]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 CM |
6 | #include <linux/sched.h> |
7 | #include <linux/bio.h> | |
5a0e3ad6 | 8 | #include <linux/slab.h> |
8a4b83cc | 9 | #include <linux/buffer_head.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
0b86a832 CM |
17 | #include "ctree.h" |
18 | #include "extent_map.h" | |
19 | #include "disk-io.h" | |
20 | #include "transaction.h" | |
21 | #include "print-tree.h" | |
22 | #include "volumes.h" | |
53b381b3 | 23 | #include "raid56.h" |
8b712842 | 24 | #include "async-thread.h" |
21adbd5c | 25 | #include "check-integrity.h" |
606686ee | 26 | #include "rcu-string.h" |
3fed40cc | 27 | #include "math.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
0b86a832 | 31 | |
af902047 ZL |
32 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
33 | [BTRFS_RAID_RAID10] = { | |
34 | .sub_stripes = 2, | |
35 | .dev_stripes = 1, | |
36 | .devs_max = 0, /* 0 == as many as possible */ | |
37 | .devs_min = 4, | |
8789f4fe | 38 | .tolerated_failures = 1, |
af902047 ZL |
39 | .devs_increment = 2, |
40 | .ncopies = 2, | |
b50836ed | 41 | .nparity = 0, |
ed23467b | 42 | .raid_name = "raid10", |
41a6e891 | 43 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 44 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
45 | }, |
46 | [BTRFS_RAID_RAID1] = { | |
47 | .sub_stripes = 1, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 2, | |
50 | .devs_min = 2, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
b50836ed | 54 | .nparity = 0, |
ed23467b | 55 | .raid_name = "raid1", |
41a6e891 | 56 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 57 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 ZL |
58 | }, |
59 | [BTRFS_RAID_DUP] = { | |
60 | .sub_stripes = 1, | |
61 | .dev_stripes = 2, | |
62 | .devs_max = 1, | |
63 | .devs_min = 1, | |
8789f4fe | 64 | .tolerated_failures = 0, |
af902047 ZL |
65 | .devs_increment = 1, |
66 | .ncopies = 2, | |
b50836ed | 67 | .nparity = 0, |
ed23467b | 68 | .raid_name = "dup", |
41a6e891 | 69 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 70 | .mindev_error = 0, |
af902047 ZL |
71 | }, |
72 | [BTRFS_RAID_RAID0] = { | |
73 | .sub_stripes = 1, | |
74 | .dev_stripes = 1, | |
75 | .devs_max = 0, | |
76 | .devs_min = 2, | |
8789f4fe | 77 | .tolerated_failures = 0, |
af902047 ZL |
78 | .devs_increment = 1, |
79 | .ncopies = 1, | |
b50836ed | 80 | .nparity = 0, |
ed23467b | 81 | .raid_name = "raid0", |
41a6e891 | 82 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 83 | .mindev_error = 0, |
af902047 ZL |
84 | }, |
85 | [BTRFS_RAID_SINGLE] = { | |
86 | .sub_stripes = 1, | |
87 | .dev_stripes = 1, | |
88 | .devs_max = 1, | |
89 | .devs_min = 1, | |
8789f4fe | 90 | .tolerated_failures = 0, |
af902047 ZL |
91 | .devs_increment = 1, |
92 | .ncopies = 1, | |
b50836ed | 93 | .nparity = 0, |
ed23467b | 94 | .raid_name = "single", |
41a6e891 | 95 | .bg_flag = 0, |
f9fbcaa2 | 96 | .mindev_error = 0, |
af902047 ZL |
97 | }, |
98 | [BTRFS_RAID_RAID5] = { | |
99 | .sub_stripes = 1, | |
100 | .dev_stripes = 1, | |
101 | .devs_max = 0, | |
102 | .devs_min = 2, | |
8789f4fe | 103 | .tolerated_failures = 1, |
af902047 | 104 | .devs_increment = 1, |
da612e31 | 105 | .ncopies = 1, |
b50836ed | 106 | .nparity = 1, |
ed23467b | 107 | .raid_name = "raid5", |
41a6e891 | 108 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 109 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
110 | }, |
111 | [BTRFS_RAID_RAID6] = { | |
112 | .sub_stripes = 1, | |
113 | .dev_stripes = 1, | |
114 | .devs_max = 0, | |
115 | .devs_min = 3, | |
8789f4fe | 116 | .tolerated_failures = 2, |
af902047 | 117 | .devs_increment = 1, |
da612e31 | 118 | .ncopies = 1, |
b50836ed | 119 | .nparity = 2, |
ed23467b | 120 | .raid_name = "raid6", |
41a6e891 | 121 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 122 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
123 | }, |
124 | }; | |
125 | ||
ed23467b AJ |
126 | const char *get_raid_name(enum btrfs_raid_types type) |
127 | { | |
128 | if (type >= BTRFS_NR_RAID_TYPES) | |
129 | return NULL; | |
130 | ||
131 | return btrfs_raid_array[type].raid_name; | |
132 | } | |
133 | ||
f89e09cf AJ |
134 | /* |
135 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
136 | * bytes including terminating null byte. | |
137 | */ | |
138 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
139 | { | |
140 | int i; | |
141 | int ret; | |
142 | char *bp = buf; | |
143 | u64 flags = bg_flags; | |
144 | u32 size_bp = size_buf; | |
145 | ||
146 | if (!flags) { | |
147 | strcpy(bp, "NONE"); | |
148 | return; | |
149 | } | |
150 | ||
151 | #define DESCRIBE_FLAG(flag, desc) \ | |
152 | do { \ | |
153 | if (flags & (flag)) { \ | |
154 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
155 | if (ret < 0 || ret >= size_bp) \ | |
156 | goto out_overflow; \ | |
157 | size_bp -= ret; \ | |
158 | bp += ret; \ | |
159 | flags &= ~(flag); \ | |
160 | } \ | |
161 | } while (0) | |
162 | ||
163 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
164 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
165 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
166 | ||
167 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
168 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
169 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
170 | btrfs_raid_array[i].raid_name); | |
171 | #undef DESCRIBE_FLAG | |
172 | ||
173 | if (flags) { | |
174 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
175 | size_bp -= ret; | |
176 | } | |
177 | ||
178 | if (size_bp < size_buf) | |
179 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
180 | ||
181 | /* | |
182 | * The text is trimmed, it's up to the caller to provide sufficiently | |
183 | * large buffer | |
184 | */ | |
185 | out_overflow:; | |
186 | } | |
187 | ||
6f8e0fc7 | 188 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 189 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 190 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 191 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 192 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
193 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
194 | enum btrfs_map_op op, | |
195 | u64 logical, u64 *length, | |
196 | struct btrfs_bio **bbio_ret, | |
197 | int mirror_num, int need_raid_map); | |
2b82032c | 198 | |
9c6b1c4d DS |
199 | /* |
200 | * Device locking | |
201 | * ============== | |
202 | * | |
203 | * There are several mutexes that protect manipulation of devices and low-level | |
204 | * structures like chunks but not block groups, extents or files | |
205 | * | |
206 | * uuid_mutex (global lock) | |
207 | * ------------------------ | |
208 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
209 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
210 | * device) or requested by the device= mount option | |
211 | * | |
212 | * the mutex can be very coarse and can cover long-running operations | |
213 | * | |
214 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 215 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
216 | * |
217 | * global::fs_devs - add, remove, updates to the global list | |
218 | * | |
219 | * does not protect: manipulation of the fs_devices::devices list! | |
220 | * | |
221 | * btrfs_device::name - renames (write side), read is RCU | |
222 | * | |
223 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
224 | * ------------------------------------------------ | |
225 | * protects updates to fs_devices::devices, ie. adding and deleting | |
226 | * | |
227 | * simple list traversal with read-only actions can be done with RCU protection | |
228 | * | |
229 | * may be used to exclude some operations from running concurrently without any | |
230 | * modifications to the list (see write_all_supers) | |
231 | * | |
9c6b1c4d DS |
232 | * balance_mutex |
233 | * ------------- | |
234 | * protects balance structures (status, state) and context accessed from | |
235 | * several places (internally, ioctl) | |
236 | * | |
237 | * chunk_mutex | |
238 | * ----------- | |
239 | * protects chunks, adding or removing during allocation, trim or when a new | |
240 | * device is added/removed | |
241 | * | |
242 | * cleaner_mutex | |
243 | * ------------- | |
244 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
245 | * cleaning together with relocation or delayed iputs | |
246 | * | |
247 | * | |
248 | * Lock nesting | |
249 | * ============ | |
250 | * | |
251 | * uuid_mutex | |
252 | * volume_mutex | |
253 | * device_list_mutex | |
254 | * chunk_mutex | |
255 | * balance_mutex | |
89595e80 AJ |
256 | * |
257 | * | |
258 | * Exclusive operations, BTRFS_FS_EXCL_OP | |
259 | * ====================================== | |
260 | * | |
261 | * Maintains the exclusivity of the following operations that apply to the | |
262 | * whole filesystem and cannot run in parallel. | |
263 | * | |
264 | * - Balance (*) | |
265 | * - Device add | |
266 | * - Device remove | |
267 | * - Device replace (*) | |
268 | * - Resize | |
269 | * | |
270 | * The device operations (as above) can be in one of the following states: | |
271 | * | |
272 | * - Running state | |
273 | * - Paused state | |
274 | * - Completed state | |
275 | * | |
276 | * Only device operations marked with (*) can go into the Paused state for the | |
277 | * following reasons: | |
278 | * | |
279 | * - ioctl (only Balance can be Paused through ioctl) | |
280 | * - filesystem remounted as read-only | |
281 | * - filesystem unmounted and mounted as read-only | |
282 | * - system power-cycle and filesystem mounted as read-only | |
283 | * - filesystem or device errors leading to forced read-only | |
284 | * | |
285 | * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations. | |
286 | * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set. | |
287 | * A device operation in Paused or Running state can be canceled or resumed | |
288 | * either by ioctl (Balance only) or when remounted as read-write. | |
289 | * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or | |
290 | * completed. | |
9c6b1c4d DS |
291 | */ |
292 | ||
67a2c45e | 293 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 294 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
295 | struct list_head *btrfs_get_fs_uuids(void) |
296 | { | |
297 | return &fs_uuids; | |
298 | } | |
8a4b83cc | 299 | |
2dfeca9b DS |
300 | /* |
301 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
302 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
303 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
304 | * |
305 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
306 | * The returned struct is not linked onto any lists and can be destroyed with | |
307 | * kfree() right away. | |
308 | */ | |
7239ff4b NB |
309 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
310 | const u8 *metadata_fsid) | |
2208a378 ID |
311 | { |
312 | struct btrfs_fs_devices *fs_devs; | |
313 | ||
78f2c9e6 | 314 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
315 | if (!fs_devs) |
316 | return ERR_PTR(-ENOMEM); | |
317 | ||
318 | mutex_init(&fs_devs->device_list_mutex); | |
319 | ||
320 | INIT_LIST_HEAD(&fs_devs->devices); | |
321 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 322 | INIT_LIST_HEAD(&fs_devs->fs_list); |
2208a378 ID |
323 | if (fsid) |
324 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 325 | |
7239ff4b NB |
326 | if (metadata_fsid) |
327 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
328 | else if (fsid) | |
329 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
330 | ||
2208a378 ID |
331 | return fs_devs; |
332 | } | |
333 | ||
a425f9d4 | 334 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 335 | { |
bbbf7243 | 336 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 337 | rcu_string_free(device->name); |
1c11b63e | 338 | extent_io_tree_release(&device->alloc_state); |
48dae9cf DS |
339 | bio_put(device->flush_bio); |
340 | kfree(device); | |
341 | } | |
342 | ||
e4404d6e YZ |
343 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
344 | { | |
345 | struct btrfs_device *device; | |
346 | WARN_ON(fs_devices->opened); | |
347 | while (!list_empty(&fs_devices->devices)) { | |
348 | device = list_entry(fs_devices->devices.next, | |
349 | struct btrfs_device, dev_list); | |
350 | list_del(&device->dev_list); | |
a425f9d4 | 351 | btrfs_free_device(device); |
e4404d6e YZ |
352 | } |
353 | kfree(fs_devices); | |
354 | } | |
355 | ||
b8b8ff59 LC |
356 | static void btrfs_kobject_uevent(struct block_device *bdev, |
357 | enum kobject_action action) | |
358 | { | |
359 | int ret; | |
360 | ||
361 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
362 | if (ret) | |
efe120a0 | 363 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
364 | action, |
365 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
366 | &disk_to_dev(bdev->bd_disk)->kobj); | |
367 | } | |
368 | ||
ffc5a379 | 369 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
370 | { |
371 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 372 | |
2b82032c YZ |
373 | while (!list_empty(&fs_uuids)) { |
374 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
375 | struct btrfs_fs_devices, fs_list); |
376 | list_del(&fs_devices->fs_list); | |
e4404d6e | 377 | free_fs_devices(fs_devices); |
8a4b83cc | 378 | } |
8a4b83cc CM |
379 | } |
380 | ||
48dae9cf DS |
381 | /* |
382 | * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error. | |
383 | * Returned struct is not linked onto any lists and must be destroyed using | |
a425f9d4 | 384 | * btrfs_free_device. |
48dae9cf | 385 | */ |
12bd2fc0 ID |
386 | static struct btrfs_device *__alloc_device(void) |
387 | { | |
388 | struct btrfs_device *dev; | |
389 | ||
78f2c9e6 | 390 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
391 | if (!dev) |
392 | return ERR_PTR(-ENOMEM); | |
393 | ||
e0ae9994 DS |
394 | /* |
395 | * Preallocate a bio that's always going to be used for flushing device | |
396 | * barriers and matches the device lifespan | |
397 | */ | |
398 | dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL); | |
399 | if (!dev->flush_bio) { | |
400 | kfree(dev); | |
401 | return ERR_PTR(-ENOMEM); | |
402 | } | |
e0ae9994 | 403 | |
12bd2fc0 ID |
404 | INIT_LIST_HEAD(&dev->dev_list); |
405 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
bbbf7243 | 406 | INIT_LIST_HEAD(&dev->post_commit_list); |
12bd2fc0 ID |
407 | |
408 | spin_lock_init(&dev->io_lock); | |
409 | ||
12bd2fc0 | 410 | atomic_set(&dev->reada_in_flight, 0); |
addc3fa7 | 411 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 412 | btrfs_device_data_ordered_init(dev); |
9bcaaea7 | 413 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
d0164adc | 414 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
1c11b63e | 415 | extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL); |
12bd2fc0 ID |
416 | |
417 | return dev; | |
418 | } | |
419 | ||
7239ff4b NB |
420 | static noinline struct btrfs_fs_devices *find_fsid( |
421 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 422 | { |
8a4b83cc CM |
423 | struct btrfs_fs_devices *fs_devices; |
424 | ||
7239ff4b NB |
425 | ASSERT(fsid); |
426 | ||
7a62d0f0 NB |
427 | if (metadata_fsid) { |
428 | /* | |
429 | * Handle scanned device having completed its fsid change but | |
430 | * belonging to a fs_devices that was created by first scanning | |
431 | * a device which didn't have its fsid/metadata_uuid changed | |
432 | * at all and the CHANGING_FSID_V2 flag set. | |
433 | */ | |
434 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
435 | if (fs_devices->fsid_change && | |
436 | memcmp(metadata_fsid, fs_devices->fsid, | |
437 | BTRFS_FSID_SIZE) == 0 && | |
438 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
439 | BTRFS_FSID_SIZE) == 0) { | |
440 | return fs_devices; | |
441 | } | |
442 | } | |
cc5de4e7 NB |
443 | /* |
444 | * Handle scanned device having completed its fsid change but | |
445 | * belonging to a fs_devices that was created by a device that | |
446 | * has an outdated pair of fsid/metadata_uuid and | |
447 | * CHANGING_FSID_V2 flag set. | |
448 | */ | |
449 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
450 | if (fs_devices->fsid_change && | |
451 | memcmp(fs_devices->metadata_uuid, | |
452 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
453 | memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
454 | BTRFS_FSID_SIZE) == 0) { | |
455 | return fs_devices; | |
456 | } | |
457 | } | |
7a62d0f0 NB |
458 | } |
459 | ||
460 | /* Handle non-split brain cases */ | |
c4babc5e | 461 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
462 | if (metadata_fsid) { |
463 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
464 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
465 | BTRFS_FSID_SIZE) == 0) | |
466 | return fs_devices; | |
467 | } else { | |
468 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
469 | return fs_devices; | |
470 | } | |
8a4b83cc CM |
471 | } |
472 | return NULL; | |
473 | } | |
474 | ||
beaf8ab3 SB |
475 | static int |
476 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
477 | int flush, struct block_device **bdev, | |
478 | struct buffer_head **bh) | |
479 | { | |
480 | int ret; | |
481 | ||
482 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
483 | ||
484 | if (IS_ERR(*bdev)) { | |
485 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
486 | goto error; |
487 | } | |
488 | ||
489 | if (flush) | |
490 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 491 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
492 | if (ret) { |
493 | blkdev_put(*bdev, flags); | |
494 | goto error; | |
495 | } | |
496 | invalidate_bdev(*bdev); | |
497 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
498 | if (IS_ERR(*bh)) { |
499 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
500 | blkdev_put(*bdev, flags); |
501 | goto error; | |
502 | } | |
503 | ||
504 | return 0; | |
505 | ||
506 | error: | |
507 | *bdev = NULL; | |
508 | *bh = NULL; | |
509 | return ret; | |
510 | } | |
511 | ||
ffbd517d CM |
512 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
513 | struct bio *head, struct bio *tail) | |
514 | { | |
515 | ||
516 | struct bio *old_head; | |
517 | ||
518 | old_head = pending_bios->head; | |
519 | pending_bios->head = head; | |
520 | if (pending_bios->tail) | |
521 | tail->bi_next = old_head; | |
522 | else | |
523 | pending_bios->tail = tail; | |
524 | } | |
525 | ||
8b712842 CM |
526 | /* |
527 | * we try to collect pending bios for a device so we don't get a large | |
528 | * number of procs sending bios down to the same device. This greatly | |
529 | * improves the schedulers ability to collect and merge the bios. | |
530 | * | |
531 | * But, it also turns into a long list of bios to process and that is sure | |
532 | * to eventually make the worker thread block. The solution here is to | |
533 | * make some progress and then put this work struct back at the end of | |
534 | * the list if the block device is congested. This way, multiple devices | |
535 | * can make progress from a single worker thread. | |
536 | */ | |
143bede5 | 537 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 538 | { |
0b246afa | 539 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
540 | struct bio *pending; |
541 | struct backing_dev_info *bdi; | |
ffbd517d | 542 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
543 | struct bio *tail; |
544 | struct bio *cur; | |
545 | int again = 0; | |
ffbd517d | 546 | unsigned long num_run; |
d644d8a1 | 547 | unsigned long batch_run = 0; |
b765ead5 | 548 | unsigned long last_waited = 0; |
d84275c9 | 549 | int force_reg = 0; |
0e588859 | 550 | int sync_pending = 0; |
211588ad CM |
551 | struct blk_plug plug; |
552 | ||
553 | /* | |
554 | * this function runs all the bios we've collected for | |
555 | * a particular device. We don't want to wander off to | |
556 | * another device without first sending all of these down. | |
557 | * So, setup a plug here and finish it off before we return | |
558 | */ | |
559 | blk_start_plug(&plug); | |
8b712842 | 560 | |
efa7c9f9 | 561 | bdi = device->bdev->bd_bdi; |
b64a2851 | 562 | |
8b712842 CM |
563 | loop: |
564 | spin_lock(&device->io_lock); | |
565 | ||
a6837051 | 566 | loop_lock: |
d84275c9 | 567 | num_run = 0; |
ffbd517d | 568 | |
8b712842 CM |
569 | /* take all the bios off the list at once and process them |
570 | * later on (without the lock held). But, remember the | |
571 | * tail and other pointers so the bios can be properly reinserted | |
572 | * into the list if we hit congestion | |
573 | */ | |
d84275c9 | 574 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 575 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
576 | force_reg = 1; |
577 | } else { | |
ffbd517d | 578 | pending_bios = &device->pending_bios; |
d84275c9 CM |
579 | force_reg = 0; |
580 | } | |
ffbd517d CM |
581 | |
582 | pending = pending_bios->head; | |
583 | tail = pending_bios->tail; | |
8b712842 | 584 | WARN_ON(pending && !tail); |
8b712842 CM |
585 | |
586 | /* | |
587 | * if pending was null this time around, no bios need processing | |
588 | * at all and we can stop. Otherwise it'll loop back up again | |
589 | * and do an additional check so no bios are missed. | |
590 | * | |
591 | * device->running_pending is used to synchronize with the | |
592 | * schedule_bio code. | |
593 | */ | |
ffbd517d CM |
594 | if (device->pending_sync_bios.head == NULL && |
595 | device->pending_bios.head == NULL) { | |
8b712842 CM |
596 | again = 0; |
597 | device->running_pending = 0; | |
ffbd517d CM |
598 | } else { |
599 | again = 1; | |
600 | device->running_pending = 1; | |
8b712842 | 601 | } |
ffbd517d CM |
602 | |
603 | pending_bios->head = NULL; | |
604 | pending_bios->tail = NULL; | |
605 | ||
8b712842 CM |
606 | spin_unlock(&device->io_lock); |
607 | ||
d397712b | 608 | while (pending) { |
ffbd517d CM |
609 | |
610 | rmb(); | |
d84275c9 CM |
611 | /* we want to work on both lists, but do more bios on the |
612 | * sync list than the regular list | |
613 | */ | |
614 | if ((num_run > 32 && | |
615 | pending_bios != &device->pending_sync_bios && | |
616 | device->pending_sync_bios.head) || | |
617 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
618 | device->pending_bios.head)) { | |
ffbd517d CM |
619 | spin_lock(&device->io_lock); |
620 | requeue_list(pending_bios, pending, tail); | |
621 | goto loop_lock; | |
622 | } | |
623 | ||
8b712842 CM |
624 | cur = pending; |
625 | pending = pending->bi_next; | |
626 | cur->bi_next = NULL; | |
b64a2851 | 627 | |
dac56212 | 628 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 629 | |
2ab1ba68 CM |
630 | /* |
631 | * if we're doing the sync list, record that our | |
632 | * plug has some sync requests on it | |
633 | * | |
634 | * If we're doing the regular list and there are | |
635 | * sync requests sitting around, unplug before | |
636 | * we add more | |
637 | */ | |
638 | if (pending_bios == &device->pending_sync_bios) { | |
639 | sync_pending = 1; | |
640 | } else if (sync_pending) { | |
641 | blk_finish_plug(&plug); | |
642 | blk_start_plug(&plug); | |
643 | sync_pending = 0; | |
644 | } | |
645 | ||
4e49ea4a | 646 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
647 | num_run++; |
648 | batch_run++; | |
853d8ec4 DS |
649 | |
650 | cond_resched(); | |
8b712842 CM |
651 | |
652 | /* | |
653 | * we made progress, there is more work to do and the bdi | |
654 | * is now congested. Back off and let other work structs | |
655 | * run instead | |
656 | */ | |
57fd5a5f | 657 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 658 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 659 | struct io_context *ioc; |
8b712842 | 660 | |
b765ead5 CM |
661 | ioc = current->io_context; |
662 | ||
663 | /* | |
664 | * the main goal here is that we don't want to | |
665 | * block if we're going to be able to submit | |
666 | * more requests without blocking. | |
667 | * | |
668 | * This code does two great things, it pokes into | |
669 | * the elevator code from a filesystem _and_ | |
670 | * it makes assumptions about how batching works. | |
671 | */ | |
672 | if (ioc && ioc->nr_batch_requests > 0 && | |
673 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
674 | (last_waited == 0 || | |
675 | ioc->last_waited == last_waited)) { | |
676 | /* | |
677 | * we want to go through our batch of | |
678 | * requests and stop. So, we copy out | |
679 | * the ioc->last_waited time and test | |
680 | * against it before looping | |
681 | */ | |
682 | last_waited = ioc->last_waited; | |
853d8ec4 | 683 | cond_resched(); |
b765ead5 CM |
684 | continue; |
685 | } | |
8b712842 | 686 | spin_lock(&device->io_lock); |
ffbd517d | 687 | requeue_list(pending_bios, pending, tail); |
a6837051 | 688 | device->running_pending = 1; |
8b712842 CM |
689 | |
690 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
691 | btrfs_queue_work(fs_info->submit_workers, |
692 | &device->work); | |
8b712842 CM |
693 | goto done; |
694 | } | |
695 | } | |
ffbd517d | 696 | |
51684082 CM |
697 | cond_resched(); |
698 | if (again) | |
699 | goto loop; | |
700 | ||
701 | spin_lock(&device->io_lock); | |
702 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
703 | goto loop_lock; | |
704 | spin_unlock(&device->io_lock); | |
705 | ||
8b712842 | 706 | done: |
211588ad | 707 | blk_finish_plug(&plug); |
8b712842 CM |
708 | } |
709 | ||
b2950863 | 710 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
711 | { |
712 | struct btrfs_device *device; | |
713 | ||
714 | device = container_of(work, struct btrfs_device, work); | |
715 | run_scheduled_bios(device); | |
716 | } | |
717 | ||
70bc7088 AJ |
718 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
719 | { | |
720 | int found; | |
721 | ||
722 | rcu_read_lock(); | |
723 | found = strcmp(rcu_str_deref(device->name), path); | |
724 | rcu_read_unlock(); | |
725 | ||
726 | return found == 0; | |
727 | } | |
728 | ||
d8367db3 AJ |
729 | /* |
730 | * Search and remove all stale (devices which are not mounted) devices. | |
731 | * When both inputs are NULL, it will search and release all stale devices. | |
732 | * path: Optional. When provided will it release all unmounted devices | |
733 | * matching this path only. | |
734 | * skip_dev: Optional. Will skip this device when searching for the stale | |
735 | * devices. | |
70bc7088 AJ |
736 | * Return: 0 for success or if @path is NULL. |
737 | * -EBUSY if @path is a mounted device. | |
738 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 739 | */ |
70bc7088 | 740 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 741 | struct btrfs_device *skip_device) |
4fde46f0 | 742 | { |
fa6d2ae5 AJ |
743 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
744 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
745 | int ret = 0; |
746 | ||
747 | if (path) | |
748 | ret = -ENOENT; | |
4fde46f0 | 749 | |
fa6d2ae5 | 750 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 751 | |
70bc7088 | 752 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
753 | list_for_each_entry_safe(device, tmp_device, |
754 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 755 | if (skip_device && skip_device == device) |
d8367db3 | 756 | continue; |
fa6d2ae5 | 757 | if (path && !device->name) |
4fde46f0 | 758 | continue; |
70bc7088 | 759 | if (path && !device_path_matched(path, device)) |
38cf665d | 760 | continue; |
70bc7088 AJ |
761 | if (fs_devices->opened) { |
762 | /* for an already deleted device return 0 */ | |
763 | if (path && ret != 0) | |
764 | ret = -EBUSY; | |
765 | break; | |
766 | } | |
4fde46f0 | 767 | |
4fde46f0 | 768 | /* delete the stale device */ |
7bcb8164 AJ |
769 | fs_devices->num_devices--; |
770 | list_del(&device->dev_list); | |
771 | btrfs_free_device(device); | |
772 | ||
70bc7088 | 773 | ret = 0; |
7bcb8164 | 774 | if (fs_devices->num_devices == 0) |
fd649f10 | 775 | break; |
7bcb8164 AJ |
776 | } |
777 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 778 | |
7bcb8164 AJ |
779 | if (fs_devices->num_devices == 0) { |
780 | btrfs_sysfs_remove_fsid(fs_devices); | |
781 | list_del(&fs_devices->fs_list); | |
782 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
783 | } |
784 | } | |
70bc7088 AJ |
785 | |
786 | return ret; | |
4fde46f0 AJ |
787 | } |
788 | ||
0fb08bcc AJ |
789 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
790 | struct btrfs_device *device, fmode_t flags, | |
791 | void *holder) | |
792 | { | |
793 | struct request_queue *q; | |
794 | struct block_device *bdev; | |
795 | struct buffer_head *bh; | |
796 | struct btrfs_super_block *disk_super; | |
797 | u64 devid; | |
798 | int ret; | |
799 | ||
800 | if (device->bdev) | |
801 | return -EINVAL; | |
802 | if (!device->name) | |
803 | return -EINVAL; | |
804 | ||
805 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
806 | &bdev, &bh); | |
807 | if (ret) | |
808 | return ret; | |
809 | ||
810 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
811 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
812 | if (devid != device->devid) | |
813 | goto error_brelse; | |
814 | ||
815 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
816 | goto error_brelse; | |
817 | ||
818 | device->generation = btrfs_super_generation(disk_super); | |
819 | ||
820 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
821 | if (btrfs_super_incompat_flags(disk_super) & |
822 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
823 | pr_err( | |
824 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
825 | goto error_brelse; | |
826 | } | |
827 | ||
ebbede42 | 828 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0fb08bcc AJ |
829 | fs_devices->seeding = 1; |
830 | } else { | |
ebbede42 AJ |
831 | if (bdev_read_only(bdev)) |
832 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
833 | else | |
834 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
835 | } |
836 | ||
837 | q = bdev_get_queue(bdev); | |
0fb08bcc AJ |
838 | if (!blk_queue_nonrot(q)) |
839 | fs_devices->rotating = 1; | |
840 | ||
841 | device->bdev = bdev; | |
e12c9621 | 842 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
843 | device->mode = flags; |
844 | ||
845 | fs_devices->open_devices++; | |
ebbede42 AJ |
846 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
847 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 848 | fs_devices->rw_devices++; |
b1b8e386 | 849 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc AJ |
850 | } |
851 | brelse(bh); | |
852 | ||
853 | return 0; | |
854 | ||
855 | error_brelse: | |
856 | brelse(bh); | |
857 | blkdev_put(bdev, flags); | |
858 | ||
859 | return -EINVAL; | |
860 | } | |
861 | ||
7a62d0f0 NB |
862 | /* |
863 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
864 | * being created with a disk that has already completed its fsid change. | |
865 | */ | |
866 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
867 | struct btrfs_super_block *disk_super) | |
868 | { | |
869 | struct btrfs_fs_devices *fs_devices; | |
870 | ||
871 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
872 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
873 | BTRFS_FSID_SIZE) != 0 && | |
874 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
875 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
876 | return fs_devices; | |
877 | } | |
878 | } | |
879 | ||
880 | return NULL; | |
881 | } | |
882 | ||
cc5de4e7 NB |
883 | |
884 | static struct btrfs_fs_devices *find_fsid_changed( | |
885 | struct btrfs_super_block *disk_super) | |
886 | { | |
887 | struct btrfs_fs_devices *fs_devices; | |
888 | ||
889 | /* | |
890 | * Handles the case where scanned device is part of an fs that had | |
891 | * multiple successful changes of FSID but curently device didn't | |
892 | * observe it. Meaning our fsid will be different than theirs. | |
893 | */ | |
894 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
895 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
896 | BTRFS_FSID_SIZE) != 0 && | |
897 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
898 | BTRFS_FSID_SIZE) == 0 && | |
899 | memcmp(fs_devices->fsid, disk_super->fsid, | |
900 | BTRFS_FSID_SIZE) != 0) { | |
901 | return fs_devices; | |
902 | } | |
903 | } | |
904 | ||
905 | return NULL; | |
906 | } | |
60999ca4 DS |
907 | /* |
908 | * Add new device to list of registered devices | |
909 | * | |
910 | * Returns: | |
e124ece5 AJ |
911 | * device pointer which was just added or updated when successful |
912 | * error pointer when failed | |
60999ca4 | 913 | */ |
e124ece5 | 914 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
915 | struct btrfs_super_block *disk_super, |
916 | bool *new_device_added) | |
8a4b83cc CM |
917 | { |
918 | struct btrfs_device *device; | |
7a62d0f0 | 919 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 920 | struct rcu_string *name; |
8a4b83cc | 921 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 922 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
923 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
924 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
925 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
926 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 927 | |
cc5de4e7 NB |
928 | if (fsid_change_in_progress) { |
929 | if (!has_metadata_uuid) { | |
930 | /* | |
931 | * When we have an image which has CHANGING_FSID_V2 set | |
932 | * it might belong to either a filesystem which has | |
933 | * disks with completed fsid change or it might belong | |
934 | * to fs with no UUID changes in effect, handle both. | |
935 | */ | |
936 | fs_devices = find_fsid_inprogress(disk_super); | |
937 | if (!fs_devices) | |
938 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
939 | } else { | |
940 | fs_devices = find_fsid_changed(disk_super); | |
941 | } | |
7a62d0f0 NB |
942 | } else if (has_metadata_uuid) { |
943 | fs_devices = find_fsid(disk_super->fsid, | |
944 | disk_super->metadata_uuid); | |
945 | } else { | |
7239ff4b | 946 | fs_devices = find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
947 | } |
948 | ||
8a4b83cc | 949 | |
8a4b83cc | 950 | if (!fs_devices) { |
7239ff4b NB |
951 | if (has_metadata_uuid) |
952 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
953 | disk_super->metadata_uuid); | |
954 | else | |
955 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
956 | ||
2208a378 | 957 | if (IS_ERR(fs_devices)) |
e124ece5 | 958 | return ERR_CAST(fs_devices); |
2208a378 | 959 | |
92900e51 AV |
960 | fs_devices->fsid_change = fsid_change_in_progress; |
961 | ||
9c6d173e | 962 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 963 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 964 | |
8a4b83cc CM |
965 | device = NULL; |
966 | } else { | |
9c6d173e | 967 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 AJ |
968 | device = btrfs_find_device(fs_devices, devid, |
969 | disk_super->dev_item.uuid, NULL, false); | |
7a62d0f0 NB |
970 | |
971 | /* | |
972 | * If this disk has been pulled into an fs devices created by | |
973 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
974 | * metadata_uuid/fsid values of the fs_devices. | |
975 | */ | |
976 | if (has_metadata_uuid && fs_devices->fsid_change && | |
977 | found_transid > fs_devices->latest_generation) { | |
978 | memcpy(fs_devices->fsid, disk_super->fsid, | |
979 | BTRFS_FSID_SIZE); | |
980 | memcpy(fs_devices->metadata_uuid, | |
981 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
982 | ||
983 | fs_devices->fsid_change = false; | |
984 | } | |
8a4b83cc | 985 | } |
443f24fe | 986 | |
8a4b83cc | 987 | if (!device) { |
9c6d173e AJ |
988 | if (fs_devices->opened) { |
989 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 990 | return ERR_PTR(-EBUSY); |
9c6d173e | 991 | } |
2b82032c | 992 | |
12bd2fc0 ID |
993 | device = btrfs_alloc_device(NULL, &devid, |
994 | disk_super->dev_item.uuid); | |
995 | if (IS_ERR(device)) { | |
9c6d173e | 996 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 997 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 998 | return device; |
8a4b83cc | 999 | } |
606686ee JB |
1000 | |
1001 | name = rcu_string_strdup(path, GFP_NOFS); | |
1002 | if (!name) { | |
a425f9d4 | 1003 | btrfs_free_device(device); |
9c6d173e | 1004 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1005 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 1006 | } |
606686ee | 1007 | rcu_assign_pointer(device->name, name); |
90519d66 | 1008 | |
1f78160c | 1009 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 1010 | fs_devices->num_devices++; |
e5e9a520 | 1011 | |
2b82032c | 1012 | device->fs_devices = fs_devices; |
4306a974 | 1013 | *new_device_added = true; |
327f18cc AJ |
1014 | |
1015 | if (disk_super->label[0]) | |
1016 | pr_info("BTRFS: device label %s devid %llu transid %llu %s\n", | |
1017 | disk_super->label, devid, found_transid, path); | |
1018 | else | |
1019 | pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n", | |
1020 | disk_super->fsid, devid, found_transid, path); | |
1021 | ||
606686ee | 1022 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
1023 | /* |
1024 | * When FS is already mounted. | |
1025 | * 1. If you are here and if the device->name is NULL that | |
1026 | * means this device was missing at time of FS mount. | |
1027 | * 2. If you are here and if the device->name is different | |
1028 | * from 'path' that means either | |
1029 | * a. The same device disappeared and reappeared with | |
1030 | * different name. or | |
1031 | * b. The missing-disk-which-was-replaced, has | |
1032 | * reappeared now. | |
1033 | * | |
1034 | * We must allow 1 and 2a above. But 2b would be a spurious | |
1035 | * and unintentional. | |
1036 | * | |
1037 | * Further in case of 1 and 2a above, the disk at 'path' | |
1038 | * would have missed some transaction when it was away and | |
1039 | * in case of 2a the stale bdev has to be updated as well. | |
1040 | * 2b must not be allowed at all time. | |
1041 | */ | |
1042 | ||
1043 | /* | |
0f23ae74 CM |
1044 | * For now, we do allow update to btrfs_fs_device through the |
1045 | * btrfs dev scan cli after FS has been mounted. We're still | |
1046 | * tracking a problem where systems fail mount by subvolume id | |
1047 | * when we reject replacement on a mounted FS. | |
b96de000 | 1048 | */ |
0f23ae74 | 1049 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
1050 | /* |
1051 | * That is if the FS is _not_ mounted and if you | |
1052 | * are here, that means there is more than one | |
1053 | * disk with same uuid and devid.We keep the one | |
1054 | * with larger generation number or the last-in if | |
1055 | * generation are equal. | |
1056 | */ | |
9c6d173e | 1057 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1058 | return ERR_PTR(-EEXIST); |
77bdae4d | 1059 | } |
b96de000 | 1060 | |
a9261d41 AJ |
1061 | /* |
1062 | * We are going to replace the device path for a given devid, | |
1063 | * make sure it's the same device if the device is mounted | |
1064 | */ | |
1065 | if (device->bdev) { | |
1066 | struct block_device *path_bdev; | |
1067 | ||
1068 | path_bdev = lookup_bdev(path); | |
1069 | if (IS_ERR(path_bdev)) { | |
1070 | mutex_unlock(&fs_devices->device_list_mutex); | |
1071 | return ERR_CAST(path_bdev); | |
1072 | } | |
1073 | ||
1074 | if (device->bdev != path_bdev) { | |
1075 | bdput(path_bdev); | |
1076 | mutex_unlock(&fs_devices->device_list_mutex); | |
1077 | btrfs_warn_in_rcu(device->fs_info, | |
1078 | "duplicate device fsid:devid for %pU:%llu old:%s new:%s", | |
1079 | disk_super->fsid, devid, | |
1080 | rcu_str_deref(device->name), path); | |
1081 | return ERR_PTR(-EEXIST); | |
1082 | } | |
1083 | bdput(path_bdev); | |
1084 | btrfs_info_in_rcu(device->fs_info, | |
1085 | "device fsid %pU devid %llu moved old:%s new:%s", | |
1086 | disk_super->fsid, devid, | |
1087 | rcu_str_deref(device->name), path); | |
1088 | } | |
1089 | ||
606686ee | 1090 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
1091 | if (!name) { |
1092 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 1093 | return ERR_PTR(-ENOMEM); |
9c6d173e | 1094 | } |
606686ee JB |
1095 | rcu_string_free(device->name); |
1096 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 1097 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 1098 | fs_devices->missing_devices--; |
e6e674bd | 1099 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 1100 | } |
8a4b83cc CM |
1101 | } |
1102 | ||
77bdae4d AJ |
1103 | /* |
1104 | * Unmount does not free the btrfs_device struct but would zero | |
1105 | * generation along with most of the other members. So just update | |
1106 | * it back. We need it to pick the disk with largest generation | |
1107 | * (as above). | |
1108 | */ | |
d1a63002 | 1109 | if (!fs_devices->opened) { |
77bdae4d | 1110 | device->generation = found_transid; |
d1a63002 NB |
1111 | fs_devices->latest_generation = max_t(u64, found_transid, |
1112 | fs_devices->latest_generation); | |
1113 | } | |
77bdae4d | 1114 | |
f2788d2f AJ |
1115 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
1116 | ||
9c6d173e | 1117 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 1118 | return device; |
8a4b83cc CM |
1119 | } |
1120 | ||
e4404d6e YZ |
1121 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
1122 | { | |
1123 | struct btrfs_fs_devices *fs_devices; | |
1124 | struct btrfs_device *device; | |
1125 | struct btrfs_device *orig_dev; | |
1126 | ||
7239ff4b | 1127 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
1128 | if (IS_ERR(fs_devices)) |
1129 | return fs_devices; | |
e4404d6e | 1130 | |
adbbb863 | 1131 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 1132 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
1133 | |
1134 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
1135 | struct rcu_string *name; |
1136 | ||
12bd2fc0 ID |
1137 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1138 | orig_dev->uuid); | |
1139 | if (IS_ERR(device)) | |
e4404d6e YZ |
1140 | goto error; |
1141 | ||
606686ee JB |
1142 | /* |
1143 | * This is ok to do without rcu read locked because we hold the | |
1144 | * uuid mutex so nothing we touch in here is going to disappear. | |
1145 | */ | |
e755f780 | 1146 | if (orig_dev->name) { |
78f2c9e6 DS |
1147 | name = rcu_string_strdup(orig_dev->name->str, |
1148 | GFP_KERNEL); | |
e755f780 | 1149 | if (!name) { |
a425f9d4 | 1150 | btrfs_free_device(device); |
e755f780 AJ |
1151 | goto error; |
1152 | } | |
1153 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1154 | } |
e4404d6e | 1155 | |
e4404d6e YZ |
1156 | list_add(&device->dev_list, &fs_devices->devices); |
1157 | device->fs_devices = fs_devices; | |
1158 | fs_devices->num_devices++; | |
1159 | } | |
adbbb863 | 1160 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1161 | return fs_devices; |
1162 | error: | |
adbbb863 | 1163 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1164 | free_fs_devices(fs_devices); |
1165 | return ERR_PTR(-ENOMEM); | |
1166 | } | |
1167 | ||
9b99b115 AJ |
1168 | /* |
1169 | * After we have read the system tree and know devids belonging to | |
1170 | * this filesystem, remove the device which does not belong there. | |
1171 | */ | |
1172 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step) | |
dfe25020 | 1173 | { |
c6e30871 | 1174 | struct btrfs_device *device, *next; |
443f24fe | 1175 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 1176 | |
dfe25020 CM |
1177 | mutex_lock(&uuid_mutex); |
1178 | again: | |
46224705 | 1179 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1180 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
e12c9621 AJ |
1181 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
1182 | &device->dev_state)) { | |
401e29c1 AJ |
1183 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1184 | &device->dev_state) && | |
1185 | (!latest_dev || | |
1186 | device->generation > latest_dev->generation)) { | |
443f24fe | 1187 | latest_dev = device; |
a6b0d5c8 | 1188 | } |
2b82032c | 1189 | continue; |
a6b0d5c8 | 1190 | } |
2b82032c | 1191 | |
8dabb742 SB |
1192 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
1193 | /* | |
1194 | * In the first step, keep the device which has | |
1195 | * the correct fsid and the devid that is used | |
1196 | * for the dev_replace procedure. | |
1197 | * In the second step, the dev_replace state is | |
1198 | * read from the device tree and it is known | |
1199 | * whether the procedure is really active or | |
1200 | * not, which means whether this device is | |
1201 | * used or whether it should be removed. | |
1202 | */ | |
401e29c1 AJ |
1203 | if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1204 | &device->dev_state)) { | |
8dabb742 SB |
1205 | continue; |
1206 | } | |
1207 | } | |
2b82032c | 1208 | if (device->bdev) { |
d4d77629 | 1209 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1210 | device->bdev = NULL; |
1211 | fs_devices->open_devices--; | |
1212 | } | |
ebbede42 | 1213 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1214 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1215 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
401e29c1 AJ |
1216 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
1217 | &device->dev_state)) | |
8dabb742 | 1218 | fs_devices->rw_devices--; |
2b82032c | 1219 | } |
e4404d6e YZ |
1220 | list_del_init(&device->dev_list); |
1221 | fs_devices->num_devices--; | |
a425f9d4 | 1222 | btrfs_free_device(device); |
dfe25020 | 1223 | } |
2b82032c YZ |
1224 | |
1225 | if (fs_devices->seed) { | |
1226 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
1227 | goto again; |
1228 | } | |
1229 | ||
443f24fe | 1230 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1231 | |
dfe25020 | 1232 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1233 | } |
a0af469b | 1234 | |
14238819 AJ |
1235 | static void btrfs_close_bdev(struct btrfs_device *device) |
1236 | { | |
08ffcae8 DS |
1237 | if (!device->bdev) |
1238 | return; | |
1239 | ||
ebbede42 | 1240 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1241 | sync_blockdev(device->bdev); |
1242 | invalidate_bdev(device->bdev); | |
1243 | } | |
1244 | ||
08ffcae8 | 1245 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1246 | } |
1247 | ||
959b1c04 | 1248 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1249 | { |
1250 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
1251 | struct btrfs_device *new_device; | |
1252 | struct rcu_string *name; | |
1253 | ||
1254 | if (device->bdev) | |
1255 | fs_devices->open_devices--; | |
1256 | ||
ebbede42 | 1257 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1258 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1259 | list_del_init(&device->dev_alloc_list); | |
1260 | fs_devices->rw_devices--; | |
1261 | } | |
1262 | ||
e6e674bd | 1263 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1264 | fs_devices->missing_devices--; |
1265 | ||
959b1c04 NB |
1266 | btrfs_close_bdev(device); |
1267 | ||
f448341a AJ |
1268 | new_device = btrfs_alloc_device(NULL, &device->devid, |
1269 | device->uuid); | |
1270 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
1271 | ||
1272 | /* Safe because we are under uuid_mutex */ | |
1273 | if (device->name) { | |
1274 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
1275 | BUG_ON(!name); /* -ENOMEM */ | |
1276 | rcu_assign_pointer(new_device->name, name); | |
1277 | } | |
1278 | ||
1279 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
1280 | new_device->fs_devices = device->fs_devices; | |
959b1c04 | 1281 | |
8e75fd89 NB |
1282 | synchronize_rcu(); |
1283 | btrfs_free_device(device); | |
f448341a AJ |
1284 | } |
1285 | ||
0226e0eb | 1286 | static int close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1287 | { |
2037a093 | 1288 | struct btrfs_device *device, *tmp; |
e4404d6e | 1289 | |
2b82032c YZ |
1290 | if (--fs_devices->opened > 0) |
1291 | return 0; | |
8a4b83cc | 1292 | |
c9513edb | 1293 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 1294 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
959b1c04 | 1295 | btrfs_close_one_device(device); |
8a4b83cc | 1296 | } |
c9513edb XG |
1297 | mutex_unlock(&fs_devices->device_list_mutex); |
1298 | ||
e4404d6e YZ |
1299 | WARN_ON(fs_devices->open_devices); |
1300 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
1301 | fs_devices->opened = 0; |
1302 | fs_devices->seeding = 0; | |
2b82032c | 1303 | |
8a4b83cc CM |
1304 | return 0; |
1305 | } | |
1306 | ||
2b82032c YZ |
1307 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
1308 | { | |
e4404d6e | 1309 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
1310 | int ret; |
1311 | ||
1312 | mutex_lock(&uuid_mutex); | |
0226e0eb | 1313 | ret = close_fs_devices(fs_devices); |
e4404d6e YZ |
1314 | if (!fs_devices->opened) { |
1315 | seed_devices = fs_devices->seed; | |
1316 | fs_devices->seed = NULL; | |
1317 | } | |
2b82032c | 1318 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
1319 | |
1320 | while (seed_devices) { | |
1321 | fs_devices = seed_devices; | |
1322 | seed_devices = fs_devices->seed; | |
0226e0eb | 1323 | close_fs_devices(fs_devices); |
e4404d6e YZ |
1324 | free_fs_devices(fs_devices); |
1325 | } | |
2b82032c YZ |
1326 | return ret; |
1327 | } | |
1328 | ||
897fb573 | 1329 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1330 | fmode_t flags, void *holder) |
8a4b83cc | 1331 | { |
8a4b83cc | 1332 | struct btrfs_device *device; |
443f24fe | 1333 | struct btrfs_device *latest_dev = NULL; |
a0af469b | 1334 | int ret = 0; |
8a4b83cc | 1335 | |
d4d77629 TH |
1336 | flags |= FMODE_EXCL; |
1337 | ||
f117e290 | 1338 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
f63e0cca | 1339 | /* Just open everything we can; ignore failures here */ |
0fb08bcc | 1340 | if (btrfs_open_one_device(fs_devices, device, flags, holder)) |
beaf8ab3 | 1341 | continue; |
a0af469b | 1342 | |
9f050db4 AJ |
1343 | if (!latest_dev || |
1344 | device->generation > latest_dev->generation) | |
1345 | latest_dev = device; | |
8a4b83cc | 1346 | } |
a0af469b | 1347 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1348 | ret = -EINVAL; |
a0af469b CM |
1349 | goto out; |
1350 | } | |
2b82032c | 1351 | fs_devices->opened = 1; |
443f24fe | 1352 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1353 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1354 | out: |
2b82032c YZ |
1355 | return ret; |
1356 | } | |
1357 | ||
f8e10cd3 AJ |
1358 | static int devid_cmp(void *priv, struct list_head *a, struct list_head *b) |
1359 | { | |
1360 | struct btrfs_device *dev1, *dev2; | |
1361 | ||
1362 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1363 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1364 | ||
1365 | if (dev1->devid < dev2->devid) | |
1366 | return -1; | |
1367 | else if (dev1->devid > dev2->devid) | |
1368 | return 1; | |
1369 | return 0; | |
1370 | } | |
1371 | ||
2b82032c | 1372 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1373 | fmode_t flags, void *holder) |
2b82032c YZ |
1374 | { |
1375 | int ret; | |
1376 | ||
f5194e34 DS |
1377 | lockdep_assert_held(&uuid_mutex); |
1378 | ||
542c5908 | 1379 | mutex_lock(&fs_devices->device_list_mutex); |
2b82032c | 1380 | if (fs_devices->opened) { |
e4404d6e YZ |
1381 | fs_devices->opened++; |
1382 | ret = 0; | |
2b82032c | 1383 | } else { |
f8e10cd3 | 1384 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1385 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1386 | } |
542c5908 AJ |
1387 | mutex_unlock(&fs_devices->device_list_mutex); |
1388 | ||
8a4b83cc CM |
1389 | return ret; |
1390 | } | |
1391 | ||
c9162bdf | 1392 | static void btrfs_release_disk_super(struct page *page) |
6cf86a00 AJ |
1393 | { |
1394 | kunmap(page); | |
1395 | put_page(page); | |
1396 | } | |
1397 | ||
c9162bdf OS |
1398 | static int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, |
1399 | struct page **page, | |
1400 | struct btrfs_super_block **disk_super) | |
6cf86a00 AJ |
1401 | { |
1402 | void *p; | |
1403 | pgoff_t index; | |
1404 | ||
1405 | /* make sure our super fits in the device */ | |
1406 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1407 | return 1; | |
1408 | ||
1409 | /* make sure our super fits in the page */ | |
1410 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1411 | return 1; | |
1412 | ||
1413 | /* make sure our super doesn't straddle pages on disk */ | |
1414 | index = bytenr >> PAGE_SHIFT; | |
1415 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1416 | return 1; | |
1417 | ||
1418 | /* pull in the page with our super */ | |
1419 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1420 | index, GFP_KERNEL); | |
1421 | ||
1422 | if (IS_ERR_OR_NULL(*page)) | |
1423 | return 1; | |
1424 | ||
1425 | p = kmap(*page); | |
1426 | ||
1427 | /* align our pointer to the offset of the super block */ | |
7073017a | 1428 | *disk_super = p + offset_in_page(bytenr); |
6cf86a00 AJ |
1429 | |
1430 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1431 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1432 | btrfs_release_disk_super(*page); | |
1433 | return 1; | |
1434 | } | |
1435 | ||
1436 | if ((*disk_super)->label[0] && | |
1437 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1438 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1439 | ||
1440 | return 0; | |
1441 | } | |
1442 | ||
228a73ab AJ |
1443 | int btrfs_forget_devices(const char *path) |
1444 | { | |
1445 | int ret; | |
1446 | ||
1447 | mutex_lock(&uuid_mutex); | |
1448 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1449 | mutex_unlock(&uuid_mutex); | |
1450 | ||
1451 | return ret; | |
1452 | } | |
1453 | ||
6f60cbd3 DS |
1454 | /* |
1455 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1456 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1457 | * is read via pagecache | |
1458 | */ | |
36350e95 GJ |
1459 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1460 | void *holder) | |
8a4b83cc CM |
1461 | { |
1462 | struct btrfs_super_block *disk_super; | |
4306a974 | 1463 | bool new_device_added = false; |
36350e95 | 1464 | struct btrfs_device *device = NULL; |
8a4b83cc | 1465 | struct block_device *bdev; |
6f60cbd3 | 1466 | struct page *page; |
6f60cbd3 | 1467 | u64 bytenr; |
8a4b83cc | 1468 | |
899f9307 DS |
1469 | lockdep_assert_held(&uuid_mutex); |
1470 | ||
6f60cbd3 DS |
1471 | /* |
1472 | * we would like to check all the supers, but that would make | |
1473 | * a btrfs mount succeed after a mkfs from a different FS. | |
1474 | * So, we need to add a special mount option to scan for | |
1475 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1476 | */ | |
1477 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1478 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1479 | |
1480 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1481 | if (IS_ERR(bdev)) |
36350e95 | 1482 | return ERR_CAST(bdev); |
6f60cbd3 | 1483 | |
05a5c55d | 1484 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) { |
36350e95 | 1485 | device = ERR_PTR(-EINVAL); |
6f60cbd3 | 1486 | goto error_bdev_put; |
05a5c55d | 1487 | } |
6f60cbd3 | 1488 | |
4306a974 | 1489 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1490 | if (!IS_ERR(device)) { |
4306a974 AJ |
1491 | if (new_device_added) |
1492 | btrfs_free_stale_devices(path, device); | |
1493 | } | |
6f60cbd3 | 1494 | |
6cf86a00 | 1495 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1496 | |
1497 | error_bdev_put: | |
d4d77629 | 1498 | blkdev_put(bdev, flags); |
b6ed73bc | 1499 | |
36350e95 | 1500 | return device; |
8a4b83cc | 1501 | } |
0b86a832 | 1502 | |
1c11b63e JM |
1503 | /* |
1504 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1505 | * such is found, record the end of it in *start | |
1506 | */ | |
1c11b63e JM |
1507 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1508 | u64 len) | |
6df9a95e | 1509 | { |
1c11b63e | 1510 | u64 physical_start, physical_end; |
6df9a95e | 1511 | |
1c11b63e | 1512 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1513 | |
1c11b63e JM |
1514 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1515 | &physical_start, &physical_end, | |
1516 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1517 | |
1c11b63e JM |
1518 | if (in_range(physical_start, *start, len) || |
1519 | in_range(*start, physical_start, | |
1520 | physical_end - physical_start)) { | |
1521 | *start = physical_end + 1; | |
1522 | return true; | |
6df9a95e JB |
1523 | } |
1524 | } | |
1c11b63e | 1525 | return false; |
6df9a95e JB |
1526 | } |
1527 | ||
1528 | ||
0b86a832 | 1529 | /* |
499f377f JM |
1530 | * find_free_dev_extent_start - find free space in the specified device |
1531 | * @device: the device which we search the free space in | |
1532 | * @num_bytes: the size of the free space that we need | |
1533 | * @search_start: the position from which to begin the search | |
1534 | * @start: store the start of the free space. | |
1535 | * @len: the size of the free space. that we find, or the size | |
1536 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1537 | * |
0b86a832 CM |
1538 | * this uses a pretty simple search, the expectation is that it is |
1539 | * called very infrequently and that a given device has a small number | |
1540 | * of extents | |
7bfc837d MX |
1541 | * |
1542 | * @start is used to store the start of the free space if we find. But if we | |
1543 | * don't find suitable free space, it will be used to store the start position | |
1544 | * of the max free space. | |
1545 | * | |
1546 | * @len is used to store the size of the free space that we find. | |
1547 | * But if we don't find suitable free space, it is used to store the size of | |
1548 | * the max free space. | |
0b86a832 | 1549 | */ |
60dfdf25 | 1550 | int find_free_dev_extent_start(struct btrfs_device *device, u64 num_bytes, |
499f377f | 1551 | u64 search_start, u64 *start, u64 *len) |
0b86a832 | 1552 | { |
0b246afa JM |
1553 | struct btrfs_fs_info *fs_info = device->fs_info; |
1554 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1555 | struct btrfs_key key; |
7bfc837d | 1556 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1557 | struct btrfs_path *path; |
7bfc837d MX |
1558 | u64 hole_size; |
1559 | u64 max_hole_start; | |
1560 | u64 max_hole_size; | |
1561 | u64 extent_end; | |
0b86a832 CM |
1562 | u64 search_end = device->total_bytes; |
1563 | int ret; | |
7bfc837d | 1564 | int slot; |
0b86a832 | 1565 | struct extent_buffer *l; |
8cdc7c5b FM |
1566 | |
1567 | /* | |
1568 | * We don't want to overwrite the superblock on the drive nor any area | |
1569 | * used by the boot loader (grub for example), so we make sure to start | |
1570 | * at an offset of at least 1MB. | |
1571 | */ | |
0d0c71b3 | 1572 | search_start = max_t(u64, search_start, SZ_1M); |
0b86a832 | 1573 | |
6df9a95e JB |
1574 | path = btrfs_alloc_path(); |
1575 | if (!path) | |
1576 | return -ENOMEM; | |
f2ab7618 | 1577 | |
7bfc837d MX |
1578 | max_hole_start = search_start; |
1579 | max_hole_size = 0; | |
1580 | ||
f2ab7618 | 1581 | again: |
401e29c1 AJ |
1582 | if (search_start >= search_end || |
1583 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1584 | ret = -ENOSPC; |
6df9a95e | 1585 | goto out; |
7bfc837d MX |
1586 | } |
1587 | ||
e4058b54 | 1588 | path->reada = READA_FORWARD; |
6df9a95e JB |
1589 | path->search_commit_root = 1; |
1590 | path->skip_locking = 1; | |
7bfc837d | 1591 | |
0b86a832 CM |
1592 | key.objectid = device->devid; |
1593 | key.offset = search_start; | |
1594 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1595 | |
125ccb0a | 1596 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1597 | if (ret < 0) |
7bfc837d | 1598 | goto out; |
1fcbac58 YZ |
1599 | if (ret > 0) { |
1600 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1601 | if (ret < 0) | |
7bfc837d | 1602 | goto out; |
1fcbac58 | 1603 | } |
7bfc837d | 1604 | |
0b86a832 CM |
1605 | while (1) { |
1606 | l = path->nodes[0]; | |
1607 | slot = path->slots[0]; | |
1608 | if (slot >= btrfs_header_nritems(l)) { | |
1609 | ret = btrfs_next_leaf(root, path); | |
1610 | if (ret == 0) | |
1611 | continue; | |
1612 | if (ret < 0) | |
7bfc837d MX |
1613 | goto out; |
1614 | ||
1615 | break; | |
0b86a832 CM |
1616 | } |
1617 | btrfs_item_key_to_cpu(l, &key, slot); | |
1618 | ||
1619 | if (key.objectid < device->devid) | |
1620 | goto next; | |
1621 | ||
1622 | if (key.objectid > device->devid) | |
7bfc837d | 1623 | break; |
0b86a832 | 1624 | |
962a298f | 1625 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1626 | goto next; |
9779b72f | 1627 | |
7bfc837d MX |
1628 | if (key.offset > search_start) { |
1629 | hole_size = key.offset - search_start; | |
9779b72f | 1630 | |
6df9a95e JB |
1631 | /* |
1632 | * Have to check before we set max_hole_start, otherwise | |
1633 | * we could end up sending back this offset anyway. | |
1634 | */ | |
1c11b63e | 1635 | if (contains_pending_extent(device, &search_start, |
1b984508 | 1636 | hole_size)) { |
1c11b63e | 1637 | if (key.offset >= search_start) |
1b984508 | 1638 | hole_size = key.offset - search_start; |
1c11b63e | 1639 | else |
1b984508 | 1640 | hole_size = 0; |
1b984508 | 1641 | } |
6df9a95e | 1642 | |
7bfc837d MX |
1643 | if (hole_size > max_hole_size) { |
1644 | max_hole_start = search_start; | |
1645 | max_hole_size = hole_size; | |
1646 | } | |
9779b72f | 1647 | |
7bfc837d MX |
1648 | /* |
1649 | * If this free space is greater than which we need, | |
1650 | * it must be the max free space that we have found | |
1651 | * until now, so max_hole_start must point to the start | |
1652 | * of this free space and the length of this free space | |
1653 | * is stored in max_hole_size. Thus, we return | |
1654 | * max_hole_start and max_hole_size and go back to the | |
1655 | * caller. | |
1656 | */ | |
1657 | if (hole_size >= num_bytes) { | |
1658 | ret = 0; | |
1659 | goto out; | |
0b86a832 CM |
1660 | } |
1661 | } | |
0b86a832 | 1662 | |
0b86a832 | 1663 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1664 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1665 | dev_extent); | |
1666 | if (extent_end > search_start) | |
1667 | search_start = extent_end; | |
0b86a832 CM |
1668 | next: |
1669 | path->slots[0]++; | |
1670 | cond_resched(); | |
1671 | } | |
0b86a832 | 1672 | |
38c01b96 | 1673 | /* |
1674 | * At this point, search_start should be the end of | |
1675 | * allocated dev extents, and when shrinking the device, | |
1676 | * search_end may be smaller than search_start. | |
1677 | */ | |
f2ab7618 | 1678 | if (search_end > search_start) { |
38c01b96 | 1679 | hole_size = search_end - search_start; |
1680 | ||
1c11b63e | 1681 | if (contains_pending_extent(device, &search_start, hole_size)) { |
f2ab7618 ZL |
1682 | btrfs_release_path(path); |
1683 | goto again; | |
1684 | } | |
0b86a832 | 1685 | |
f2ab7618 ZL |
1686 | if (hole_size > max_hole_size) { |
1687 | max_hole_start = search_start; | |
1688 | max_hole_size = hole_size; | |
1689 | } | |
6df9a95e JB |
1690 | } |
1691 | ||
7bfc837d | 1692 | /* See above. */ |
f2ab7618 | 1693 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1694 | ret = -ENOSPC; |
1695 | else | |
1696 | ret = 0; | |
1697 | ||
1698 | out: | |
2b82032c | 1699 | btrfs_free_path(path); |
7bfc837d | 1700 | *start = max_hole_start; |
b2117a39 | 1701 | if (len) |
7bfc837d | 1702 | *len = max_hole_size; |
0b86a832 CM |
1703 | return ret; |
1704 | } | |
1705 | ||
60dfdf25 | 1706 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1707 | u64 *start, u64 *len) |
1708 | { | |
499f377f | 1709 | /* FIXME use last free of some kind */ |
60dfdf25 | 1710 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1711 | } |
1712 | ||
b2950863 | 1713 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1714 | struct btrfs_device *device, |
2196d6e8 | 1715 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1716 | { |
0b246afa JM |
1717 | struct btrfs_fs_info *fs_info = device->fs_info; |
1718 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1719 | int ret; |
1720 | struct btrfs_path *path; | |
8f18cf13 | 1721 | struct btrfs_key key; |
a061fc8d CM |
1722 | struct btrfs_key found_key; |
1723 | struct extent_buffer *leaf = NULL; | |
1724 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1725 | |
1726 | path = btrfs_alloc_path(); | |
1727 | if (!path) | |
1728 | return -ENOMEM; | |
1729 | ||
1730 | key.objectid = device->devid; | |
1731 | key.offset = start; | |
1732 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1733 | again: |
8f18cf13 | 1734 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1735 | if (ret > 0) { |
1736 | ret = btrfs_previous_item(root, path, key.objectid, | |
1737 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1738 | if (ret) |
1739 | goto out; | |
a061fc8d CM |
1740 | leaf = path->nodes[0]; |
1741 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1742 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1743 | struct btrfs_dev_extent); | |
1744 | BUG_ON(found_key.offset > start || found_key.offset + | |
1745 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1746 | key = found_key; |
1747 | btrfs_release_path(path); | |
1748 | goto again; | |
a061fc8d CM |
1749 | } else if (ret == 0) { |
1750 | leaf = path->nodes[0]; | |
1751 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1752 | struct btrfs_dev_extent); | |
79787eaa | 1753 | } else { |
0b246afa | 1754 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1755 | goto out; |
a061fc8d | 1756 | } |
8f18cf13 | 1757 | |
2196d6e8 MX |
1758 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1759 | ||
8f18cf13 | 1760 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1761 | if (ret) { |
0b246afa JM |
1762 | btrfs_handle_fs_error(fs_info, ret, |
1763 | "Failed to remove dev extent item"); | |
13212b54 | 1764 | } else { |
3204d33c | 1765 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1766 | } |
b0b802d7 | 1767 | out: |
8f18cf13 CM |
1768 | btrfs_free_path(path); |
1769 | return ret; | |
1770 | } | |
1771 | ||
48a3b636 ES |
1772 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1773 | struct btrfs_device *device, | |
48a3b636 | 1774 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1775 | { |
1776 | int ret; | |
1777 | struct btrfs_path *path; | |
0b246afa JM |
1778 | struct btrfs_fs_info *fs_info = device->fs_info; |
1779 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1780 | struct btrfs_dev_extent *extent; |
1781 | struct extent_buffer *leaf; | |
1782 | struct btrfs_key key; | |
1783 | ||
e12c9621 | 1784 | WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)); |
401e29c1 | 1785 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
0b86a832 CM |
1786 | path = btrfs_alloc_path(); |
1787 | if (!path) | |
1788 | return -ENOMEM; | |
1789 | ||
0b86a832 | 1790 | key.objectid = device->devid; |
2b82032c | 1791 | key.offset = start; |
0b86a832 CM |
1792 | key.type = BTRFS_DEV_EXTENT_KEY; |
1793 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1794 | sizeof(*extent)); | |
2cdcecbc MF |
1795 | if (ret) |
1796 | goto out; | |
0b86a832 CM |
1797 | |
1798 | leaf = path->nodes[0]; | |
1799 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1800 | struct btrfs_dev_extent); | |
b5d9071c NB |
1801 | btrfs_set_dev_extent_chunk_tree(leaf, extent, |
1802 | BTRFS_CHUNK_TREE_OBJECTID); | |
0ca00afb NB |
1803 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, |
1804 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
e17cade2 CM |
1805 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); |
1806 | ||
0b86a832 CM |
1807 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1808 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1809 | out: |
0b86a832 CM |
1810 | btrfs_free_path(path); |
1811 | return ret; | |
1812 | } | |
1813 | ||
6df9a95e | 1814 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1815 | { |
6df9a95e JB |
1816 | struct extent_map_tree *em_tree; |
1817 | struct extent_map *em; | |
1818 | struct rb_node *n; | |
1819 | u64 ret = 0; | |
0b86a832 | 1820 | |
6df9a95e JB |
1821 | em_tree = &fs_info->mapping_tree.map_tree; |
1822 | read_lock(&em_tree->lock); | |
07e1ce09 | 1823 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1824 | if (n) { |
1825 | em = rb_entry(n, struct extent_map, rb_node); | |
1826 | ret = em->start + em->len; | |
0b86a832 | 1827 | } |
6df9a95e JB |
1828 | read_unlock(&em_tree->lock); |
1829 | ||
0b86a832 CM |
1830 | return ret; |
1831 | } | |
1832 | ||
53f10659 ID |
1833 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1834 | u64 *devid_ret) | |
0b86a832 CM |
1835 | { |
1836 | int ret; | |
1837 | struct btrfs_key key; | |
1838 | struct btrfs_key found_key; | |
2b82032c YZ |
1839 | struct btrfs_path *path; |
1840 | ||
2b82032c YZ |
1841 | path = btrfs_alloc_path(); |
1842 | if (!path) | |
1843 | return -ENOMEM; | |
0b86a832 CM |
1844 | |
1845 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1846 | key.type = BTRFS_DEV_ITEM_KEY; | |
1847 | key.offset = (u64)-1; | |
1848 | ||
53f10659 | 1849 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1850 | if (ret < 0) |
1851 | goto error; | |
1852 | ||
79787eaa | 1853 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1854 | |
53f10659 ID |
1855 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1856 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1857 | BTRFS_DEV_ITEM_KEY); |
1858 | if (ret) { | |
53f10659 | 1859 | *devid_ret = 1; |
0b86a832 CM |
1860 | } else { |
1861 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1862 | path->slots[0]); | |
53f10659 | 1863 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1864 | } |
1865 | ret = 0; | |
1866 | error: | |
2b82032c | 1867 | btrfs_free_path(path); |
0b86a832 CM |
1868 | return ret; |
1869 | } | |
1870 | ||
1871 | /* | |
1872 | * the device information is stored in the chunk root | |
1873 | * the btrfs_device struct should be fully filled in | |
1874 | */ | |
c74a0b02 | 1875 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1876 | struct btrfs_device *device) |
0b86a832 CM |
1877 | { |
1878 | int ret; | |
1879 | struct btrfs_path *path; | |
1880 | struct btrfs_dev_item *dev_item; | |
1881 | struct extent_buffer *leaf; | |
1882 | struct btrfs_key key; | |
1883 | unsigned long ptr; | |
0b86a832 | 1884 | |
0b86a832 CM |
1885 | path = btrfs_alloc_path(); |
1886 | if (!path) | |
1887 | return -ENOMEM; | |
1888 | ||
0b86a832 CM |
1889 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1890 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1891 | key.offset = device->devid; |
0b86a832 | 1892 | |
8e87e856 NB |
1893 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1894 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1895 | if (ret) |
1896 | goto out; | |
1897 | ||
1898 | leaf = path->nodes[0]; | |
1899 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1900 | ||
1901 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1902 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1903 | btrfs_set_device_type(leaf, dev_item, device->type); |
1904 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1905 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1906 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1907 | btrfs_set_device_total_bytes(leaf, dev_item, |
1908 | btrfs_device_get_disk_total_bytes(device)); | |
1909 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1910 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1911 | btrfs_set_device_group(leaf, dev_item, 0); |
1912 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1913 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1914 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1915 | |
410ba3a2 | 1916 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1917 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1918 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1919 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1920 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1921 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1922 | |
2b82032c | 1923 | ret = 0; |
0b86a832 CM |
1924 | out: |
1925 | btrfs_free_path(path); | |
1926 | return ret; | |
1927 | } | |
8f18cf13 | 1928 | |
5a1972bd QW |
1929 | /* |
1930 | * Function to update ctime/mtime for a given device path. | |
1931 | * Mainly used for ctime/mtime based probe like libblkid. | |
1932 | */ | |
da353f6b | 1933 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1934 | { |
1935 | struct file *filp; | |
1936 | ||
1937 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1938 | if (IS_ERR(filp)) |
5a1972bd QW |
1939 | return; |
1940 | file_update_time(filp); | |
1941 | filp_close(filp, NULL); | |
5a1972bd QW |
1942 | } |
1943 | ||
5b4aacef | 1944 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1945 | struct btrfs_device *device) |
1946 | { | |
5b4aacef | 1947 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1948 | int ret; |
1949 | struct btrfs_path *path; | |
a061fc8d | 1950 | struct btrfs_key key; |
a061fc8d CM |
1951 | struct btrfs_trans_handle *trans; |
1952 | ||
a061fc8d CM |
1953 | path = btrfs_alloc_path(); |
1954 | if (!path) | |
1955 | return -ENOMEM; | |
1956 | ||
a22285a6 | 1957 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1958 | if (IS_ERR(trans)) { |
1959 | btrfs_free_path(path); | |
1960 | return PTR_ERR(trans); | |
1961 | } | |
a061fc8d CM |
1962 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1963 | key.type = BTRFS_DEV_ITEM_KEY; | |
1964 | key.offset = device->devid; | |
1965 | ||
1966 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1967 | if (ret) { |
1968 | if (ret > 0) | |
1969 | ret = -ENOENT; | |
1970 | btrfs_abort_transaction(trans, ret); | |
1971 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1972 | goto out; |
1973 | } | |
1974 | ||
1975 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1976 | if (ret) { |
1977 | btrfs_abort_transaction(trans, ret); | |
1978 | btrfs_end_transaction(trans); | |
1979 | } | |
1980 | ||
a061fc8d CM |
1981 | out: |
1982 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1983 | if (!ret) |
1984 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1985 | return ret; |
1986 | } | |
1987 | ||
3cc31a0d DS |
1988 | /* |
1989 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1990 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1991 | * replace. | |
1992 | */ | |
1993 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1994 | u64 num_devices) | |
a061fc8d | 1995 | { |
a061fc8d | 1996 | u64 all_avail; |
de98ced9 | 1997 | unsigned seq; |
418775a2 | 1998 | int i; |
a061fc8d | 1999 | |
de98ced9 | 2000 | do { |
bd45ffbc | 2001 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 2002 | |
bd45ffbc AJ |
2003 | all_avail = fs_info->avail_data_alloc_bits | |
2004 | fs_info->avail_system_alloc_bits | | |
2005 | fs_info->avail_metadata_alloc_bits; | |
2006 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2007 | |
418775a2 | 2008 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2009 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2010 | continue; |
a061fc8d | 2011 | |
418775a2 | 2012 | if (num_devices < btrfs_raid_array[i].devs_min) { |
f9fbcaa2 | 2013 | int ret = btrfs_raid_array[i].mindev_error; |
bd45ffbc | 2014 | |
418775a2 DS |
2015 | if (ret) |
2016 | return ret; | |
2017 | } | |
53b381b3 DW |
2018 | } |
2019 | ||
bd45ffbc | 2020 | return 0; |
f1fa7f26 AJ |
2021 | } |
2022 | ||
c9162bdf OS |
2023 | static struct btrfs_device * btrfs_find_next_active_device( |
2024 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2025 | { |
2b82032c | 2026 | struct btrfs_device *next_device; |
88acff64 AJ |
2027 | |
2028 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2029 | if (next_device != device && | |
e6e674bd AJ |
2030 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2031 | && next_device->bdev) | |
88acff64 AJ |
2032 | return next_device; |
2033 | } | |
2034 | ||
2035 | return NULL; | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
2040 | * and replace it with the provided or the next active device, in the context | |
2041 | * where this function called, there should be always be another device (or | |
2042 | * this_dev) which is active. | |
2043 | */ | |
d6507cf1 NB |
2044 | void btrfs_assign_next_active_device(struct btrfs_device *device, |
2045 | struct btrfs_device *this_dev) | |
88acff64 | 2046 | { |
d6507cf1 | 2047 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 AJ |
2048 | struct btrfs_device *next_device; |
2049 | ||
2050 | if (this_dev) | |
2051 | next_device = this_dev; | |
2052 | else | |
2053 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
2054 | device); | |
2055 | ASSERT(next_device); | |
2056 | ||
2057 | if (fs_info->sb->s_bdev && | |
2058 | (fs_info->sb->s_bdev == device->bdev)) | |
2059 | fs_info->sb->s_bdev = next_device->bdev; | |
2060 | ||
2061 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
2062 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
2063 | } | |
2064 | ||
1da73967 AJ |
2065 | /* |
2066 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2067 | * currently replaced. | |
2068 | */ | |
2069 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2070 | { | |
2071 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2072 | ||
cb5583dd | 2073 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2074 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2075 | ASSERT(num_devices > 1); | |
2076 | num_devices--; | |
2077 | } | |
cb5583dd | 2078 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2079 | |
2080 | return num_devices; | |
2081 | } | |
2082 | ||
da353f6b DS |
2083 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
2084 | u64 devid) | |
f1fa7f26 AJ |
2085 | { |
2086 | struct btrfs_device *device; | |
1f78160c | 2087 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2088 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2089 | u64 num_devices; |
a061fc8d CM |
2090 | int ret = 0; |
2091 | ||
a061fc8d CM |
2092 | mutex_lock(&uuid_mutex); |
2093 | ||
1da73967 | 2094 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2095 | |
0b246afa | 2096 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2097 | if (ret) |
a061fc8d | 2098 | goto out; |
a061fc8d | 2099 | |
a27a94c2 NB |
2100 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2101 | ||
2102 | if (IS_ERR(device)) { | |
2103 | if (PTR_ERR(device) == -ENOENT && | |
2104 | strcmp(device_path, "missing") == 0) | |
2105 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2106 | else | |
2107 | ret = PTR_ERR(device); | |
53b381b3 | 2108 | goto out; |
a27a94c2 | 2109 | } |
dfe25020 | 2110 | |
eede2bf3 OS |
2111 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2112 | btrfs_warn_in_rcu(fs_info, | |
2113 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2114 | rcu_str_deref(device->name), device->devid); | |
2115 | ret = -ETXTBSY; | |
2116 | goto out; | |
2117 | } | |
2118 | ||
401e29c1 | 2119 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2120 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2121 | goto out; |
63a212ab SB |
2122 | } |
2123 | ||
ebbede42 AJ |
2124 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2125 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2126 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2127 | goto out; |
2b82032c YZ |
2128 | } |
2129 | ||
ebbede42 | 2130 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2131 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2132 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2133 | device->fs_devices->rw_devices--; |
34441361 | 2134 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2135 | } |
a061fc8d | 2136 | |
d7901554 | 2137 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2138 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 2139 | mutex_lock(&uuid_mutex); |
a061fc8d | 2140 | if (ret) |
9b3517e9 | 2141 | goto error_undo; |
a061fc8d | 2142 | |
63a212ab SB |
2143 | /* |
2144 | * TODO: the superblock still includes this device in its num_devices | |
2145 | * counter although write_all_supers() is not locked out. This | |
2146 | * could give a filesystem state which requires a degraded mount. | |
2147 | */ | |
0b246afa | 2148 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 2149 | if (ret) |
9b3517e9 | 2150 | goto error_undo; |
a061fc8d | 2151 | |
e12c9621 | 2152 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0b246afa | 2153 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
2154 | |
2155 | /* | |
2156 | * the device list mutex makes sure that we don't change | |
2157 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2158 | * the device supers. Whoever is writing all supers, should |
2159 | * lock the device list mutex before getting the number of | |
2160 | * devices in the super block (super_copy). Conversely, | |
2161 | * whoever updates the number of devices in the super block | |
2162 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2163 | */ |
1f78160c | 2164 | |
41a52a0f AJ |
2165 | /* |
2166 | * In normal cases the cur_devices == fs_devices. But in case | |
2167 | * of deleting a seed device, the cur_devices should point to | |
2168 | * its own fs_devices listed under the fs_devices->seed. | |
2169 | */ | |
1f78160c | 2170 | cur_devices = device->fs_devices; |
b5185197 | 2171 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2172 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2173 | |
41a52a0f AJ |
2174 | cur_devices->num_devices--; |
2175 | cur_devices->total_devices--; | |
b4993e64 AJ |
2176 | /* Update total_devices of the parent fs_devices if it's seed */ |
2177 | if (cur_devices != fs_devices) | |
2178 | fs_devices->total_devices--; | |
2b82032c | 2179 | |
e6e674bd | 2180 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2181 | cur_devices->missing_devices--; |
cd02dca5 | 2182 | |
d6507cf1 | 2183 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2184 | |
0bfaa9c5 | 2185 | if (device->bdev) { |
41a52a0f | 2186 | cur_devices->open_devices--; |
0bfaa9c5 | 2187 | /* remove sysfs entry */ |
b5185197 | 2188 | btrfs_sysfs_rm_device_link(fs_devices, device); |
0bfaa9c5 | 2189 | } |
99994cde | 2190 | |
0b246afa JM |
2191 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2192 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2193 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2194 | |
cea67ab9 JM |
2195 | /* |
2196 | * at this point, the device is zero sized and detached from | |
2197 | * the devices list. All that's left is to zero out the old | |
2198 | * supers and free the device. | |
2199 | */ | |
ebbede42 | 2200 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
cea67ab9 JM |
2201 | btrfs_scratch_superblocks(device->bdev, device->name->str); |
2202 | ||
2203 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2204 | synchronize_rcu(); |
2205 | btrfs_free_device(device); | |
cea67ab9 | 2206 | |
1f78160c | 2207 | if (cur_devices->open_devices == 0) { |
e4404d6e | 2208 | while (fs_devices) { |
8321cf25 RS |
2209 | if (fs_devices->seed == cur_devices) { |
2210 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 2211 | break; |
8321cf25 | 2212 | } |
e4404d6e | 2213 | fs_devices = fs_devices->seed; |
2b82032c | 2214 | } |
1f78160c | 2215 | cur_devices->seed = NULL; |
0226e0eb | 2216 | close_fs_devices(cur_devices); |
1f78160c | 2217 | free_fs_devices(cur_devices); |
2b82032c YZ |
2218 | } |
2219 | ||
a061fc8d CM |
2220 | out: |
2221 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2222 | return ret; |
24fc572f | 2223 | |
9b3517e9 | 2224 | error_undo: |
ebbede42 | 2225 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2226 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2227 | list_add(&device->dev_alloc_list, |
b5185197 | 2228 | &fs_devices->alloc_list); |
c3929c36 | 2229 | device->fs_devices->rw_devices++; |
34441361 | 2230 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2231 | } |
24fc572f | 2232 | goto out; |
a061fc8d CM |
2233 | } |
2234 | ||
68a9db5f | 2235 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2236 | { |
d51908ce AJ |
2237 | struct btrfs_fs_devices *fs_devices; |
2238 | ||
68a9db5f | 2239 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2240 | |
25e8e911 AJ |
2241 | /* |
2242 | * in case of fs with no seed, srcdev->fs_devices will point | |
2243 | * to fs_devices of fs_info. However when the dev being replaced is | |
2244 | * a seed dev it will point to the seed's local fs_devices. In short | |
2245 | * srcdev will have its correct fs_devices in both the cases. | |
2246 | */ | |
2247 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2248 | |
e93c89c1 | 2249 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2250 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2251 | fs_devices->num_devices--; |
e6e674bd | 2252 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2253 | fs_devices->missing_devices--; |
e93c89c1 | 2254 | |
ebbede42 | 2255 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2256 | fs_devices->rw_devices--; |
1357272f | 2257 | |
82372bc8 | 2258 | if (srcdev->bdev) |
d51908ce | 2259 | fs_devices->open_devices--; |
084b6e7c QW |
2260 | } |
2261 | ||
2262 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2263 | struct btrfs_device *srcdev) | |
2264 | { | |
2265 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2266 | |
ebbede42 | 2267 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) { |
48b3b9d4 AJ |
2268 | /* zero out the old super if it is writable */ |
2269 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2270 | } | |
14238819 AJ |
2271 | |
2272 | btrfs_close_bdev(srcdev); | |
8e75fd89 NB |
2273 | synchronize_rcu(); |
2274 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2275 | |
94d5f0c2 AJ |
2276 | /* if this is no devs we rather delete the fs_devices */ |
2277 | if (!fs_devices->num_devices) { | |
2278 | struct btrfs_fs_devices *tmp_fs_devices; | |
2279 | ||
6dd38f81 AJ |
2280 | /* |
2281 | * On a mounted FS, num_devices can't be zero unless it's a | |
2282 | * seed. In case of a seed device being replaced, the replace | |
2283 | * target added to the sprout FS, so there will be no more | |
2284 | * device left under the seed FS. | |
2285 | */ | |
2286 | ASSERT(fs_devices->seeding); | |
2287 | ||
94d5f0c2 AJ |
2288 | tmp_fs_devices = fs_info->fs_devices; |
2289 | while (tmp_fs_devices) { | |
2290 | if (tmp_fs_devices->seed == fs_devices) { | |
2291 | tmp_fs_devices->seed = fs_devices->seed; | |
2292 | break; | |
2293 | } | |
2294 | tmp_fs_devices = tmp_fs_devices->seed; | |
2295 | } | |
2296 | fs_devices->seed = NULL; | |
0226e0eb | 2297 | close_fs_devices(fs_devices); |
8bef8401 | 2298 | free_fs_devices(fs_devices); |
94d5f0c2 | 2299 | } |
e93c89c1 SB |
2300 | } |
2301 | ||
4f5ad7bd | 2302 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2303 | { |
4f5ad7bd | 2304 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2305 | |
e93c89c1 | 2306 | WARN_ON(!tgtdev); |
d9a071f0 | 2307 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2308 | |
d9a071f0 | 2309 | btrfs_sysfs_rm_device_link(fs_devices, tgtdev); |
d2ff1b20 | 2310 | |
779bf3fe | 2311 | if (tgtdev->bdev) |
d9a071f0 | 2312 | fs_devices->open_devices--; |
779bf3fe | 2313 | |
d9a071f0 | 2314 | fs_devices->num_devices--; |
e93c89c1 | 2315 | |
d6507cf1 | 2316 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2317 | |
e93c89c1 | 2318 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2319 | |
d9a071f0 | 2320 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2321 | |
2322 | /* | |
2323 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2324 | * may lead to a call to btrfs_show_devname() which will try | |
2325 | * to hold device_list_mutex. And here this device | |
2326 | * is already out of device list, so we don't have to hold | |
2327 | * the device_list_mutex lock. | |
2328 | */ | |
2329 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2330 | |
2331 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2332 | synchronize_rcu(); |
2333 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2334 | } |
2335 | ||
b444ad46 NB |
2336 | static struct btrfs_device *btrfs_find_device_by_path( |
2337 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2338 | { |
2339 | int ret = 0; | |
2340 | struct btrfs_super_block *disk_super; | |
2341 | u64 devid; | |
2342 | u8 *dev_uuid; | |
2343 | struct block_device *bdev; | |
2344 | struct buffer_head *bh; | |
b444ad46 | 2345 | struct btrfs_device *device; |
7ba15b7d | 2346 | |
7ba15b7d | 2347 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
0b246afa | 2348 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d | 2349 | if (ret) |
b444ad46 | 2350 | return ERR_PTR(ret); |
7ba15b7d SB |
2351 | disk_super = (struct btrfs_super_block *)bh->b_data; |
2352 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2353 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2354 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2355 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2356 | disk_super->metadata_uuid, true); |
7239ff4b | 2357 | else |
e4319cd9 | 2358 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2359 | disk_super->fsid, true); |
7239ff4b | 2360 | |
7ba15b7d | 2361 | brelse(bh); |
b444ad46 NB |
2362 | if (!device) |
2363 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2364 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2365 | return device; |
7ba15b7d SB |
2366 | } |
2367 | ||
5c5c0df0 DS |
2368 | /* |
2369 | * Lookup a device given by device id, or the path if the id is 0. | |
2370 | */ | |
a27a94c2 | 2371 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2372 | struct btrfs_fs_info *fs_info, u64 devid, |
2373 | const char *device_path) | |
24e0474b | 2374 | { |
a27a94c2 | 2375 | struct btrfs_device *device; |
24e0474b | 2376 | |
5c5c0df0 | 2377 | if (devid) { |
e4319cd9 | 2378 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
09ba3bc9 | 2379 | NULL, true); |
a27a94c2 NB |
2380 | if (!device) |
2381 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2382 | return device; |
2383 | } | |
2384 | ||
2385 | if (!device_path || !device_path[0]) | |
2386 | return ERR_PTR(-EINVAL); | |
2387 | ||
2388 | if (strcmp(device_path, "missing") == 0) { | |
2389 | /* Find first missing device */ | |
2390 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2391 | dev_list) { | |
2392 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2393 | &device->dev_state) && !device->bdev) | |
2394 | return device; | |
d95a830c | 2395 | } |
6e927ceb | 2396 | return ERR_PTR(-ENOENT); |
24e0474b | 2397 | } |
6e927ceb AJ |
2398 | |
2399 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2400 | } |
2401 | ||
2b82032c YZ |
2402 | /* |
2403 | * does all the dirty work required for changing file system's UUID. | |
2404 | */ | |
2ff7e61e | 2405 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2406 | { |
0b246afa | 2407 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2408 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2409 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2410 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2411 | struct btrfs_device *device; |
2412 | u64 super_flags; | |
2413 | ||
a32bf9a3 | 2414 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2415 | if (!fs_devices->seeding) |
2b82032c YZ |
2416 | return -EINVAL; |
2417 | ||
7239ff4b | 2418 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2419 | if (IS_ERR(seed_devices)) |
2420 | return PTR_ERR(seed_devices); | |
2b82032c | 2421 | |
e4404d6e YZ |
2422 | old_devices = clone_fs_devices(fs_devices); |
2423 | if (IS_ERR(old_devices)) { | |
2424 | kfree(seed_devices); | |
2425 | return PTR_ERR(old_devices); | |
2b82032c | 2426 | } |
e4404d6e | 2427 | |
c4babc5e | 2428 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2429 | |
e4404d6e YZ |
2430 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2431 | seed_devices->opened = 1; | |
2432 | INIT_LIST_HEAD(&seed_devices->devices); | |
2433 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2434 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2435 | |
321a4bf7 | 2436 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2437 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2438 | synchronize_rcu); | |
2196d6e8 MX |
2439 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2440 | device->fs_devices = seed_devices; | |
c9513edb | 2441 | |
34441361 | 2442 | mutex_lock(&fs_info->chunk_mutex); |
e4404d6e | 2443 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
34441361 | 2444 | mutex_unlock(&fs_info->chunk_mutex); |
e4404d6e | 2445 | |
2b82032c YZ |
2446 | fs_devices->seeding = 0; |
2447 | fs_devices->num_devices = 0; | |
2448 | fs_devices->open_devices = 0; | |
69611ac8 | 2449 | fs_devices->missing_devices = 0; |
69611ac8 | 2450 | fs_devices->rotating = 0; |
e4404d6e | 2451 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2452 | |
2453 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2454 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2455 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2456 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2457 | |
2b82032c YZ |
2458 | super_flags = btrfs_super_flags(disk_super) & |
2459 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2460 | btrfs_set_super_flags(disk_super, super_flags); | |
2461 | ||
2462 | return 0; | |
2463 | } | |
2464 | ||
2465 | /* | |
01327610 | 2466 | * Store the expected generation for seed devices in device items. |
2b82032c | 2467 | */ |
5c466629 | 2468 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2469 | { |
5c466629 | 2470 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2471 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2472 | struct btrfs_path *path; |
2473 | struct extent_buffer *leaf; | |
2474 | struct btrfs_dev_item *dev_item; | |
2475 | struct btrfs_device *device; | |
2476 | struct btrfs_key key; | |
44880fdc | 2477 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2478 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2479 | u64 devid; | |
2480 | int ret; | |
2481 | ||
2482 | path = btrfs_alloc_path(); | |
2483 | if (!path) | |
2484 | return -ENOMEM; | |
2485 | ||
2b82032c YZ |
2486 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2487 | key.offset = 0; | |
2488 | key.type = BTRFS_DEV_ITEM_KEY; | |
2489 | ||
2490 | while (1) { | |
2491 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2492 | if (ret < 0) | |
2493 | goto error; | |
2494 | ||
2495 | leaf = path->nodes[0]; | |
2496 | next_slot: | |
2497 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2498 | ret = btrfs_next_leaf(root, path); | |
2499 | if (ret > 0) | |
2500 | break; | |
2501 | if (ret < 0) | |
2502 | goto error; | |
2503 | leaf = path->nodes[0]; | |
2504 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2505 | btrfs_release_path(path); |
2b82032c YZ |
2506 | continue; |
2507 | } | |
2508 | ||
2509 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2510 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2511 | key.type != BTRFS_DEV_ITEM_KEY) | |
2512 | break; | |
2513 | ||
2514 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2515 | struct btrfs_dev_item); | |
2516 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2517 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2518 | BTRFS_UUID_SIZE); |
1473b24e | 2519 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2520 | BTRFS_FSID_SIZE); |
e4319cd9 | 2521 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 2522 | fs_uuid, true); |
79787eaa | 2523 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2524 | |
2525 | if (device->fs_devices->seeding) { | |
2526 | btrfs_set_device_generation(leaf, dev_item, | |
2527 | device->generation); | |
2528 | btrfs_mark_buffer_dirty(leaf); | |
2529 | } | |
2530 | ||
2531 | path->slots[0]++; | |
2532 | goto next_slot; | |
2533 | } | |
2534 | ret = 0; | |
2535 | error: | |
2536 | btrfs_free_path(path); | |
2537 | return ret; | |
2538 | } | |
2539 | ||
da353f6b | 2540 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2541 | { |
5112febb | 2542 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2543 | struct request_queue *q; |
788f20eb CM |
2544 | struct btrfs_trans_handle *trans; |
2545 | struct btrfs_device *device; | |
2546 | struct block_device *bdev; | |
0b246afa | 2547 | struct super_block *sb = fs_info->sb; |
606686ee | 2548 | struct rcu_string *name; |
5da54bc1 | 2549 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2550 | u64 orig_super_total_bytes; |
2551 | u64 orig_super_num_devices; | |
2b82032c | 2552 | int seeding_dev = 0; |
788f20eb | 2553 | int ret = 0; |
7132a262 | 2554 | bool unlocked = false; |
788f20eb | 2555 | |
5da54bc1 | 2556 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2557 | return -EROFS; |
788f20eb | 2558 | |
a5d16333 | 2559 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2560 | fs_info->bdev_holder); |
7f59203a JB |
2561 | if (IS_ERR(bdev)) |
2562 | return PTR_ERR(bdev); | |
a2135011 | 2563 | |
5da54bc1 | 2564 | if (fs_devices->seeding) { |
2b82032c YZ |
2565 | seeding_dev = 1; |
2566 | down_write(&sb->s_umount); | |
2567 | mutex_lock(&uuid_mutex); | |
2568 | } | |
2569 | ||
8c8bee1d | 2570 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2571 | |
5da54bc1 | 2572 | mutex_lock(&fs_devices->device_list_mutex); |
694c51fb | 2573 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
788f20eb CM |
2574 | if (device->bdev == bdev) { |
2575 | ret = -EEXIST; | |
d25628bd | 2576 | mutex_unlock( |
5da54bc1 | 2577 | &fs_devices->device_list_mutex); |
2b82032c | 2578 | goto error; |
788f20eb CM |
2579 | } |
2580 | } | |
5da54bc1 | 2581 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2582 | |
0b246afa | 2583 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2584 | if (IS_ERR(device)) { |
788f20eb | 2585 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2586 | ret = PTR_ERR(device); |
2b82032c | 2587 | goto error; |
788f20eb CM |
2588 | } |
2589 | ||
78f2c9e6 | 2590 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2591 | if (!name) { |
2b82032c | 2592 | ret = -ENOMEM; |
5c4cf6c9 | 2593 | goto error_free_device; |
788f20eb | 2594 | } |
606686ee | 2595 | rcu_assign_pointer(device->name, name); |
2b82032c | 2596 | |
a22285a6 | 2597 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2598 | if (IS_ERR(trans)) { |
98d5dc13 | 2599 | ret = PTR_ERR(trans); |
5c4cf6c9 | 2600 | goto error_free_device; |
98d5dc13 TI |
2601 | } |
2602 | ||
d5e2003c | 2603 | q = bdev_get_queue(bdev); |
ebbede42 | 2604 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2605 | device->generation = trans->transid; |
0b246afa JM |
2606 | device->io_width = fs_info->sectorsize; |
2607 | device->io_align = fs_info->sectorsize; | |
2608 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2609 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2610 | fs_info->sectorsize); | |
2cc3c559 | 2611 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2612 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2613 | device->fs_info = fs_info; |
788f20eb | 2614 | device->bdev = bdev; |
e12c9621 | 2615 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2616 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2617 | device->mode = FMODE_EXCL; |
27087f37 | 2618 | device->dev_stats_valid = 1; |
9f6d2510 | 2619 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2620 | |
2b82032c | 2621 | if (seeding_dev) { |
1751e8a6 | 2622 | sb->s_flags &= ~SB_RDONLY; |
2ff7e61e | 2623 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2624 | if (ret) { |
2625 | btrfs_abort_transaction(trans, ret); | |
2626 | goto error_trans; | |
2627 | } | |
2b82032c | 2628 | } |
788f20eb | 2629 | |
5da54bc1 | 2630 | device->fs_devices = fs_devices; |
e5e9a520 | 2631 | |
5da54bc1 | 2632 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2633 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2634 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2635 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2636 | fs_devices->num_devices++; | |
2637 | fs_devices->open_devices++; | |
2638 | fs_devices->rw_devices++; | |
2639 | fs_devices->total_devices++; | |
2640 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2641 | |
a5ed45f8 | 2642 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2643 | |
e884f4f0 | 2644 | if (!blk_queue_nonrot(q)) |
5da54bc1 | 2645 | fs_devices->rotating = 1; |
c289811c | 2646 | |
39379faa | 2647 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2648 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2649 | round_down(orig_super_total_bytes + device->total_bytes, |
2650 | fs_info->sectorsize)); | |
788f20eb | 2651 | |
39379faa NA |
2652 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2653 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2654 | orig_super_num_devices + 1); | |
0d39376a AJ |
2655 | |
2656 | /* add sysfs device entry */ | |
5da54bc1 | 2657 | btrfs_sysfs_add_device_link(fs_devices, device); |
0d39376a | 2658 | |
2196d6e8 MX |
2659 | /* |
2660 | * we've got more storage, clear any full flags on the space | |
2661 | * infos | |
2662 | */ | |
0b246afa | 2663 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2664 | |
34441361 | 2665 | mutex_unlock(&fs_info->chunk_mutex); |
5da54bc1 | 2666 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2667 | |
2b82032c | 2668 | if (seeding_dev) { |
34441361 | 2669 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2670 | ret = init_first_rw_device(trans); |
34441361 | 2671 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2672 | if (ret) { |
66642832 | 2673 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2674 | goto error_sysfs; |
005d6427 | 2675 | } |
2196d6e8 MX |
2676 | } |
2677 | ||
8e87e856 | 2678 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2679 | if (ret) { |
66642832 | 2680 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2681 | goto error_sysfs; |
2196d6e8 MX |
2682 | } |
2683 | ||
2684 | if (seeding_dev) { | |
2685 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2686 | ||
5c466629 | 2687 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2688 | if (ret) { |
66642832 | 2689 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2690 | goto error_sysfs; |
005d6427 | 2691 | } |
b2373f25 AJ |
2692 | |
2693 | /* Sprouting would change fsid of the mounted root, | |
2694 | * so rename the fsid on the sysfs | |
2695 | */ | |
2696 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
de37aa51 | 2697 | fs_info->fs_devices->fsid); |
5da54bc1 | 2698 | if (kobject_rename(&fs_devices->fsid_kobj, fsid_buf)) |
0b246afa JM |
2699 | btrfs_warn(fs_info, |
2700 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2701 | } |
2702 | ||
3a45bb20 | 2703 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2704 | |
2b82032c YZ |
2705 | if (seeding_dev) { |
2706 | mutex_unlock(&uuid_mutex); | |
2707 | up_write(&sb->s_umount); | |
7132a262 | 2708 | unlocked = true; |
788f20eb | 2709 | |
79787eaa JM |
2710 | if (ret) /* transaction commit */ |
2711 | return ret; | |
2712 | ||
2ff7e61e | 2713 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2714 | if (ret < 0) |
0b246afa | 2715 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2716 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2717 | trans = btrfs_attach_transaction(root); |
2718 | if (IS_ERR(trans)) { | |
2719 | if (PTR_ERR(trans) == -ENOENT) | |
2720 | return 0; | |
7132a262 AJ |
2721 | ret = PTR_ERR(trans); |
2722 | trans = NULL; | |
2723 | goto error_sysfs; | |
671415b7 | 2724 | } |
3a45bb20 | 2725 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2726 | } |
c9e9f97b | 2727 | |
5a1972bd QW |
2728 | /* Update ctime/mtime for libblkid */ |
2729 | update_dev_time(device_path); | |
2b82032c | 2730 | return ret; |
79787eaa | 2731 | |
d31c32f6 | 2732 | error_sysfs: |
5da54bc1 | 2733 | btrfs_sysfs_rm_device_link(fs_devices, device); |
39379faa NA |
2734 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2735 | mutex_lock(&fs_info->chunk_mutex); | |
2736 | list_del_rcu(&device->dev_list); | |
2737 | list_del(&device->dev_alloc_list); | |
2738 | fs_info->fs_devices->num_devices--; | |
2739 | fs_info->fs_devices->open_devices--; | |
2740 | fs_info->fs_devices->rw_devices--; | |
2741 | fs_info->fs_devices->total_devices--; | |
2742 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2743 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2744 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2745 | orig_super_total_bytes); | |
2746 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2747 | orig_super_num_devices); | |
2748 | mutex_unlock(&fs_info->chunk_mutex); | |
2749 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2750 | error_trans: |
0af2c4bf | 2751 | if (seeding_dev) |
1751e8a6 | 2752 | sb->s_flags |= SB_RDONLY; |
7132a262 AJ |
2753 | if (trans) |
2754 | btrfs_end_transaction(trans); | |
5c4cf6c9 | 2755 | error_free_device: |
a425f9d4 | 2756 | btrfs_free_device(device); |
2b82032c | 2757 | error: |
e525fd89 | 2758 | blkdev_put(bdev, FMODE_EXCL); |
7132a262 | 2759 | if (seeding_dev && !unlocked) { |
2b82032c YZ |
2760 | mutex_unlock(&uuid_mutex); |
2761 | up_write(&sb->s_umount); | |
2762 | } | |
c9e9f97b | 2763 | return ret; |
788f20eb CM |
2764 | } |
2765 | ||
d397712b CM |
2766 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2767 | struct btrfs_device *device) | |
0b86a832 CM |
2768 | { |
2769 | int ret; | |
2770 | struct btrfs_path *path; | |
0b246afa | 2771 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2772 | struct btrfs_dev_item *dev_item; |
2773 | struct extent_buffer *leaf; | |
2774 | struct btrfs_key key; | |
2775 | ||
0b86a832 CM |
2776 | path = btrfs_alloc_path(); |
2777 | if (!path) | |
2778 | return -ENOMEM; | |
2779 | ||
2780 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2781 | key.type = BTRFS_DEV_ITEM_KEY; | |
2782 | key.offset = device->devid; | |
2783 | ||
2784 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2785 | if (ret < 0) | |
2786 | goto out; | |
2787 | ||
2788 | if (ret > 0) { | |
2789 | ret = -ENOENT; | |
2790 | goto out; | |
2791 | } | |
2792 | ||
2793 | leaf = path->nodes[0]; | |
2794 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2795 | ||
2796 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2797 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2798 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2799 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2800 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2801 | btrfs_set_device_total_bytes(leaf, dev_item, |
2802 | btrfs_device_get_disk_total_bytes(device)); | |
2803 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2804 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2805 | btrfs_mark_buffer_dirty(leaf); |
2806 | ||
2807 | out: | |
2808 | btrfs_free_path(path); | |
2809 | return ret; | |
2810 | } | |
2811 | ||
2196d6e8 | 2812 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2813 | struct btrfs_device *device, u64 new_size) |
2814 | { | |
0b246afa JM |
2815 | struct btrfs_fs_info *fs_info = device->fs_info; |
2816 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2817 | u64 old_total; |
2818 | u64 diff; | |
8f18cf13 | 2819 | |
ebbede42 | 2820 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2821 | return -EACCES; |
2196d6e8 | 2822 | |
7dfb8be1 NB |
2823 | new_size = round_down(new_size, fs_info->sectorsize); |
2824 | ||
34441361 | 2825 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2826 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2827 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2828 | |
63a212ab | 2829 | if (new_size <= device->total_bytes || |
401e29c1 | 2830 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2831 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2832 | return -EINVAL; |
2196d6e8 | 2833 | } |
2b82032c | 2834 | |
7dfb8be1 NB |
2835 | btrfs_set_super_total_bytes(super_copy, |
2836 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2837 | device->fs_devices->total_rw_bytes += diff; |
2838 | ||
7cc8e58d MX |
2839 | btrfs_device_set_total_bytes(device, new_size); |
2840 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2841 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2842 | if (list_empty(&device->post_commit_list)) |
2843 | list_add_tail(&device->post_commit_list, | |
2844 | &trans->transaction->dev_update_list); | |
34441361 | 2845 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2846 | |
8f18cf13 CM |
2847 | return btrfs_update_device(trans, device); |
2848 | } | |
2849 | ||
f4208794 | 2850 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2851 | { |
f4208794 | 2852 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2853 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2854 | int ret; |
2855 | struct btrfs_path *path; | |
2856 | struct btrfs_key key; | |
2857 | ||
8f18cf13 CM |
2858 | path = btrfs_alloc_path(); |
2859 | if (!path) | |
2860 | return -ENOMEM; | |
2861 | ||
408fbf19 | 2862 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2863 | key.offset = chunk_offset; |
2864 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2865 | ||
2866 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2867 | if (ret < 0) |
2868 | goto out; | |
2869 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2870 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2871 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2872 | ret = -ENOENT; |
2873 | goto out; | |
2874 | } | |
8f18cf13 CM |
2875 | |
2876 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2877 | if (ret < 0) |
0b246afa JM |
2878 | btrfs_handle_fs_error(fs_info, ret, |
2879 | "Failed to delete chunk item."); | |
79787eaa | 2880 | out: |
8f18cf13 | 2881 | btrfs_free_path(path); |
65a246c5 | 2882 | return ret; |
8f18cf13 CM |
2883 | } |
2884 | ||
408fbf19 | 2885 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2886 | { |
0b246afa | 2887 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2888 | struct btrfs_disk_key *disk_key; |
2889 | struct btrfs_chunk *chunk; | |
2890 | u8 *ptr; | |
2891 | int ret = 0; | |
2892 | u32 num_stripes; | |
2893 | u32 array_size; | |
2894 | u32 len = 0; | |
2895 | u32 cur; | |
2896 | struct btrfs_key key; | |
2897 | ||
34441361 | 2898 | mutex_lock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2899 | array_size = btrfs_super_sys_array_size(super_copy); |
2900 | ||
2901 | ptr = super_copy->sys_chunk_array; | |
2902 | cur = 0; | |
2903 | ||
2904 | while (cur < array_size) { | |
2905 | disk_key = (struct btrfs_disk_key *)ptr; | |
2906 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2907 | ||
2908 | len = sizeof(*disk_key); | |
2909 | ||
2910 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2911 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2912 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2913 | len += btrfs_chunk_item_size(num_stripes); | |
2914 | } else { | |
2915 | ret = -EIO; | |
2916 | break; | |
2917 | } | |
408fbf19 | 2918 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2919 | key.offset == chunk_offset) { |
2920 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2921 | array_size -= len; | |
2922 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2923 | } else { | |
2924 | ptr += len; | |
2925 | cur += len; | |
2926 | } | |
2927 | } | |
34441361 | 2928 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 CM |
2929 | return ret; |
2930 | } | |
2931 | ||
60ca842e OS |
2932 | /* |
2933 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2934 | * @logical: Logical block offset in bytes. | |
2935 | * @length: Length of extent in bytes. | |
2936 | * | |
2937 | * Return: Chunk mapping or ERR_PTR. | |
2938 | */ | |
2939 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2940 | u64 logical, u64 length) | |
592d92ee LB |
2941 | { |
2942 | struct extent_map_tree *em_tree; | |
2943 | struct extent_map *em; | |
2944 | ||
2945 | em_tree = &fs_info->mapping_tree.map_tree; | |
2946 | read_lock(&em_tree->lock); | |
2947 | em = lookup_extent_mapping(em_tree, logical, length); | |
2948 | read_unlock(&em_tree->lock); | |
2949 | ||
2950 | if (!em) { | |
2951 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2952 | logical, length); | |
2953 | return ERR_PTR(-EINVAL); | |
2954 | } | |
2955 | ||
2956 | if (em->start > logical || em->start + em->len < logical) { | |
2957 | btrfs_crit(fs_info, | |
2958 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2959 | logical, length, em->start, em->start + em->len); | |
2960 | free_extent_map(em); | |
2961 | return ERR_PTR(-EINVAL); | |
2962 | } | |
2963 | ||
2964 | /* callers are responsible for dropping em's ref. */ | |
2965 | return em; | |
2966 | } | |
2967 | ||
97aff912 | 2968 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2969 | { |
97aff912 | 2970 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2971 | struct extent_map *em; |
2972 | struct map_lookup *map; | |
2196d6e8 | 2973 | u64 dev_extent_len = 0; |
47ab2a6c | 2974 | int i, ret = 0; |
0b246afa | 2975 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2976 | |
60ca842e | 2977 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2978 | if (IS_ERR(em)) { |
47ab2a6c JB |
2979 | /* |
2980 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2981 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2982 | * do anything we still error out. |
2983 | */ | |
2984 | ASSERT(0); | |
592d92ee | 2985 | return PTR_ERR(em); |
47ab2a6c | 2986 | } |
95617d69 | 2987 | map = em->map_lookup; |
34441361 | 2988 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 2989 | check_system_chunk(trans, map->type); |
34441361 | 2990 | mutex_unlock(&fs_info->chunk_mutex); |
8f18cf13 | 2991 | |
57ba4cb8 FM |
2992 | /* |
2993 | * Take the device list mutex to prevent races with the final phase of | |
2994 | * a device replace operation that replaces the device object associated | |
2995 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2996 | */ | |
2997 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2998 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2999 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3000 | ret = btrfs_free_dev_extent(trans, device, |
3001 | map->stripes[i].physical, | |
3002 | &dev_extent_len); | |
47ab2a6c | 3003 | if (ret) { |
57ba4cb8 | 3004 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3005 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3006 | goto out; |
3007 | } | |
a061fc8d | 3008 | |
2196d6e8 | 3009 | if (device->bytes_used > 0) { |
34441361 | 3010 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3011 | btrfs_device_set_bytes_used(device, |
3012 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3013 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3014 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3015 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3016 | } |
a061fc8d | 3017 | |
64bc6c2a NB |
3018 | ret = btrfs_update_device(trans, device); |
3019 | if (ret) { | |
3020 | mutex_unlock(&fs_devices->device_list_mutex); | |
3021 | btrfs_abort_transaction(trans, ret); | |
3022 | goto out; | |
dfe25020 | 3023 | } |
8f18cf13 | 3024 | } |
57ba4cb8 FM |
3025 | mutex_unlock(&fs_devices->device_list_mutex); |
3026 | ||
f4208794 | 3027 | ret = btrfs_free_chunk(trans, chunk_offset); |
47ab2a6c | 3028 | if (ret) { |
66642832 | 3029 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3030 | goto out; |
3031 | } | |
8f18cf13 | 3032 | |
6bccf3ab | 3033 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3034 | |
8f18cf13 | 3035 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3036 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3037 | if (ret) { |
66642832 | 3038 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3039 | goto out; |
3040 | } | |
8f18cf13 CM |
3041 | } |
3042 | ||
5a98ec01 | 3043 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3044 | if (ret) { |
66642832 | 3045 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3046 | goto out; |
3047 | } | |
2b82032c | 3048 | |
47ab2a6c | 3049 | out: |
2b82032c YZ |
3050 | /* once for us */ |
3051 | free_extent_map(em); | |
47ab2a6c JB |
3052 | return ret; |
3053 | } | |
2b82032c | 3054 | |
5b4aacef | 3055 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3056 | { |
5b4aacef | 3057 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3058 | struct btrfs_trans_handle *trans; |
47ab2a6c | 3059 | int ret; |
2b82032c | 3060 | |
67c5e7d4 FM |
3061 | /* |
3062 | * Prevent races with automatic removal of unused block groups. | |
3063 | * After we relocate and before we remove the chunk with offset | |
3064 | * chunk_offset, automatic removal of the block group can kick in, | |
3065 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3066 | * | |
3067 | * Make sure to acquire this mutex before doing a tree search (dev | |
3068 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3069 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3070 | * we release the path used to search the chunk/dev tree and before | |
3071 | * the current task acquires this mutex and calls us. | |
3072 | */ | |
a32bf9a3 | 3073 | lockdep_assert_held(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 3074 | |
0b246afa | 3075 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
3076 | if (ret) |
3077 | return -ENOSPC; | |
3078 | ||
3079 | /* step one, relocate all the extents inside this chunk */ | |
2ff7e61e | 3080 | btrfs_scrub_pause(fs_info); |
0b246afa | 3081 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3082 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3083 | if (ret) |
3084 | return ret; | |
3085 | ||
75cb379d JM |
3086 | /* |
3087 | * We add the kobjects here (and after forcing data chunk creation) | |
3088 | * since relocation is the only place we'll create chunks of a new | |
3089 | * type at runtime. The only place where we'll remove the last | |
3090 | * chunk of a type is the call immediately below this one. Even | |
3091 | * so, we're protected against races with the cleaner thread since | |
3092 | * we're covered by the delete_unused_bgs_mutex. | |
3093 | */ | |
3094 | btrfs_add_raid_kobjects(fs_info); | |
3095 | ||
19c4d2f9 CM |
3096 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3097 | chunk_offset); | |
3098 | if (IS_ERR(trans)) { | |
3099 | ret = PTR_ERR(trans); | |
3100 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3101 | return ret; | |
3102 | } | |
3103 | ||
47ab2a6c | 3104 | /* |
19c4d2f9 CM |
3105 | * step two, delete the device extents and the |
3106 | * chunk tree entries | |
47ab2a6c | 3107 | */ |
97aff912 | 3108 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3109 | btrfs_end_transaction(trans); |
19c4d2f9 | 3110 | return ret; |
2b82032c YZ |
3111 | } |
3112 | ||
2ff7e61e | 3113 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3114 | { |
0b246afa | 3115 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3116 | struct btrfs_path *path; |
3117 | struct extent_buffer *leaf; | |
3118 | struct btrfs_chunk *chunk; | |
3119 | struct btrfs_key key; | |
3120 | struct btrfs_key found_key; | |
2b82032c | 3121 | u64 chunk_type; |
ba1bf481 JB |
3122 | bool retried = false; |
3123 | int failed = 0; | |
2b82032c YZ |
3124 | int ret; |
3125 | ||
3126 | path = btrfs_alloc_path(); | |
3127 | if (!path) | |
3128 | return -ENOMEM; | |
3129 | ||
ba1bf481 | 3130 | again: |
2b82032c YZ |
3131 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3132 | key.offset = (u64)-1; | |
3133 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3134 | ||
3135 | while (1) { | |
0b246afa | 3136 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3137 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3138 | if (ret < 0) { |
0b246afa | 3139 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 3140 | goto error; |
67c5e7d4 | 3141 | } |
79787eaa | 3142 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3143 | |
3144 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3145 | key.type); | |
67c5e7d4 | 3146 | if (ret) |
0b246afa | 3147 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
3148 | if (ret < 0) |
3149 | goto error; | |
3150 | if (ret > 0) | |
3151 | break; | |
1a40e23b | 3152 | |
2b82032c YZ |
3153 | leaf = path->nodes[0]; |
3154 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3155 | |
2b82032c YZ |
3156 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3157 | struct btrfs_chunk); | |
3158 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3159 | btrfs_release_path(path); |
8f18cf13 | 3160 | |
2b82032c | 3161 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3162 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3163 | if (ret == -ENOSPC) |
3164 | failed++; | |
14586651 HS |
3165 | else |
3166 | BUG_ON(ret); | |
2b82032c | 3167 | } |
0b246afa | 3168 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3169 | |
2b82032c YZ |
3170 | if (found_key.offset == 0) |
3171 | break; | |
3172 | key.offset = found_key.offset - 1; | |
3173 | } | |
3174 | ret = 0; | |
ba1bf481 JB |
3175 | if (failed && !retried) { |
3176 | failed = 0; | |
3177 | retried = true; | |
3178 | goto again; | |
fae7f21c | 3179 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3180 | ret = -ENOSPC; |
3181 | } | |
2b82032c YZ |
3182 | error: |
3183 | btrfs_free_path(path); | |
3184 | return ret; | |
8f18cf13 CM |
3185 | } |
3186 | ||
a6f93c71 LB |
3187 | /* |
3188 | * return 1 : allocate a data chunk successfully, | |
3189 | * return <0: errors during allocating a data chunk, | |
3190 | * return 0 : no need to allocate a data chunk. | |
3191 | */ | |
3192 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3193 | u64 chunk_offset) | |
3194 | { | |
3195 | struct btrfs_block_group_cache *cache; | |
3196 | u64 bytes_used; | |
3197 | u64 chunk_type; | |
3198 | ||
3199 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3200 | ASSERT(cache); | |
3201 | chunk_type = cache->flags; | |
3202 | btrfs_put_block_group(cache); | |
3203 | ||
3204 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) { | |
3205 | spin_lock(&fs_info->data_sinfo->lock); | |
3206 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3207 | spin_unlock(&fs_info->data_sinfo->lock); | |
3208 | ||
3209 | if (!bytes_used) { | |
3210 | struct btrfs_trans_handle *trans; | |
3211 | int ret; | |
3212 | ||
3213 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3214 | if (IS_ERR(trans)) | |
3215 | return PTR_ERR(trans); | |
3216 | ||
43a7e99d | 3217 | ret = btrfs_force_chunk_alloc(trans, |
a6f93c71 LB |
3218 | BTRFS_BLOCK_GROUP_DATA); |
3219 | btrfs_end_transaction(trans); | |
3220 | if (ret < 0) | |
3221 | return ret; | |
3222 | ||
75cb379d JM |
3223 | btrfs_add_raid_kobjects(fs_info); |
3224 | ||
a6f93c71 LB |
3225 | return 1; |
3226 | } | |
3227 | } | |
3228 | return 0; | |
3229 | } | |
3230 | ||
6bccf3ab | 3231 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3232 | struct btrfs_balance_control *bctl) |
3233 | { | |
6bccf3ab | 3234 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3235 | struct btrfs_trans_handle *trans; |
3236 | struct btrfs_balance_item *item; | |
3237 | struct btrfs_disk_balance_args disk_bargs; | |
3238 | struct btrfs_path *path; | |
3239 | struct extent_buffer *leaf; | |
3240 | struct btrfs_key key; | |
3241 | int ret, err; | |
3242 | ||
3243 | path = btrfs_alloc_path(); | |
3244 | if (!path) | |
3245 | return -ENOMEM; | |
3246 | ||
3247 | trans = btrfs_start_transaction(root, 0); | |
3248 | if (IS_ERR(trans)) { | |
3249 | btrfs_free_path(path); | |
3250 | return PTR_ERR(trans); | |
3251 | } | |
3252 | ||
3253 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3254 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3255 | key.offset = 0; |
3256 | ||
3257 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3258 | sizeof(*item)); | |
3259 | if (ret) | |
3260 | goto out; | |
3261 | ||
3262 | leaf = path->nodes[0]; | |
3263 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3264 | ||
b159fa28 | 3265 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3266 | |
3267 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3268 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3269 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3270 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3271 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3272 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3273 | ||
3274 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3275 | ||
3276 | btrfs_mark_buffer_dirty(leaf); | |
3277 | out: | |
3278 | btrfs_free_path(path); | |
3a45bb20 | 3279 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3280 | if (err && !ret) |
3281 | ret = err; | |
3282 | return ret; | |
3283 | } | |
3284 | ||
6bccf3ab | 3285 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3286 | { |
6bccf3ab | 3287 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3288 | struct btrfs_trans_handle *trans; |
3289 | struct btrfs_path *path; | |
3290 | struct btrfs_key key; | |
3291 | int ret, err; | |
3292 | ||
3293 | path = btrfs_alloc_path(); | |
3294 | if (!path) | |
3295 | return -ENOMEM; | |
3296 | ||
3297 | trans = btrfs_start_transaction(root, 0); | |
3298 | if (IS_ERR(trans)) { | |
3299 | btrfs_free_path(path); | |
3300 | return PTR_ERR(trans); | |
3301 | } | |
3302 | ||
3303 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3304 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3305 | key.offset = 0; |
3306 | ||
3307 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3308 | if (ret < 0) | |
3309 | goto out; | |
3310 | if (ret > 0) { | |
3311 | ret = -ENOENT; | |
3312 | goto out; | |
3313 | } | |
3314 | ||
3315 | ret = btrfs_del_item(trans, root, path); | |
3316 | out: | |
3317 | btrfs_free_path(path); | |
3a45bb20 | 3318 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3319 | if (err && !ret) |
3320 | ret = err; | |
3321 | return ret; | |
3322 | } | |
3323 | ||
59641015 ID |
3324 | /* |
3325 | * This is a heuristic used to reduce the number of chunks balanced on | |
3326 | * resume after balance was interrupted. | |
3327 | */ | |
3328 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3329 | { | |
3330 | /* | |
3331 | * Turn on soft mode for chunk types that were being converted. | |
3332 | */ | |
3333 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3334 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3335 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3336 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3337 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3338 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3339 | ||
3340 | /* | |
3341 | * Turn on usage filter if is not already used. The idea is | |
3342 | * that chunks that we have already balanced should be | |
3343 | * reasonably full. Don't do it for chunks that are being | |
3344 | * converted - that will keep us from relocating unconverted | |
3345 | * (albeit full) chunks. | |
3346 | */ | |
3347 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3348 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3349 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3350 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3351 | bctl->data.usage = 90; | |
3352 | } | |
3353 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3354 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3355 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3356 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3357 | bctl->sys.usage = 90; | |
3358 | } | |
3359 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3360 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3361 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3362 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3363 | bctl->meta.usage = 90; | |
3364 | } | |
3365 | } | |
3366 | ||
149196a2 DS |
3367 | /* |
3368 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3369 | */ | |
3370 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3371 | { |
3372 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3373 | int ret; |
c9e9f97b ID |
3374 | |
3375 | BUG_ON(!fs_info->balance_ctl); | |
3376 | ||
3377 | spin_lock(&fs_info->balance_lock); | |
3378 | fs_info->balance_ctl = NULL; | |
3379 | spin_unlock(&fs_info->balance_lock); | |
3380 | ||
3381 | kfree(bctl); | |
149196a2 DS |
3382 | ret = del_balance_item(fs_info); |
3383 | if (ret) | |
3384 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3385 | } |
3386 | ||
ed25e9b2 ID |
3387 | /* |
3388 | * Balance filters. Return 1 if chunk should be filtered out | |
3389 | * (should not be balanced). | |
3390 | */ | |
899c81ea | 3391 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3392 | struct btrfs_balance_args *bargs) |
3393 | { | |
899c81ea ID |
3394 | chunk_type = chunk_to_extended(chunk_type) & |
3395 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3396 | |
899c81ea | 3397 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3398 | return 0; |
3399 | ||
3400 | return 1; | |
3401 | } | |
3402 | ||
dba72cb3 | 3403 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3404 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3405 | { |
3406 | struct btrfs_block_group_cache *cache; | |
3407 | u64 chunk_used; | |
3408 | u64 user_thresh_min; | |
3409 | u64 user_thresh_max; | |
3410 | int ret = 1; | |
3411 | ||
3412 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3413 | chunk_used = btrfs_block_group_used(&cache->item); | |
3414 | ||
3415 | if (bargs->usage_min == 0) | |
3416 | user_thresh_min = 0; | |
3417 | else | |
3418 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3419 | bargs->usage_min); | |
3420 | ||
3421 | if (bargs->usage_max == 0) | |
3422 | user_thresh_max = 1; | |
3423 | else if (bargs->usage_max > 100) | |
3424 | user_thresh_max = cache->key.offset; | |
3425 | else | |
3426 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3427 | bargs->usage_max); | |
3428 | ||
3429 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3430 | ret = 0; | |
3431 | ||
3432 | btrfs_put_block_group(cache); | |
3433 | return ret; | |
3434 | } | |
3435 | ||
dba72cb3 | 3436 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3437 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3438 | { |
3439 | struct btrfs_block_group_cache *cache; | |
3440 | u64 chunk_used, user_thresh; | |
3441 | int ret = 1; | |
3442 | ||
3443 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3444 | chunk_used = btrfs_block_group_used(&cache->item); | |
3445 | ||
bc309467 | 3446 | if (bargs->usage_min == 0) |
3e39cea6 | 3447 | user_thresh = 1; |
a105bb88 ID |
3448 | else if (bargs->usage > 100) |
3449 | user_thresh = cache->key.offset; | |
3450 | else | |
3451 | user_thresh = div_factor_fine(cache->key.offset, | |
3452 | bargs->usage); | |
3453 | ||
5ce5b3c0 ID |
3454 | if (chunk_used < user_thresh) |
3455 | ret = 0; | |
3456 | ||
3457 | btrfs_put_block_group(cache); | |
3458 | return ret; | |
3459 | } | |
3460 | ||
409d404b ID |
3461 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3462 | struct btrfs_chunk *chunk, | |
3463 | struct btrfs_balance_args *bargs) | |
3464 | { | |
3465 | struct btrfs_stripe *stripe; | |
3466 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3467 | int i; | |
3468 | ||
3469 | for (i = 0; i < num_stripes; i++) { | |
3470 | stripe = btrfs_stripe_nr(chunk, i); | |
3471 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3472 | return 0; | |
3473 | } | |
3474 | ||
3475 | return 1; | |
3476 | } | |
3477 | ||
94e60d5a ID |
3478 | /* [pstart, pend) */ |
3479 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3480 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3481 | struct btrfs_balance_args *bargs) |
3482 | { | |
3483 | struct btrfs_stripe *stripe; | |
3484 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3485 | u64 stripe_offset; | |
3486 | u64 stripe_length; | |
3487 | int factor; | |
3488 | int i; | |
3489 | ||
3490 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3491 | return 0; | |
3492 | ||
3493 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3494 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3495 | factor = num_stripes / 2; | |
3496 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3497 | factor = num_stripes - 1; | |
3498 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3499 | factor = num_stripes - 2; | |
3500 | } else { | |
3501 | factor = num_stripes; | |
3502 | } | |
94e60d5a ID |
3503 | |
3504 | for (i = 0; i < num_stripes; i++) { | |
3505 | stripe = btrfs_stripe_nr(chunk, i); | |
3506 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3507 | continue; | |
3508 | ||
3509 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3510 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3511 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3512 | |
3513 | if (stripe_offset < bargs->pend && | |
3514 | stripe_offset + stripe_length > bargs->pstart) | |
3515 | return 0; | |
3516 | } | |
3517 | ||
3518 | return 1; | |
3519 | } | |
3520 | ||
ea67176a ID |
3521 | /* [vstart, vend) */ |
3522 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3523 | struct btrfs_chunk *chunk, | |
3524 | u64 chunk_offset, | |
3525 | struct btrfs_balance_args *bargs) | |
3526 | { | |
3527 | if (chunk_offset < bargs->vend && | |
3528 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3529 | /* at least part of the chunk is inside this vrange */ | |
3530 | return 0; | |
3531 | ||
3532 | return 1; | |
3533 | } | |
3534 | ||
dee32d0a GAP |
3535 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3536 | struct btrfs_chunk *chunk, | |
3537 | struct btrfs_balance_args *bargs) | |
3538 | { | |
3539 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3540 | ||
3541 | if (bargs->stripes_min <= num_stripes | |
3542 | && num_stripes <= bargs->stripes_max) | |
3543 | return 0; | |
3544 | ||
3545 | return 1; | |
3546 | } | |
3547 | ||
899c81ea | 3548 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3549 | struct btrfs_balance_args *bargs) |
3550 | { | |
3551 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3552 | return 0; | |
3553 | ||
899c81ea ID |
3554 | chunk_type = chunk_to_extended(chunk_type) & |
3555 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3556 | |
899c81ea | 3557 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3558 | return 1; |
3559 | ||
3560 | return 0; | |
3561 | } | |
3562 | ||
6ec0896c | 3563 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3564 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3565 | { | |
6ec0896c | 3566 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3567 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3568 | struct btrfs_balance_args *bargs = NULL; |
3569 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3570 | ||
3571 | /* type filter */ | |
3572 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3573 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3574 | return 0; | |
3575 | } | |
3576 | ||
3577 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3578 | bargs = &bctl->data; | |
3579 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3580 | bargs = &bctl->sys; | |
3581 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3582 | bargs = &bctl->meta; | |
3583 | ||
ed25e9b2 ID |
3584 | /* profiles filter */ |
3585 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3586 | chunk_profiles_filter(chunk_type, bargs)) { | |
3587 | return 0; | |
5ce5b3c0 ID |
3588 | } |
3589 | ||
3590 | /* usage filter */ | |
3591 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3592 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3593 | return 0; |
bc309467 | 3594 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3595 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3596 | return 0; |
409d404b ID |
3597 | } |
3598 | ||
3599 | /* devid filter */ | |
3600 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3601 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3602 | return 0; | |
94e60d5a ID |
3603 | } |
3604 | ||
3605 | /* drange filter, makes sense only with devid filter */ | |
3606 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3607 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3608 | return 0; |
ea67176a ID |
3609 | } |
3610 | ||
3611 | /* vrange filter */ | |
3612 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3613 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3614 | return 0; | |
ed25e9b2 ID |
3615 | } |
3616 | ||
dee32d0a GAP |
3617 | /* stripes filter */ |
3618 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3619 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3620 | return 0; | |
3621 | } | |
3622 | ||
cfa4c961 ID |
3623 | /* soft profile changing mode */ |
3624 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3625 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3626 | return 0; | |
3627 | } | |
3628 | ||
7d824b6f DS |
3629 | /* |
3630 | * limited by count, must be the last filter | |
3631 | */ | |
3632 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3633 | if (bargs->limit == 0) | |
3634 | return 0; | |
3635 | else | |
3636 | bargs->limit--; | |
12907fc7 DS |
3637 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3638 | /* | |
3639 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3640 | * determined here because we do not have the global information |
12907fc7 DS |
3641 | * about the count of all chunks that satisfy the filters. |
3642 | */ | |
3643 | if (bargs->limit_max == 0) | |
3644 | return 0; | |
3645 | else | |
3646 | bargs->limit_max--; | |
7d824b6f DS |
3647 | } |
3648 | ||
f43ffb60 ID |
3649 | return 1; |
3650 | } | |
3651 | ||
c9e9f97b | 3652 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3653 | { |
19a39dce | 3654 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3655 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3656 | u64 chunk_type; |
f43ffb60 | 3657 | struct btrfs_chunk *chunk; |
5a488b9d | 3658 | struct btrfs_path *path = NULL; |
ec44a35c | 3659 | struct btrfs_key key; |
ec44a35c | 3660 | struct btrfs_key found_key; |
f43ffb60 ID |
3661 | struct extent_buffer *leaf; |
3662 | int slot; | |
c9e9f97b ID |
3663 | int ret; |
3664 | int enospc_errors = 0; | |
19a39dce | 3665 | bool counting = true; |
12907fc7 | 3666 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3667 | u64 limit_data = bctl->data.limit; |
3668 | u64 limit_meta = bctl->meta.limit; | |
3669 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3670 | u32 count_data = 0; |
3671 | u32 count_meta = 0; | |
3672 | u32 count_sys = 0; | |
2c9fe835 | 3673 | int chunk_reserved = 0; |
ec44a35c | 3674 | |
ec44a35c | 3675 | path = btrfs_alloc_path(); |
17e9f796 MF |
3676 | if (!path) { |
3677 | ret = -ENOMEM; | |
3678 | goto error; | |
3679 | } | |
19a39dce ID |
3680 | |
3681 | /* zero out stat counters */ | |
3682 | spin_lock(&fs_info->balance_lock); | |
3683 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3684 | spin_unlock(&fs_info->balance_lock); | |
3685 | again: | |
7d824b6f | 3686 | if (!counting) { |
12907fc7 DS |
3687 | /* |
3688 | * The single value limit and min/max limits use the same bytes | |
3689 | * in the | |
3690 | */ | |
7d824b6f DS |
3691 | bctl->data.limit = limit_data; |
3692 | bctl->meta.limit = limit_meta; | |
3693 | bctl->sys.limit = limit_sys; | |
3694 | } | |
ec44a35c CM |
3695 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3696 | key.offset = (u64)-1; | |
3697 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3698 | ||
d397712b | 3699 | while (1) { |
19a39dce | 3700 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3701 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3702 | ret = -ECANCELED; |
3703 | goto error; | |
3704 | } | |
3705 | ||
67c5e7d4 | 3706 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3707 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3708 | if (ret < 0) { |
3709 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3710 | goto error; |
67c5e7d4 | 3711 | } |
ec44a35c CM |
3712 | |
3713 | /* | |
3714 | * this shouldn't happen, it means the last relocate | |
3715 | * failed | |
3716 | */ | |
3717 | if (ret == 0) | |
c9e9f97b | 3718 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3719 | |
3720 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3721 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3722 | if (ret) { |
67c5e7d4 | 3723 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3724 | ret = 0; |
ec44a35c | 3725 | break; |
c9e9f97b | 3726 | } |
7d9eb12c | 3727 | |
f43ffb60 ID |
3728 | leaf = path->nodes[0]; |
3729 | slot = path->slots[0]; | |
3730 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3731 | |
67c5e7d4 FM |
3732 | if (found_key.objectid != key.objectid) { |
3733 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3734 | break; |
67c5e7d4 | 3735 | } |
7d9eb12c | 3736 | |
f43ffb60 | 3737 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3738 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3739 | |
19a39dce ID |
3740 | if (!counting) { |
3741 | spin_lock(&fs_info->balance_lock); | |
3742 | bctl->stat.considered++; | |
3743 | spin_unlock(&fs_info->balance_lock); | |
3744 | } | |
3745 | ||
6ec0896c | 3746 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3747 | |
b3b4aa74 | 3748 | btrfs_release_path(path); |
67c5e7d4 FM |
3749 | if (!ret) { |
3750 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3751 | goto loop; |
67c5e7d4 | 3752 | } |
f43ffb60 | 3753 | |
19a39dce | 3754 | if (counting) { |
67c5e7d4 | 3755 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3756 | spin_lock(&fs_info->balance_lock); |
3757 | bctl->stat.expected++; | |
3758 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3759 | |
3760 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3761 | count_data++; | |
3762 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3763 | count_sys++; | |
3764 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3765 | count_meta++; | |
3766 | ||
3767 | goto loop; | |
3768 | } | |
3769 | ||
3770 | /* | |
3771 | * Apply limit_min filter, no need to check if the LIMITS | |
3772 | * filter is used, limit_min is 0 by default | |
3773 | */ | |
3774 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3775 | count_data < bctl->data.limit_min) | |
3776 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3777 | count_meta < bctl->meta.limit_min) | |
3778 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3779 | count_sys < bctl->sys.limit_min)) { | |
3780 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3781 | goto loop; |
3782 | } | |
3783 | ||
a6f93c71 LB |
3784 | if (!chunk_reserved) { |
3785 | /* | |
3786 | * We may be relocating the only data chunk we have, | |
3787 | * which could potentially end up with losing data's | |
3788 | * raid profile, so lets allocate an empty one in | |
3789 | * advance. | |
3790 | */ | |
3791 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3792 | found_key.offset); | |
2c9fe835 ZL |
3793 | if (ret < 0) { |
3794 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3795 | goto error; | |
a6f93c71 LB |
3796 | } else if (ret == 1) { |
3797 | chunk_reserved = 1; | |
2c9fe835 | 3798 | } |
2c9fe835 ZL |
3799 | } |
3800 | ||
5b4aacef | 3801 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3802 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce | 3803 | if (ret == -ENOSPC) { |
c9e9f97b | 3804 | enospc_errors++; |
eede2bf3 OS |
3805 | } else if (ret == -ETXTBSY) { |
3806 | btrfs_info(fs_info, | |
3807 | "skipping relocation of block group %llu due to active swapfile", | |
3808 | found_key.offset); | |
3809 | ret = 0; | |
3810 | } else if (ret) { | |
3811 | goto error; | |
19a39dce ID |
3812 | } else { |
3813 | spin_lock(&fs_info->balance_lock); | |
3814 | bctl->stat.completed++; | |
3815 | spin_unlock(&fs_info->balance_lock); | |
3816 | } | |
f43ffb60 | 3817 | loop: |
795a3321 ID |
3818 | if (found_key.offset == 0) |
3819 | break; | |
ba1bf481 | 3820 | key.offset = found_key.offset - 1; |
ec44a35c | 3821 | } |
c9e9f97b | 3822 | |
19a39dce ID |
3823 | if (counting) { |
3824 | btrfs_release_path(path); | |
3825 | counting = false; | |
3826 | goto again; | |
3827 | } | |
ec44a35c CM |
3828 | error: |
3829 | btrfs_free_path(path); | |
c9e9f97b | 3830 | if (enospc_errors) { |
efe120a0 | 3831 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3832 | enospc_errors); |
c9e9f97b ID |
3833 | if (!ret) |
3834 | ret = -ENOSPC; | |
3835 | } | |
3836 | ||
ec44a35c CM |
3837 | return ret; |
3838 | } | |
3839 | ||
0c460c0d ID |
3840 | /** |
3841 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3842 | * @flags: profile to validate | |
3843 | * @extended: if true @flags is treated as an extended profile | |
3844 | */ | |
3845 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3846 | { | |
3847 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3848 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3849 | ||
3850 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3851 | ||
3852 | /* 1) check that all other bits are zeroed */ | |
3853 | if (flags & ~mask) | |
3854 | return 0; | |
3855 | ||
3856 | /* 2) see if profile is reduced */ | |
3857 | if (flags == 0) | |
3858 | return !extended; /* "0" is valid for usual profiles */ | |
3859 | ||
3860 | /* true if exactly one bit set */ | |
818255fe | 3861 | return is_power_of_2(flags); |
0c460c0d ID |
3862 | } |
3863 | ||
837d5b6e ID |
3864 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3865 | { | |
a7e99c69 ID |
3866 | /* cancel requested || normal exit path */ |
3867 | return atomic_read(&fs_info->balance_cancel_req) || | |
3868 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3869 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3870 | } |
3871 | ||
bdcd3c97 AM |
3872 | /* Non-zero return value signifies invalidity */ |
3873 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3874 | u64 allowed) | |
3875 | { | |
3876 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3877 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3878 | (bctl_arg->target & ~allowed))); | |
3879 | } | |
3880 | ||
56fc37d9 AJ |
3881 | /* |
3882 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3883 | * @size_buf including the terminating null. The output may be trimmed if it | |
3884 | * does not fit into the provided buffer. | |
3885 | */ | |
3886 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3887 | u32 size_buf) | |
3888 | { | |
3889 | int ret; | |
3890 | u32 size_bp = size_buf; | |
3891 | char *bp = buf; | |
3892 | u64 flags = bargs->flags; | |
3893 | char tmp_buf[128] = {'\0'}; | |
3894 | ||
3895 | if (!flags) | |
3896 | return; | |
3897 | ||
3898 | #define CHECK_APPEND_NOARG(a) \ | |
3899 | do { \ | |
3900 | ret = snprintf(bp, size_bp, (a)); \ | |
3901 | if (ret < 0 || ret >= size_bp) \ | |
3902 | goto out_overflow; \ | |
3903 | size_bp -= ret; \ | |
3904 | bp += ret; \ | |
3905 | } while (0) | |
3906 | ||
3907 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3908 | do { \ | |
3909 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
3910 | if (ret < 0 || ret >= size_bp) \ | |
3911 | goto out_overflow; \ | |
3912 | size_bp -= ret; \ | |
3913 | bp += ret; \ | |
3914 | } while (0) | |
3915 | ||
3916 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
3917 | do { \ | |
3918 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
3919 | if (ret < 0 || ret >= size_bp) \ | |
3920 | goto out_overflow; \ | |
3921 | size_bp -= ret; \ | |
3922 | bp += ret; \ | |
3923 | } while (0) | |
3924 | ||
3925 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3926 | int index = btrfs_bg_flags_to_raid_index(bargs->target); | |
3927 | ||
3928 | CHECK_APPEND_1ARG("convert=%s,", get_raid_name(index)); | |
3929 | } | |
3930 | ||
3931 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
3932 | CHECK_APPEND_NOARG("soft,"); | |
3933 | ||
3934 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
3935 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
3936 | sizeof(tmp_buf)); | |
3937 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
3938 | } | |
3939 | ||
3940 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
3941 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
3942 | ||
3943 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
3944 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
3945 | bargs->usage_min, bargs->usage_max); | |
3946 | ||
3947 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
3948 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
3949 | ||
3950 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
3951 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
3952 | bargs->pstart, bargs->pend); | |
3953 | ||
3954 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
3955 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
3956 | bargs->vstart, bargs->vend); | |
3957 | ||
3958 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
3959 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
3960 | ||
3961 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
3962 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
3963 | bargs->limit_min, bargs->limit_max); | |
3964 | ||
3965 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
3966 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
3967 | bargs->stripes_min, bargs->stripes_max); | |
3968 | ||
3969 | #undef CHECK_APPEND_2ARG | |
3970 | #undef CHECK_APPEND_1ARG | |
3971 | #undef CHECK_APPEND_NOARG | |
3972 | ||
3973 | out_overflow: | |
3974 | ||
3975 | if (size_bp < size_buf) | |
3976 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
3977 | else | |
3978 | buf[0] = '\0'; | |
3979 | } | |
3980 | ||
3981 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
3982 | { | |
3983 | u32 size_buf = 1024; | |
3984 | char tmp_buf[192] = {'\0'}; | |
3985 | char *buf; | |
3986 | char *bp; | |
3987 | u32 size_bp = size_buf; | |
3988 | int ret; | |
3989 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3990 | ||
3991 | buf = kzalloc(size_buf, GFP_KERNEL); | |
3992 | if (!buf) | |
3993 | return; | |
3994 | ||
3995 | bp = buf; | |
3996 | ||
3997 | #define CHECK_APPEND_1ARG(a, v1) \ | |
3998 | do { \ | |
3999 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4000 | if (ret < 0 || ret >= size_bp) \ | |
4001 | goto out_overflow; \ | |
4002 | size_bp -= ret; \ | |
4003 | bp += ret; \ | |
4004 | } while (0) | |
4005 | ||
4006 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4007 | CHECK_APPEND_1ARG("%s", "-f "); | |
4008 | ||
4009 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4010 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4011 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4012 | } | |
4013 | ||
4014 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4015 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4016 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4017 | } | |
4018 | ||
4019 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4020 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4021 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4022 | } | |
4023 | ||
4024 | #undef CHECK_APPEND_1ARG | |
4025 | ||
4026 | out_overflow: | |
4027 | ||
4028 | if (size_bp < size_buf) | |
4029 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4030 | btrfs_info(fs_info, "balance: %s %s", | |
4031 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4032 | "resume" : "start", buf); | |
4033 | ||
4034 | kfree(buf); | |
4035 | } | |
4036 | ||
c9e9f97b | 4037 | /* |
dccdb07b | 4038 | * Should be called with balance mutexe held |
c9e9f97b | 4039 | */ |
6fcf6e2b DS |
4040 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4041 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4042 | struct btrfs_ioctl_balance_args *bargs) |
4043 | { | |
14506127 | 4044 | u64 meta_target, data_target; |
f43ffb60 | 4045 | u64 allowed; |
e4837f8f | 4046 | int mixed = 0; |
c9e9f97b | 4047 | int ret; |
8dabb742 | 4048 | u64 num_devices; |
de98ced9 | 4049 | unsigned seq; |
5a8067c0 | 4050 | bool reducing_integrity; |
c9e9f97b | 4051 | |
837d5b6e | 4052 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
4053 | atomic_read(&fs_info->balance_pause_req) || |
4054 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
4055 | ret = -EINVAL; |
4056 | goto out; | |
4057 | } | |
4058 | ||
e4837f8f ID |
4059 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4060 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4061 | mixed = 1; | |
4062 | ||
f43ffb60 ID |
4063 | /* |
4064 | * In case of mixed groups both data and meta should be picked, | |
4065 | * and identical options should be given for both of them. | |
4066 | */ | |
e4837f8f ID |
4067 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4068 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4069 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4070 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4071 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4072 | btrfs_err(fs_info, |
6dac13f8 | 4073 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4074 | ret = -EINVAL; |
4075 | goto out; | |
4076 | } | |
4077 | } | |
4078 | ||
1da73967 AJ |
4079 | num_devices = btrfs_num_devices(fs_info); |
4080 | ||
88be159c AH |
4081 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
4082 | if (num_devices > 1) | |
e4d8ec0f | 4083 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
4084 | if (num_devices > 2) |
4085 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
4086 | if (num_devices > 3) | |
4087 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
4088 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 4089 | if (validate_convert_profile(&bctl->data, allowed)) { |
6dac13f8 AJ |
4090 | int index = btrfs_bg_flags_to_raid_index(bctl->data.target); |
4091 | ||
5d163e0e | 4092 | btrfs_err(fs_info, |
6dac13f8 AJ |
4093 | "balance: invalid convert data profile %s", |
4094 | get_raid_name(index)); | |
e4d8ec0f ID |
4095 | ret = -EINVAL; |
4096 | goto out; | |
4097 | } | |
bdcd3c97 | 4098 | if (validate_convert_profile(&bctl->meta, allowed)) { |
6dac13f8 AJ |
4099 | int index = btrfs_bg_flags_to_raid_index(bctl->meta.target); |
4100 | ||
efe120a0 | 4101 | btrfs_err(fs_info, |
6dac13f8 AJ |
4102 | "balance: invalid convert metadata profile %s", |
4103 | get_raid_name(index)); | |
e4d8ec0f ID |
4104 | ret = -EINVAL; |
4105 | goto out; | |
4106 | } | |
bdcd3c97 | 4107 | if (validate_convert_profile(&bctl->sys, allowed)) { |
6dac13f8 AJ |
4108 | int index = btrfs_bg_flags_to_raid_index(bctl->sys.target); |
4109 | ||
efe120a0 | 4110 | btrfs_err(fs_info, |
6dac13f8 AJ |
4111 | "balance: invalid convert system profile %s", |
4112 | get_raid_name(index)); | |
e4d8ec0f ID |
4113 | ret = -EINVAL; |
4114 | goto out; | |
4115 | } | |
4116 | ||
e4d8ec0f ID |
4117 | /* allow to reduce meta or sys integrity only if force set */ |
4118 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
4119 | BTRFS_BLOCK_GROUP_RAID10 | |
4120 | BTRFS_BLOCK_GROUP_RAID5 | | |
4121 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
4122 | do { |
4123 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4124 | ||
4125 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4126 | (fs_info->avail_system_alloc_bits & allowed) && | |
4127 | !(bctl->sys.target & allowed)) || | |
4128 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4129 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 FM |
4130 | !(bctl->meta.target & allowed))) |
4131 | reducing_integrity = true; | |
4132 | else | |
4133 | reducing_integrity = false; | |
4134 | ||
4135 | /* if we're not converting, the target field is uninitialized */ | |
4136 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4137 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4138 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4139 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4140 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4141 | |
5a8067c0 FM |
4142 | if (reducing_integrity) { |
4143 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
4144 | btrfs_info(fs_info, | |
4145 | "balance: force reducing metadata integrity"); | |
4146 | } else { | |
4147 | btrfs_err(fs_info, | |
4148 | "balance: reduces metadata integrity, use --force if you want this"); | |
4149 | ret = -EINVAL; | |
4150 | goto out; | |
4151 | } | |
4152 | } | |
4153 | ||
14506127 AB |
4154 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4155 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
6dac13f8 AJ |
4156 | int meta_index = btrfs_bg_flags_to_raid_index(meta_target); |
4157 | int data_index = btrfs_bg_flags_to_raid_index(data_target); | |
4158 | ||
ee592d07 | 4159 | btrfs_warn(fs_info, |
6dac13f8 AJ |
4160 | "balance: metadata profile %s has lower redundancy than data profile %s", |
4161 | get_raid_name(meta_index), get_raid_name(data_index)); | |
ee592d07 ST |
4162 | } |
4163 | ||
6bccf3ab | 4164 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4165 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4166 | goto out; |
4167 | ||
59641015 ID |
4168 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4169 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4170 | BUG_ON(fs_info->balance_ctl); |
4171 | spin_lock(&fs_info->balance_lock); | |
4172 | fs_info->balance_ctl = bctl; | |
4173 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4174 | } else { |
4175 | BUG_ON(ret != -EEXIST); | |
4176 | spin_lock(&fs_info->balance_lock); | |
4177 | update_balance_args(bctl); | |
4178 | spin_unlock(&fs_info->balance_lock); | |
4179 | } | |
c9e9f97b | 4180 | |
3009a62f DS |
4181 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4182 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4183 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4184 | mutex_unlock(&fs_info->balance_mutex); |
4185 | ||
4186 | ret = __btrfs_balance(fs_info); | |
4187 | ||
4188 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4189 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4190 | btrfs_info(fs_info, "balance: paused"); | |
4191 | else if (ret == -ECANCELED && atomic_read(&fs_info->balance_cancel_req)) | |
4192 | btrfs_info(fs_info, "balance: canceled"); | |
4193 | else | |
4194 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4195 | ||
3009a62f | 4196 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4197 | |
4198 | if (bargs) { | |
4199 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4200 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4201 | } |
4202 | ||
3a01aa7a ID |
4203 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4204 | balance_need_close(fs_info)) { | |
149196a2 | 4205 | reset_balance_state(fs_info); |
a17c95df | 4206 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
3a01aa7a ID |
4207 | } |
4208 | ||
837d5b6e | 4209 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4210 | |
4211 | return ret; | |
4212 | out: | |
59641015 | 4213 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4214 | reset_balance_state(fs_info); |
a17c95df | 4215 | else |
59641015 | 4216 | kfree(bctl); |
a17c95df DS |
4217 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
4218 | ||
59641015 ID |
4219 | return ret; |
4220 | } | |
4221 | ||
4222 | static int balance_kthread(void *data) | |
4223 | { | |
2b6ba629 | 4224 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4225 | int ret = 0; |
59641015 | 4226 | |
59641015 | 4227 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4228 | if (fs_info->balance_ctl) |
6fcf6e2b | 4229 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4230 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4231 | |
59641015 ID |
4232 | return ret; |
4233 | } | |
4234 | ||
2b6ba629 ID |
4235 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4236 | { | |
4237 | struct task_struct *tsk; | |
4238 | ||
1354e1a1 | 4239 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4240 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4241 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4242 | return 0; |
4243 | } | |
1354e1a1 | 4244 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4245 | |
3cdde224 | 4246 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4247 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4248 | return 0; |
4249 | } | |
4250 | ||
02ee654d AJ |
4251 | /* |
4252 | * A ro->rw remount sequence should continue with the paused balance | |
4253 | * regardless of who pauses it, system or the user as of now, so set | |
4254 | * the resume flag. | |
4255 | */ | |
4256 | spin_lock(&fs_info->balance_lock); | |
4257 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4258 | spin_unlock(&fs_info->balance_lock); | |
4259 | ||
2b6ba629 | 4260 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4261 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4262 | } |
4263 | ||
68310a5e | 4264 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4265 | { |
59641015 ID |
4266 | struct btrfs_balance_control *bctl; |
4267 | struct btrfs_balance_item *item; | |
4268 | struct btrfs_disk_balance_args disk_bargs; | |
4269 | struct btrfs_path *path; | |
4270 | struct extent_buffer *leaf; | |
4271 | struct btrfs_key key; | |
4272 | int ret; | |
4273 | ||
4274 | path = btrfs_alloc_path(); | |
4275 | if (!path) | |
4276 | return -ENOMEM; | |
4277 | ||
59641015 | 4278 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4279 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4280 | key.offset = 0; |
4281 | ||
68310a5e | 4282 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4283 | if (ret < 0) |
68310a5e | 4284 | goto out; |
59641015 ID |
4285 | if (ret > 0) { /* ret = -ENOENT; */ |
4286 | ret = 0; | |
68310a5e ID |
4287 | goto out; |
4288 | } | |
4289 | ||
4290 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4291 | if (!bctl) { | |
4292 | ret = -ENOMEM; | |
4293 | goto out; | |
59641015 ID |
4294 | } |
4295 | ||
4296 | leaf = path->nodes[0]; | |
4297 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4298 | ||
68310a5e ID |
4299 | bctl->flags = btrfs_balance_flags(leaf, item); |
4300 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4301 | |
4302 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4303 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4304 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4305 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4306 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4307 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4308 | ||
eee95e3f DS |
4309 | /* |
4310 | * This should never happen, as the paused balance state is recovered | |
4311 | * during mount without any chance of other exclusive ops to collide. | |
4312 | * | |
4313 | * This gives the exclusive op status to balance and keeps in paused | |
4314 | * state until user intervention (cancel or umount). If the ownership | |
4315 | * cannot be assigned, show a message but do not fail. The balance | |
4316 | * is in a paused state and must have fs_info::balance_ctl properly | |
4317 | * set up. | |
4318 | */ | |
4319 | if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) | |
4320 | btrfs_warn(fs_info, | |
6dac13f8 | 4321 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4322 | |
68310a5e | 4323 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4324 | BUG_ON(fs_info->balance_ctl); |
4325 | spin_lock(&fs_info->balance_lock); | |
4326 | fs_info->balance_ctl = bctl; | |
4327 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4328 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4329 | out: |
4330 | btrfs_free_path(path); | |
ec44a35c CM |
4331 | return ret; |
4332 | } | |
4333 | ||
837d5b6e ID |
4334 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4335 | { | |
4336 | int ret = 0; | |
4337 | ||
4338 | mutex_lock(&fs_info->balance_mutex); | |
4339 | if (!fs_info->balance_ctl) { | |
4340 | mutex_unlock(&fs_info->balance_mutex); | |
4341 | return -ENOTCONN; | |
4342 | } | |
4343 | ||
3009a62f | 4344 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4345 | atomic_inc(&fs_info->balance_pause_req); |
4346 | mutex_unlock(&fs_info->balance_mutex); | |
4347 | ||
4348 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4349 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4350 | |
4351 | mutex_lock(&fs_info->balance_mutex); | |
4352 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4353 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4354 | atomic_dec(&fs_info->balance_pause_req); |
4355 | } else { | |
4356 | ret = -ENOTCONN; | |
4357 | } | |
4358 | ||
4359 | mutex_unlock(&fs_info->balance_mutex); | |
4360 | return ret; | |
4361 | } | |
4362 | ||
a7e99c69 ID |
4363 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4364 | { | |
4365 | mutex_lock(&fs_info->balance_mutex); | |
4366 | if (!fs_info->balance_ctl) { | |
4367 | mutex_unlock(&fs_info->balance_mutex); | |
4368 | return -ENOTCONN; | |
4369 | } | |
4370 | ||
cf7d20f4 DS |
4371 | /* |
4372 | * A paused balance with the item stored on disk can be resumed at | |
4373 | * mount time if the mount is read-write. Otherwise it's still paused | |
4374 | * and we must not allow cancelling as it deletes the item. | |
4375 | */ | |
4376 | if (sb_rdonly(fs_info->sb)) { | |
4377 | mutex_unlock(&fs_info->balance_mutex); | |
4378 | return -EROFS; | |
4379 | } | |
4380 | ||
a7e99c69 ID |
4381 | atomic_inc(&fs_info->balance_cancel_req); |
4382 | /* | |
4383 | * if we are running just wait and return, balance item is | |
4384 | * deleted in btrfs_balance in this case | |
4385 | */ | |
3009a62f | 4386 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4387 | mutex_unlock(&fs_info->balance_mutex); |
4388 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4389 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4390 | mutex_lock(&fs_info->balance_mutex); |
4391 | } else { | |
a7e99c69 | 4392 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4393 | /* |
4394 | * Lock released to allow other waiters to continue, we'll | |
4395 | * reexamine the status again. | |
4396 | */ | |
a7e99c69 ID |
4397 | mutex_lock(&fs_info->balance_mutex); |
4398 | ||
a17c95df | 4399 | if (fs_info->balance_ctl) { |
149196a2 | 4400 | reset_balance_state(fs_info); |
a17c95df | 4401 | clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags); |
6dac13f8 | 4402 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4403 | } |
a7e99c69 ID |
4404 | } |
4405 | ||
3009a62f DS |
4406 | BUG_ON(fs_info->balance_ctl || |
4407 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4408 | atomic_dec(&fs_info->balance_cancel_req); |
4409 | mutex_unlock(&fs_info->balance_mutex); | |
4410 | return 0; | |
4411 | } | |
4412 | ||
803b2f54 SB |
4413 | static int btrfs_uuid_scan_kthread(void *data) |
4414 | { | |
4415 | struct btrfs_fs_info *fs_info = data; | |
4416 | struct btrfs_root *root = fs_info->tree_root; | |
4417 | struct btrfs_key key; | |
803b2f54 SB |
4418 | struct btrfs_path *path = NULL; |
4419 | int ret = 0; | |
4420 | struct extent_buffer *eb; | |
4421 | int slot; | |
4422 | struct btrfs_root_item root_item; | |
4423 | u32 item_size; | |
f45388f3 | 4424 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4425 | |
4426 | path = btrfs_alloc_path(); | |
4427 | if (!path) { | |
4428 | ret = -ENOMEM; | |
4429 | goto out; | |
4430 | } | |
4431 | ||
4432 | key.objectid = 0; | |
4433 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4434 | key.offset = 0; | |
4435 | ||
803b2f54 | 4436 | while (1) { |
7c829b72 AJ |
4437 | ret = btrfs_search_forward(root, &key, path, |
4438 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4439 | if (ret) { |
4440 | if (ret > 0) | |
4441 | ret = 0; | |
4442 | break; | |
4443 | } | |
4444 | ||
4445 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4446 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4447 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4448 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4449 | goto skip; | |
4450 | ||
4451 | eb = path->nodes[0]; | |
4452 | slot = path->slots[0]; | |
4453 | item_size = btrfs_item_size_nr(eb, slot); | |
4454 | if (item_size < sizeof(root_item)) | |
4455 | goto skip; | |
4456 | ||
803b2f54 SB |
4457 | read_extent_buffer(eb, &root_item, |
4458 | btrfs_item_ptr_offset(eb, slot), | |
4459 | (int)sizeof(root_item)); | |
4460 | if (btrfs_root_refs(&root_item) == 0) | |
4461 | goto skip; | |
f45388f3 FDBM |
4462 | |
4463 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4464 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4465 | if (trans) | |
4466 | goto update_tree; | |
4467 | ||
4468 | btrfs_release_path(path); | |
803b2f54 SB |
4469 | /* |
4470 | * 1 - subvol uuid item | |
4471 | * 1 - received_subvol uuid item | |
4472 | */ | |
4473 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4474 | if (IS_ERR(trans)) { | |
4475 | ret = PTR_ERR(trans); | |
4476 | break; | |
4477 | } | |
f45388f3 FDBM |
4478 | continue; |
4479 | } else { | |
4480 | goto skip; | |
4481 | } | |
4482 | update_tree: | |
4483 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
cdb345a8 | 4484 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4485 | BTRFS_UUID_KEY_SUBVOL, |
4486 | key.objectid); | |
4487 | if (ret < 0) { | |
efe120a0 | 4488 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4489 | ret); |
803b2f54 SB |
4490 | break; |
4491 | } | |
4492 | } | |
4493 | ||
4494 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4495 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4496 | root_item.received_uuid, |
4497 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4498 | key.objectid); | |
4499 | if (ret < 0) { | |
efe120a0 | 4500 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4501 | ret); |
803b2f54 SB |
4502 | break; |
4503 | } | |
4504 | } | |
4505 | ||
f45388f3 | 4506 | skip: |
803b2f54 | 4507 | if (trans) { |
3a45bb20 | 4508 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4509 | trans = NULL; |
803b2f54 SB |
4510 | if (ret) |
4511 | break; | |
4512 | } | |
4513 | ||
803b2f54 SB |
4514 | btrfs_release_path(path); |
4515 | if (key.offset < (u64)-1) { | |
4516 | key.offset++; | |
4517 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4518 | key.offset = 0; | |
4519 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4520 | } else if (key.objectid < (u64)-1) { | |
4521 | key.offset = 0; | |
4522 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4523 | key.objectid++; | |
4524 | } else { | |
4525 | break; | |
4526 | } | |
4527 | cond_resched(); | |
4528 | } | |
4529 | ||
4530 | out: | |
4531 | btrfs_free_path(path); | |
f45388f3 | 4532 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4533 | btrfs_end_transaction(trans); |
803b2f54 | 4534 | if (ret) |
efe120a0 | 4535 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4536 | else |
afcdd129 | 4537 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4538 | up(&fs_info->uuid_tree_rescan_sem); |
4539 | return 0; | |
4540 | } | |
4541 | ||
70f80175 SB |
4542 | /* |
4543 | * Callback for btrfs_uuid_tree_iterate(). | |
4544 | * returns: | |
4545 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4546 | * < 0 if an error occurred. |
70f80175 SB |
4547 | * > 0 if the check failed, which means the caller shall remove the entry. |
4548 | */ | |
4549 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4550 | u8 *uuid, u8 type, u64 subid) | |
4551 | { | |
4552 | struct btrfs_key key; | |
4553 | int ret = 0; | |
4554 | struct btrfs_root *subvol_root; | |
4555 | ||
4556 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4557 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4558 | goto out; | |
4559 | ||
4560 | key.objectid = subid; | |
4561 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4562 | key.offset = (u64)-1; | |
4563 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4564 | if (IS_ERR(subvol_root)) { | |
4565 | ret = PTR_ERR(subvol_root); | |
4566 | if (ret == -ENOENT) | |
4567 | ret = 1; | |
4568 | goto out; | |
4569 | } | |
4570 | ||
4571 | switch (type) { | |
4572 | case BTRFS_UUID_KEY_SUBVOL: | |
4573 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4574 | ret = 1; | |
4575 | break; | |
4576 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4577 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4578 | BTRFS_UUID_SIZE)) | |
4579 | ret = 1; | |
4580 | break; | |
4581 | } | |
4582 | ||
4583 | out: | |
4584 | return ret; | |
4585 | } | |
4586 | ||
4587 | static int btrfs_uuid_rescan_kthread(void *data) | |
4588 | { | |
4589 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4590 | int ret; | |
4591 | ||
4592 | /* | |
4593 | * 1st step is to iterate through the existing UUID tree and | |
4594 | * to delete all entries that contain outdated data. | |
4595 | * 2nd step is to add all missing entries to the UUID tree. | |
4596 | */ | |
4597 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4598 | if (ret < 0) { | |
efe120a0 | 4599 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4600 | up(&fs_info->uuid_tree_rescan_sem); |
4601 | return ret; | |
4602 | } | |
4603 | return btrfs_uuid_scan_kthread(data); | |
4604 | } | |
4605 | ||
f7a81ea4 SB |
4606 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4607 | { | |
4608 | struct btrfs_trans_handle *trans; | |
4609 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4610 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4611 | struct task_struct *task; |
4612 | int ret; | |
f7a81ea4 SB |
4613 | |
4614 | /* | |
4615 | * 1 - root node | |
4616 | * 1 - root item | |
4617 | */ | |
4618 | trans = btrfs_start_transaction(tree_root, 2); | |
4619 | if (IS_ERR(trans)) | |
4620 | return PTR_ERR(trans); | |
4621 | ||
9b7a2440 | 4622 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4623 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4624 | ret = PTR_ERR(uuid_root); |
66642832 | 4625 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4626 | btrfs_end_transaction(trans); |
6d13f549 | 4627 | return ret; |
f7a81ea4 SB |
4628 | } |
4629 | ||
4630 | fs_info->uuid_root = uuid_root; | |
4631 | ||
3a45bb20 | 4632 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4633 | if (ret) |
4634 | return ret; | |
4635 | ||
4636 | down(&fs_info->uuid_tree_rescan_sem); | |
4637 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4638 | if (IS_ERR(task)) { | |
70f80175 | 4639 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4640 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4641 | up(&fs_info->uuid_tree_rescan_sem); |
4642 | return PTR_ERR(task); | |
4643 | } | |
4644 | ||
4645 | return 0; | |
f7a81ea4 | 4646 | } |
803b2f54 | 4647 | |
70f80175 SB |
4648 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4649 | { | |
4650 | struct task_struct *task; | |
4651 | ||
4652 | down(&fs_info->uuid_tree_rescan_sem); | |
4653 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4654 | if (IS_ERR(task)) { | |
4655 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4656 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4657 | up(&fs_info->uuid_tree_rescan_sem); |
4658 | return PTR_ERR(task); | |
4659 | } | |
4660 | ||
4661 | return 0; | |
4662 | } | |
4663 | ||
8f18cf13 CM |
4664 | /* |
4665 | * shrinking a device means finding all of the device extents past | |
4666 | * the new size, and then following the back refs to the chunks. | |
4667 | * The chunk relocation code actually frees the device extent | |
4668 | */ | |
4669 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4670 | { | |
0b246afa JM |
4671 | struct btrfs_fs_info *fs_info = device->fs_info; |
4672 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4673 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4674 | struct btrfs_dev_extent *dev_extent = NULL; |
4675 | struct btrfs_path *path; | |
4676 | u64 length; | |
8f18cf13 CM |
4677 | u64 chunk_offset; |
4678 | int ret; | |
4679 | int slot; | |
ba1bf481 JB |
4680 | int failed = 0; |
4681 | bool retried = false; | |
8f18cf13 CM |
4682 | struct extent_buffer *l; |
4683 | struct btrfs_key key; | |
0b246afa | 4684 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4685 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4686 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4687 | u64 diff; |
61d0d0d2 | 4688 | u64 start; |
7dfb8be1 NB |
4689 | |
4690 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4691 | start = new_size; |
0e4324a4 | 4692 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4693 | |
401e29c1 | 4694 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4695 | return -EINVAL; |
4696 | ||
8f18cf13 CM |
4697 | path = btrfs_alloc_path(); |
4698 | if (!path) | |
4699 | return -ENOMEM; | |
4700 | ||
0338dff6 | 4701 | path->reada = READA_BACK; |
8f18cf13 | 4702 | |
61d0d0d2 NB |
4703 | trans = btrfs_start_transaction(root, 0); |
4704 | if (IS_ERR(trans)) { | |
4705 | btrfs_free_path(path); | |
4706 | return PTR_ERR(trans); | |
4707 | } | |
4708 | ||
34441361 | 4709 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4710 | |
7cc8e58d | 4711 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4712 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4713 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4714 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4715 | } |
61d0d0d2 NB |
4716 | |
4717 | /* | |
4718 | * Once the device's size has been set to the new size, ensure all | |
4719 | * in-memory chunks are synced to disk so that the loop below sees them | |
4720 | * and relocates them accordingly. | |
4721 | */ | |
1c11b63e | 4722 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4723 | mutex_unlock(&fs_info->chunk_mutex); |
4724 | ret = btrfs_commit_transaction(trans); | |
4725 | if (ret) | |
4726 | goto done; | |
4727 | } else { | |
4728 | mutex_unlock(&fs_info->chunk_mutex); | |
4729 | btrfs_end_transaction(trans); | |
4730 | } | |
8f18cf13 | 4731 | |
ba1bf481 | 4732 | again: |
8f18cf13 CM |
4733 | key.objectid = device->devid; |
4734 | key.offset = (u64)-1; | |
4735 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4736 | ||
213e64da | 4737 | do { |
0b246afa | 4738 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4739 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4740 | if (ret < 0) { |
0b246afa | 4741 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4742 | goto done; |
67c5e7d4 | 4743 | } |
8f18cf13 CM |
4744 | |
4745 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4746 | if (ret) |
0b246afa | 4747 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4748 | if (ret < 0) |
4749 | goto done; | |
4750 | if (ret) { | |
4751 | ret = 0; | |
b3b4aa74 | 4752 | btrfs_release_path(path); |
bf1fb512 | 4753 | break; |
8f18cf13 CM |
4754 | } |
4755 | ||
4756 | l = path->nodes[0]; | |
4757 | slot = path->slots[0]; | |
4758 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4759 | ||
ba1bf481 | 4760 | if (key.objectid != device->devid) { |
0b246afa | 4761 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4762 | btrfs_release_path(path); |
bf1fb512 | 4763 | break; |
ba1bf481 | 4764 | } |
8f18cf13 CM |
4765 | |
4766 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4767 | length = btrfs_dev_extent_length(l, dev_extent); | |
4768 | ||
ba1bf481 | 4769 | if (key.offset + length <= new_size) { |
0b246afa | 4770 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4771 | btrfs_release_path(path); |
d6397bae | 4772 | break; |
ba1bf481 | 4773 | } |
8f18cf13 | 4774 | |
8f18cf13 | 4775 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4776 | btrfs_release_path(path); |
8f18cf13 | 4777 | |
a6f93c71 LB |
4778 | /* |
4779 | * We may be relocating the only data chunk we have, | |
4780 | * which could potentially end up with losing data's | |
4781 | * raid profile, so lets allocate an empty one in | |
4782 | * advance. | |
4783 | */ | |
4784 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4785 | if (ret < 0) { | |
4786 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
4787 | goto done; | |
4788 | } | |
4789 | ||
0b246afa JM |
4790 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4791 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
eede2bf3 | 4792 | if (ret == -ENOSPC) { |
ba1bf481 | 4793 | failed++; |
eede2bf3 OS |
4794 | } else if (ret) { |
4795 | if (ret == -ETXTBSY) { | |
4796 | btrfs_warn(fs_info, | |
4797 | "could not shrink block group %llu due to active swapfile", | |
4798 | chunk_offset); | |
4799 | } | |
4800 | goto done; | |
4801 | } | |
213e64da | 4802 | } while (key.offset-- > 0); |
ba1bf481 JB |
4803 | |
4804 | if (failed && !retried) { | |
4805 | failed = 0; | |
4806 | retried = true; | |
4807 | goto again; | |
4808 | } else if (failed && retried) { | |
4809 | ret = -ENOSPC; | |
ba1bf481 | 4810 | goto done; |
8f18cf13 CM |
4811 | } |
4812 | ||
d6397bae | 4813 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4814 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4815 | if (IS_ERR(trans)) { |
4816 | ret = PTR_ERR(trans); | |
4817 | goto done; | |
4818 | } | |
4819 | ||
34441361 | 4820 | mutex_lock(&fs_info->chunk_mutex); |
7cc8e58d | 4821 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4822 | if (list_empty(&device->post_commit_list)) |
4823 | list_add_tail(&device->post_commit_list, | |
4824 | &trans->transaction->dev_update_list); | |
d6397bae | 4825 | |
d6397bae | 4826 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4827 | btrfs_set_super_total_bytes(super_copy, |
4828 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4829 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4830 | |
4831 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4832 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4833 | if (ret < 0) { |
4834 | btrfs_abort_transaction(trans, ret); | |
4835 | btrfs_end_transaction(trans); | |
4836 | } else { | |
4837 | ret = btrfs_commit_transaction(trans); | |
4838 | } | |
8f18cf13 CM |
4839 | done: |
4840 | btrfs_free_path(path); | |
53e489bc | 4841 | if (ret) { |
34441361 | 4842 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4843 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4844 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4845 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4846 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4847 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4848 | } |
8f18cf13 CM |
4849 | return ret; |
4850 | } | |
4851 | ||
2ff7e61e | 4852 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4853 | struct btrfs_key *key, |
4854 | struct btrfs_chunk *chunk, int item_size) | |
4855 | { | |
0b246afa | 4856 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4857 | struct btrfs_disk_key disk_key; |
4858 | u32 array_size; | |
4859 | u8 *ptr; | |
4860 | ||
34441361 | 4861 | mutex_lock(&fs_info->chunk_mutex); |
0b86a832 | 4862 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4863 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4864 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
34441361 | 4865 | mutex_unlock(&fs_info->chunk_mutex); |
0b86a832 | 4866 | return -EFBIG; |
fe48a5c0 | 4867 | } |
0b86a832 CM |
4868 | |
4869 | ptr = super_copy->sys_chunk_array + array_size; | |
4870 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4871 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4872 | ptr += sizeof(disk_key); | |
4873 | memcpy(ptr, chunk, item_size); | |
4874 | item_size += sizeof(disk_key); | |
4875 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
34441361 | 4876 | mutex_unlock(&fs_info->chunk_mutex); |
fe48a5c0 | 4877 | |
0b86a832 CM |
4878 | return 0; |
4879 | } | |
4880 | ||
73c5de00 AJ |
4881 | /* |
4882 | * sort the devices in descending order by max_avail, total_avail | |
4883 | */ | |
4884 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4885 | { |
73c5de00 AJ |
4886 | const struct btrfs_device_info *di_a = a; |
4887 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4888 | |
73c5de00 | 4889 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4890 | return -1; |
73c5de00 | 4891 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4892 | return 1; |
73c5de00 AJ |
4893 | if (di_a->total_avail > di_b->total_avail) |
4894 | return -1; | |
4895 | if (di_a->total_avail < di_b->total_avail) | |
4896 | return 1; | |
4897 | return 0; | |
b2117a39 | 4898 | } |
0b86a832 | 4899 | |
53b381b3 DW |
4900 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4901 | { | |
ffe2d203 | 4902 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4903 | return; |
4904 | ||
ceda0864 | 4905 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4906 | } |
4907 | ||
73c5de00 | 4908 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
72b468c8 | 4909 | u64 start, u64 type) |
b2117a39 | 4910 | { |
2ff7e61e | 4911 | struct btrfs_fs_info *info = trans->fs_info; |
73c5de00 | 4912 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
ebcc9301 | 4913 | struct btrfs_device *device; |
73c5de00 AJ |
4914 | struct map_lookup *map = NULL; |
4915 | struct extent_map_tree *em_tree; | |
4916 | struct extent_map *em; | |
4917 | struct btrfs_device_info *devices_info = NULL; | |
4918 | u64 total_avail; | |
4919 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4920 | int data_stripes; /* number of stripes that count for |
4921 | block group size */ | |
73c5de00 AJ |
4922 | int sub_stripes; /* sub_stripes info for map */ |
4923 | int dev_stripes; /* stripes per dev */ | |
4924 | int devs_max; /* max devs to use */ | |
4925 | int devs_min; /* min devs needed */ | |
4926 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4927 | int ncopies; /* how many copies to data has */ | |
b50836ed HK |
4928 | int nparity; /* number of stripes worth of bytes to |
4929 | store parity information */ | |
73c5de00 AJ |
4930 | int ret; |
4931 | u64 max_stripe_size; | |
4932 | u64 max_chunk_size; | |
4933 | u64 stripe_size; | |
23f0ff1e | 4934 | u64 chunk_size; |
73c5de00 AJ |
4935 | int ndevs; |
4936 | int i; | |
4937 | int j; | |
31e50229 | 4938 | int index; |
593060d7 | 4939 | |
0c460c0d | 4940 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4941 | |
4117f207 QW |
4942 | if (list_empty(&fs_devices->alloc_list)) { |
4943 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
4944 | btrfs_debug(info, "%s: no writable device", __func__); | |
73c5de00 | 4945 | return -ENOSPC; |
4117f207 | 4946 | } |
b2117a39 | 4947 | |
3e72ee88 | 4948 | index = btrfs_bg_flags_to_raid_index(type); |
73c5de00 | 4949 | |
31e50229 LB |
4950 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4951 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4952 | devs_max = btrfs_raid_array[index].devs_max; | |
4953 | devs_min = btrfs_raid_array[index].devs_min; | |
4954 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4955 | ncopies = btrfs_raid_array[index].ncopies; | |
b50836ed | 4956 | nparity = btrfs_raid_array[index].nparity; |
b2117a39 | 4957 | |
9b3f68b9 | 4958 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4959 | max_stripe_size = SZ_1G; |
fce466ea | 4960 | max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; |
23f8f9b7 | 4961 | if (!devs_max) |
062d4d1f | 4962 | devs_max = BTRFS_MAX_DEVS(info); |
9b3f68b9 | 4963 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4964 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4965 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4966 | max_stripe_size = SZ_1G; | |
1100373f | 4967 | else |
ee22184b | 4968 | max_stripe_size = SZ_256M; |
73c5de00 | 4969 | max_chunk_size = max_stripe_size; |
23f8f9b7 | 4970 | if (!devs_max) |
062d4d1f | 4971 | devs_max = BTRFS_MAX_DEVS(info); |
a40a90a0 | 4972 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4973 | max_stripe_size = SZ_32M; |
73c5de00 | 4974 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4975 | if (!devs_max) |
4976 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4977 | } else { |
351fd353 | 4978 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 | 4979 | type); |
290342f6 | 4980 | BUG(); |
9b3f68b9 CM |
4981 | } |
4982 | ||
52042d8e | 4983 | /* We don't want a chunk larger than 10% of writable space */ |
2b82032c YZ |
4984 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), |
4985 | max_chunk_size); | |
9b3f68b9 | 4986 | |
31e818fe | 4987 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4988 | GFP_NOFS); |
4989 | if (!devices_info) | |
4990 | return -ENOMEM; | |
0cad8a11 | 4991 | |
9f680ce0 | 4992 | /* |
73c5de00 AJ |
4993 | * in the first pass through the devices list, we gather information |
4994 | * about the available holes on each device. | |
9f680ce0 | 4995 | */ |
73c5de00 | 4996 | ndevs = 0; |
ebcc9301 | 4997 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
73c5de00 AJ |
4998 | u64 max_avail; |
4999 | u64 dev_offset; | |
b2117a39 | 5000 | |
ebbede42 | 5001 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5002 | WARN(1, KERN_ERR |
efe120a0 | 5003 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5004 | continue; |
5005 | } | |
b2117a39 | 5006 | |
e12c9621 AJ |
5007 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5008 | &device->dev_state) || | |
401e29c1 | 5009 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5010 | continue; |
b2117a39 | 5011 | |
73c5de00 AJ |
5012 | if (device->total_bytes > device->bytes_used) |
5013 | total_avail = device->total_bytes - device->bytes_used; | |
5014 | else | |
5015 | total_avail = 0; | |
38c01b96 | 5016 | |
5017 | /* If there is no space on this device, skip it. */ | |
5018 | if (total_avail == 0) | |
5019 | continue; | |
b2117a39 | 5020 | |
60dfdf25 | 5021 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
5022 | max_stripe_size * dev_stripes, |
5023 | &dev_offset, &max_avail); | |
5024 | if (ret && ret != -ENOSPC) | |
5025 | goto error; | |
b2117a39 | 5026 | |
73c5de00 AJ |
5027 | if (ret == 0) |
5028 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 5029 | |
4117f207 QW |
5030 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) { |
5031 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5032 | btrfs_debug(info, | |
5033 | "%s: devid %llu has no free space, have=%llu want=%u", | |
5034 | __func__, device->devid, max_avail, | |
5035 | BTRFS_STRIPE_LEN * dev_stripes); | |
73c5de00 | 5036 | continue; |
4117f207 | 5037 | } |
b2117a39 | 5038 | |
063d006f ES |
5039 | if (ndevs == fs_devices->rw_devices) { |
5040 | WARN(1, "%s: found more than %llu devices\n", | |
5041 | __func__, fs_devices->rw_devices); | |
5042 | break; | |
5043 | } | |
73c5de00 AJ |
5044 | devices_info[ndevs].dev_offset = dev_offset; |
5045 | devices_info[ndevs].max_avail = max_avail; | |
5046 | devices_info[ndevs].total_avail = total_avail; | |
5047 | devices_info[ndevs].dev = device; | |
5048 | ++ndevs; | |
5049 | } | |
b2117a39 | 5050 | |
73c5de00 AJ |
5051 | /* |
5052 | * now sort the devices by hole size / available space | |
5053 | */ | |
5054 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
5055 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 5056 | |
73c5de00 | 5057 | /* round down to number of usable stripes */ |
e5600fd6 | 5058 | ndevs = round_down(ndevs, devs_increment); |
b2117a39 | 5059 | |
ba89b802 | 5060 | if (ndevs < devs_min) { |
73c5de00 | 5061 | ret = -ENOSPC; |
4117f207 QW |
5062 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
5063 | btrfs_debug(info, | |
5064 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
ba89b802 | 5065 | __func__, ndevs, devs_min); |
4117f207 | 5066 | } |
73c5de00 | 5067 | goto error; |
b2117a39 | 5068 | } |
9f680ce0 | 5069 | |
f148ef4d NB |
5070 | ndevs = min(ndevs, devs_max); |
5071 | ||
73c5de00 | 5072 | /* |
92e222df HK |
5073 | * The primary goal is to maximize the number of stripes, so use as |
5074 | * many devices as possible, even if the stripes are not maximum sized. | |
5075 | * | |
5076 | * The DUP profile stores more than one stripe per device, the | |
5077 | * max_avail is the total size so we have to adjust. | |
73c5de00 | 5078 | */ |
92e222df | 5079 | stripe_size = div_u64(devices_info[ndevs - 1].max_avail, dev_stripes); |
73c5de00 | 5080 | num_stripes = ndevs * dev_stripes; |
b2117a39 | 5081 | |
53b381b3 DW |
5082 | /* |
5083 | * this will have to be fixed for RAID1 and RAID10 over | |
5084 | * more drives | |
5085 | */ | |
b50836ed | 5086 | data_stripes = (num_stripes - nparity) / ncopies; |
86db2578 CM |
5087 | |
5088 | /* | |
5089 | * Use the number of data stripes to figure out how big this chunk | |
5090 | * is really going to be in terms of logical address space, | |
baf92114 HK |
5091 | * and compare that answer with the max chunk size. If it's higher, |
5092 | * we try to reduce stripe_size. | |
86db2578 CM |
5093 | */ |
5094 | if (stripe_size * data_stripes > max_chunk_size) { | |
793ff2c8 | 5095 | /* |
baf92114 HK |
5096 | * Reduce stripe_size, round it up to a 16MB boundary again and |
5097 | * then use it, unless it ends up being even bigger than the | |
5098 | * previous value we had already. | |
86db2578 | 5099 | */ |
baf92114 HK |
5100 | stripe_size = min(round_up(div_u64(max_chunk_size, |
5101 | data_stripes), SZ_16M), | |
793ff2c8 | 5102 | stripe_size); |
86db2578 CM |
5103 | } |
5104 | ||
37db63a4 | 5105 | /* align to BTRFS_STRIPE_LEN */ |
500ceed8 | 5106 | stripe_size = round_down(stripe_size, BTRFS_STRIPE_LEN); |
b2117a39 MX |
5107 | |
5108 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
5109 | if (!map) { | |
5110 | ret = -ENOMEM; | |
5111 | goto error; | |
5112 | } | |
5113 | map->num_stripes = num_stripes; | |
9b3f68b9 | 5114 | |
73c5de00 AJ |
5115 | for (i = 0; i < ndevs; ++i) { |
5116 | for (j = 0; j < dev_stripes; ++j) { | |
5117 | int s = i * dev_stripes + j; | |
5118 | map->stripes[s].dev = devices_info[i].dev; | |
5119 | map->stripes[s].physical = devices_info[i].dev_offset + | |
5120 | j * stripe_size; | |
6324fbf3 | 5121 | } |
6324fbf3 | 5122 | } |
500ceed8 NB |
5123 | map->stripe_len = BTRFS_STRIPE_LEN; |
5124 | map->io_align = BTRFS_STRIPE_LEN; | |
5125 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5126 | map->type = type; |
2b82032c | 5127 | map->sub_stripes = sub_stripes; |
0b86a832 | 5128 | |
23f0ff1e | 5129 | chunk_size = stripe_size * data_stripes; |
0b86a832 | 5130 | |
23f0ff1e | 5131 | trace_btrfs_chunk_alloc(info, map, start, chunk_size); |
1abe9b8a | 5132 | |
172ddd60 | 5133 | em = alloc_extent_map(); |
2b82032c | 5134 | if (!em) { |
298a8f9c | 5135 | kfree(map); |
b2117a39 MX |
5136 | ret = -ENOMEM; |
5137 | goto error; | |
593060d7 | 5138 | } |
298a8f9c | 5139 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5140 | em->map_lookup = map; |
2b82032c | 5141 | em->start = start; |
23f0ff1e | 5142 | em->len = chunk_size; |
2b82032c YZ |
5143 | em->block_start = 0; |
5144 | em->block_len = em->len; | |
6df9a95e | 5145 | em->orig_block_len = stripe_size; |
593060d7 | 5146 | |
0b246afa | 5147 | em_tree = &info->mapping_tree.map_tree; |
890871be | 5148 | write_lock(&em_tree->lock); |
09a2a8f9 | 5149 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5150 | if (ret) { |
1efb72a3 | 5151 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5152 | free_extent_map(em); |
1dd4602f | 5153 | goto error; |
0f5d42b2 | 5154 | } |
1efb72a3 NB |
5155 | write_unlock(&em_tree->lock); |
5156 | ||
23f0ff1e | 5157 | ret = btrfs_make_block_group(trans, 0, type, start, chunk_size); |
6df9a95e JB |
5158 | if (ret) |
5159 | goto error_del_extent; | |
2b82032c | 5160 | |
bbbf7243 NB |
5161 | for (i = 0; i < map->num_stripes; i++) { |
5162 | struct btrfs_device *dev = map->stripes[i].dev; | |
5163 | ||
5164 | btrfs_device_set_bytes_used(dev, dev->bytes_used + stripe_size); | |
5165 | if (list_empty(&dev->post_commit_list)) | |
5166 | list_add_tail(&dev->post_commit_list, | |
5167 | &trans->transaction->dev_update_list); | |
5168 | } | |
43530c46 | 5169 | |
a5ed45f8 | 5170 | atomic64_sub(stripe_size * map->num_stripes, &info->free_chunk_space); |
1c116187 | 5171 | |
0f5d42b2 | 5172 | free_extent_map(em); |
0b246afa | 5173 | check_raid56_incompat_flag(info, type); |
53b381b3 | 5174 | |
b2117a39 | 5175 | kfree(devices_info); |
2b82032c | 5176 | return 0; |
b2117a39 | 5177 | |
6df9a95e | 5178 | error_del_extent: |
0f5d42b2 JB |
5179 | write_lock(&em_tree->lock); |
5180 | remove_extent_mapping(em_tree, em); | |
5181 | write_unlock(&em_tree->lock); | |
5182 | ||
5183 | /* One for our allocation */ | |
5184 | free_extent_map(em); | |
5185 | /* One for the tree reference */ | |
5186 | free_extent_map(em); | |
b2117a39 | 5187 | error: |
b2117a39 MX |
5188 | kfree(devices_info); |
5189 | return ret; | |
2b82032c YZ |
5190 | } |
5191 | ||
6df9a95e | 5192 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
97aff912 | 5193 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 5194 | { |
97aff912 | 5195 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab JM |
5196 | struct btrfs_root *extent_root = fs_info->extent_root; |
5197 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 5198 | struct btrfs_key key; |
2b82032c YZ |
5199 | struct btrfs_device *device; |
5200 | struct btrfs_chunk *chunk; | |
5201 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
5202 | struct extent_map *em; |
5203 | struct map_lookup *map; | |
5204 | size_t item_size; | |
5205 | u64 dev_offset; | |
5206 | u64 stripe_size; | |
5207 | int i = 0; | |
140e639f | 5208 | int ret = 0; |
2b82032c | 5209 | |
60ca842e | 5210 | em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size); |
592d92ee LB |
5211 | if (IS_ERR(em)) |
5212 | return PTR_ERR(em); | |
6df9a95e | 5213 | |
95617d69 | 5214 | map = em->map_lookup; |
6df9a95e JB |
5215 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5216 | stripe_size = em->orig_block_len; | |
5217 | ||
2b82032c | 5218 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
5219 | if (!chunk) { |
5220 | ret = -ENOMEM; | |
5221 | goto out; | |
5222 | } | |
5223 | ||
50460e37 FM |
5224 | /* |
5225 | * Take the device list mutex to prevent races with the final phase of | |
5226 | * a device replace operation that replaces the device object associated | |
5227 | * with the map's stripes, because the device object's id can change | |
5228 | * at any time during that final phase of the device replace operation | |
5229 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
5230 | */ | |
0b246afa | 5231 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
5232 | for (i = 0; i < map->num_stripes; i++) { |
5233 | device = map->stripes[i].dev; | |
5234 | dev_offset = map->stripes[i].physical; | |
2b82032c | 5235 | |
0b86a832 | 5236 | ret = btrfs_update_device(trans, device); |
3acd3953 | 5237 | if (ret) |
50460e37 | 5238 | break; |
b5d9071c NB |
5239 | ret = btrfs_alloc_dev_extent(trans, device, chunk_offset, |
5240 | dev_offset, stripe_size); | |
6df9a95e | 5241 | if (ret) |
50460e37 FM |
5242 | break; |
5243 | } | |
5244 | if (ret) { | |
0b246afa | 5245 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 5246 | goto out; |
2b82032c YZ |
5247 | } |
5248 | ||
2b82032c | 5249 | stripe = &chunk->stripe; |
6df9a95e JB |
5250 | for (i = 0; i < map->num_stripes; i++) { |
5251 | device = map->stripes[i].dev; | |
5252 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 5253 | |
e17cade2 CM |
5254 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5255 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5256 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5257 | stripe++; |
0b86a832 | 5258 | } |
0b246afa | 5259 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 5260 | |
2b82032c | 5261 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 5262 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5263 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5264 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5265 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5266 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5267 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5268 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5269 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5270 | |
2b82032c YZ |
5271 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5272 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
5273 | key.offset = chunk_offset; | |
0b86a832 | 5274 | |
2b82032c | 5275 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
5276 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
5277 | /* | |
5278 | * TODO: Cleanup of inserted chunk root in case of | |
5279 | * failure. | |
5280 | */ | |
2ff7e61e | 5281 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
8f18cf13 | 5282 | } |
1abe9b8a | 5283 | |
6df9a95e | 5284 | out: |
0b86a832 | 5285 | kfree(chunk); |
6df9a95e | 5286 | free_extent_map(em); |
4ed1d16e | 5287 | return ret; |
2b82032c | 5288 | } |
0b86a832 | 5289 | |
2b82032c | 5290 | /* |
52042d8e AG |
5291 | * Chunk allocation falls into two parts. The first part does work |
5292 | * that makes the new allocated chunk usable, but does not do any operation | |
5293 | * that modifies the chunk tree. The second part does the work that | |
5294 | * requires modifying the chunk tree. This division is important for the | |
2b82032c YZ |
5295 | * bootstrap process of adding storage to a seed btrfs. |
5296 | */ | |
c216b203 | 5297 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type) |
2b82032c YZ |
5298 | { |
5299 | u64 chunk_offset; | |
2b82032c | 5300 | |
c216b203 NB |
5301 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
5302 | chunk_offset = find_next_chunk(trans->fs_info); | |
72b468c8 | 5303 | return __btrfs_alloc_chunk(trans, chunk_offset, type); |
2b82032c YZ |
5304 | } |
5305 | ||
6f8e0fc7 | 5306 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5307 | { |
6f8e0fc7 | 5308 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c YZ |
5309 | u64 chunk_offset; |
5310 | u64 sys_chunk_offset; | |
2b82032c | 5311 | u64 alloc_profile; |
2b82032c YZ |
5312 | int ret; |
5313 | ||
6df9a95e | 5314 | chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5315 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
72b468c8 | 5316 | ret = __btrfs_alloc_chunk(trans, chunk_offset, alloc_profile); |
79787eaa JM |
5317 | if (ret) |
5318 | return ret; | |
2b82032c | 5319 | |
0b246afa | 5320 | sys_chunk_offset = find_next_chunk(fs_info); |
1b86826d | 5321 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
72b468c8 | 5322 | ret = __btrfs_alloc_chunk(trans, sys_chunk_offset, alloc_profile); |
79787eaa | 5323 | return ret; |
2b82032c YZ |
5324 | } |
5325 | ||
d20983b4 MX |
5326 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5327 | { | |
5328 | int max_errors; | |
5329 | ||
5330 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5331 | BTRFS_BLOCK_GROUP_RAID10 | | |
5332 | BTRFS_BLOCK_GROUP_RAID5 | | |
5333 | BTRFS_BLOCK_GROUP_DUP)) { | |
5334 | max_errors = 1; | |
5335 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5336 | max_errors = 2; | |
5337 | } else { | |
5338 | max_errors = 0; | |
005d6427 | 5339 | } |
2b82032c | 5340 | |
d20983b4 | 5341 | return max_errors; |
2b82032c YZ |
5342 | } |
5343 | ||
2ff7e61e | 5344 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5345 | { |
5346 | struct extent_map *em; | |
5347 | struct map_lookup *map; | |
2b82032c | 5348 | int readonly = 0; |
d20983b4 | 5349 | int miss_ndevs = 0; |
2b82032c YZ |
5350 | int i; |
5351 | ||
60ca842e | 5352 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5353 | if (IS_ERR(em)) |
2b82032c YZ |
5354 | return 1; |
5355 | ||
95617d69 | 5356 | map = em->map_lookup; |
2b82032c | 5357 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5358 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5359 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5360 | miss_ndevs++; |
5361 | continue; | |
5362 | } | |
ebbede42 AJ |
5363 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5364 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5365 | readonly = 1; |
d20983b4 | 5366 | goto end; |
2b82032c YZ |
5367 | } |
5368 | } | |
d20983b4 MX |
5369 | |
5370 | /* | |
5371 | * If the number of missing devices is larger than max errors, | |
5372 | * we can not write the data into that chunk successfully, so | |
5373 | * set it readonly. | |
5374 | */ | |
5375 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5376 | readonly = 1; | |
5377 | end: | |
0b86a832 | 5378 | free_extent_map(em); |
2b82032c | 5379 | return readonly; |
0b86a832 CM |
5380 | } |
5381 | ||
5382 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5383 | { | |
a8067e02 | 5384 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5385 | } |
5386 | ||
5387 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5388 | { | |
5389 | struct extent_map *em; | |
5390 | ||
d397712b | 5391 | while (1) { |
890871be | 5392 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5393 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5394 | if (em) | |
5395 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5396 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5397 | if (!em) |
5398 | break; | |
0b86a832 CM |
5399 | /* once for us */ |
5400 | free_extent_map(em); | |
5401 | /* once for the tree */ | |
5402 | free_extent_map(em); | |
5403 | } | |
5404 | } | |
5405 | ||
5d964051 | 5406 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5407 | { |
5408 | struct extent_map *em; | |
5409 | struct map_lookup *map; | |
f188591e CM |
5410 | int ret; |
5411 | ||
60ca842e | 5412 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5413 | if (IS_ERR(em)) |
5414 | /* | |
5415 | * We could return errors for these cases, but that could get | |
5416 | * ugly and we'd probably do the same thing which is just not do | |
5417 | * anything else and exit, so return 1 so the callers don't try | |
5418 | * to use other copies. | |
5419 | */ | |
fb7669b5 | 5420 | return 1; |
fb7669b5 | 5421 | |
95617d69 | 5422 | map = em->map_lookup; |
f188591e CM |
5423 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5424 | ret = map->num_stripes; | |
321aecc6 CM |
5425 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5426 | ret = map->sub_stripes; | |
53b381b3 DW |
5427 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5428 | ret = 2; | |
5429 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5430 | /* |
5431 | * There could be two corrupted data stripes, we need | |
5432 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5433 | * |
8810f751 LB |
5434 | * Fail a stripe at a time on every retry except the |
5435 | * stripe under reconstruction. | |
5436 | */ | |
5437 | ret = map->num_stripes; | |
f188591e CM |
5438 | else |
5439 | ret = 1; | |
5440 | free_extent_map(em); | |
ad6d620e | 5441 | |
cb5583dd | 5442 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5443 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5444 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5445 | ret++; |
cb5583dd | 5446 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5447 | |
f188591e CM |
5448 | return ret; |
5449 | } | |
5450 | ||
2ff7e61e | 5451 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5452 | u64 logical) |
5453 | { | |
5454 | struct extent_map *em; | |
5455 | struct map_lookup *map; | |
0b246afa | 5456 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5457 | |
60ca842e | 5458 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5459 | |
69f03f13 NB |
5460 | if (!WARN_ON(IS_ERR(em))) { |
5461 | map = em->map_lookup; | |
5462 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5463 | len = map->stripe_len * nr_data_stripes(map); | |
5464 | free_extent_map(em); | |
5465 | } | |
53b381b3 DW |
5466 | return len; |
5467 | } | |
5468 | ||
e4ff5fb5 | 5469 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5470 | { |
5471 | struct extent_map *em; | |
5472 | struct map_lookup *map; | |
53b381b3 DW |
5473 | int ret = 0; |
5474 | ||
60ca842e | 5475 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5476 | |
69f03f13 NB |
5477 | if(!WARN_ON(IS_ERR(em))) { |
5478 | map = em->map_lookup; | |
5479 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5480 | ret = 1; | |
5481 | free_extent_map(em); | |
5482 | } | |
53b381b3 DW |
5483 | return ret; |
5484 | } | |
5485 | ||
30d9861f | 5486 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5487 | struct map_lookup *map, int first, |
8ba0ae78 | 5488 | int dev_replace_is_ongoing) |
dfe25020 CM |
5489 | { |
5490 | int i; | |
99f92a7c | 5491 | int num_stripes; |
8ba0ae78 | 5492 | int preferred_mirror; |
30d9861f SB |
5493 | int tolerance; |
5494 | struct btrfs_device *srcdev; | |
5495 | ||
99f92a7c AJ |
5496 | ASSERT((map->type & |
5497 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))); | |
5498 | ||
5499 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5500 | num_stripes = map->sub_stripes; | |
5501 | else | |
5502 | num_stripes = map->num_stripes; | |
5503 | ||
8ba0ae78 AJ |
5504 | preferred_mirror = first + current->pid % num_stripes; |
5505 | ||
30d9861f SB |
5506 | if (dev_replace_is_ongoing && |
5507 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5508 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5509 | srcdev = fs_info->dev_replace.srcdev; | |
5510 | else | |
5511 | srcdev = NULL; | |
5512 | ||
5513 | /* | |
5514 | * try to avoid the drive that is the source drive for a | |
5515 | * dev-replace procedure, only choose it if no other non-missing | |
5516 | * mirror is available | |
5517 | */ | |
5518 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5519 | if (map->stripes[preferred_mirror].dev->bdev && |
5520 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5521 | return preferred_mirror; | |
99f92a7c | 5522 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5523 | if (map->stripes[i].dev->bdev && |
5524 | (tolerance || map->stripes[i].dev != srcdev)) | |
5525 | return i; | |
5526 | } | |
dfe25020 | 5527 | } |
30d9861f | 5528 | |
dfe25020 CM |
5529 | /* we couldn't find one that doesn't fail. Just return something |
5530 | * and the io error handling code will clean up eventually | |
5531 | */ | |
8ba0ae78 | 5532 | return preferred_mirror; |
dfe25020 CM |
5533 | } |
5534 | ||
53b381b3 DW |
5535 | static inline int parity_smaller(u64 a, u64 b) |
5536 | { | |
5537 | return a > b; | |
5538 | } | |
5539 | ||
5540 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5541 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5542 | { |
5543 | struct btrfs_bio_stripe s; | |
5544 | int i; | |
5545 | u64 l; | |
5546 | int again = 1; | |
5547 | ||
5548 | while (again) { | |
5549 | again = 0; | |
cc7539ed | 5550 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5551 | if (parity_smaller(bbio->raid_map[i], |
5552 | bbio->raid_map[i+1])) { | |
53b381b3 | 5553 | s = bbio->stripes[i]; |
8e5cfb55 | 5554 | l = bbio->raid_map[i]; |
53b381b3 | 5555 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5556 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5557 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5558 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5559 | |
53b381b3 DW |
5560 | again = 1; |
5561 | } | |
5562 | } | |
5563 | } | |
5564 | } | |
5565 | ||
6e9606d2 ZL |
5566 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5567 | { | |
5568 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5569 | /* the size of the btrfs_bio */ |
6e9606d2 | 5570 | sizeof(struct btrfs_bio) + |
e57cf21e | 5571 | /* plus the variable array for the stripes */ |
6e9606d2 | 5572 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5573 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5574 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5575 | /* |
5576 | * plus the raid_map, which includes both the tgt dev | |
5577 | * and the stripes | |
5578 | */ | |
5579 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5580 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5581 | |
5582 | atomic_set(&bbio->error, 0); | |
140475ae | 5583 | refcount_set(&bbio->refs, 1); |
6e9606d2 ZL |
5584 | |
5585 | return bbio; | |
5586 | } | |
5587 | ||
5588 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5589 | { | |
140475ae ER |
5590 | WARN_ON(!refcount_read(&bbio->refs)); |
5591 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5592 | } |
5593 | ||
5594 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5595 | { | |
5596 | if (!bbio) | |
5597 | return; | |
140475ae | 5598 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5599 | kfree(bbio); |
5600 | } | |
5601 | ||
0b3d4cd3 LB |
5602 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5603 | /* | |
5604 | * Please note that, discard won't be sent to target device of device | |
5605 | * replace. | |
5606 | */ | |
5607 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
5608 | u64 logical, u64 length, | |
5609 | struct btrfs_bio **bbio_ret) | |
5610 | { | |
5611 | struct extent_map *em; | |
5612 | struct map_lookup *map; | |
5613 | struct btrfs_bio *bbio; | |
5614 | u64 offset; | |
5615 | u64 stripe_nr; | |
5616 | u64 stripe_nr_end; | |
5617 | u64 stripe_end_offset; | |
5618 | u64 stripe_cnt; | |
5619 | u64 stripe_len; | |
5620 | u64 stripe_offset; | |
5621 | u64 num_stripes; | |
5622 | u32 stripe_index; | |
5623 | u32 factor = 0; | |
5624 | u32 sub_stripes = 0; | |
5625 | u64 stripes_per_dev = 0; | |
5626 | u32 remaining_stripes = 0; | |
5627 | u32 last_stripe = 0; | |
5628 | int ret = 0; | |
5629 | int i; | |
5630 | ||
5631 | /* discard always return a bbio */ | |
5632 | ASSERT(bbio_ret); | |
5633 | ||
60ca842e | 5634 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5635 | if (IS_ERR(em)) |
5636 | return PTR_ERR(em); | |
5637 | ||
5638 | map = em->map_lookup; | |
5639 | /* we don't discard raid56 yet */ | |
5640 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5641 | ret = -EOPNOTSUPP; | |
5642 | goto out; | |
5643 | } | |
5644 | ||
5645 | offset = logical - em->start; | |
5646 | length = min_t(u64, em->len - offset, length); | |
5647 | ||
5648 | stripe_len = map->stripe_len; | |
5649 | /* | |
5650 | * stripe_nr counts the total number of stripes we have to stride | |
5651 | * to get to this block | |
5652 | */ | |
5653 | stripe_nr = div64_u64(offset, stripe_len); | |
5654 | ||
5655 | /* stripe_offset is the offset of this block in its stripe */ | |
5656 | stripe_offset = offset - stripe_nr * stripe_len; | |
5657 | ||
5658 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5659 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5660 | stripe_cnt = stripe_nr_end - stripe_nr; |
5661 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5662 | (offset + length); | |
5663 | /* | |
5664 | * after this, stripe_nr is the number of stripes on this | |
5665 | * device we have to walk to find the data, and stripe_index is | |
5666 | * the number of our device in the stripe array | |
5667 | */ | |
5668 | num_stripes = 1; | |
5669 | stripe_index = 0; | |
5670 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5671 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5672 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5673 | sub_stripes = 1; | |
5674 | else | |
5675 | sub_stripes = map->sub_stripes; | |
5676 | ||
5677 | factor = map->num_stripes / sub_stripes; | |
5678 | num_stripes = min_t(u64, map->num_stripes, | |
5679 | sub_stripes * stripe_cnt); | |
5680 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5681 | stripe_index *= sub_stripes; | |
5682 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5683 | &remaining_stripes); | |
5684 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5685 | last_stripe *= sub_stripes; | |
5686 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5687 | BTRFS_BLOCK_GROUP_DUP)) { | |
5688 | num_stripes = map->num_stripes; | |
5689 | } else { | |
5690 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5691 | &stripe_index); | |
5692 | } | |
5693 | ||
5694 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5695 | if (!bbio) { | |
5696 | ret = -ENOMEM; | |
5697 | goto out; | |
5698 | } | |
5699 | ||
5700 | for (i = 0; i < num_stripes; i++) { | |
5701 | bbio->stripes[i].physical = | |
5702 | map->stripes[stripe_index].physical + | |
5703 | stripe_offset + stripe_nr * map->stripe_len; | |
5704 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5705 | ||
5706 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5707 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5708 | bbio->stripes[i].length = stripes_per_dev * | |
5709 | map->stripe_len; | |
5710 | ||
5711 | if (i / sub_stripes < remaining_stripes) | |
5712 | bbio->stripes[i].length += | |
5713 | map->stripe_len; | |
5714 | ||
5715 | /* | |
5716 | * Special for the first stripe and | |
5717 | * the last stripe: | |
5718 | * | |
5719 | * |-------|...|-------| | |
5720 | * |----------| | |
5721 | * off end_off | |
5722 | */ | |
5723 | if (i < sub_stripes) | |
5724 | bbio->stripes[i].length -= | |
5725 | stripe_offset; | |
5726 | ||
5727 | if (stripe_index >= last_stripe && | |
5728 | stripe_index <= (last_stripe + | |
5729 | sub_stripes - 1)) | |
5730 | bbio->stripes[i].length -= | |
5731 | stripe_end_offset; | |
5732 | ||
5733 | if (i == sub_stripes - 1) | |
5734 | stripe_offset = 0; | |
5735 | } else { | |
5736 | bbio->stripes[i].length = length; | |
5737 | } | |
5738 | ||
5739 | stripe_index++; | |
5740 | if (stripe_index == map->num_stripes) { | |
5741 | stripe_index = 0; | |
5742 | stripe_nr++; | |
5743 | } | |
5744 | } | |
5745 | ||
5746 | *bbio_ret = bbio; | |
5747 | bbio->map_type = map->type; | |
5748 | bbio->num_stripes = num_stripes; | |
5749 | out: | |
5750 | free_extent_map(em); | |
5751 | return ret; | |
5752 | } | |
5753 | ||
5ab56090 LB |
5754 | /* |
5755 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5756 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5757 | * left cursor can also be read from the target drive. | |
5758 | * | |
5759 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5760 | * array of stripes. | |
5761 | * For READ, it also needs to be supported using the same mirror number. | |
5762 | * | |
5763 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5764 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5765 | * case. | |
5766 | */ | |
5767 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5768 | u64 logical, u64 length, | |
5769 | u64 srcdev_devid, int *mirror_num, | |
5770 | u64 *physical) | |
5771 | { | |
5772 | struct btrfs_bio *bbio = NULL; | |
5773 | int num_stripes; | |
5774 | int index_srcdev = 0; | |
5775 | int found = 0; | |
5776 | u64 physical_of_found = 0; | |
5777 | int i; | |
5778 | int ret = 0; | |
5779 | ||
5780 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
5781 | logical, &length, &bbio, 0, 0); | |
5782 | if (ret) { | |
5783 | ASSERT(bbio == NULL); | |
5784 | return ret; | |
5785 | } | |
5786 | ||
5787 | num_stripes = bbio->num_stripes; | |
5788 | if (*mirror_num > num_stripes) { | |
5789 | /* | |
5790 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
5791 | * that means that the requested area is not left of the left | |
5792 | * cursor | |
5793 | */ | |
5794 | btrfs_put_bbio(bbio); | |
5795 | return -EIO; | |
5796 | } | |
5797 | ||
5798 | /* | |
5799 | * process the rest of the function using the mirror_num of the source | |
5800 | * drive. Therefore look it up first. At the end, patch the device | |
5801 | * pointer to the one of the target drive. | |
5802 | */ | |
5803 | for (i = 0; i < num_stripes; i++) { | |
5804 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
5805 | continue; | |
5806 | ||
5807 | /* | |
5808 | * In case of DUP, in order to keep it simple, only add the | |
5809 | * mirror with the lowest physical address | |
5810 | */ | |
5811 | if (found && | |
5812 | physical_of_found <= bbio->stripes[i].physical) | |
5813 | continue; | |
5814 | ||
5815 | index_srcdev = i; | |
5816 | found = 1; | |
5817 | physical_of_found = bbio->stripes[i].physical; | |
5818 | } | |
5819 | ||
5820 | btrfs_put_bbio(bbio); | |
5821 | ||
5822 | ASSERT(found); | |
5823 | if (!found) | |
5824 | return -EIO; | |
5825 | ||
5826 | *mirror_num = index_srcdev + 1; | |
5827 | *physical = physical_of_found; | |
5828 | return ret; | |
5829 | } | |
5830 | ||
73c0f228 LB |
5831 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
5832 | struct btrfs_bio **bbio_ret, | |
5833 | struct btrfs_dev_replace *dev_replace, | |
5834 | int *num_stripes_ret, int *max_errors_ret) | |
5835 | { | |
5836 | struct btrfs_bio *bbio = *bbio_ret; | |
5837 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5838 | int tgtdev_indexes = 0; | |
5839 | int num_stripes = *num_stripes_ret; | |
5840 | int max_errors = *max_errors_ret; | |
5841 | int i; | |
5842 | ||
5843 | if (op == BTRFS_MAP_WRITE) { | |
5844 | int index_where_to_add; | |
5845 | ||
5846 | /* | |
5847 | * duplicate the write operations while the dev replace | |
5848 | * procedure is running. Since the copying of the old disk to | |
5849 | * the new disk takes place at run time while the filesystem is | |
5850 | * mounted writable, the regular write operations to the old | |
5851 | * disk have to be duplicated to go to the new disk as well. | |
5852 | * | |
5853 | * Note that device->missing is handled by the caller, and that | |
5854 | * the write to the old disk is already set up in the stripes | |
5855 | * array. | |
5856 | */ | |
5857 | index_where_to_add = num_stripes; | |
5858 | for (i = 0; i < num_stripes; i++) { | |
5859 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5860 | /* write to new disk, too */ | |
5861 | struct btrfs_bio_stripe *new = | |
5862 | bbio->stripes + index_where_to_add; | |
5863 | struct btrfs_bio_stripe *old = | |
5864 | bbio->stripes + i; | |
5865 | ||
5866 | new->physical = old->physical; | |
5867 | new->length = old->length; | |
5868 | new->dev = dev_replace->tgtdev; | |
5869 | bbio->tgtdev_map[i] = index_where_to_add; | |
5870 | index_where_to_add++; | |
5871 | max_errors++; | |
5872 | tgtdev_indexes++; | |
5873 | } | |
5874 | } | |
5875 | num_stripes = index_where_to_add; | |
5876 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
5877 | int index_srcdev = 0; | |
5878 | int found = 0; | |
5879 | u64 physical_of_found = 0; | |
5880 | ||
5881 | /* | |
5882 | * During the dev-replace procedure, the target drive can also | |
5883 | * be used to read data in case it is needed to repair a corrupt | |
5884 | * block elsewhere. This is possible if the requested area is | |
5885 | * left of the left cursor. In this area, the target drive is a | |
5886 | * full copy of the source drive. | |
5887 | */ | |
5888 | for (i = 0; i < num_stripes; i++) { | |
5889 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5890 | /* | |
5891 | * In case of DUP, in order to keep it simple, | |
5892 | * only add the mirror with the lowest physical | |
5893 | * address | |
5894 | */ | |
5895 | if (found && | |
5896 | physical_of_found <= | |
5897 | bbio->stripes[i].physical) | |
5898 | continue; | |
5899 | index_srcdev = i; | |
5900 | found = 1; | |
5901 | physical_of_found = bbio->stripes[i].physical; | |
5902 | } | |
5903 | } | |
5904 | if (found) { | |
5905 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5906 | bbio->stripes + num_stripes; | |
5907 | ||
5908 | tgtdev_stripe->physical = physical_of_found; | |
5909 | tgtdev_stripe->length = | |
5910 | bbio->stripes[index_srcdev].length; | |
5911 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5912 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5913 | ||
5914 | tgtdev_indexes++; | |
5915 | num_stripes++; | |
5916 | } | |
5917 | } | |
5918 | ||
5919 | *num_stripes_ret = num_stripes; | |
5920 | *max_errors_ret = max_errors; | |
5921 | bbio->num_tgtdevs = tgtdev_indexes; | |
5922 | *bbio_ret = bbio; | |
5923 | } | |
5924 | ||
2b19a1fe LB |
5925 | static bool need_full_stripe(enum btrfs_map_op op) |
5926 | { | |
5927 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
5928 | } | |
5929 | ||
cf8cddd3 CH |
5930 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5931 | enum btrfs_map_op op, | |
f2d8d74d | 5932 | u64 logical, u64 *length, |
a1d3c478 | 5933 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5934 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5935 | { |
5936 | struct extent_map *em; | |
5937 | struct map_lookup *map; | |
0b86a832 | 5938 | u64 offset; |
593060d7 CM |
5939 | u64 stripe_offset; |
5940 | u64 stripe_nr; | |
53b381b3 | 5941 | u64 stripe_len; |
9d644a62 | 5942 | u32 stripe_index; |
cea9e445 | 5943 | int i; |
de11cc12 | 5944 | int ret = 0; |
f2d8d74d | 5945 | int num_stripes; |
a236aed1 | 5946 | int max_errors = 0; |
2c8cdd6e | 5947 | int tgtdev_indexes = 0; |
a1d3c478 | 5948 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5949 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5950 | int dev_replace_is_ongoing = 0; | |
5951 | int num_alloc_stripes; | |
ad6d620e SB |
5952 | int patch_the_first_stripe_for_dev_replace = 0; |
5953 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5954 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5955 | |
0b3d4cd3 LB |
5956 | if (op == BTRFS_MAP_DISCARD) |
5957 | return __btrfs_map_block_for_discard(fs_info, logical, | |
5958 | *length, bbio_ret); | |
5959 | ||
60ca842e | 5960 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
592d92ee LB |
5961 | if (IS_ERR(em)) |
5962 | return PTR_ERR(em); | |
0b86a832 | 5963 | |
95617d69 | 5964 | map = em->map_lookup; |
0b86a832 | 5965 | offset = logical - em->start; |
593060d7 | 5966 | |
53b381b3 | 5967 | stripe_len = map->stripe_len; |
593060d7 CM |
5968 | stripe_nr = offset; |
5969 | /* | |
5970 | * stripe_nr counts the total number of stripes we have to stride | |
5971 | * to get to this block | |
5972 | */ | |
47c5713f | 5973 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5974 | |
53b381b3 | 5975 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5976 | if (offset < stripe_offset) { |
5d163e0e JM |
5977 | btrfs_crit(fs_info, |
5978 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5979 | stripe_offset, offset, em->start, logical, |
5980 | stripe_len); | |
5981 | free_extent_map(em); | |
5982 | return -EINVAL; | |
5983 | } | |
593060d7 CM |
5984 | |
5985 | /* stripe_offset is the offset of this block in its stripe*/ | |
5986 | stripe_offset = offset - stripe_offset; | |
5987 | ||
53b381b3 | 5988 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5989 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5990 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5991 | raid56_full_stripe_start = offset; | |
5992 | ||
5993 | /* allow a write of a full stripe, but make sure we don't | |
5994 | * allow straddling of stripes | |
5995 | */ | |
47c5713f DS |
5996 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5997 | full_stripe_len); | |
53b381b3 DW |
5998 | raid56_full_stripe_start *= full_stripe_len; |
5999 | } | |
6000 | ||
0b3d4cd3 | 6001 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
53b381b3 DW |
6002 | u64 max_len; |
6003 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
6004 | For other RAID types and for RAID[56] reads, just allow a single | |
6005 | stripe (on a single disk). */ | |
ffe2d203 | 6006 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 6007 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
6008 | max_len = stripe_len * nr_data_stripes(map) - |
6009 | (offset - raid56_full_stripe_start); | |
6010 | } else { | |
6011 | /* we limit the length of each bio to what fits in a stripe */ | |
6012 | max_len = stripe_len - stripe_offset; | |
6013 | } | |
6014 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
6015 | } else { |
6016 | *length = em->len - offset; | |
6017 | } | |
f2d8d74d | 6018 | |
da12fe54 NB |
6019 | /* |
6020 | * This is for when we're called from btrfs_bio_fits_in_stripe and all | |
6021 | * it cares about is the length | |
6022 | */ | |
a1d3c478 | 6023 | if (!bbio_ret) |
cea9e445 CM |
6024 | goto out; |
6025 | ||
cb5583dd | 6026 | down_read(&dev_replace->rwsem); |
472262f3 | 6027 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6028 | /* |
6029 | * Hold the semaphore for read during the whole operation, write is | |
6030 | * requested at commit time but must wait. | |
6031 | */ | |
472262f3 | 6032 | if (!dev_replace_is_ongoing) |
cb5583dd | 6033 | up_read(&dev_replace->rwsem); |
472262f3 | 6034 | |
ad6d620e | 6035 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6036 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6037 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6038 | dev_replace->srcdev->devid, | |
6039 | &mirror_num, | |
6040 | &physical_to_patch_in_first_stripe); | |
6041 | if (ret) | |
ad6d620e | 6042 | goto out; |
5ab56090 LB |
6043 | else |
6044 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6045 | } else if (mirror_num > map->num_stripes) { |
6046 | mirror_num = 0; | |
6047 | } | |
6048 | ||
f2d8d74d | 6049 | num_stripes = 1; |
cea9e445 | 6050 | stripe_index = 0; |
fce3bb9a | 6051 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6052 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6053 | &stripe_index); | |
de483734 | 6054 | if (!need_full_stripe(op)) |
28e1cc7d | 6055 | mirror_num = 1; |
fce3bb9a | 6056 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
de483734 | 6057 | if (need_full_stripe(op)) |
f2d8d74d | 6058 | num_stripes = map->num_stripes; |
2fff734f | 6059 | else if (mirror_num) |
f188591e | 6060 | stripe_index = mirror_num - 1; |
dfe25020 | 6061 | else { |
30d9861f | 6062 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6063 | dev_replace_is_ongoing); |
a1d3c478 | 6064 | mirror_num = stripe_index + 1; |
dfe25020 | 6065 | } |
2fff734f | 6066 | |
611f0e00 | 6067 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6068 | if (need_full_stripe(op)) { |
f2d8d74d | 6069 | num_stripes = map->num_stripes; |
a1d3c478 | 6070 | } else if (mirror_num) { |
f188591e | 6071 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6072 | } else { |
6073 | mirror_num = 1; | |
6074 | } | |
2fff734f | 6075 | |
321aecc6 | 6076 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6077 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6078 | |
47c5713f | 6079 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6080 | stripe_index *= map->sub_stripes; |
6081 | ||
de483734 | 6082 | if (need_full_stripe(op)) |
f2d8d74d | 6083 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6084 | else if (mirror_num) |
6085 | stripe_index += mirror_num - 1; | |
dfe25020 | 6086 | else { |
3e74317a | 6087 | int old_stripe_index = stripe_index; |
30d9861f SB |
6088 | stripe_index = find_live_mirror(fs_info, map, |
6089 | stripe_index, | |
30d9861f | 6090 | dev_replace_is_ongoing); |
3e74317a | 6091 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6092 | } |
53b381b3 | 6093 | |
ffe2d203 | 6094 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6095 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6096 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6097 | stripe_nr = div64_u64(raid56_full_stripe_start, |
b8b93add | 6098 | stripe_len * nr_data_stripes(map)); |
53b381b3 DW |
6099 | |
6100 | /* RAID[56] write or recovery. Return all stripes */ | |
6101 | num_stripes = map->num_stripes; | |
6102 | max_errors = nr_parity_stripes(map); | |
6103 | ||
53b381b3 DW |
6104 | *length = map->stripe_len; |
6105 | stripe_index = 0; | |
6106 | stripe_offset = 0; | |
6107 | } else { | |
6108 | /* | |
6109 | * Mirror #0 or #1 means the original data block. | |
6110 | * Mirror #2 is RAID5 parity block. | |
6111 | * Mirror #3 is RAID6 Q block. | |
6112 | */ | |
47c5713f DS |
6113 | stripe_nr = div_u64_rem(stripe_nr, |
6114 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
6115 | if (mirror_num > 1) |
6116 | stripe_index = nr_data_stripes(map) + | |
6117 | mirror_num - 2; | |
6118 | ||
6119 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6120 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6121 | &stripe_index); | |
de483734 | 6122 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6123 | mirror_num = 1; |
53b381b3 | 6124 | } |
8790d502 CM |
6125 | } else { |
6126 | /* | |
47c5713f DS |
6127 | * after this, stripe_nr is the number of stripes on this |
6128 | * device we have to walk to find the data, and stripe_index is | |
6129 | * the number of our device in the stripe array | |
8790d502 | 6130 | */ |
47c5713f DS |
6131 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6132 | &stripe_index); | |
a1d3c478 | 6133 | mirror_num = stripe_index + 1; |
8790d502 | 6134 | } |
e042d1ec | 6135 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6136 | btrfs_crit(fs_info, |
6137 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6138 | stripe_index, map->num_stripes); |
6139 | ret = -EINVAL; | |
6140 | goto out; | |
6141 | } | |
cea9e445 | 6142 | |
472262f3 | 6143 | num_alloc_stripes = num_stripes; |
6fad823f | 6144 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6145 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6146 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6147 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6148 | num_alloc_stripes++; |
2c8cdd6e | 6149 | tgtdev_indexes = num_stripes; |
ad6d620e | 6150 | } |
2c8cdd6e | 6151 | |
6e9606d2 | 6152 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6153 | if (!bbio) { |
6154 | ret = -ENOMEM; | |
6155 | goto out; | |
6156 | } | |
6fad823f | 6157 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) |
2c8cdd6e | 6158 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); |
de11cc12 | 6159 | |
8e5cfb55 | 6160 | /* build raid_map */ |
2b19a1fe LB |
6161 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6162 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6163 | u64 tmp; |
9d644a62 | 6164 | unsigned rot; |
8e5cfb55 ZL |
6165 | |
6166 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
6167 | sizeof(struct btrfs_bio_stripe) * | |
6168 | num_alloc_stripes + | |
6169 | sizeof(int) * tgtdev_indexes); | |
6170 | ||
6171 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 6172 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6173 | |
6174 | /* Fill in the logical address of each stripe */ | |
6175 | tmp = stripe_nr * nr_data_stripes(map); | |
6176 | for (i = 0; i < nr_data_stripes(map); i++) | |
6177 | bbio->raid_map[(i+rot) % num_stripes] = | |
6178 | em->start + (tmp + i) * map->stripe_len; | |
6179 | ||
6180 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6181 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6182 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6183 | RAID6_Q_STRIPE; | |
6184 | } | |
6185 | ||
b89203f7 | 6186 | |
0b3d4cd3 LB |
6187 | for (i = 0; i < num_stripes; i++) { |
6188 | bbio->stripes[i].physical = | |
6189 | map->stripes[stripe_index].physical + | |
6190 | stripe_offset + | |
6191 | stripe_nr * map->stripe_len; | |
6192 | bbio->stripes[i].dev = | |
6193 | map->stripes[stripe_index].dev; | |
6194 | stripe_index++; | |
593060d7 | 6195 | } |
de11cc12 | 6196 | |
2b19a1fe | 6197 | if (need_full_stripe(op)) |
d20983b4 | 6198 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6199 | |
8e5cfb55 ZL |
6200 | if (bbio->raid_map) |
6201 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 6202 | |
73c0f228 | 6203 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6204 | need_full_stripe(op)) { |
73c0f228 LB |
6205 | handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes, |
6206 | &max_errors); | |
472262f3 SB |
6207 | } |
6208 | ||
de11cc12 | 6209 | *bbio_ret = bbio; |
10f11900 | 6210 | bbio->map_type = map->type; |
de11cc12 LZ |
6211 | bbio->num_stripes = num_stripes; |
6212 | bbio->max_errors = max_errors; | |
6213 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6214 | |
6215 | /* | |
6216 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6217 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6218 | * available as a mirror | |
6219 | */ | |
6220 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6221 | WARN_ON(num_stripes > 1); | |
6222 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6223 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6224 | bbio->mirror_num = map->num_stripes + 1; | |
6225 | } | |
cea9e445 | 6226 | out: |
73beece9 | 6227 | if (dev_replace_is_ongoing) { |
53176dde DS |
6228 | lockdep_assert_held(&dev_replace->rwsem); |
6229 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6230 | up_read(&dev_replace->rwsem); |
73beece9 | 6231 | } |
0b86a832 | 6232 | free_extent_map(em); |
de11cc12 | 6233 | return ret; |
0b86a832 CM |
6234 | } |
6235 | ||
cf8cddd3 | 6236 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6237 | u64 logical, u64 *length, |
a1d3c478 | 6238 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6239 | { |
b3d3fa51 | 6240 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6241 | mirror_num, 0); |
f2d8d74d CM |
6242 | } |
6243 | ||
af8e2d1d | 6244 | /* For Scrub/replace */ |
cf8cddd3 | 6245 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6246 | u64 logical, u64 *length, |
825ad4c9 | 6247 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6248 | { |
825ad4c9 | 6249 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6250 | } |
6251 | ||
63a9c7b9 NB |
6252 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, u64 chunk_start, |
6253 | u64 physical, u64 **logical, int *naddrs, int *stripe_len) | |
a512bbf8 | 6254 | { |
a512bbf8 YZ |
6255 | struct extent_map *em; |
6256 | struct map_lookup *map; | |
6257 | u64 *buf; | |
6258 | u64 bytenr; | |
6259 | u64 length; | |
6260 | u64 stripe_nr; | |
53b381b3 | 6261 | u64 rmap_len; |
a512bbf8 YZ |
6262 | int i, j, nr = 0; |
6263 | ||
60ca842e | 6264 | em = btrfs_get_chunk_map(fs_info, chunk_start, 1); |
592d92ee | 6265 | if (IS_ERR(em)) |
835d974f | 6266 | return -EIO; |
835d974f | 6267 | |
95617d69 | 6268 | map = em->map_lookup; |
a512bbf8 | 6269 | length = em->len; |
53b381b3 DW |
6270 | rmap_len = map->stripe_len; |
6271 | ||
a512bbf8 | 6272 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 6273 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 6274 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 6275 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 6276 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 6277 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
6278 | rmap_len = map->stripe_len * nr_data_stripes(map); |
6279 | } | |
a512bbf8 | 6280 | |
31e818fe | 6281 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 6282 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
6283 | |
6284 | for (i = 0; i < map->num_stripes; i++) { | |
a512bbf8 YZ |
6285 | if (map->stripes[i].physical > physical || |
6286 | map->stripes[i].physical + length <= physical) | |
6287 | continue; | |
6288 | ||
6289 | stripe_nr = physical - map->stripes[i].physical; | |
42c61ab6 | 6290 | stripe_nr = div64_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
6291 | |
6292 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
6293 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 6294 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
6295 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6296 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
6297 | } /* else if RAID[56], multiply by nr_data_stripes(). |
6298 | * Alternatively, just use rmap_len below instead of | |
6299 | * map->stripe_len */ | |
6300 | ||
6301 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 6302 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
6303 | for (j = 0; j < nr; j++) { |
6304 | if (buf[j] == bytenr) | |
6305 | break; | |
6306 | } | |
934d375b CM |
6307 | if (j == nr) { |
6308 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 6309 | buf[nr++] = bytenr; |
934d375b | 6310 | } |
a512bbf8 YZ |
6311 | } |
6312 | ||
a512bbf8 YZ |
6313 | *logical = buf; |
6314 | *naddrs = nr; | |
53b381b3 | 6315 | *stripe_len = rmap_len; |
a512bbf8 YZ |
6316 | |
6317 | free_extent_map(em); | |
6318 | return 0; | |
f2d8d74d CM |
6319 | } |
6320 | ||
4246a0b6 | 6321 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6322 | { |
326e1dbb MS |
6323 | bio->bi_private = bbio->private; |
6324 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6325 | bio_endio(bio); |
326e1dbb | 6326 | |
6e9606d2 | 6327 | btrfs_put_bbio(bbio); |
8408c716 MX |
6328 | } |
6329 | ||
4246a0b6 | 6330 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6331 | { |
9be3395b | 6332 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6333 | int is_orig_bio = 0; |
8790d502 | 6334 | |
4e4cbee9 | 6335 | if (bio->bi_status) { |
a1d3c478 | 6336 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6337 | if (bio->bi_status == BLK_STS_IOERR || |
6338 | bio->bi_status == BLK_STS_TARGET) { | |
442a4f63 | 6339 | unsigned int stripe_index = |
9be3395b | 6340 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6341 | struct btrfs_device *dev; |
442a4f63 SB |
6342 | |
6343 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6344 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6345 | if (dev->bdev) { |
37226b21 | 6346 | if (bio_op(bio) == REQ_OP_WRITE) |
1cb34c8e | 6347 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6348 | BTRFS_DEV_STAT_WRITE_ERRS); |
0cc068e6 | 6349 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
1cb34c8e | 6350 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6351 | BTRFS_DEV_STAT_READ_ERRS); |
70fd7614 | 6352 | if (bio->bi_opf & REQ_PREFLUSH) |
1cb34c8e | 6353 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6354 | BTRFS_DEV_STAT_FLUSH_ERRS); |
597a60fa | 6355 | } |
442a4f63 SB |
6356 | } |
6357 | } | |
8790d502 | 6358 | |
a1d3c478 | 6359 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6360 | is_orig_bio = 1; |
6361 | ||
c404e0dc MX |
6362 | btrfs_bio_counter_dec(bbio->fs_info); |
6363 | ||
a1d3c478 | 6364 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6365 | if (!is_orig_bio) { |
6366 | bio_put(bio); | |
a1d3c478 | 6367 | bio = bbio->orig_bio; |
7d2b4daa | 6368 | } |
c7b22bb1 | 6369 | |
9be3395b | 6370 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6371 | /* only send an error to the higher layers if it is |
53b381b3 | 6372 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6373 | */ |
a1d3c478 | 6374 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6375 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6376 | } else { |
1259ab75 CM |
6377 | /* |
6378 | * this bio is actually up to date, we didn't | |
6379 | * go over the max number of errors | |
6380 | */ | |
2dbe0c77 | 6381 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6382 | } |
c55f1396 | 6383 | |
4246a0b6 | 6384 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6385 | } else if (!is_orig_bio) { |
8790d502 CM |
6386 | bio_put(bio); |
6387 | } | |
8790d502 CM |
6388 | } |
6389 | ||
8b712842 CM |
6390 | /* |
6391 | * see run_scheduled_bios for a description of why bios are collected for | |
6392 | * async submit. | |
6393 | * | |
6394 | * This will add one bio to the pending list for a device and make sure | |
6395 | * the work struct is scheduled. | |
6396 | */ | |
2ff7e61e | 6397 | static noinline void btrfs_schedule_bio(struct btrfs_device *device, |
4e49ea4a | 6398 | struct bio *bio) |
8b712842 | 6399 | { |
0b246afa | 6400 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6401 | int should_queue = 1; |
ffbd517d | 6402 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
6403 | |
6404 | /* don't bother with additional async steps for reads, right now */ | |
37226b21 | 6405 | if (bio_op(bio) == REQ_OP_READ) { |
4e49ea4a | 6406 | btrfsic_submit_bio(bio); |
143bede5 | 6407 | return; |
8b712842 CM |
6408 | } |
6409 | ||
492bb6de | 6410 | WARN_ON(bio->bi_next); |
8b712842 | 6411 | bio->bi_next = NULL; |
8b712842 CM |
6412 | |
6413 | spin_lock(&device->io_lock); | |
67f055c7 | 6414 | if (op_is_sync(bio->bi_opf)) |
ffbd517d CM |
6415 | pending_bios = &device->pending_sync_bios; |
6416 | else | |
6417 | pending_bios = &device->pending_bios; | |
8b712842 | 6418 | |
ffbd517d CM |
6419 | if (pending_bios->tail) |
6420 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6421 | |
ffbd517d CM |
6422 | pending_bios->tail = bio; |
6423 | if (!pending_bios->head) | |
6424 | pending_bios->head = bio; | |
8b712842 CM |
6425 | if (device->running_pending) |
6426 | should_queue = 0; | |
6427 | ||
6428 | spin_unlock(&device->io_lock); | |
6429 | ||
6430 | if (should_queue) | |
0b246afa | 6431 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6432 | } |
6433 | ||
2ff7e61e JM |
6434 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
6435 | u64 physical, int dev_nr, int async) | |
de1ee92a JB |
6436 | { |
6437 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
2ff7e61e | 6438 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6439 | |
6440 | bio->bi_private = bbio; | |
9be3395b | 6441 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6442 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6443 | bio->bi_iter.bi_sector = physical >> 9; |
672d5990 MT |
6444 | btrfs_debug_in_rcu(fs_info, |
6445 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6446 | bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector, | |
6447 | (u_long)dev->bdev->bd_dev, rcu_str_deref(dev->name), dev->devid, | |
6448 | bio->bi_iter.bi_size); | |
74d46992 | 6449 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6450 | |
2ff7e61e | 6451 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6452 | |
de1ee92a | 6453 | if (async) |
2ff7e61e | 6454 | btrfs_schedule_bio(dev, bio); |
de1ee92a | 6455 | else |
4e49ea4a | 6456 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6457 | } |
6458 | ||
de1ee92a JB |
6459 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6460 | { | |
6461 | atomic_inc(&bbio->error); | |
6462 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6463 | /* Should be the original bio. */ |
8408c716 MX |
6464 | WARN_ON(bio != bbio->orig_bio); |
6465 | ||
9be3395b | 6466 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6467 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6468 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6469 | bio->bi_status = BLK_STS_IOERR; | |
6470 | else | |
6471 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6472 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6473 | } |
6474 | } | |
6475 | ||
58efbc9f OS |
6476 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
6477 | int mirror_num, int async_submit) | |
0b86a832 | 6478 | { |
0b86a832 | 6479 | struct btrfs_device *dev; |
8790d502 | 6480 | struct bio *first_bio = bio; |
4f024f37 | 6481 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6482 | u64 length = 0; |
6483 | u64 map_length; | |
0b86a832 | 6484 | int ret; |
08da757d ZL |
6485 | int dev_nr; |
6486 | int total_devs; | |
a1d3c478 | 6487 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6488 | |
4f024f37 | 6489 | length = bio->bi_iter.bi_size; |
0b86a832 | 6490 | map_length = length; |
cea9e445 | 6491 | |
0b246afa | 6492 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6493 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6494 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6495 | if (ret) { |
0b246afa | 6496 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6497 | return errno_to_blk_status(ret); |
c404e0dc | 6498 | } |
cea9e445 | 6499 | |
a1d3c478 | 6500 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6501 | bbio->orig_bio = first_bio; |
6502 | bbio->private = first_bio->bi_private; | |
6503 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6504 | bbio->fs_info = fs_info; |
53b381b3 DW |
6505 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6506 | ||
ad1ba2a0 | 6507 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6508 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6509 | /* In this case, map_length has been set to the length of |
6510 | a single stripe; not the whole write */ | |
37226b21 | 6511 | if (bio_op(bio) == REQ_OP_WRITE) { |
2ff7e61e JM |
6512 | ret = raid56_parity_write(fs_info, bio, bbio, |
6513 | map_length); | |
53b381b3 | 6514 | } else { |
2ff7e61e JM |
6515 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6516 | map_length, mirror_num, 1); | |
53b381b3 | 6517 | } |
4245215d | 6518 | |
0b246afa | 6519 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6520 | return errno_to_blk_status(ret); |
53b381b3 DW |
6521 | } |
6522 | ||
cea9e445 | 6523 | if (map_length < length) { |
0b246afa | 6524 | btrfs_crit(fs_info, |
5d163e0e JM |
6525 | "mapping failed logical %llu bio len %llu len %llu", |
6526 | logical, length, map_length); | |
cea9e445 CM |
6527 | BUG(); |
6528 | } | |
a1d3c478 | 6529 | |
08da757d | 6530 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6531 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6532 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6533 | &dev->dev_state) || | |
ebbede42 AJ |
6534 | (bio_op(first_bio) == REQ_OP_WRITE && |
6535 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { | |
de1ee92a | 6536 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6537 | continue; |
6538 | } | |
6539 | ||
3aa8e074 | 6540 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6541 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6542 | else |
a1d3c478 | 6543 | bio = first_bio; |
de1ee92a | 6544 | |
2ff7e61e JM |
6545 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, |
6546 | dev_nr, async_submit); | |
8790d502 | 6547 | } |
0b246afa | 6548 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6549 | return BLK_STS_OK; |
0b86a832 CM |
6550 | } |
6551 | ||
09ba3bc9 AJ |
6552 | /* |
6553 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6554 | * return NULL. | |
6555 | * | |
6556 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6557 | * only devid is used. | |
6558 | * | |
6559 | * If @seed is true, traverse through the seed devices. | |
6560 | */ | |
e4319cd9 | 6561 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
09ba3bc9 AJ |
6562 | u64 devid, u8 *uuid, u8 *fsid, |
6563 | bool seed) | |
0b86a832 | 6564 | { |
2b82032c | 6565 | struct btrfs_device *device; |
2b82032c | 6566 | |
e4319cd9 | 6567 | while (fs_devices) { |
2b82032c | 6568 | if (!fsid || |
e4319cd9 | 6569 | !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
09ba3bc9 AJ |
6570 | list_for_each_entry(device, &fs_devices->devices, |
6571 | dev_list) { | |
6572 | if (device->devid == devid && | |
6573 | (!uuid || memcmp(device->uuid, uuid, | |
6574 | BTRFS_UUID_SIZE) == 0)) | |
6575 | return device; | |
6576 | } | |
2b82032c | 6577 | } |
09ba3bc9 AJ |
6578 | if (seed) |
6579 | fs_devices = fs_devices->seed; | |
6580 | else | |
6581 | return NULL; | |
2b82032c YZ |
6582 | } |
6583 | return NULL; | |
0b86a832 CM |
6584 | } |
6585 | ||
2ff7e61e | 6586 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6587 | u64 devid, u8 *dev_uuid) |
6588 | { | |
6589 | struct btrfs_device *device; | |
dfe25020 | 6590 | |
12bd2fc0 ID |
6591 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6592 | if (IS_ERR(device)) | |
adfb69af | 6593 | return device; |
12bd2fc0 ID |
6594 | |
6595 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6596 | device->fs_devices = fs_devices; |
dfe25020 | 6597 | fs_devices->num_devices++; |
12bd2fc0 | 6598 | |
e6e674bd | 6599 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6600 | fs_devices->missing_devices++; |
12bd2fc0 | 6601 | |
dfe25020 CM |
6602 | return device; |
6603 | } | |
6604 | ||
12bd2fc0 ID |
6605 | /** |
6606 | * btrfs_alloc_device - allocate struct btrfs_device | |
6607 | * @fs_info: used only for generating a new devid, can be NULL if | |
6608 | * devid is provided (i.e. @devid != NULL). | |
6609 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6610 | * is generated. | |
6611 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6612 | * is generated. | |
6613 | * | |
6614 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6615 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6616 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6617 | */ |
6618 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6619 | const u64 *devid, | |
6620 | const u8 *uuid) | |
6621 | { | |
6622 | struct btrfs_device *dev; | |
6623 | u64 tmp; | |
6624 | ||
fae7f21c | 6625 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6626 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6627 | |
6628 | dev = __alloc_device(); | |
6629 | if (IS_ERR(dev)) | |
6630 | return dev; | |
6631 | ||
6632 | if (devid) | |
6633 | tmp = *devid; | |
6634 | else { | |
6635 | int ret; | |
6636 | ||
6637 | ret = find_next_devid(fs_info, &tmp); | |
6638 | if (ret) { | |
a425f9d4 | 6639 | btrfs_free_device(dev); |
12bd2fc0 ID |
6640 | return ERR_PTR(ret); |
6641 | } | |
6642 | } | |
6643 | dev->devid = tmp; | |
6644 | ||
6645 | if (uuid) | |
6646 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6647 | else | |
6648 | generate_random_uuid(dev->uuid); | |
6649 | ||
9e0af237 LB |
6650 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6651 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6652 | |
6653 | return dev; | |
6654 | } | |
6655 | ||
5a2b8e60 | 6656 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6657 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6658 | { |
2b902dfc AJ |
6659 | if (error) |
6660 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6661 | devid, uuid); | |
6662 | else | |
6663 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6664 | devid, uuid); | |
5a2b8e60 AJ |
6665 | } |
6666 | ||
39e264a4 NB |
6667 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6668 | { | |
6669 | int index = btrfs_bg_flags_to_raid_index(type); | |
6670 | int ncopies = btrfs_raid_array[index].ncopies; | |
6671 | int data_stripes; | |
6672 | ||
6673 | switch (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
6674 | case BTRFS_BLOCK_GROUP_RAID5: | |
6675 | data_stripes = num_stripes - 1; | |
6676 | break; | |
6677 | case BTRFS_BLOCK_GROUP_RAID6: | |
6678 | data_stripes = num_stripes - 2; | |
6679 | break; | |
6680 | default: | |
6681 | data_stripes = num_stripes / ncopies; | |
6682 | break; | |
6683 | } | |
6684 | return div_u64(chunk_len, data_stripes); | |
6685 | } | |
6686 | ||
9690ac09 | 6687 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6688 | struct btrfs_chunk *chunk) |
6689 | { | |
9690ac09 | 6690 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6691 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6692 | struct map_lookup *map; |
6693 | struct extent_map *em; | |
6694 | u64 logical; | |
6695 | u64 length; | |
e06cd3dd LB |
6696 | u64 devid; |
6697 | u8 uuid[BTRFS_UUID_SIZE]; | |
6698 | int num_stripes; | |
6699 | int ret; | |
6700 | int i; | |
6701 | ||
6702 | logical = key->offset; | |
6703 | length = btrfs_chunk_length(leaf, chunk); | |
e06cd3dd LB |
6704 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6705 | ||
075cb3c7 QW |
6706 | /* |
6707 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6708 | * as chunk item in tree block is already verified by tree-checker. | |
6709 | */ | |
6710 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6711 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6712 | if (ret) |
6713 | return ret; | |
6714 | } | |
a061fc8d | 6715 | |
890871be | 6716 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6717 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6718 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6719 | |
6720 | /* already mapped? */ | |
6721 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6722 | free_extent_map(em); | |
0b86a832 CM |
6723 | return 0; |
6724 | } else if (em) { | |
6725 | free_extent_map(em); | |
6726 | } | |
0b86a832 | 6727 | |
172ddd60 | 6728 | em = alloc_extent_map(); |
0b86a832 CM |
6729 | if (!em) |
6730 | return -ENOMEM; | |
593060d7 | 6731 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6732 | if (!map) { |
6733 | free_extent_map(em); | |
6734 | return -ENOMEM; | |
6735 | } | |
6736 | ||
298a8f9c | 6737 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6738 | em->map_lookup = map; |
0b86a832 CM |
6739 | em->start = logical; |
6740 | em->len = length; | |
70c8a91c | 6741 | em->orig_start = 0; |
0b86a832 | 6742 | em->block_start = 0; |
c8b97818 | 6743 | em->block_len = em->len; |
0b86a832 | 6744 | |
593060d7 CM |
6745 | map->num_stripes = num_stripes; |
6746 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6747 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 CM |
6748 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6749 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6750 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6751 | map->verified_stripes = 0; |
39e264a4 NB |
6752 | em->orig_block_len = calc_stripe_length(map->type, em->len, |
6753 | map->num_stripes); | |
593060d7 CM |
6754 | for (i = 0; i < num_stripes; i++) { |
6755 | map->stripes[i].physical = | |
6756 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6757 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6758 | read_extent_buffer(leaf, uuid, (unsigned long) |
6759 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6760 | BTRFS_UUID_SIZE); | |
e4319cd9 | 6761 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
09ba3bc9 | 6762 | devid, uuid, NULL, true); |
3cdde224 | 6763 | if (!map->stripes[i].dev && |
0b246afa | 6764 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 6765 | free_extent_map(em); |
2b902dfc | 6766 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 6767 | return -ENOENT; |
593060d7 | 6768 | } |
dfe25020 CM |
6769 | if (!map->stripes[i].dev) { |
6770 | map->stripes[i].dev = | |
2ff7e61e JM |
6771 | add_missing_dev(fs_info->fs_devices, devid, |
6772 | uuid); | |
adfb69af | 6773 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 6774 | free_extent_map(em); |
adfb69af AJ |
6775 | btrfs_err(fs_info, |
6776 | "failed to init missing dev %llu: %ld", | |
6777 | devid, PTR_ERR(map->stripes[i].dev)); | |
6778 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 6779 | } |
2b902dfc | 6780 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 6781 | } |
e12c9621 AJ |
6782 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6783 | &(map->stripes[i].dev->dev_state)); | |
6784 | ||
0b86a832 CM |
6785 | } |
6786 | ||
890871be | 6787 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6788 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6789 | write_unlock(&map_tree->map_tree.lock); |
64f64f43 QW |
6790 | if (ret < 0) { |
6791 | btrfs_err(fs_info, | |
6792 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6793 | em->start, em->len, ret); | |
6794 | } | |
0b86a832 CM |
6795 | free_extent_map(em); |
6796 | ||
64f64f43 | 6797 | return ret; |
0b86a832 CM |
6798 | } |
6799 | ||
143bede5 | 6800 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6801 | struct btrfs_dev_item *dev_item, |
6802 | struct btrfs_device *device) | |
6803 | { | |
6804 | unsigned long ptr; | |
0b86a832 CM |
6805 | |
6806 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6807 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6808 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6809 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6810 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6811 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6812 | device->type = btrfs_device_type(leaf, dev_item); |
6813 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6814 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6815 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6816 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6817 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6818 | |
410ba3a2 | 6819 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6820 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6821 | } |
6822 | ||
2ff7e61e | 6823 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6824 | u8 *fsid) |
2b82032c YZ |
6825 | { |
6826 | struct btrfs_fs_devices *fs_devices; | |
6827 | int ret; | |
6828 | ||
a32bf9a3 | 6829 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6830 | ASSERT(fsid); |
2b82032c | 6831 | |
0b246afa | 6832 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6833 | while (fs_devices) { |
44880fdc | 6834 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6835 | return fs_devices; |
6836 | ||
2b82032c YZ |
6837 | fs_devices = fs_devices->seed; |
6838 | } | |
6839 | ||
7239ff4b | 6840 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6841 | if (!fs_devices) { |
0b246afa | 6842 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6843 | return ERR_PTR(-ENOENT); |
6844 | ||
7239ff4b | 6845 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6846 | if (IS_ERR(fs_devices)) |
6847 | return fs_devices; | |
6848 | ||
6849 | fs_devices->seeding = 1; | |
6850 | fs_devices->opened = 1; | |
6851 | return fs_devices; | |
2b82032c | 6852 | } |
e4404d6e YZ |
6853 | |
6854 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6855 | if (IS_ERR(fs_devices)) |
6856 | return fs_devices; | |
2b82032c | 6857 | |
897fb573 | 6858 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
6859 | if (ret) { |
6860 | free_fs_devices(fs_devices); | |
5f375835 | 6861 | fs_devices = ERR_PTR(ret); |
2b82032c | 6862 | goto out; |
48d28232 | 6863 | } |
2b82032c YZ |
6864 | |
6865 | if (!fs_devices->seeding) { | |
0226e0eb | 6866 | close_fs_devices(fs_devices); |
e4404d6e | 6867 | free_fs_devices(fs_devices); |
5f375835 | 6868 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6869 | goto out; |
6870 | } | |
6871 | ||
0b246afa JM |
6872 | fs_devices->seed = fs_info->fs_devices->seed; |
6873 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6874 | out: |
5f375835 | 6875 | return fs_devices; |
2b82032c YZ |
6876 | } |
6877 | ||
17850759 | 6878 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6879 | struct btrfs_dev_item *dev_item) |
6880 | { | |
17850759 | 6881 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6882 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6883 | struct btrfs_device *device; |
6884 | u64 devid; | |
6885 | int ret; | |
44880fdc | 6886 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6887 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6888 | ||
0b86a832 | 6889 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6890 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6891 | BTRFS_UUID_SIZE); |
1473b24e | 6892 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6893 | BTRFS_FSID_SIZE); |
2b82032c | 6894 | |
de37aa51 | 6895 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6896 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6897 | if (IS_ERR(fs_devices)) |
6898 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6899 | } |
6900 | ||
e4319cd9 | 6901 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
09ba3bc9 | 6902 | fs_uuid, true); |
5f375835 | 6903 | if (!device) { |
c5502451 | 6904 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6905 | btrfs_report_missing_device(fs_info, devid, |
6906 | dev_uuid, true); | |
45dbdbc9 | 6907 | return -ENOENT; |
c5502451 | 6908 | } |
2b82032c | 6909 | |
2ff7e61e | 6910 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6911 | if (IS_ERR(device)) { |
6912 | btrfs_err(fs_info, | |
6913 | "failed to add missing dev %llu: %ld", | |
6914 | devid, PTR_ERR(device)); | |
6915 | return PTR_ERR(device); | |
6916 | } | |
2b902dfc | 6917 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6918 | } else { |
c5502451 | 6919 | if (!device->bdev) { |
2b902dfc AJ |
6920 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6921 | btrfs_report_missing_device(fs_info, | |
6922 | devid, dev_uuid, true); | |
45dbdbc9 | 6923 | return -ENOENT; |
2b902dfc AJ |
6924 | } |
6925 | btrfs_report_missing_device(fs_info, devid, | |
6926 | dev_uuid, false); | |
c5502451 | 6927 | } |
5f375835 | 6928 | |
e6e674bd AJ |
6929 | if (!device->bdev && |
6930 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6931 | /* |
6932 | * this happens when a device that was properly setup | |
6933 | * in the device info lists suddenly goes bad. | |
6934 | * device->bdev is NULL, and so we have to set | |
6935 | * device->missing to one here | |
6936 | */ | |
5f375835 | 6937 | device->fs_devices->missing_devices++; |
e6e674bd | 6938 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6939 | } |
5f375835 MX |
6940 | |
6941 | /* Move the device to its own fs_devices */ | |
6942 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6943 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6944 | &device->dev_state)); | |
5f375835 MX |
6945 | |
6946 | list_move(&device->dev_list, &fs_devices->devices); | |
6947 | device->fs_devices->num_devices--; | |
6948 | fs_devices->num_devices++; | |
6949 | ||
6950 | device->fs_devices->missing_devices--; | |
6951 | fs_devices->missing_devices++; | |
6952 | ||
6953 | device->fs_devices = fs_devices; | |
6954 | } | |
2b82032c YZ |
6955 | } |
6956 | ||
0b246afa | 6957 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 6958 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
6959 | if (device->generation != |
6960 | btrfs_device_generation(leaf, dev_item)) | |
6961 | return -EINVAL; | |
6324fbf3 | 6962 | } |
0b86a832 CM |
6963 | |
6964 | fill_device_from_item(leaf, dev_item, device); | |
e12c9621 | 6965 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 6966 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 6967 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 6968 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
6969 | atomic64_add(device->total_bytes - device->bytes_used, |
6970 | &fs_info->free_chunk_space); | |
2bf64758 | 6971 | } |
0b86a832 | 6972 | ret = 0; |
0b86a832 CM |
6973 | return ret; |
6974 | } | |
6975 | ||
6bccf3ab | 6976 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6977 | { |
6bccf3ab | 6978 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6979 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6980 | struct extent_buffer *sb; |
0b86a832 | 6981 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6982 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6983 | u8 *array_ptr; |
6984 | unsigned long sb_array_offset; | |
84eed90f | 6985 | int ret = 0; |
0b86a832 CM |
6986 | u32 num_stripes; |
6987 | u32 array_size; | |
6988 | u32 len = 0; | |
1ffb22cf | 6989 | u32 cur_offset; |
e06cd3dd | 6990 | u64 type; |
84eed90f | 6991 | struct btrfs_key key; |
0b86a832 | 6992 | |
0b246afa | 6993 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6994 | /* |
6995 | * This will create extent buffer of nodesize, superblock size is | |
6996 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6997 | * overallocate but we can keep it as-is, only the first page is used. | |
6998 | */ | |
2ff7e61e | 6999 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET); |
c871b0f2 LB |
7000 | if (IS_ERR(sb)) |
7001 | return PTR_ERR(sb); | |
4db8c528 | 7002 | set_extent_buffer_uptodate(sb); |
85d4e461 | 7003 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 7004 | /* |
01327610 | 7005 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7006 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7007 | * pages up-to-date when the page is larger: extent does not cover the |
7008 | * whole page and consequently check_page_uptodate does not find all | |
7009 | * the page's extents up-to-date (the hole beyond sb), | |
7010 | * write_extent_buffer then triggers a WARN_ON. | |
7011 | * | |
7012 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7013 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7014 | * to silence the warning eg. on PowerPC 64. | |
7015 | */ | |
09cbfeaf | 7016 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7017 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7018 | |
a061fc8d | 7019 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7020 | array_size = btrfs_super_sys_array_size(super_copy); |
7021 | ||
1ffb22cf DS |
7022 | array_ptr = super_copy->sys_chunk_array; |
7023 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7024 | cur_offset = 0; | |
0b86a832 | 7025 | |
1ffb22cf DS |
7026 | while (cur_offset < array_size) { |
7027 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7028 | len = sizeof(*disk_key); |
7029 | if (cur_offset + len > array_size) | |
7030 | goto out_short_read; | |
7031 | ||
0b86a832 CM |
7032 | btrfs_disk_key_to_cpu(&key, disk_key); |
7033 | ||
1ffb22cf DS |
7034 | array_ptr += len; |
7035 | sb_array_offset += len; | |
7036 | cur_offset += len; | |
0b86a832 | 7037 | |
0d81ba5d | 7038 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 7039 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
7040 | /* |
7041 | * At least one btrfs_chunk with one stripe must be | |
7042 | * present, exact stripe count check comes afterwards | |
7043 | */ | |
7044 | len = btrfs_chunk_item_size(1); | |
7045 | if (cur_offset + len > array_size) | |
7046 | goto out_short_read; | |
7047 | ||
7048 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 7049 | if (!num_stripes) { |
ab8d0fc4 JM |
7050 | btrfs_err(fs_info, |
7051 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
7052 | num_stripes, cur_offset); |
7053 | ret = -EIO; | |
7054 | break; | |
7055 | } | |
7056 | ||
e06cd3dd LB |
7057 | type = btrfs_chunk_type(sb, chunk); |
7058 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7059 | btrfs_err(fs_info, |
e06cd3dd LB |
7060 | "invalid chunk type %llu in sys_array at offset %u", |
7061 | type, cur_offset); | |
7062 | ret = -EIO; | |
7063 | break; | |
7064 | } | |
7065 | ||
e3540eab DS |
7066 | len = btrfs_chunk_item_size(num_stripes); |
7067 | if (cur_offset + len > array_size) | |
7068 | goto out_short_read; | |
7069 | ||
9690ac09 | 7070 | ret = read_one_chunk(&key, sb, chunk); |
84eed90f CM |
7071 | if (ret) |
7072 | break; | |
0b86a832 | 7073 | } else { |
ab8d0fc4 JM |
7074 | btrfs_err(fs_info, |
7075 | "unexpected item type %u in sys_array at offset %u", | |
7076 | (u32)key.type, cur_offset); | |
84eed90f CM |
7077 | ret = -EIO; |
7078 | break; | |
0b86a832 | 7079 | } |
1ffb22cf DS |
7080 | array_ptr += len; |
7081 | sb_array_offset += len; | |
7082 | cur_offset += len; | |
0b86a832 | 7083 | } |
d865177a | 7084 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7085 | free_extent_buffer_stale(sb); |
84eed90f | 7086 | return ret; |
e3540eab DS |
7087 | |
7088 | out_short_read: | |
ab8d0fc4 | 7089 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7090 | len, cur_offset); |
d865177a | 7091 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7092 | free_extent_buffer_stale(sb); |
e3540eab | 7093 | return -EIO; |
0b86a832 CM |
7094 | } |
7095 | ||
21634a19 QW |
7096 | /* |
7097 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7098 | * | |
6528b99d AJ |
7099 | * If the @failing_dev is specified, it's accounted as missing. |
7100 | * | |
21634a19 QW |
7101 | * Return true if all chunks meet the minimal RW mount requirements. |
7102 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7103 | */ | |
6528b99d AJ |
7104 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7105 | struct btrfs_device *failing_dev) | |
21634a19 QW |
7106 | { |
7107 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
7108 | struct extent_map *em; | |
7109 | u64 next_start = 0; | |
7110 | bool ret = true; | |
7111 | ||
7112 | read_lock(&map_tree->map_tree.lock); | |
7113 | em = lookup_extent_mapping(&map_tree->map_tree, 0, (u64)-1); | |
7114 | read_unlock(&map_tree->map_tree.lock); | |
7115 | /* No chunk at all? Return false anyway */ | |
7116 | if (!em) { | |
7117 | ret = false; | |
7118 | goto out; | |
7119 | } | |
7120 | while (em) { | |
7121 | struct map_lookup *map; | |
7122 | int missing = 0; | |
7123 | int max_tolerated; | |
7124 | int i; | |
7125 | ||
7126 | map = em->map_lookup; | |
7127 | max_tolerated = | |
7128 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7129 | map->type); | |
7130 | for (i = 0; i < map->num_stripes; i++) { | |
7131 | struct btrfs_device *dev = map->stripes[i].dev; | |
7132 | ||
e6e674bd AJ |
7133 | if (!dev || !dev->bdev || |
7134 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7135 | dev->last_flush_error) |
7136 | missing++; | |
6528b99d AJ |
7137 | else if (failing_dev && failing_dev == dev) |
7138 | missing++; | |
21634a19 QW |
7139 | } |
7140 | if (missing > max_tolerated) { | |
6528b99d AJ |
7141 | if (!failing_dev) |
7142 | btrfs_warn(fs_info, | |
52042d8e | 7143 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7144 | em->start, missing, max_tolerated); |
7145 | free_extent_map(em); | |
7146 | ret = false; | |
7147 | goto out; | |
7148 | } | |
7149 | next_start = extent_map_end(em); | |
7150 | free_extent_map(em); | |
7151 | ||
7152 | read_lock(&map_tree->map_tree.lock); | |
7153 | em = lookup_extent_mapping(&map_tree->map_tree, next_start, | |
7154 | (u64)(-1) - next_start); | |
7155 | read_unlock(&map_tree->map_tree.lock); | |
7156 | } | |
7157 | out: | |
7158 | return ret; | |
7159 | } | |
7160 | ||
5b4aacef | 7161 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7162 | { |
5b4aacef | 7163 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7164 | struct btrfs_path *path; |
7165 | struct extent_buffer *leaf; | |
7166 | struct btrfs_key key; | |
7167 | struct btrfs_key found_key; | |
7168 | int ret; | |
7169 | int slot; | |
99e3ecfc | 7170 | u64 total_dev = 0; |
0b86a832 | 7171 | |
0b86a832 CM |
7172 | path = btrfs_alloc_path(); |
7173 | if (!path) | |
7174 | return -ENOMEM; | |
7175 | ||
3dd0f7a3 AJ |
7176 | /* |
7177 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7178 | * otherwise we don't need it. | |
7179 | */ | |
b367e47f | 7180 | mutex_lock(&uuid_mutex); |
34441361 | 7181 | mutex_lock(&fs_info->chunk_mutex); |
b367e47f | 7182 | |
395927a9 FDBM |
7183 | /* |
7184 | * Read all device items, and then all the chunk items. All | |
7185 | * device items are found before any chunk item (their object id | |
7186 | * is smaller than the lowest possible object id for a chunk | |
7187 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7188 | */ |
7189 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7190 | key.offset = 0; | |
7191 | key.type = 0; | |
0b86a832 | 7192 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7193 | if (ret < 0) |
7194 | goto error; | |
d397712b | 7195 | while (1) { |
0b86a832 CM |
7196 | leaf = path->nodes[0]; |
7197 | slot = path->slots[0]; | |
7198 | if (slot >= btrfs_header_nritems(leaf)) { | |
7199 | ret = btrfs_next_leaf(root, path); | |
7200 | if (ret == 0) | |
7201 | continue; | |
7202 | if (ret < 0) | |
7203 | goto error; | |
7204 | break; | |
7205 | } | |
7206 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
7207 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7208 | struct btrfs_dev_item *dev_item; | |
7209 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7210 | struct btrfs_dev_item); |
17850759 | 7211 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7212 | if (ret) |
7213 | goto error; | |
99e3ecfc | 7214 | total_dev++; |
0b86a832 CM |
7215 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7216 | struct btrfs_chunk *chunk; | |
7217 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
9690ac09 | 7218 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7219 | if (ret) |
7220 | goto error; | |
0b86a832 CM |
7221 | } |
7222 | path->slots[0]++; | |
7223 | } | |
99e3ecfc LB |
7224 | |
7225 | /* | |
7226 | * After loading chunk tree, we've got all device information, | |
7227 | * do another round of validation checks. | |
7228 | */ | |
0b246afa JM |
7229 | if (total_dev != fs_info->fs_devices->total_devices) { |
7230 | btrfs_err(fs_info, | |
99e3ecfc | 7231 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7232 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7233 | total_dev); |
7234 | ret = -EINVAL; | |
7235 | goto error; | |
7236 | } | |
0b246afa JM |
7237 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7238 | fs_info->fs_devices->total_rw_bytes) { | |
7239 | btrfs_err(fs_info, | |
99e3ecfc | 7240 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7241 | btrfs_super_total_bytes(fs_info->super_copy), |
7242 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7243 | ret = -EINVAL; |
7244 | goto error; | |
7245 | } | |
0b86a832 CM |
7246 | ret = 0; |
7247 | error: | |
34441361 | 7248 | mutex_unlock(&fs_info->chunk_mutex); |
b367e47f LZ |
7249 | mutex_unlock(&uuid_mutex); |
7250 | ||
2b82032c | 7251 | btrfs_free_path(path); |
0b86a832 CM |
7252 | return ret; |
7253 | } | |
442a4f63 | 7254 | |
cb517eab MX |
7255 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7256 | { | |
7257 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7258 | struct btrfs_device *device; | |
7259 | ||
29cc83f6 LB |
7260 | while (fs_devices) { |
7261 | mutex_lock(&fs_devices->device_list_mutex); | |
7262 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 7263 | device->fs_info = fs_info; |
29cc83f6 LB |
7264 | mutex_unlock(&fs_devices->device_list_mutex); |
7265 | ||
7266 | fs_devices = fs_devices->seed; | |
7267 | } | |
cb517eab MX |
7268 | } |
7269 | ||
733f4fbb SB |
7270 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
7271 | { | |
7272 | int i; | |
7273 | ||
7274 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7275 | btrfs_dev_stat_reset(dev, i); | |
7276 | } | |
7277 | ||
7278 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7279 | { | |
7280 | struct btrfs_key key; | |
7281 | struct btrfs_key found_key; | |
7282 | struct btrfs_root *dev_root = fs_info->dev_root; | |
7283 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7284 | struct extent_buffer *eb; | |
7285 | int slot; | |
7286 | int ret = 0; | |
7287 | struct btrfs_device *device; | |
7288 | struct btrfs_path *path = NULL; | |
7289 | int i; | |
7290 | ||
7291 | path = btrfs_alloc_path(); | |
7292 | if (!path) { | |
7293 | ret = -ENOMEM; | |
7294 | goto out; | |
7295 | } | |
7296 | ||
7297 | mutex_lock(&fs_devices->device_list_mutex); | |
7298 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
7299 | int item_size; | |
7300 | struct btrfs_dev_stats_item *ptr; | |
7301 | ||
242e2956 DS |
7302 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7303 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7304 | key.offset = device->devid; |
7305 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
7306 | if (ret) { | |
733f4fbb SB |
7307 | __btrfs_reset_dev_stats(device); |
7308 | device->dev_stats_valid = 1; | |
7309 | btrfs_release_path(path); | |
7310 | continue; | |
7311 | } | |
7312 | slot = path->slots[0]; | |
7313 | eb = path->nodes[0]; | |
7314 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
7315 | item_size = btrfs_item_size_nr(eb, slot); | |
7316 | ||
7317 | ptr = btrfs_item_ptr(eb, slot, | |
7318 | struct btrfs_dev_stats_item); | |
7319 | ||
7320 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7321 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7322 | btrfs_dev_stat_set(device, i, | |
7323 | btrfs_dev_stats_value(eb, ptr, i)); | |
7324 | else | |
7325 | btrfs_dev_stat_reset(device, i); | |
7326 | } | |
7327 | ||
7328 | device->dev_stats_valid = 1; | |
7329 | btrfs_dev_stat_print_on_load(device); | |
7330 | btrfs_release_path(path); | |
7331 | } | |
7332 | mutex_unlock(&fs_devices->device_list_mutex); | |
7333 | ||
7334 | out: | |
7335 | btrfs_free_path(path); | |
7336 | return ret < 0 ? ret : 0; | |
7337 | } | |
7338 | ||
7339 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7340 | struct btrfs_device *device) |
7341 | { | |
5495f195 | 7342 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7343 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7344 | struct btrfs_path *path; |
7345 | struct btrfs_key key; | |
7346 | struct extent_buffer *eb; | |
7347 | struct btrfs_dev_stats_item *ptr; | |
7348 | int ret; | |
7349 | int i; | |
7350 | ||
242e2956 DS |
7351 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7352 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7353 | key.offset = device->devid; |
7354 | ||
7355 | path = btrfs_alloc_path(); | |
fa252992 DS |
7356 | if (!path) |
7357 | return -ENOMEM; | |
733f4fbb SB |
7358 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7359 | if (ret < 0) { | |
0b246afa | 7360 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7361 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7362 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7363 | goto out; |
7364 | } | |
7365 | ||
7366 | if (ret == 0 && | |
7367 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7368 | /* need to delete old one and insert a new one */ | |
7369 | ret = btrfs_del_item(trans, dev_root, path); | |
7370 | if (ret != 0) { | |
0b246afa | 7371 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7372 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7373 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7374 | goto out; |
7375 | } | |
7376 | ret = 1; | |
7377 | } | |
7378 | ||
7379 | if (ret == 1) { | |
7380 | /* need to insert a new item */ | |
7381 | btrfs_release_path(path); | |
7382 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7383 | &key, sizeof(*ptr)); | |
7384 | if (ret < 0) { | |
0b246afa | 7385 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7386 | "insert dev_stats item for device %s failed %d", |
7387 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7388 | goto out; |
7389 | } | |
7390 | } | |
7391 | ||
7392 | eb = path->nodes[0]; | |
7393 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7394 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7395 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7396 | btrfs_dev_stat_read(device, i)); | |
7397 | btrfs_mark_buffer_dirty(eb); | |
7398 | ||
7399 | out: | |
7400 | btrfs_free_path(path); | |
7401 | return ret; | |
7402 | } | |
7403 | ||
7404 | /* | |
7405 | * called from commit_transaction. Writes all changed device stats to disk. | |
7406 | */ | |
196c9d8d | 7407 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7408 | { |
196c9d8d | 7409 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7410 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7411 | struct btrfs_device *device; | |
addc3fa7 | 7412 | int stats_cnt; |
733f4fbb SB |
7413 | int ret = 0; |
7414 | ||
7415 | mutex_lock(&fs_devices->device_list_mutex); | |
7416 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7417 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7418 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7419 | continue; |
7420 | ||
9deae968 NB |
7421 | |
7422 | /* | |
7423 | * There is a LOAD-LOAD control dependency between the value of | |
7424 | * dev_stats_ccnt and updating the on-disk values which requires | |
7425 | * reading the in-memory counters. Such control dependencies | |
7426 | * require explicit read memory barriers. | |
7427 | * | |
7428 | * This memory barriers pairs with smp_mb__before_atomic in | |
7429 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7430 | * barrier implied by atomic_xchg in | |
7431 | * btrfs_dev_stats_read_and_reset | |
7432 | */ | |
7433 | smp_rmb(); | |
7434 | ||
5495f195 | 7435 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7436 | if (!ret) |
addc3fa7 | 7437 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7438 | } |
7439 | mutex_unlock(&fs_devices->device_list_mutex); | |
7440 | ||
7441 | return ret; | |
7442 | } | |
7443 | ||
442a4f63 SB |
7444 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7445 | { | |
7446 | btrfs_dev_stat_inc(dev, index); | |
7447 | btrfs_dev_stat_print_on_error(dev); | |
7448 | } | |
7449 | ||
48a3b636 | 7450 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7451 | { |
733f4fbb SB |
7452 | if (!dev->dev_stats_valid) |
7453 | return; | |
fb456252 | 7454 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7455 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7456 | rcu_str_deref(dev->name), |
442a4f63 SB |
7457 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7458 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7459 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7460 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7461 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7462 | } |
c11d2c23 | 7463 | |
733f4fbb SB |
7464 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7465 | { | |
a98cdb85 SB |
7466 | int i; |
7467 | ||
7468 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7469 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7470 | break; | |
7471 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7472 | return; /* all values == 0, suppress message */ | |
7473 | ||
fb456252 | 7474 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7475 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7476 | rcu_str_deref(dev->name), |
733f4fbb SB |
7477 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7478 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7479 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7480 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7481 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7482 | } | |
7483 | ||
2ff7e61e | 7484 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7485 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7486 | { |
7487 | struct btrfs_device *dev; | |
0b246afa | 7488 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7489 | int i; |
7490 | ||
7491 | mutex_lock(&fs_devices->device_list_mutex); | |
09ba3bc9 AJ |
7492 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL, |
7493 | true); | |
c11d2c23 SB |
7494 | mutex_unlock(&fs_devices->device_list_mutex); |
7495 | ||
7496 | if (!dev) { | |
0b246afa | 7497 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7498 | return -ENODEV; |
733f4fbb | 7499 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7500 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7501 | return -ENODEV; |
b27f7c0c | 7502 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7503 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7504 | if (stats->nr_items > i) | |
7505 | stats->values[i] = | |
7506 | btrfs_dev_stat_read_and_reset(dev, i); | |
7507 | else | |
7508 | btrfs_dev_stat_reset(dev, i); | |
7509 | } | |
7510 | } else { | |
7511 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7512 | if (stats->nr_items > i) | |
7513 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7514 | } | |
7515 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7516 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7517 | return 0; | |
7518 | } | |
a8a6dab7 | 7519 | |
da353f6b | 7520 | void btrfs_scratch_superblocks(struct block_device *bdev, const char *device_path) |
a8a6dab7 SB |
7521 | { |
7522 | struct buffer_head *bh; | |
7523 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7524 | int copy_num; |
a8a6dab7 | 7525 | |
12b1c263 AJ |
7526 | if (!bdev) |
7527 | return; | |
a8a6dab7 | 7528 | |
12b1c263 AJ |
7529 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7530 | copy_num++) { | |
a8a6dab7 | 7531 | |
12b1c263 AJ |
7532 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7533 | continue; | |
7534 | ||
7535 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7536 | ||
7537 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7538 | set_buffer_dirty(bh); | |
7539 | sync_dirty_buffer(bh); | |
7540 | brelse(bh); | |
7541 | } | |
7542 | ||
7543 | /* Notify udev that device has changed */ | |
7544 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7545 | ||
7546 | /* Update ctime/mtime for device path for libblkid */ | |
7547 | update_dev_time(device_path); | |
a8a6dab7 | 7548 | } |
935e5cc9 MX |
7549 | |
7550 | /* | |
bbbf7243 NB |
7551 | * Update the size and bytes used for each device where it changed. This is |
7552 | * delayed since we would otherwise get errors while writing out the | |
7553 | * superblocks. | |
7554 | * | |
7555 | * Must be invoked during transaction commit. | |
935e5cc9 | 7556 | */ |
bbbf7243 | 7557 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7558 | { |
935e5cc9 MX |
7559 | struct btrfs_device *curr, *next; |
7560 | ||
bbbf7243 | 7561 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7562 | |
bbbf7243 | 7563 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7564 | return; |
7565 | ||
bbbf7243 NB |
7566 | /* |
7567 | * We don't need the device_list_mutex here. This list is owned by the | |
7568 | * transaction and the transaction must complete before the device is | |
7569 | * released. | |
7570 | */ | |
7571 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7572 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7573 | post_commit_list) { | |
7574 | list_del_init(&curr->post_commit_list); | |
7575 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7576 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7577 | } |
bbbf7243 | 7578 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7579 | } |
5a13f430 AJ |
7580 | |
7581 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7582 | { | |
7583 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7584 | while (fs_devices) { | |
7585 | fs_devices->fs_info = fs_info; | |
7586 | fs_devices = fs_devices->seed; | |
7587 | } | |
7588 | } | |
7589 | ||
7590 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7591 | { | |
7592 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7593 | while (fs_devices) { | |
7594 | fs_devices->fs_info = NULL; | |
7595 | fs_devices = fs_devices->seed; | |
7596 | } | |
7597 | } | |
46df06b8 DS |
7598 | |
7599 | /* | |
7600 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7601 | */ | |
7602 | int btrfs_bg_type_to_factor(u64 flags) | |
7603 | { | |
7604 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
7605 | BTRFS_BLOCK_GROUP_RAID10)) | |
7606 | return 2; | |
7607 | return 1; | |
7608 | } | |
cf90d884 QW |
7609 | |
7610 | ||
cf90d884 QW |
7611 | |
7612 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7613 | u64 chunk_offset, u64 devid, | |
7614 | u64 physical_offset, u64 physical_len) | |
7615 | { | |
7616 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7617 | struct extent_map *em; | |
7618 | struct map_lookup *map; | |
05a37c48 | 7619 | struct btrfs_device *dev; |
cf90d884 QW |
7620 | u64 stripe_len; |
7621 | bool found = false; | |
7622 | int ret = 0; | |
7623 | int i; | |
7624 | ||
7625 | read_lock(&em_tree->lock); | |
7626 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7627 | read_unlock(&em_tree->lock); | |
7628 | ||
7629 | if (!em) { | |
7630 | btrfs_err(fs_info, | |
7631 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7632 | physical_offset, devid); | |
7633 | ret = -EUCLEAN; | |
7634 | goto out; | |
7635 | } | |
7636 | ||
7637 | map = em->map_lookup; | |
7638 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7639 | if (physical_len != stripe_len) { | |
7640 | btrfs_err(fs_info, | |
7641 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7642 | physical_offset, devid, em->start, physical_len, | |
7643 | stripe_len); | |
7644 | ret = -EUCLEAN; | |
7645 | goto out; | |
7646 | } | |
7647 | ||
7648 | for (i = 0; i < map->num_stripes; i++) { | |
7649 | if (map->stripes[i].dev->devid == devid && | |
7650 | map->stripes[i].physical == physical_offset) { | |
7651 | found = true; | |
7652 | if (map->verified_stripes >= map->num_stripes) { | |
7653 | btrfs_err(fs_info, | |
7654 | "too many dev extents for chunk %llu found", | |
7655 | em->start); | |
7656 | ret = -EUCLEAN; | |
7657 | goto out; | |
7658 | } | |
7659 | map->verified_stripes++; | |
7660 | break; | |
7661 | } | |
7662 | } | |
7663 | if (!found) { | |
7664 | btrfs_err(fs_info, | |
7665 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7666 | physical_offset, devid); | |
7667 | ret = -EUCLEAN; | |
7668 | } | |
05a37c48 QW |
7669 | |
7670 | /* Make sure no dev extent is beyond device bondary */ | |
09ba3bc9 | 7671 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true); |
05a37c48 QW |
7672 | if (!dev) { |
7673 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7674 | ret = -EUCLEAN; | |
7675 | goto out; | |
7676 | } | |
1b3922a8 QW |
7677 | |
7678 | /* It's possible this device is a dummy for seed device */ | |
7679 | if (dev->disk_total_bytes == 0) { | |
09ba3bc9 AJ |
7680 | dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL, |
7681 | NULL, false); | |
1b3922a8 QW |
7682 | if (!dev) { |
7683 | btrfs_err(fs_info, "failed to find seed devid %llu", | |
7684 | devid); | |
7685 | ret = -EUCLEAN; | |
7686 | goto out; | |
7687 | } | |
7688 | } | |
7689 | ||
05a37c48 QW |
7690 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7691 | btrfs_err(fs_info, | |
7692 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7693 | devid, physical_offset, physical_len, | |
7694 | dev->disk_total_bytes); | |
7695 | ret = -EUCLEAN; | |
7696 | goto out; | |
7697 | } | |
cf90d884 QW |
7698 | out: |
7699 | free_extent_map(em); | |
7700 | return ret; | |
7701 | } | |
7702 | ||
7703 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7704 | { | |
7705 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; | |
7706 | struct extent_map *em; | |
7707 | struct rb_node *node; | |
7708 | int ret = 0; | |
7709 | ||
7710 | read_lock(&em_tree->lock); | |
07e1ce09 | 7711 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7712 | em = rb_entry(node, struct extent_map, rb_node); |
7713 | if (em->map_lookup->num_stripes != | |
7714 | em->map_lookup->verified_stripes) { | |
7715 | btrfs_err(fs_info, | |
7716 | "chunk %llu has missing dev extent, have %d expect %d", | |
7717 | em->start, em->map_lookup->verified_stripes, | |
7718 | em->map_lookup->num_stripes); | |
7719 | ret = -EUCLEAN; | |
7720 | goto out; | |
7721 | } | |
7722 | } | |
7723 | out: | |
7724 | read_unlock(&em_tree->lock); | |
7725 | return ret; | |
7726 | } | |
7727 | ||
7728 | /* | |
7729 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7730 | * later chunk allocation/free would cause unexpected behavior. | |
7731 | * | |
7732 | * NOTE: This will iterate through the whole device tree, which should be of | |
7733 | * the same size level as the chunk tree. This slightly increases mount time. | |
7734 | */ | |
7735 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7736 | { | |
7737 | struct btrfs_path *path; | |
7738 | struct btrfs_root *root = fs_info->dev_root; | |
7739 | struct btrfs_key key; | |
5eb19381 QW |
7740 | u64 prev_devid = 0; |
7741 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7742 | int ret = 0; |
7743 | ||
7744 | key.objectid = 1; | |
7745 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7746 | key.offset = 0; | |
7747 | ||
7748 | path = btrfs_alloc_path(); | |
7749 | if (!path) | |
7750 | return -ENOMEM; | |
7751 | ||
7752 | path->reada = READA_FORWARD; | |
7753 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7754 | if (ret < 0) | |
7755 | goto out; | |
7756 | ||
7757 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
7758 | ret = btrfs_next_item(root, path); | |
7759 | if (ret < 0) | |
7760 | goto out; | |
7761 | /* No dev extents at all? Not good */ | |
7762 | if (ret > 0) { | |
7763 | ret = -EUCLEAN; | |
7764 | goto out; | |
7765 | } | |
7766 | } | |
7767 | while (1) { | |
7768 | struct extent_buffer *leaf = path->nodes[0]; | |
7769 | struct btrfs_dev_extent *dext; | |
7770 | int slot = path->slots[0]; | |
7771 | u64 chunk_offset; | |
7772 | u64 physical_offset; | |
7773 | u64 physical_len; | |
7774 | u64 devid; | |
7775 | ||
7776 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7777 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7778 | break; | |
7779 | devid = key.objectid; | |
7780 | physical_offset = key.offset; | |
7781 | ||
7782 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7783 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7784 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7785 | ||
5eb19381 QW |
7786 | /* Check if this dev extent overlaps with the previous one */ |
7787 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7788 | btrfs_err(fs_info, | |
7789 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7790 | devid, physical_offset, prev_dev_ext_end); | |
7791 | ret = -EUCLEAN; | |
7792 | goto out; | |
7793 | } | |
7794 | ||
cf90d884 QW |
7795 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7796 | physical_offset, physical_len); | |
7797 | if (ret < 0) | |
7798 | goto out; | |
5eb19381 QW |
7799 | prev_devid = devid; |
7800 | prev_dev_ext_end = physical_offset + physical_len; | |
7801 | ||
cf90d884 QW |
7802 | ret = btrfs_next_item(root, path); |
7803 | if (ret < 0) | |
7804 | goto out; | |
7805 | if (ret > 0) { | |
7806 | ret = 0; | |
7807 | break; | |
7808 | } | |
7809 | } | |
7810 | ||
7811 | /* Ensure all chunks have corresponding dev extents */ | |
7812 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7813 | out: | |
7814 | btrfs_free_path(path); | |
7815 | return ret; | |
7816 | } | |
eede2bf3 OS |
7817 | |
7818 | /* | |
7819 | * Check whether the given block group or device is pinned by any inode being | |
7820 | * used as a swapfile. | |
7821 | */ | |
7822 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7823 | { | |
7824 | struct btrfs_swapfile_pin *sp; | |
7825 | struct rb_node *node; | |
7826 | ||
7827 | spin_lock(&fs_info->swapfile_pins_lock); | |
7828 | node = fs_info->swapfile_pins.rb_node; | |
7829 | while (node) { | |
7830 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7831 | if (ptr < sp->ptr) | |
7832 | node = node->rb_left; | |
7833 | else if (ptr > sp->ptr) | |
7834 | node = node->rb_right; | |
7835 | else | |
7836 | break; | |
7837 | } | |
7838 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7839 | return node != NULL; | |
7840 | } |