errors; | |
5888 | bbio->num_tgtdevs = tgtdev_indexes; | |
5889 | *bbio_ret = bbio; | |
5890 | } | |
5891 | ||
2b19a1fe LB |
5892 | static bool need_full_stripe(enum btrfs_map_op op) |
5893 | { | |
5894 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5895 | } | |
5896 | ||
5f141126 NB |
5897 | /* |
5898 | * btrfs_get_io_geometry - calculates the geomery of a particular (address, len) | |
5899 | * tuple. This information is used to calculate how big a | |
5900 | * particular bio can get before it straddles a stripe. | |
5901 | * | |
5902 | * @fs_info - the filesystem | |
5903 | * @logical - address that we want to figure out the geometry of | |
5904 | * @len - the length of IO we are going to perform, starting at @logical | |
5905 | * @op - type of operation - write or read | |
5906 | * @io_geom - pointer used to return values | |
5907 | * | |
5908 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
5909 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
5910 | */ | |
5911 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, | |
89b798ad | 5912 | u64 logical, u64 len, struct btrfs_io_geometry *io_geom) |
5f141126 NB |
5913 | { |
5914 | struct extent_map *em; | |
5915 | struct map_lookup *map; | |
5916 | u64 offset; | |
5917 | u64 stripe_offset; | |
5918 | u64 stripe_nr; | |
5919 | u64 stripe_len; | |
5920 | u64 raid56_full_stripe_start = (u64)-1; | |
5921 | int data_stripes; | |
373c3b80 | 5922 | int ret = 0; |
5f141126 NB |
5923 | |
5924 | ASSERT(op != BTRFS_MAP_DISCARD); | |
5925 | ||
5926 | em = btrfs_get_chunk_map(fs_info, logical, len); | |
5927 | if (IS_ERR(em)) | |
5928 | return PTR_ERR(em); | |
5929 | ||
5930 | map = em->map_lookup; | |
5931 | /* Offset of this logical address in the chunk */ | |
5932 | offset = logical - em->start; | |
5933 | /* Len of a stripe in a chunk */ | |
5934 | stripe_len = map->stripe_len; | |
5935 | /* Stripe wher this block falls in */ | |
5936 | stripe_nr = div64_u64(offset, stripe_len); | |
5937 | /* Offset of stripe in the chunk */ | |
5938 | stripe_offset = stripe_nr * stripe_len; | |
5939 | if (offset < stripe_offset) { | |
5940 | btrfs_crit(fs_info, | |
5941 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
5942 | stripe_offset, offset, em->start, logical, stripe_len); | |
373c3b80 JT |
5943 | ret = -EINVAL; |
5944 | goto out; | |
5f141126 NB |
5945 | } |
5946 | ||
5947 | /* stripe_offset is the offset of this block in its stripe */ | |
5948 | stripe_offset = offset - stripe_offset; | |
5949 | data_stripes = nr_data_stripes(map); | |
5950 | ||
5951 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5952 | u64 max_len = stripe_len - stripe_offset; | |
5953 | ||
5954 | /* | |
5955 | * In case of raid56, we need to know the stripe aligned start | |
5956 | */ | |
5957 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5958 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
5959 | raid56_full_stripe_start = offset; | |
5960 | ||
5961 | /* | |
5962 | * Allow a write of a full stripe, but make sure we | |
5963 | * don't allow straddling of stripes | |
5964 | */ | |
5965 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5966 | full_stripe_len); | |
5967 | raid56_full_stripe_start *= full_stripe_len; | |
5968 | ||
5969 | /* | |
5970 | * For writes to RAID[56], allow a full stripeset across | |
5971 | * all disks. For other RAID types and for RAID[56] | |
5972 | * reads, just allow a single stripe (on a single disk). | |
5973 | */ | |
5974 | if (op == BTRFS_MAP_WRITE) { | |
5975 | max_len = stripe_len * data_stripes - | |
5976 | (offset - raid56_full_stripe_start); | |
5977 | } | |
5978 | } | |
5979 | len = min_t(u64, em->len - offset, max_len); | |
5980 | } else { | |
5981 | len = em->len - offset; | |
5982 | } | |
5983 | ||
5984 | io_geom->len = len; | |
5985 | io_geom->offset = offset; | |
5986 | io_geom->stripe_len = stripe_len; | |
5987 | io_geom->stripe_nr = stripe_nr; | |
5988 | io_geom->stripe_offset = stripe_offset; | |
5989 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
5990 | ||
373c3b80 JT |
5991 | out: |
5992 | /* once for us */ | |
5993 | free_extent_map(em); | |
5994 | return ret; | |
5f141126 NB |
5995 | } |
5996 | ||
cf8cddd3 CH |
5997 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5998 | enum btrfs_map_op op, | |
f2d8d74d | 5999 | u64 logical, u64 *length, |
a1d3c478 | 6000 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6001 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6002 | { |
6003 | struct extent_map *em; | |
6004 | struct map_lookup *map; | |
593060d7 CM |
6005 | u64 stripe_offset; |
6006 | u64 stripe_nr; | |
53b381b3 | 6007 | u64 stripe_len; |
9d644a62 | 6008 | u32 stripe_index; |
cff82672 | 6009 | int data_stripes; |
cea9e445 | 6010 | int i; |
de11cc12 | 6011 | int ret = 0; |
f2d8d74d | 6012 | int num_stripes; |
a236aed1 | 6013 | int max_errors = 0; |
2c8cdd6e | 6014 | int tgtdev_indexes = 0; |
a1d3c478 | 6015 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6016 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6017 | int dev_replace_is_ongoing = 0; | |
6018 | int num_alloc_stripes; | |
ad6d620e SB |
6019 | int patch_the_first_stripe_for_dev_replace = 0; |
6020 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6021 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6022 | struct btrfs_io_geometry geom; |
6023 | ||
6024 | ASSERT(bbio_ret); | |
0b86a832 | 6025 | |
0b3d4cd3 LB |
6026 | if (op == BTRFS_MAP_DISCARD) |
6027 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6028 | *length, bbio_ret); | |
6029 | ||
89b798ad NB |
6030 | ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom); |
6031 | if (ret < 0) | |
6032 | return ret; | |
0b86a832 | 6033 | |
89b798ad | 6034 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
f1136989 | 6035 | ASSERT(!IS_ERR(em)); |
95617d69 | 6036 | map = em->map_lookup; |
593060d7 | 6037 | |
89b798ad | 6038 | *length = geom.len; |
89b798ad NB |
6039 | stripe_len = geom.stripe_len; |
6040 | stripe_nr = geom.stripe_nr; | |
6041 | stripe_offset = geom.stripe_offset; | |
6042 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6043 | data_stripes = nr_data_stripes(map); |
593060d7 | 6044 | |
cb5583dd | 6045 | down_read(&dev_replace->rwsem); |
472262f3 | 6046 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6047 | /* |
6048 | * Hold the semaphore for read during the whole operation, write is | |
6049 | * requested at commit time but must wait. | |
6050 | */ | |
472262f3 | 6051 | if (!dev_replace_is_ongoing) |
cb5583dd | 6052 | up_read(&dev_replace->rwsem); |
472262f3 | 6053 | |
ad6d620e | 6054 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6055 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6056 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6057 | dev_replace->srcdev->devid, | |
6058 | &mirror_num, | |
6059 | &physical_to_patch_in_first_stripe); | |
6060 | if (ret) | |
ad6d620e | 6061 | goto out; |
5ab56090 LB |
6062 | else |
6063 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6064 | } else if (mirror_num > map->num_stripes) { |
6065 | mirror_num = 0; | |
6066 | } | |
6067 | ||
f2d8d74d | 6068 | num_stripes = 1; |
cea9e445 | 6069 | stripe_index = 0; |
fce3bb9a | 6070 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6071 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6072 | &stripe_index); | |
de483734 | 6073 | if (!need_full_stripe(op)) |
28e1cc7d | 6074 | mirror_num = 1; |
c7369b3f | 6075 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6076 | if (need_full_stripe(op)) |
f2d8d74d | 6077 | num_stripes = map->num_stripes; |
2fff734f | 6078 | else if (mirror_num) |
f188591e | 6079 | stripe_index = mirror_num - 1; |
dfe25020 | 6080 | else { |
30d9861f | 6081 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6082 | dev_replace_is_ongoing); |
a1d3c478 | 6083 | mirror_num = stripe_index + 1; |
dfe25020 | 6084 | } |
2fff734f | 6085 | |
611f0e00 | 6086 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6087 | if (need_full_stripe(op)) { |
f2d8d74d | 6088 | num_stripes = map->num_stripes; |
a1d3c478 | 6089 | } else if (mirror_num) { |
f188591e | 6090 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6091 | } else { |
6092 | mirror_num = 1; | |
6093 | } | |
2fff734f | 6094 | |
321aecc6 | 6095 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6096 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6097 | |
47c5713f | 6098 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6099 | stripe_index *= map->sub_stripes; |
6100 | ||
de483734 | 6101 | if (need_full_stripe(op)) |
f2d8d74d | 6102 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6103 | else if (mirror_num) |
6104 | stripe_index += mirror_num - 1; | |
dfe25020 | 6105 | else { |
3e74317a | 6106 | int old_stripe_index = stripe_index; |
30d9861f SB |
6107 | stripe_index = find_live_mirror(fs_info, map, |
6108 | stripe_index, | |
30d9861f | 6109 | dev_replace_is_ongoing); |
3e74317a | 6110 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6111 | } |
53b381b3 | 6112 | |
ffe2d203 | 6113 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6114 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6115 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6116 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6117 | stripe_len * data_stripes); |
53b381b3 DW |
6118 | |
6119 | /* RAID[56] write or recovery. Return all stripes */ | |
6120 | num_stripes = map->num_stripes; | |
6121 | max_errors = nr_parity_stripes(map); | |
6122 | ||
53b381b3 DW |
6123 | *length = map->stripe_len; |
6124 | stripe_index = 0; | |
6125 | stripe_offset = 0; | |
6126 | } else { | |
6127 | /* | |
6128 | * Mirror #0 or #1 means the original data block. | |
6129 | * Mirror #2 is RAID5 parity block. | |
6130 | * Mirror #3 is RAID6 Q block. | |
6131 | */ | |
47c5713f | 6132 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6133 | data_stripes, &stripe_index); |
53b381b3 | 6134 | if (mirror_num > 1) |
cff82672 | 6135 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6136 | |
6137 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6138 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6139 | &stripe_index); | |
de483734 | 6140 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6141 | mirror_num = 1; |
53b381b3 | 6142 | } |
8790d502 CM |
6143 | } else { |
6144 | /* | |
47c5713f DS |
6145 | * after this, stripe_nr is the number of stripes on this |
6146 | * device we have to walk to find the data, and stripe_index is | |
6147 | * the number of our device in the stripe array | |
8790d502 | 6148 | */ |
47c5713f DS |
6149 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6150 | &stripe_index); | |
a1d3c478 | 6151 | mirror_num = stripe_index + 1; |
8790d502 | 6152 | } |
e042d1ec | 6153 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6154 | btrfs_crit(fs_info, |
6155 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6156 | stripe_index, map->num_stripes); |
6157 | ret = -EINVAL; | |
6158 | goto out; | |
6159 | } | |
cea9e445 | 6160 | |
472262f3 | 6161 | num_alloc_stripes = num_stripes; |
6fad823f | 6162 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6163 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6164 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6165 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6166 | num_alloc_stripes++; |
2c8cdd6e | 6167 | tgtdev_indexes = num_stripes; |
ad6d620e | 6168 | } |
2c8cdd6e | 6169 | |
6e9606d2 | 6170 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6171 | if (!bbio) { |
6172 | ret = -ENOMEM; | |
6173 | goto out; | |
6174 | } | |
6fad823f | 6175 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6176 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6177 | |
8e5cfb55 | 6178 | /* build raid_map */ |
2b19a1fe LB |
6179 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6180 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6181 | u64 tmp; |
9d644a62 | 6182 | unsigned rot; |
8e5cfb55 ZL |
6183 | |
6184 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6185 | sizeof(struct btrfs_bio_stripe) * | |
6186 | num_alloc_stripes + | |
6187 | sizeof(int) * tgtdev_indexes); | |
6188 | ||
6189 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6190 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6191 | |
6192 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6193 | tmp = stripe_nr * data_stripes; |
6194 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6195 | bbio->raid_map[(i+rot) % num_stripes] = |
6196 | em->start + (tmp + i) * map->stripe_len; | |
6197 | ||
6198 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6199 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6200 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6201 | RAID6_Q_STRIPE; | |
6202 | } | |
6203 | ||
b89203f7 | 6204 | |
0b3d4cd3 LB |
6205 | for (i = 0; i < num_stripes; i++) { |
6206 | bbio->stripes[i].physical = | |
6207 | map->stripes[stripe_index].physical + | |
6208 | stripe_offset + | |
6209 | stripe_nr * map->stripe_len; | |
6210 | bbio->stripes[i].dev = | |
6211 | map->stripes[stripe_index].dev; | |
6212 | stripe_index++; | |
593060d7 | 6213 | } |
de11cc12 | 6214 | |
2b19a1fe | 6215 | if (need_full_stripe(op)) |
d20983b4 | 6216 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6217 | |
8e5cfb55 ZL |
6218 | if (bbio->raid_map) |
6219 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6220 | |
73c0f228 | 6221 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6222 | need_full_stripe(op)) { |
73c0f228 LB |
6223 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6224 | &max_errors); | |
472262f3 SB |
6225 | } |
6226 | ||
de11cc12 | 6227 | *bbio_ret = bbio; |
10f11900 | 6228 | bbio->map_type = map->type; |
de11cc12 LZ |
6229 | bbio->num_stripes = num_stripes; |
6230 | bbio->max_errors = max_errors; | |
6231 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6232 | |
6233 | /* | |
6234 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6235 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6236 | * available as a mirror | |
6237 | */ | |
6238 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6239 | WARN_ON(num_stripes > 1); | |
6240 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6241 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6242 | bbio->mirror_num = map->num_stripes + 1; | |
6243 | } | |
cea9e445 | 6244 | out: |
73beece9 | 6245 | if (dev_replace_is_ongoing) { |
53176dde DS |
6246 | lockdep_assert_held(&dev_replace->rwsem); |
6247 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6248 | up_read(&dev_replace->rwsem); |
73beece9 | 6249 | } |
0b86a832 | 6250 | free_extent_map(em); |
de11cc12 | 6251 | return ret; |
0b86a832 CM |
6252 | } |
6253 | ||
cf8cddd3 | 6254 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6255 | u64 logical, u64 *length, |
a1d3c478 | 6256 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6257 | { |
b3d3fa51 | 6258 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6259 | mirror_num, 0); |
f2d8d74d CM |
6260 | } |
6261 | ||
af8e2d1d | 6262 | /* For Scrub/replace */ |
cf8cddd3 | 6263 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6264 | u64 logical, u64 *length, |
825ad4c9 | 6265 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6266 | { |
825ad4c9 | 6267 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6268 | } |
6269 | ||
63a9c7b9 NB |
6270 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6271 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6272 | { |
a512bbf8 YZ |
6273 | struct extent_map *em; |
6274 | struct map_lookup *map; | |
6275 | u64 *buf; | |
6276 | u64 bytenr; | |
6277 | u64 length; | |
6278 | u64 stripe_nr; | |
53b381b3 | 6279 | u64 rmap_len; |
a512bbf8 YZ |
6280 | int i, j, nr = 0; |
6281 | ||
60ca842e | 6282 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6283 | if (IS_ERR(em)) |
835d974f | 6284 | return -EIO; |
835d974f | 6285 | |
95617d69 | 6286 | map = em->map_lookup; |
a512bbf8 | 6287 | length = em->len; |
53b381b3 DW |
6288 | rmap_len = map->stripe_len; |
6289 | ||
a512bbf8 | 6290 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6291 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6292 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6293 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6294 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6295 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6296 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6297 | } | |
a512bbf8 | 6298 | |
31e818fe | 6299 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6300 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6301 | |
6302 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6303 | if (map->stripes[i].physical > physical || |
6304 | map->stripes[i].physical + length <= physical) | |
6305 | continue; | |
6306 | ||
6307 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6308 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6309 | |
6310 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6311 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6312 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6313 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6314 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6315 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6316 | * Alternatively, just use rmap_len below instead of | |
6317 | * map->stripe_len */ | |
6318 | ||
6319 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6320 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6321 | for (j = 0; j < nr; j++) { |
6322 | if (buf[j] == bytenr) | |
6323 | break; | |
6324 | } | |
934d375b CM |
6325 | if (j == nr) { |
6326 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6327 | buf[nr++] = bytenr; |
934d375b | 6328 | } |
a512bbf8 YZ |
6329 | } |
6330 | ||
a512bbf8 YZ |
6331 | *logical = buf; |
6332 | *naddrs = nr; | |
53b381b3 | 6333 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6334 | |
6335 | free_extent_map(em); | |
6336 | return 0; | |
f2d8d74d CM |
6337 | } |
6338 | ||
4246a0b6 | 6339 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6340 | { |
326e1dbb MS |
6341 | bio->bi_private = bbio->private; |
6342 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6343 | bio_endio(bio); |
326e1dbb | 6344 | |
6e9606d2 | 6345 | btrfs_put_bbio(bbio); |
8408c716 MX |
6346 | } |
6347 | ||
4246a0b6 | 6348 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6349 | { |
9be3395b | 6350 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6351 | int is_orig_bio = 0; |
8790d502 | 6352 | |
4e4cbee9 | 6353 | if (bio->bi_status) { |
a1d3c478 | 6354 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6355 | if (bio->bi_status == BLK_STS_IOERR || |
6356 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6357 | unsigned int stripe_index = |
9be3395b | 6358 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6359 | struct btrfs_device *dev; |
442a4f63 SB |
6360 | |
6361 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6362 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6363 | if (dev->bdev) { |
37226b21 | 6364 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6365 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6366 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6367 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6368 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6369 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6370 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6371 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6372 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6373 | } |
442a4f63 SB |
6374 | } |
6375 | } | |
8790d502 | 6376 | |
a1d3c478 | 6377 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6378 | is_orig_bio = 1; |
6379 | ||
c404e0dc MX |
6380 | btrfs_bio_counter_dec(bbio->fs_info); |
6381 | ||
a1d3c478 | 6382 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6383 | if (!is_orig_bio) { |
6384 | bio_put(bio); | |
a1d3c478 | 6385 | bio = bbio->orig_bio; |
7d2b4daa | 6386 | } |
c7b22bb1 | 6387 | |
9be3395b | 6388 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6389 | /* only send an error to the higher layers if it is |
53b381b3 | 6390 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6391 | */ |
a1d3c478 | 6392 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6393 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6394 | } else { |
1259ab75 CM |
6395 | /* |
6396 | * this bio is actually up to date, we didn't | |
6397 | * go over the max number of errors | |
6398 | */ | |
2dbe0c77 | 6399 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6400 | } |
c55f1396 | 6401 | |
4246a0b6 | 6402 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6403 | } else if (!is_orig_bio) { |
8790d502 CM |
6404 | bio_put(bio); |
6405 | } | |
8790d502 CM |
6406 | } |
6407 | ||
8b712842 CM |
6408 | /* |
6409 | * see run_scheduled_bios for a description of why bios are collected for | |
6410 | * async submit. | |
6411 | * | |
6412 | * This will add one bio to the pending list for a device and make sure | |
6413 | * the work struct is scheduled. | |
6414 | */ | |
2ff7e61e | 6415 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6416 | struct bio *bio) |
8b712842 | 6417 | { |
0b246afa | 6418 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6419 | int should_queue = 1; |
ffbd517d | 6420 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6421 | |
6422 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6423 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6424 | btrfsic_submit_bio(bio); |
143bede5 | 6425 | return; |
8b712842 CM |
6426 | } |
6427 | ||
492bb6de | 6428 | WARN_ON(bio->bi_next); |
8b712842 | 6429 | bio->bi_next = NULL; |
8b712842 CM |
6430 | |
6431 | spin_lock(&device->io_lock); | |
67f055c7 | 6432 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6433 | pending_bios = &device->pending_sync_bios; |
6434 | else | |
6435 | pending_bios = &device->pending_bios; | |
8b712842 | 6436 | |
ffbd517d CM |
6437 | if (pending_bios->tail) |
6438 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6439 | |
ffbd517d CM |
6440 | pending_bios->tail = bio; |
6441 | if (!pending_bios->head) | |
6442 | pending_bios->head = bio; | |
8b712842 CM |
6443 | if (device->running_pending) |
6444 | should_queue = 0; | |
6445 | ||
6446 | spin_unlock(&device->io_lock); | |
6447 | ||
6448 | if (should_queue) | |
0b246afa | 6449 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6450 | } |
6451 | ||
2ff7e61e JM |
6452 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6453 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6454 | { |
6455 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6456 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6457 | |
6458 | bio->bi_private = bbio; | |
9be3395b | 6459 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6460 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6461 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6462 | btrfs_debug_in_rcu(fs_info, |
6463 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6464 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6465 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6466 | bio->bi_iter.bi_size); | |
74d46992 | 6467 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6468 | |
2ff7e61e | 6469 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6470 | |
de1ee92a | 6471 | if (async) |
2ff7e61e | 6472 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6473 | else |
4e49ea4a | 6474 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6475 | } |
6476 | ||
de1ee92a JB |
6477 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6478 | { | |
6479 | atomic_inc(&bbio->error); | |
6480 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6481 | /* Should be the original bio. */ |
8408c716 MX |
6482 | WARN_ON(bio != bbio->orig_bio); |
6483 | ||
9be3395b | 6484 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6485 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6486 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6487 | bio->bi_status = BLK_STS_IOERR; | |
6488 | else | |
6489 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6490 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6491 | } |
6492 | } | |
6493 | ||
58efbc9f OS |
6494 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6495 | int mirror_num, int async_submit) | |
0b86a832 | 6496 | { |
0b86a832 | 6497 | struct btrfs_device *dev; |
8790d502 | 6498 | struct bio *first_bio = bio; |
4f024f37 | 6499 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6500 | u64 length = 0; |
6501 | u64 map_length; | |
0b86a832 | 6502 | int ret; |
08da757d ZL |
6503 | int dev_nr; |
6504 | int total_devs; | |
a1d3c478 | 6505 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6506 | |
4f024f37 | 6507 | length = bio->bi_iter.bi_size; |
0b86a832 | 6508 | map_length = length; |
cea9e445 | 6509 | |
0b246afa | 6510 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6511 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6512 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6513 | if (ret) { |
0b246afa | 6514 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6515 | return errno_to_blk_status(ret); |
c404e0dc | 6516 | } |
cea9e445 | 6517 | |
a1d3c478 | 6518 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6519 | bbio->orig_bio = first_bio; |
6520 | bbio->private = first_bio->bi_private; | |
6521 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6522 | bbio->fs_info = fs_info; |
53b381b3 DW |
6523 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6524 | ||
ad1ba2a0 | 6525 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6526 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6527 | /* In this case, map_length has been set to the length of |
6528 | a single stripe; not the whole write */ | |
37226b21 | 6529 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6530 | ret = raid56_parity_write(fs_info, bio, bbio, |
6531 | map_length); | |
53b381b3 | 6532 | } else { |
2ff7e61e JM |
6533 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6534 | map_length, mirror_num, 1); | |
53b381b3 | 6535 | } |
4245215d | 6536 | |
0b246afa | 6537 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6538 | return errno_to_blk_status(ret); |
53b381b3 DW |
6539 | } |
6540 | ||
cea9e445 | 6541 | if (map_length < length) { |
0b246afa | 6542 | btrfs_crit(fs_info, |
5d163e0e JM |
6543 | "mapping failed logical %llu bio len %llu len %llu", |
6544 | logical, length, map_length); | |
cea9e445 CM |
6545 | BUG(); |
6546 | } | |
a1d3c478 | 6547 | |
08da757d | 6548 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6549 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6550 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6551 | &dev->dev_state) || | |
ebbede42 AJ |
6552 | (bio_op(first_bio) == REQ_OP_WRITE && |
6553 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6554 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6555 | continue; |
6556 | } | |
6557 | ||
3aa8e074 | 6558 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6559 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6560 | else |
a1d3c478 | 6561 | bio = first_bio; |
de1ee92a | 6562 | |
2ff7e61e JM |
6563 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6564 | dev_nr, async_submit); | |
8790d502 | 6565 | } |
0b246afa | 6566 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6567 | return BLK_STS_OK; |
0b86a832 CM |
6568 | } |
6569 | ||
09ba3bc9 AJ |
6570 | /* |
6571 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6572 | * return NULL. | |
6573 | * | |
6574 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6575 | * only devid is used. | |
6576 | * | |
6577 | * If @seed is true, traverse through the seed devices. | |
6578 | */ | |
e4319cd9 | 6579 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6580 | u64 devid, u8 *uuid, u8 *fsid, |
6581 | bool seed) | |
0b86a832 | 6582 | { |
2b82032c | 6583 | struct btrfs_device *device; |
2b82032c | 6584 | |
e4319cd9 | 6585 | while (fs_devices) { |
2b82032c | 6586 | if (!fsid || |
e4319cd9 | 6587 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6588 | list_for_each_entry(device, &fs_devices->devices, |
6589 | dev_list) { | |
6590 | if (device->devid == devid && | |
6591 | (!uuid || memcmp(device->uuid, uuid, | |
6592 | BTRFS_UUID_SIZE) == 0)) | |
6593 | return device; | |
6594 | } | |
2b82032c | 6595 | } |
09ba3bc9 AJ |
6596 | if (seed) |
6597 | fs_devices = fs_devices->seed; | |
6598 | else | |
6599 | return NULL; | |
2b82032c YZ |
6600 | } |
6601 | return NULL; | |
0b86a832 CM |
6602 | } |
6603 | ||
2ff7e61e | 6604 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6605 | u64 devid, u8 *dev_uuid) |
6606 | { | |
6607 | struct btrfs_device *device; | |
dfe25020 | 6608 | |
12bd2fc0 ID |
6609 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6610 | if (IS_ERR(device)) | |
adfb69af | 6611 | return device; |
12bd2fc0 ID |
6612 | |
6613 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6614 | device->fs_devices = fs_devices; |
dfe25020 | 6615 | fs_devices->num_devices++; |
12bd2fc0 | 6616 | |
e6e674bd | 6617 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6618 | fs_devices->missing_devices++; |
12bd2fc0 | 6619 | |
dfe25020 CM |
6620 | return device; |
6621 | } | |
6622 | ||
12bd2fc0 ID |
6623 | /** |
6624 | * btrfs_alloc_device - allocate struct btrfs_device | |
6625 | * @fs_info: used only for generating a new devid, can be NULL if | |
6626 | * devid is provided (i.e. @devid != NULL). | |
6627 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6628 | * is generated. | |
6629 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6630 | * is generated. | |
6631 | * | |
6632 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6633 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6634 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6635 | */ |
6636 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6637 | const u64 *devid, | |
6638 | const u8 *uuid) | |
6639 | { | |
6640 | struct btrfs_device *dev; | |
6641 | u64 tmp; | |
6642 | ||
fae7f21c | 6643 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6644 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6645 | |
6646 | dev = __alloc_device(); | |
6647 | if (IS_ERR(dev)) | |
6648 | return dev; | |
6649 | ||
6650 | if (devid) | |
6651 | tmp = *devid; | |
6652 | else { | |
6653 | int ret; | |
6654 | ||
6655 | ret = find_next_devid(fs_info, &tmp); | |
6656 | if (ret) { | |
a425f9d4 | 6657 | btrfs_free_device(dev); |
12bd2fc0 ID |
6658 | return ERR_PTR(ret); |
6659 | } | |
6660 | } | |
6661 | dev->devid = tmp; | |
6662 | ||
6663 | if (uuid) | |
6664 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6665 | else | |
6666 | generate_random_uuid(dev->uuid); | |
6667 | ||
9e0af237 LB |
6668 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6669 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6670 | |
6671 | return dev; | |
6672 | } | |
6673 | ||
5a2b8e60 | 6674 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6675 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6676 | { |
2b902dfc AJ |
6677 | if (error) |
6678 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6679 | devid, uuid); | |
6680 | else | |
6681 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6682 | devid, uuid); | |
5a2b8e60 AJ |
6683 | } |
6684 | ||
39e264a4 NB |
6685 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6686 | { | |
6687 | int index = btrfs_bg_flags_to_raid_index(type); | |
6688 | int ncopies = btrfs_raid_array[index].ncopies; | |
6689 | int data_stripes; | |
6690 | ||
6691 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
6692 | case BTRFS_BLOCK_GROUP_RAID5: | |
6693 | data_stripes = num_stripes - 1; | |
6694 | break; | |
6695 | case BTRFS_BLOCK_GROUP_RAID6: | |
6696 | data_stripes = num_stripes - 2; | |
6697 | break; | |
6698 | default: | |
6699 | data_stripes = num_stripes / ncopies; | |
6700 | break; | |
6701 | } | |
6702 | return div_u64(chunk_len, data_stripes); | |
6703 | } | |
6704 | ||
9690ac09 | 6705 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6706 | struct btrfs_chunk *chunk) |
6707 | { | |
9690ac09 | 6708 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6709 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6710 | struct map_lookup *map; |
6711 | struct extent_map *em; | |
6712 | u64 logical; | |
6713 | u64 length; | |
e06cd3dd LB |
6714 | u64 devid; |
6715 | u8 uuid[BTRFS_UUID_SIZE]; | |
6716 | int num_stripes; | |
6717 | int ret; | |
6718 | int i; | |
6719 | ||
6720 | logical = key->offset; | |
6721 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6722 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6723 | ||
075cb3c7 QW |
6724 | /* |
6725 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6726 | * as chunk item in tree block is already verified by tree-checker. | |
6727 | */ | |
6728 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6729 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6730 | if (ret) |
6731 | return ret; | |
6732 | } | |
a061fc8d | 6733 | |
c8bf1b67 DS |
6734 | read_lock(&map_tree->lock); |
6735 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6736 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6737 | |
6738 | /* already mapped? */ | |
6739 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6740 | free_extent_map(em); | |
0b86a832 CM |
6741 | return 0; |
6742 | } else if (em) { | |
6743 | free_extent_map(em); | |
6744 | } | |
0b86a832 | 6745 | |
172ddd60 | 6746 | em = alloc_extent_map(); |
0b86a832 CM |
6747 | if (!em) |
6748 | return -ENOMEM; | |
593060d7 | 6749 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6750 | if (!map) { |
6751 | free_extent_map(em); | |
6752 | return -ENOMEM; | |
6753 | } | |
6754 | ||
298a8f9c | 6755 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6756 | em->map_lookup = map; |
0b86a832 CM |
6757 | em->start = logical; |
6758 | em->len = length; | |
70c8a91c | 6759 | em->orig_start = 0; |
0b86a832 | 6760 | em->block_start = 0; |
c8b97818 | 6761 | em->block_len = em->len; |
0b86a832 | 6762 | |
593060d7 CM |
6763 | map->num_stripes = num_stripes; |
6764 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6765 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6766 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6767 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6768 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6769 | map->verified_stripes = 0; |
39e264a4 NB |
6770 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6771 | map->num_stripes); | |
593060d7 CM |
6772 | for (i = 0; i < num_stripes; i++) { |
6773 | map->stripes[i].physical = | |
6774 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6775 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6776 | read_extent_buffer(leaf, uuid, (unsigned long) |
6777 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6778 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6779 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6780 | devid, uuid, NULL, true); |
3cdde224 | 6781 | if (!map->stripes[i].dev && |
0b246afa | 6782 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6783 | free_extent_map(em); |
2b902dfc | 6784 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6785 | return -ENOENT; |
593060d7 | 6786 | } |
dfe25020 CM |
6787 | if (!map->stripes[i].dev) { |
6788 | map->stripes[i].dev = | |
2ff7e61e JM |
6789 | add_missing_dev(fs_info->fs_devices, devid, |
6790 | uuid); | |
adfb69af | 6791 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6792 | free_extent_map(em); |
adfb69af AJ |
6793 | btrfs_err(fs_info, |
6794 | "failed to init missing dev %llu: %ld", | |
6795 | devid, PTR_ERR(map->stripes[i].dev)); | |
6796 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6797 | } |
2b902dfc | 6798 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6799 | } |
e12c9621 AJ |
6800 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6801 | &(map->stripes[i].dev->dev_state)); | |
6802 | ||
0b86a832 CM |
6803 | } |
6804 | ||
c8bf1b67 DS |
6805 | write_lock(&map_tree->lock); |
6806 | ret = add_extent_mapping(map_tree, em, 0); | |
6807 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6808 | if (ret < 0) { |
6809 | btrfs_err(fs_info, | |
6810 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6811 | em->start, em->len, ret); | |
6812 | } | |
0b86a832 CM |
6813 | free_extent_map(em); |
6814 | ||
64f64f43 | 6815 | return ret; |
0b86a832 CM |
6816 | } |
6817 | ||
143bede5 | 6818 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6819 | struct btrfs_dev_item *dev_item, |
6820 | struct btrfs_device *device) | |
6821 | { | |
6822 | unsigned long ptr; | |
0b86a832 CM |
6823 | |
6824 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6825 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6826 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6827 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6828 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6829 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6830 | device->type = btrfs_device_type(leaf, dev_item); |
6831 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6832 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6833 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6834 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6835 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6836 | |
410ba3a2 | 6837 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6838 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6839 | } |
6840 | ||
2ff7e61e | 6841 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6842 | u8 *fsid) |
2b82032c YZ |
6843 | { |
6844 | struct btrfs_fs_devices *fs_devices; | |
6845 | int ret; | |
6846 | ||
a32bf9a3 | 6847 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6848 | ASSERT(fsid); |
2b82032c | 6849 | |
0b246afa | 6850 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6851 | while (fs_devices) { |
44880fdc | 6852 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6853 | return fs_devices; |
6854 | ||
2b82032c YZ |
6855 | fs_devices = fs_devices->seed; |
6856 | } | |
6857 | ||
7239ff4b | 6858 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6859 | if (!fs_devices) { |
0b246afa | 6860 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6861 | return ERR_PTR(-ENOENT); |
6862 | ||
7239ff4b | 6863 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6864 | if (IS_ERR(fs_devices)) |
6865 | return fs_devices; | |
6866 | ||
6867 | fs_devices->seeding = 1; | |
6868 | fs_devices->opened = 1; | |
6869 | return fs_devices; | |
2b82032c | 6870 | } |
e4404d6e YZ |
6871 | |
6872 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6873 | if (IS_ERR(fs_devices)) |
6874 | return fs_devices; | |
2b82032c | 6875 | |
897fb573 | 6876 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6877 | if (ret) { |
6878 | free_fs_devices(fs_devices); | |
5f375835 | 6879 | fs_devices = ERR_PTR(ret); |
2b82032c | 6880 | goto out; |
48d28232 | 6881 | } |
2b82032c YZ |
6882 | |
6883 | if (!fs_devices->seeding) { | |
0226e0eb | 6884 | close_fs_devices(fs_devices); |
e4404d6e | 6885 | free_fs_devices(fs_devices); |
5f375835 | 6886 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6887 | goto out; |
6888 | } | |
6889 | ||
0b246afa JM |
6890 | fs_devices->seed = fs_info->fs_devices->seed; |
6891 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6892 | out: |
5f375835 | 6893 | return fs_devices; |
2b82032c YZ |
6894 | } |
6895 | ||
17850759 | 6896 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6897 | struct btrfs_dev_item *dev_item) |
6898 | { | |
17850759 | 6899 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6900 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6901 | struct btrfs_device *device; |
6902 | u64 devid; | |
6903 | int ret; | |
44880fdc | 6904 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6905 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6906 | ||
0b86a832 | 6907 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6908 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6909 | BTRFS_UUID_SIZE); |
1473b24e | 6910 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6911 | BTRFS_FSID_SIZE); |
2b82032c | 6912 | |
de37aa51 | 6913 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6914 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6915 | if (IS_ERR(fs_devices)) |
6916 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6917 | } |
6918 | ||
e4319cd9 | 6919 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6920 | fs_uuid, true); |
5f375835 | 6921 | if (!device) { |
c5502451 | 6922 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6923 | btrfs_report_missing_device(fs_info, devid, |
6924 | dev_uuid, true); | |
45dbdbc9 | 6925 | return -ENOENT; |
c5502451 | 6926 | } |
2b82032c | 6927 | |
2ff7e61e | 6928 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6929 | if (IS_ERR(device)) { |
6930 | btrfs_err(fs_info, | |
6931 | "failed to add missing dev %llu: %ld", | |
6932 | devid, PTR_ERR(device)); | |
6933 | return PTR_ERR(device); | |
6934 | } | |
2b902dfc | 6935 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6936 | } else { |
c5502451 | 6937 | if (!device->bdev) { |
2b902dfc AJ |
6938 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6939 | btrfs_report_missing_device(fs_info, | |
6940 | devid, dev_uuid, true); | |
45dbdbc9 | 6941 | return -ENOENT; |
2b902dfc AJ |
6942 | } |
6943 | btrfs_report_missing_device(fs_info, devid, | |
6944 | dev_uuid, false); | |
c5502451 | 6945 | } |
5f375835 | 6946 | |
e6e674bd AJ |
6947 | if (!device->bdev && |
6948 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6949 | /* |
6950 | * this happens when a device that was properly setup | |
6951 | * in the device info lists suddenly goes bad. | |
6952 | * device->bdev is NULL, and so we have to set | |
6953 | * device->missing to one here | |
6954 | */ | |
5f375835 | 6955 | device->fs_devices->missing_devices++; |
e6e674bd | 6956 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6957 | } |
5f375835 MX |
6958 | |
6959 | /* Move the device to its own fs_devices */ | |
6960 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6961 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6962 | &device->dev_state)); | |
5f375835 MX |
6963 | |
6964 | list_move(&device->dev_list, &fs_devices->devices); | |
6965 | device->fs_devices->num_devices--; | |
6966 | fs_devices->num_devices++; | |
6967 | ||
6968 | device->fs_devices->missing_devices--; | |
6969 | fs_devices->missing_devices++; | |
6970 | ||
6971 | device->fs_devices = fs_devices; | |
6972 | } | |
2b82032c YZ |
6973 | } |
6974 | ||
0b246afa | 6975 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6976 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6977 | if (device->generation != |
6978 | btrfs_device_generation(leaf, dev_item)) | |
6979 | return -EINVAL; | |
6324fbf3 | 6980 | } |
0b86a832 CM |
6981 | |
6982 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6983 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6984 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6985 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6986 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6987 | atomic64_add(device->total_bytes - device->bytes_used, |
6988 | &fs_info->free_chunk_space); | |
2bf64758 | 6989 | } |
0b86a832 | 6990 | ret = 0; |
0b86a832 CM |
6991 | return ret; |
6992 | } | |
6993 | ||
6bccf3ab | 6994 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6995 | { |
6bccf3ab | 6996 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6997 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6998 | struct extent_buffer *sb; |
0b86a832 | 6999 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7000 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7001 | u8 *array_ptr; |
7002 | unsigned long sb_array_offset; | |
84eed90f | 7003 | int ret = 0; |
0b86a832 CM |
7004 | u32 num_stripes; |
7005 | u32 array_size; | |
7006 | u32 len = 0; | |
1ffb22cf | 7007 | u32 cur_offset; |
e06cd3dd | 7008 | u64 type; |
84eed90f | 7009 | struct btrfs_key key; |
0b86a832 | 7010 | |
0b246afa | 7011 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7012 | /* |
7013 | * This will create extent buffer of nodesize, superblock size is | |
7014 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7015 | * overallocate but we can keep it as-is, only the first page is used. | |
7016 | */ | |
2ff7e61e | 7017 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7018 | if (IS_ERR(sb)) |
7019 | return PTR_ERR(sb); | |
4db8c528 | 7020 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7021 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7022 | /* |
01327610 | 7023 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7024 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7025 | * pages up-to-date when the page is larger: extent does not cover the |
7026 | * whole page and consequently check_page_uptodate does not find all | |
7027 | * the page's extents up-to-date (the hole beyond sb), | |
7028 | * write_extent_buffer then triggers a WARN_ON. | |
7029 | * | |
7030 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7031 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7032 | * to silence the warning eg. on PowerPC 64. | |
7033 | */ | |
09cbfeaf | 7034 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7035 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7036 | |
a061fc8d | 7037 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7038 | array_size = btrfs_super_sys_array_size(super_copy); |
7039 | ||
1ffb22cf DS |
7040 | array_ptr = super_copy->sys_chunk_array; |
7041 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7042 | cur_offset = 0; | |
0b86a832 | 7043 | |
1ffb22cf DS |
7044 | while (cur_offset < array_size) { |
7045 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7046 | len = sizeof(*disk_key); |
7047 | if (cur_offset + len > array_size) | |
7048 | goto out_short_read; | |
7049 | ||
0b86a832 CM |
7050 | btrfs_disk_key_to_cpu(&key, disk_key); |
7051 | ||
1ffb22cf DS |
7052 | array_ptr += len; |
7053 | sb_array_offset += len; | |
7054 | cur_offset += len; | |
0b86a832 | 7055 | |
0d81ba5d | 7056 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7057 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7058 | /* |
7059 | * At least one btrfs_chunk with one stripe must be | |
7060 | * present, exact stripe count check comes afterwards | |
7061 | */ | |
7062 | len = btrfs_chunk_item_size(1); | |
7063 | if (cur_offset + len > array_size) | |
7064 | goto out_short_read; | |
7065 | ||
7066 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7067 | if (!num_stripes) { |
ab8d0fc4 JM |
7068 | btrfs_err(fs_info, |
7069 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7070 | num_stripes, cur_offset); |
7071 | ret = -EIO; | |
7072 | break; | |
7073 | } | |
7074 | ||
e06cd3dd LB |
7075 | type = btrfs_chunk_type(sb, chunk); |
7076 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7077 | btrfs_err(fs_info, |
e06cd3dd LB |
7078 | "invalid chunk type %llu in sys_array at offset %u", |
7079 | type, cur_offset); | |
7080 | ret = -EIO; | |
7081 | break; | |
7082 | } | |
7083 | ||
e3540eab DS |
7084 | len = btrfs_chunk_item_size(num_stripes); |
7085 | if (cur_offset + len > array_size) | |
7086 | goto out_short_read; | |
7087 | ||
9690ac09 | 7088 | ret = read_one_chunk(&key, sb, chunk); |
84eed90f CM |
7089 | if (ret) |
7090 | break; | |
0b86a832 | 7091 | } else { |
ab8d0fc4 JM |
7092 | btrfs_err(fs_info, |
7093 | "unexpected item type %u in sys_array at offset %u", | |
7094 | (u32)key.type, cur_offset); | |
84eed90f CM |
7095 | ret = -EIO; |
7096 | break; | |
0b86a832 | 7097 | } |
1ffb22cf DS |
7098 | array_ptr += len; |
7099 | sb_array_offset += len; | |
7100 | cur_offset += len; | |
0b86a832 | 7101 | } |
d865177a | 7102 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7103 | free_extent_buffer_stale(sb); |
84eed90f | 7104 | return ret; |
e3540eab DS |
7105 | |
7106 | out_short_read: | |
ab8d0fc4 | 7107 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7108 | len, cur_offset); |
d865177a | 7109 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7110 | free_extent_buffer_stale(sb); |
e3540eab | 7111 | return -EIO; |
0b86a832 CM |
7112 | } |
7113 | ||
21634a19 QW |
7114 | /* |
7115 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7116 | * | |
6528b99d AJ |
7117 | * If the @failing_dev is specified, it's accounted as missing. |
7118 | * | |
21634a19 QW |
7119 | * Return true if all chunks meet the minimal RW mount requirements. |
7120 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7121 | */ | |
6528b99d AJ |
7122 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7123 | struct btrfs_device *failing_dev) | |
21634a19 | 7124 | { |
c8bf1b67 | 7125 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7126 | struct extent_map *em; |
7127 | u64 next_start = 0; | |
7128 | bool ret = true; | |
7129 | ||
c8bf1b67 DS |
7130 | read_lock(&map_tree->lock); |
7131 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7132 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7133 | /* No chunk at all? Return false anyway */ |
7134 | if (!em) { | |
7135 | ret = false; | |
7136 | goto out; | |
7137 | } | |
7138 | while (em) { | |
7139 | struct map_lookup *map; | |
7140 | int missing = 0; | |
7141 | int max_tolerated; | |
7142 | int i; | |
7143 | ||
7144 | map = em->map_lookup; | |
7145 | max_tolerated = | |
7146 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7147 | map->type); | |
7148 | for (i = 0; i < map->num_stripes; i++) { | |
7149 | struct btrfs_device *dev = map->stripes[i].dev; | |
7150 | ||
e6e674bd AJ |
7151 | if (!dev || !dev->bdev || |
7152 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7153 | dev->last_flush_error) |
7154 | missing++; | |
6528b99d AJ |
7155 | else if (failing_dev && failing_dev == dev) |
7156 | missing++; | |
21634a19 QW |
7157 | } |
7158 | if (missing > max_tolerated) { | |
6528b99d AJ |
7159 | if (!failing_dev) |
7160 | btrfs_warn(fs_info, | |
52042d8e | 7161 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7162 | em->start, missing, max_tolerated); |
7163 | free_extent_map(em); | |
7164 | ret = false; | |
7165 | goto out; | |
7166 | } | |
7167 | next_start = extent_map_end(em); | |
7168 | free_extent_map(em); | |
7169 | ||
c8bf1b67 DS |
7170 | read_lock(&map_tree->lock); |
7171 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7172 | (u64)(-1) - next_start); |
c8bf1b67 | 7173 | read_unlock(&map_tree->lock); |
21634a19 QW |
7174 | } |
7175 | out: | |
7176 | return ret; | |
7177 | } | |
7178 | ||
5b4aacef | 7179 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7180 | { |
5b4aacef | 7181 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7182 | struct btrfs_path *path; |
7183 | struct extent_buffer *leaf; | |
7184 | struct btrfs_key key; | |
7185 | struct btrfs_key found_key; | |
7186 | int ret; | |
7187 | int slot; | |
99e3ecfc | 7188 | u64 total_dev = 0; |
0b86a832 | 7189 | |
0b86a832 CM |
7190 | path = btrfs_alloc_path(); |
7191 | if (!path) | |
7192 | return -ENOMEM; | |
7193 | ||
3dd0f7a3 AJ |
7194 | /* |
7195 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7196 | * otherwise we don't need it. | |
7197 | */ | |
b367e47f | 7198 | mutex_lock(&uuid_mutex); |
34441361 | 7199 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7200 | |
395927a9 FDBM |
7201 | /* |
7202 | * Read all device items, and then all the chunk items. All | |
7203 | * device items are found before any chunk item (their object id | |
7204 | * is smaller than the lowest possible object id for a chunk | |
7205 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7206 | */ |
7207 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7208 | key.offset = 0; | |
7209 | key.type = 0; | |
0b86a832 | 7210 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7211 | if (ret < 0) |
7212 | goto error; | |
d397712b | 7213 | while (1) { |
0b86a832 CM |
7214 | leaf = path->nodes[0]; |
7215 | slot = path->slots[0]; | |
7216 | if (slot >= btrfs_header_nritems(leaf)) { | |
7217 | ret = btrfs_next_leaf(root, path); | |
7218 | if (ret == 0) | |
7219 | continue; | |
7220 | if (ret < 0) | |
7221 | goto error; | |
7222 | break; | |
7223 | } | |
7224 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7225 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7226 | struct btrfs_dev_item *dev_item; | |
7227 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7228 | struct btrfs_dev_item); |
17850759 | 7229 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7230 | if (ret) |
7231 | goto error; | |
99e3ecfc | 7232 | total_dev++; |
0b86a832 CM |
7233 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7234 | struct btrfs_chunk *chunk; | |
7235 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
9690ac09 | 7236 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7237 | if (ret) |
7238 | goto error; | |
0b86a832 CM |
7239 | } |
7240 | path->slots[0]++; | |
7241 | } | |
99e3ecfc LB |
7242 | |
7243 | /* | |
7244 | * After loading chunk tree, we've got all device information, | |
7245 | * do another round of validation checks. | |
7246 | */ | |
0b246afa JM |
7247 | if (total_dev != fs_info->fs_devices->total_devices) { |
7248 | btrfs_err(fs_info, | |
99e3ecfc | 7249 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7250 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7251 | total_dev); |
7252 | ret = -EINVAL; | |
7253 | goto error; | |
7254 | } | |
0b246afa JM |
7255 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7256 | fs_info->fs_devices->total_rw_bytes) { | |
7257 | btrfs_err(fs_info, | |
99e3ecfc | 7258 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7259 | btrfs_super_total_bytes(fs_info->super_copy), |
7260 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7261 | ret = -EINVAL; |
7262 | goto error; | |
7263 | } | |
0b86a832 CM |
7264 | ret = 0; |
7265 | error: | |
34441361 | 7266 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7267 | mutex_unlock(&uuid_mutex); |
7268 | ||
2b82032c | 7269 | btrfs_free_path(path); |
0b86a832 CM |
7270 | return ret; |
7271 | } | |
442a4f63 | 7272 | |
cb517eab MX |
7273 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7274 | { | |
7275 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7276 | struct btrfs_device *device; | |
7277 | ||
29cc83f6 LB |
7278 | while (fs_devices) { |
7279 | mutex_lock(&fs_devices->device_list_mutex); | |
7280 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7281 | device->fs_info = fs_info; |
29cc83f6 LB |
7282 | mutex_unlock(&fs_devices->device_list_mutex); |
7283 | ||
7284 | fs_devices = fs_devices->seed; | |
7285 | } | |
cb517eab MX |
7286 | } |
7287 | ||
1dc990df DS |
7288 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7289 | const struct btrfs_dev_stats_item *ptr, | |
7290 | int index) | |
7291 | { | |
7292 | u64 val; | |
7293 | ||
7294 | read_extent_buffer(eb, &val, | |
7295 | offsetof(struct btrfs_dev_stats_item, values) + | |
7296 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7297 | sizeof(val)); | |
7298 | return val; | |
7299 | } | |
7300 | ||
7301 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7302 | struct btrfs_dev_stats_item *ptr, | |
7303 | int index, u64 val) | |
7304 | { | |
7305 | write_extent_buffer(eb, &val, | |
7306 | offsetof(struct btrfs_dev_stats_item, values) + | |
7307 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7308 | sizeof(val)); | |
7309 | } | |
7310 | ||
733f4fbb SB |
7311 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
7312 | { | |
7313 | struct btrfs_key key; | |
733f4fbb SB |
7314 | struct btrfs_root *dev_root = fs_info->dev_root; |
7315 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7316 | struct extent_buffer *eb; | |
7317 | int slot; | |
7318 | int ret = 0; | |
7319 | struct btrfs_device *device; | |
7320 | struct btrfs_path *path = NULL; | |
7321 | int i; | |
7322 | ||
7323 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7324 | if (!path) |
7325 | return -ENOMEM; | |
733f4fbb SB |
7326 | |
7327 | mutex_lock(&fs_devices->device_list_mutex); | |
7328 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7329 | int item_size; | |
7330 | struct btrfs_dev_stats_item *ptr; | |
7331 | ||
242e2956 DS |
7332 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7333 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7334 | key.offset = device->devid; |
7335 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7336 | if (ret) { | |
ae4b9b4c AJ |
7337 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
7338 | btrfs_dev_stat_set(device, i, 0); | |
733f4fbb SB |
7339 | device->dev_stats_valid = 1; |
7340 | btrfs_release_path(path); | |
7341 | continue; | |
7342 | } | |
7343 | slot = path->slots[0]; | |
7344 | eb = path->nodes[0]; | |
733f4fbb SB |
7345 | item_size = btrfs_item_size_nr(eb, slot); |
7346 | ||
7347 | ptr = btrfs_item_ptr(eb, slot, | |
7348 | struct btrfs_dev_stats_item); | |
7349 | ||
7350 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7351 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7352 | btrfs_dev_stat_set(device, i, | |
7353 | btrfs_dev_stats_value(eb, ptr, i)); | |
7354 | else | |
4e411a7d | 7355 | btrfs_dev_stat_set(device, i, 0); |
733f4fbb SB |
7356 | } |
7357 | ||
7358 | device->dev_stats_valid = 1; | |
7359 | btrfs_dev_stat_print_on_load(device); | |
7360 | btrfs_release_path(path); | |
7361 | } | |
7362 | mutex_unlock(&fs_devices->device_list_mutex); | |
7363 | ||
733f4fbb SB |
7364 | btrfs_free_path(path); |
7365 | return ret < 0 ? ret : 0; | |
7366 | } | |
7367 | ||
7368 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7369 | struct btrfs_device *device) |
7370 | { | |
5495f195 | 7371 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7372 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7373 | struct btrfs_path *path; |
7374 | struct btrfs_key key; | |
7375 | struct extent_buffer *eb; | |
7376 | struct btrfs_dev_stats_item *ptr; | |
7377 | int ret; | |
7378 | int i; | |
7379 | ||
242e2956 DS |
7380 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7381 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7382 | key.offset = device->devid; |
7383 | ||
7384 | path = btrfs_alloc_path(); | |
fa252992 DS |
7385 | if (!path) |
7386 | return -ENOMEM; | |
733f4fbb SB |
7387 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7388 | if (ret < 0) { | |
0b246afa | 7389 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7390 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7391 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7392 | goto out; |
7393 | } | |
7394 | ||
7395 | if (ret == 0 && | |
7396 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7397 | /* need to delete old one and insert a new one */ | |
7398 | ret = btrfs_del_item(trans, dev_root, path); | |
7399 | if (ret != 0) { | |
0b246afa | 7400 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7401 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7402 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7403 | goto out; |
7404 | } | |
7405 | ret = 1; | |
7406 | } | |
7407 | ||
7408 | if (ret == 1) { | |
7409 | /* need to insert a new item */ | |
7410 | btrfs_release_path(path); | |
7411 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7412 | &key, sizeof(*ptr)); | |
7413 | if (ret < 0) { | |
0b246afa | 7414 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7415 | "insert dev_stats item for device %s failed %d", |
7416 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7417 | goto out; |
7418 | } | |
7419 | } | |
7420 | ||
7421 | eb = path->nodes[0]; | |
7422 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7423 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7424 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7425 | btrfs_dev_stat_read(device, i)); | |
7426 | btrfs_mark_buffer_dirty(eb); | |
7427 | ||
7428 | out: | |
7429 | btrfs_free_path(path); | |
7430 | return ret; | |
7431 | } | |
7432 | ||
7433 | /* | |
7434 | * called from commit_transaction. Writes all changed device stats to disk. | |
7435 | */ | |
196c9d8d | 7436 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7437 | { |
196c9d8d | 7438 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7439 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7440 | struct btrfs_device *device; | |
addc3fa7 | 7441 | int stats_cnt; |
733f4fbb SB |
7442 | int ret = 0; |
7443 | ||
7444 | mutex_lock(&fs_devices->device_list_mutex); | |
7445 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7446 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7447 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7448 | continue; |
7449 | ||
9deae968 NB |
7450 | |
7451 | /* | |
7452 | * There is a LOAD-LOAD control dependency between the value of | |
7453 | * dev_stats_ccnt and updating the on-disk values which requires | |
7454 | * reading the in-memory counters. Such control dependencies | |
7455 | * require explicit read memory barriers. | |
7456 | * | |
7457 | * This memory barriers pairs with smp_mb__before_atomic in | |
7458 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7459 | * barrier implied by atomic_xchg in | |
7460 | * btrfs_dev_stats_read_and_reset | |
7461 | */ | |
7462 | smp_rmb(); | |
7463 | ||
5495f195 | 7464 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7465 | if (!ret) |
addc3fa7 | 7466 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7467 | } |
7468 | mutex_unlock(&fs_devices->device_list_mutex); | |
7469 | ||
7470 | return ret; | |
7471 | } | |
7472 | ||
442a4f63 SB |
7473 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7474 | { | |
7475 | btrfs_dev_stat_inc(dev, index); | |
7476 | btrfs_dev_stat_print_on_error(dev); | |
7477 | } | |
7478 | ||
48a3b636 | 7479 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7480 | { |
733f4fbb SB |
7481 | if (!dev->dev_stats_valid) |
7482 | return; | |
fb456252 | 7483 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7484 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7485 | rcu_str_deref(dev->name), |
442a4f63 SB |
7486 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7487 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7488 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7489 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7490 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7491 | } |
c11d2c23 | 7492 | |
733f4fbb SB |
7493 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7494 | { | |
a98cdb85 SB |
7495 | int i; |
7496 | ||
7497 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7498 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7499 | break; | |
7500 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7501 | return; /* all values == 0, suppress message */ | |
7502 | ||
fb456252 | 7503 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7504 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7505 | rcu_str_deref(dev->name), |
733f4fbb SB |
7506 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7507 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7508 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7509 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7510 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7511 | } | |
7512 | ||
2ff7e61e | 7513 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7514 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7515 | { |
7516 | struct btrfs_device *dev; | |
0b246afa | 7517 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7518 | int i; |
7519 | ||
7520 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7521 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7522 | true); | |
c11d2c23 SB |
7523 | mutex_unlock(&fs_devices->device_list_mutex); |
7524 | ||
7525 | if (!dev) { | |
0b246afa | 7526 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7527 | return -ENODEV; |
733f4fbb | 7528 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7529 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7530 | return -ENODEV; |
b27f7c0c | 7531 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7532 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7533 | if (stats->nr_items > i) | |
7534 | stats->values[i] = | |
7535 | btrfs_dev_stat_read_and_reset(dev, i); | |
7536 | else | |
4e411a7d | 7537 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 SB |
7538 | } |
7539 | } else { | |
7540 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7541 | if (stats->nr_items > i) | |
7542 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7543 | } | |
7544 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7545 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7546 | return 0; | |
7547 | } | |
a8a6dab7 | 7548 | |
da353f6b | 7549 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7550 | { |
7551 | struct buffer_head *bh; | |
7552 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7553 | int copy_num; |
a8a6dab7 | 7554 | |
12b1c263 AJ |
7555 | if (!bdev) |
7556 | return; | |
a8a6dab7 | 7557 | |
12b1c263 AJ |
7558 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7559 | copy_num++) { | |
a8a6dab7 | 7560 | |
12b1c263 AJ |
7561 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7562 | continue; | |
7563 | ||
7564 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7565 | ||
7566 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7567 | set_buffer_dirty(bh); | |
7568 | sync_dirty_buffer(bh); | |
7569 | brelse(bh); | |
7570 | } | |
7571 | ||
7572 | /* Notify udev that device has changed */ | |
7573 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7574 | ||
7575 | /* Update ctime/mtime for device path for libblkid */ | |
7576 | update_dev_time(device_path); | |
a8a6dab7 | 7577 | } |
935e5cc9 MX |
7578 | |
7579 | /* | |
bbbf7243 NB |
7580 | * Update the size and bytes used for each device where it changed. This is |
7581 | * delayed since we would otherwise get errors while writing out the | |
7582 | * superblocks. | |
7583 | * | |
7584 | * Must be invoked during transaction commit. | |
935e5cc9 | 7585 | */ |
bbbf7243 | 7586 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7587 | { |
935e5cc9 MX |
7588 | struct btrfs_device *curr, *next; |
7589 | ||
bbbf7243 | 7590 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7591 | |
bbbf7243 | 7592 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7593 | return; |
7594 | ||
bbbf7243 NB |
7595 | /* |
7596 | * We don't need the device_list_mutex here. This list is owned by the | |
7597 | * transaction and the transaction must complete before the device is | |
7598 | * released. | |
7599 | */ | |
7600 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7601 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7602 | post_commit_list) { | |
7603 | list_del_init(&curr->post_commit_list); | |
7604 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7605 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7606 | } |
bbbf7243 | 7607 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7608 | } |
5a13f430 AJ |
7609 | |
7610 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7611 | { | |
7612 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7613 | while (fs_devices) { | |
7614 | fs_devices->fs_info = fs_info; | |
7615 | fs_devices = fs_devices->seed; | |
7616 | } | |
7617 | } | |
7618 | ||
7619 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7620 | { | |
7621 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7622 | while (fs_devices) { | |
7623 | fs_devices->fs_info = NULL; | |
7624 | fs_devices = fs_devices->seed; | |
7625 | } | |
7626 | } | |
46df06b8 DS |
7627 | |
7628 | /* | |
7629 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7630 | */ | |
7631 | int btrfs_bg_type_to_factor(u64 flags) | |
7632 | { | |
44b28ada DS |
7633 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7634 | ||
7635 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7636 | } |
cf90d884 QW |
7637 | |
7638 | ||
cf90d884 QW |
7639 | |
7640 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7641 | u64 chunk_offset, u64 devid, | |
7642 | u64 physical_offset, u64 physical_len) | |
7643 | { | |
c8bf1b67 | 7644 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7645 | struct extent_map *em; |
7646 | struct map_lookup *map; | |
05a37c48 | 7647 | struct btrfs_device *dev; |
cf90d884 QW |
7648 | u64 stripe_len; |
7649 | bool found = false; | |
7650 | int ret = 0; | |
7651 | int i; | |
7652 | ||
7653 | read_lock(&em_tree->lock); | |
7654 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7655 | read_unlock(&em_tree->lock); | |
7656 | ||
7657 | if (!em) { | |
7658 | btrfs_err(fs_info, | |
7659 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7660 | physical_offset, devid); | |
7661 | ret = -EUCLEAN; | |
7662 | goto out; | |
7663 | } | |
7664 | ||
7665 | map = em->map_lookup; | |
7666 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7667 | if (physical_len != stripe_len) { | |
7668 | btrfs_err(fs_info, | |
7669 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7670 | physical_offset, devid, em->start, physical_len, | |
7671 | stripe_len); | |
7672 | ret = -EUCLEAN; | |
7673 | goto out; | |
7674 | } | |
7675 | ||
7676 | for (i = 0; i < map->num_stripes; i++) { | |
7677 | if (map->stripes[i].dev->devid == devid && | |
7678 | map->stripes[i].physical == physical_offset) { | |
7679 | found = true; | |
7680 | if (map->verified_stripes >= map->num_stripes) { | |
7681 | btrfs_err(fs_info, | |
7682 | "too many dev extents for chunk %llu found", | |
7683 | em->start); | |
7684 | ret = -EUCLEAN; | |
7685 | goto out; | |
7686 | } | |
7687 | map->verified_stripes++; | |
7688 | break; | |
7689 | } | |
7690 | } | |
7691 | if (!found) { | |
7692 | btrfs_err(fs_info, | |
7693 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7694 | physical_offset, devid); | |
7695 | ret = -EUCLEAN; | |
7696 | } | |
05a37c48 QW |
7697 | |
7698 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7699 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7700 | if (!dev) { |
7701 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7702 | ret = -EUCLEAN; | |
7703 | goto out; | |
7704 | } | |
1b3922a8 QW |
7705 | |
7706 | /* It's possible this device is a dummy for seed device */ | |
7707 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7708 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7709 | NULL, false); | |
1b3922a8 QW |
7710 | if (!dev) { |
7711 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7712 | devid); | |
7713 | ret = -EUCLEAN; | |
7714 | goto out; | |
7715 | } | |
7716 | } | |
7717 | ||
05a37c48 QW |
7718 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7719 | btrfs_err(fs_info, | |
7720 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7721 | devid, physical_offset, physical_len, | |
7722 | dev->disk_total_bytes); | |
7723 | ret = -EUCLEAN; | |
7724 | goto out; | |
7725 | } | |
cf90d884 QW |
7726 | out: |
7727 | free_extent_map(em); | |
7728 | return ret; | |
7729 | } | |
7730 | ||
7731 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7732 | { | |
c8bf1b67 | 7733 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7734 | struct extent_map *em; |
7735 | struct rb_node *node; | |
7736 | int ret = 0; | |
7737 | ||
7738 | read_lock(&em_tree->lock); | |
07e1ce09 | 7739 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7740 | em = rb_entry(node, struct extent_map, rb_node); |
7741 | if (em->map_lookup->num_stripes != | |
7742 | em->map_lookup->verified_stripes) { | |
7743 | btrfs_err(fs_info, | |
7744 | "chunk %llu has missing dev extent, have %d expect %d", | |
7745 | em->start, em->map_lookup->verified_stripes, | |
7746 | em->map_lookup->num_stripes); | |
7747 | ret = -EUCLEAN; | |
7748 | goto out; | |
7749 | } | |
7750 | } | |
7751 | out: | |
7752 | read_unlock(&em_tree->lock); | |
7753 | return ret; | |
7754 | } | |
7755 | ||
7756 | /* | |
7757 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7758 | * later chunk allocation/free would cause unexpected behavior. | |
7759 | * | |
7760 | * NOTE: This will iterate through the whole device tree, which should be of | |
7761 | * the same size level as the chunk tree. This slightly increases mount time. | |
7762 | */ | |
7763 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7764 | { | |
7765 | struct btrfs_path *path; | |
7766 | struct btrfs_root *root = fs_info->dev_root; | |
7767 | struct btrfs_key key; | |
5eb19381 QW |
7768 | u64 prev_devid = 0; |
7769 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7770 | int ret = 0; |
7771 | ||
7772 | key.objectid = 1; | |
7773 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7774 | key.offset = 0; | |
7775 | ||
7776 | path = btrfs_alloc_path(); | |
7777 | if (!path) | |
7778 | return -ENOMEM; | |
7779 | ||
7780 | path->reada = READA_FORWARD; | |
7781 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7782 | if (ret < 0) | |
7783 | goto out; | |
7784 | ||
7785 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7786 | ret = btrfs_next_item(root, path); | |
7787 | if (ret < 0) | |
7788 | goto out; | |
7789 | /* No dev extents at all? Not good */ | |
7790 | if (ret > 0) { | |
7791 | ret = -EUCLEAN; | |
7792 | goto out; | |
7793 | } | |
7794 | } | |
7795 | while (1) { | |
7796 | struct extent_buffer *leaf = path->nodes[0]; | |
7797 | struct btrfs_dev_extent *dext; | |
7798 | int slot = path->slots[0]; | |
7799 | u64 chunk_offset; | |
7800 | u64 physical_offset; | |
7801 | u64 physical_len; | |
7802 | u64 devid; | |
7803 | ||
7804 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7805 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7806 | break; | |
7807 | devid = key.objectid; | |
7808 | physical_offset = key.offset; | |
7809 | ||
7810 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7811 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7812 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7813 | ||
5eb19381 QW |
7814 | /* Check if this dev extent overlaps with the previous one */ |
7815 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7816 | btrfs_err(fs_info, | |
7817 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7818 | devid, physical_offset, prev_dev_ext_end); | |
7819 | ret = -EUCLEAN; | |
7820 | goto out; | |
7821 | } | |
7822 | ||
cf90d884 QW |
7823 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7824 | physical_offset, physical_len); | |
7825 | if (ret < 0) | |
7826 | goto out; | |
5eb19381 QW |
7827 | prev_devid = devid; |
7828 | prev_dev_ext_end = physical_offset + physical_len; | |
7829 | ||
cf90d884 QW |
7830 | ret = btrfs_next_item(root, path); |
7831 | if (ret < 0) | |
7832 | goto out; | |
7833 | if (ret > 0) { | |
7834 | ret = 0; | |
7835 | break; | |
7836 | } | |
7837 | } | |
7838 | ||
7839 | /* Ensure all chunks have corresponding dev extents */ | |
7840 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7841 | out: | |
7842 | btrfs_free_path(path); | |
7843 | return ret; | |
7844 | } | |
eede2bf3 OS |
7845 | |
7846 | /* | |
7847 | * Check whether the given block group or device is pinned by any inode being | |
7848 | * used as a swapfile. | |
7849 | */ | |
7850 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7851 | { | |
7852 | struct btrfs_swapfile_pin *sp; | |
7853 | struct rb_node *node; | |
7854 | ||
7855 | spin_lock(&fs_info->swapfile_pins_lock); | |
7856 | node = fs_info->swapfile_pins.rb_node; | |
7857 | while (node) { | |
7858 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7859 | if (ptr < sp->ptr) | |
7860 | node = node->rb_left; | |
7861 | else if (ptr > sp->ptr) | |
7862 | node = node->rb_right; | |
7863 | else | |
7864 | break; | |
7865 | } | |
7866 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7867 | return node != NULL; | |
7868 | } |