Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 | 6 | #include <linux/sched.h> |
fccc0007 | 7 | #include <linux/sched/mm.h> |
5a0e3ad6 | 8 | #include <linux/slab.h> |
442a4f63 | 9 | #include <linux/ratelimit.h> |
59641015 | 10 | #include <linux/kthread.h> |
803b2f54 | 11 | #include <linux/semaphore.h> |
8da4b8c4 | 12 | #include <linux/uuid.h> |
f8e10cd3 | 13 | #include <linux/list_sort.h> |
54fde91f | 14 | #include <linux/namei.h> |
784352fe | 15 | #include "misc.h" |
0b86a832 | 16 | #include "ctree.h" |
0b86a832 CM |
17 | #include "disk-io.h" |
18 | #include "transaction.h" | |
0b86a832 | 19 | #include "volumes.h" |
53b381b3 | 20 | #include "raid56.h" |
606686ee | 21 | #include "rcu-string.h" |
8dabb742 | 22 | #include "dev-replace.h" |
99994cde | 23 | #include "sysfs.h" |
82fc28fb | 24 | #include "tree-checker.h" |
8719aaae | 25 | #include "space-info.h" |
aac0023c | 26 | #include "block-group.h" |
b0643e59 | 27 | #include "discard.h" |
5b316468 | 28 | #include "zoned.h" |
c7f13d42 | 29 | #include "fs.h" |
07e81dc9 | 30 | #include "accessors.h" |
c7a03b52 | 31 | #include "uuid-tree.h" |
7572dec8 | 32 | #include "ioctl.h" |
67707479 | 33 | #include "relocation.h" |
2fc6822c | 34 | #include "scrub.h" |
7f0add25 | 35 | #include "super.h" |
10e27980 | 36 | #include "raid-stripe-tree.h" |
0b86a832 | 37 | |
bf08387f QW |
38 | #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ |
39 | BTRFS_BLOCK_GROUP_RAID10 | \ | |
40 | BTRFS_BLOCK_GROUP_RAID56_MASK) | |
41 | ||
fd747f2d JT |
42 | struct btrfs_io_geometry { |
43 | u32 stripe_index; | |
44 | u32 stripe_nr; | |
45 | int mirror_num; | |
46 | int num_stripes; | |
47 | u64 stripe_offset; | |
48 | u64 raid56_full_stripe_start; | |
49 | int max_errors; | |
50 | enum btrfs_map_op op; | |
51 | }; | |
52 | ||
af902047 ZL |
53 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
54 | [BTRFS_RAID_RAID10] = { | |
55 | .sub_stripes = 2, | |
56 | .dev_stripes = 1, | |
57 | .devs_max = 0, /* 0 == as many as possible */ | |
b2f78e88 | 58 | .devs_min = 2, |
8789f4fe | 59 | .tolerated_failures = 1, |
af902047 ZL |
60 | .devs_increment = 2, |
61 | .ncopies = 2, | |
b50836ed | 62 | .nparity = 0, |
ed23467b | 63 | .raid_name = "raid10", |
41a6e891 | 64 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 65 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
66 | }, |
67 | [BTRFS_RAID_RAID1] = { | |
68 | .sub_stripes = 1, | |
69 | .dev_stripes = 1, | |
70 | .devs_max = 2, | |
71 | .devs_min = 2, | |
8789f4fe | 72 | .tolerated_failures = 1, |
af902047 ZL |
73 | .devs_increment = 2, |
74 | .ncopies = 2, | |
b50836ed | 75 | .nparity = 0, |
ed23467b | 76 | .raid_name = "raid1", |
41a6e891 | 77 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 78 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 79 | }, |
47e6f742 DS |
80 | [BTRFS_RAID_RAID1C3] = { |
81 | .sub_stripes = 1, | |
82 | .dev_stripes = 1, | |
cf93e15e | 83 | .devs_max = 3, |
47e6f742 DS |
84 | .devs_min = 3, |
85 | .tolerated_failures = 2, | |
86 | .devs_increment = 3, | |
87 | .ncopies = 3, | |
db26a024 | 88 | .nparity = 0, |
47e6f742 DS |
89 | .raid_name = "raid1c3", |
90 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
91 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
92 | }, | |
8d6fac00 DS |
93 | [BTRFS_RAID_RAID1C4] = { |
94 | .sub_stripes = 1, | |
95 | .dev_stripes = 1, | |
cf93e15e | 96 | .devs_max = 4, |
8d6fac00 DS |
97 | .devs_min = 4, |
98 | .tolerated_failures = 3, | |
99 | .devs_increment = 4, | |
100 | .ncopies = 4, | |
db26a024 | 101 | .nparity = 0, |
8d6fac00 DS |
102 | .raid_name = "raid1c4", |
103 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
104 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
105 | }, | |
af902047 ZL |
106 | [BTRFS_RAID_DUP] = { |
107 | .sub_stripes = 1, | |
108 | .dev_stripes = 2, | |
109 | .devs_max = 1, | |
110 | .devs_min = 1, | |
8789f4fe | 111 | .tolerated_failures = 0, |
af902047 ZL |
112 | .devs_increment = 1, |
113 | .ncopies = 2, | |
b50836ed | 114 | .nparity = 0, |
ed23467b | 115 | .raid_name = "dup", |
41a6e891 | 116 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 117 | .mindev_error = 0, |
af902047 ZL |
118 | }, |
119 | [BTRFS_RAID_RAID0] = { | |
120 | .sub_stripes = 1, | |
121 | .dev_stripes = 1, | |
122 | .devs_max = 0, | |
b2f78e88 | 123 | .devs_min = 1, |
8789f4fe | 124 | .tolerated_failures = 0, |
af902047 ZL |
125 | .devs_increment = 1, |
126 | .ncopies = 1, | |
b50836ed | 127 | .nparity = 0, |
ed23467b | 128 | .raid_name = "raid0", |
41a6e891 | 129 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 130 | .mindev_error = 0, |
af902047 ZL |
131 | }, |
132 | [BTRFS_RAID_SINGLE] = { | |
133 | .sub_stripes = 1, | |
134 | .dev_stripes = 1, | |
135 | .devs_max = 1, | |
136 | .devs_min = 1, | |
8789f4fe | 137 | .tolerated_failures = 0, |
af902047 ZL |
138 | .devs_increment = 1, |
139 | .ncopies = 1, | |
b50836ed | 140 | .nparity = 0, |
ed23467b | 141 | .raid_name = "single", |
41a6e891 | 142 | .bg_flag = 0, |
f9fbcaa2 | 143 | .mindev_error = 0, |
af902047 ZL |
144 | }, |
145 | [BTRFS_RAID_RAID5] = { | |
146 | .sub_stripes = 1, | |
147 | .dev_stripes = 1, | |
148 | .devs_max = 0, | |
149 | .devs_min = 2, | |
8789f4fe | 150 | .tolerated_failures = 1, |
af902047 | 151 | .devs_increment = 1, |
da612e31 | 152 | .ncopies = 1, |
b50836ed | 153 | .nparity = 1, |
ed23467b | 154 | .raid_name = "raid5", |
41a6e891 | 155 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 156 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
157 | }, |
158 | [BTRFS_RAID_RAID6] = { | |
159 | .sub_stripes = 1, | |
160 | .dev_stripes = 1, | |
161 | .devs_max = 0, | |
162 | .devs_min = 3, | |
8789f4fe | 163 | .tolerated_failures = 2, |
af902047 | 164 | .devs_increment = 1, |
da612e31 | 165 | .ncopies = 1, |
b50836ed | 166 | .nparity = 2, |
ed23467b | 167 | .raid_name = "raid6", |
41a6e891 | 168 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 169 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
170 | }, |
171 | }; | |
172 | ||
500a44c9 DS |
173 | /* |
174 | * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which | |
175 | * can be used as index to access btrfs_raid_array[]. | |
176 | */ | |
177 | enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) | |
178 | { | |
719fae89 | 179 | const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); |
500a44c9 | 180 | |
719fae89 QW |
181 | if (!profile) |
182 | return BTRFS_RAID_SINGLE; | |
183 | ||
184 | return BTRFS_BG_FLAG_TO_INDEX(profile); | |
500a44c9 DS |
185 | } |
186 | ||
158da513 | 187 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 188 | { |
158da513 DS |
189 | const int index = btrfs_bg_flags_to_raid_index(flags); |
190 | ||
191 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
192 | return NULL; |
193 | ||
158da513 | 194 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
195 | } |
196 | ||
0b30f719 QW |
197 | int btrfs_nr_parity_stripes(u64 type) |
198 | { | |
199 | enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); | |
200 | ||
201 | return btrfs_raid_array[index].nparity; | |
202 | } | |
203 | ||
f89e09cf AJ |
204 | /* |
205 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
206 | * bytes including terminating null byte. | |
207 | */ | |
208 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
209 | { | |
210 | int i; | |
211 | int ret; | |
212 | char *bp = buf; | |
213 | u64 flags = bg_flags; | |
214 | u32 size_bp = size_buf; | |
215 | ||
216 | if (!flags) { | |
217 | strcpy(bp, "NONE"); | |
218 | return; | |
219 | } | |
220 | ||
221 | #define DESCRIBE_FLAG(flag, desc) \ | |
222 | do { \ | |
223 | if (flags & (flag)) { \ | |
224 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
225 | if (ret < 0 || ret >= size_bp) \ | |
226 | goto out_overflow; \ | |
227 | size_bp -= ret; \ | |
228 | bp += ret; \ | |
229 | flags &= ~(flag); \ | |
230 | } \ | |
231 | } while (0) | |
232 | ||
233 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
234 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
235 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
236 | ||
237 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
238 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
239 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
240 | btrfs_raid_array[i].raid_name); | |
241 | #undef DESCRIBE_FLAG | |
242 | ||
243 | if (flags) { | |
244 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
245 | size_bp -= ret; | |
246 | } | |
247 | ||
248 | if (size_bp < size_buf) | |
249 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
250 | ||
251 | /* | |
252 | * The text is trimmed, it's up to the caller to provide sufficiently | |
253 | * large buffer | |
254 | */ | |
255 | out_overflow:; | |
256 | } | |
257 | ||
6f8e0fc7 | 258 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 259 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 260 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 261 | |
9c6b1c4d DS |
262 | /* |
263 | * Device locking | |
264 | * ============== | |
265 | * | |
266 | * There are several mutexes that protect manipulation of devices and low-level | |
267 | * structures like chunks but not block groups, extents or files | |
268 | * | |
269 | * uuid_mutex (global lock) | |
270 | * ------------------------ | |
271 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
272 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
273 | * device) or requested by the device= mount option | |
274 | * | |
275 | * the mutex can be very coarse and can cover long-running operations | |
276 | * | |
277 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 278 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
279 | * |
280 | * global::fs_devs - add, remove, updates to the global list | |
281 | * | |
18c850fd JB |
282 | * does not protect: manipulation of the fs_devices::devices list in general |
283 | * but in mount context it could be used to exclude list modifications by eg. | |
284 | * scan ioctl | |
9c6b1c4d DS |
285 | * |
286 | * btrfs_device::name - renames (write side), read is RCU | |
287 | * | |
288 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
289 | * ------------------------------------------------ | |
290 | * protects updates to fs_devices::devices, ie. adding and deleting | |
291 | * | |
292 | * simple list traversal with read-only actions can be done with RCU protection | |
293 | * | |
294 | * may be used to exclude some operations from running concurrently without any | |
295 | * modifications to the list (see write_all_supers) | |
296 | * | |
18c850fd JB |
297 | * Is not required at mount and close times, because our device list is |
298 | * protected by the uuid_mutex at that point. | |
299 | * | |
9c6b1c4d DS |
300 | * balance_mutex |
301 | * ------------- | |
302 | * protects balance structures (status, state) and context accessed from | |
303 | * several places (internally, ioctl) | |
304 | * | |
305 | * chunk_mutex | |
306 | * ----------- | |
307 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
308 | * device is added/removed. Additionally it also protects post_commit_list of |
309 | * individual devices, since they can be added to the transaction's | |
310 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
311 | * |
312 | * cleaner_mutex | |
313 | * ------------- | |
314 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
315 | * cleaning together with relocation or delayed iputs | |
316 | * | |
317 | * | |
318 | * Lock nesting | |
319 | * ============ | |
320 | * | |
321 | * uuid_mutex | |
ae3e715f AJ |
322 | * device_list_mutex |
323 | * chunk_mutex | |
324 | * balance_mutex | |
89595e80 AJ |
325 | * |
326 | * | |
c3e1f96c GR |
327 | * Exclusive operations |
328 | * ==================== | |
89595e80 AJ |
329 | * |
330 | * Maintains the exclusivity of the following operations that apply to the | |
331 | * whole filesystem and cannot run in parallel. | |
332 | * | |
333 | * - Balance (*) | |
334 | * - Device add | |
335 | * - Device remove | |
336 | * - Device replace (*) | |
337 | * - Resize | |
338 | * | |
339 | * The device operations (as above) can be in one of the following states: | |
340 | * | |
341 | * - Running state | |
342 | * - Paused state | |
343 | * - Completed state | |
344 | * | |
345 | * Only device operations marked with (*) can go into the Paused state for the | |
346 | * following reasons: | |
347 | * | |
348 | * - ioctl (only Balance can be Paused through ioctl) | |
349 | * - filesystem remounted as read-only | |
350 | * - filesystem unmounted and mounted as read-only | |
351 | * - system power-cycle and filesystem mounted as read-only | |
352 | * - filesystem or device errors leading to forced read-only | |
353 | * | |
c3e1f96c GR |
354 | * The status of exclusive operation is set and cleared atomically. |
355 | * During the course of Paused state, fs_info::exclusive_operation remains set. | |
89595e80 AJ |
356 | * A device operation in Paused or Running state can be canceled or resumed |
357 | * either by ioctl (Balance only) or when remounted as read-write. | |
c3e1f96c | 358 | * The exclusive status is cleared when the device operation is canceled or |
89595e80 | 359 | * completed. |
9c6b1c4d DS |
360 | */ |
361 | ||
67a2c45e | 362 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 363 | static LIST_HEAD(fs_uuids); |
4143cb8b | 364 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
365 | { |
366 | return &fs_uuids; | |
367 | } | |
8a4b83cc | 368 | |
2dfeca9b | 369 | /* |
f7361d8c AJ |
370 | * Allocate new btrfs_fs_devices structure identified by a fsid. |
371 | * | |
372 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to | |
373 | * fs_devices::metadata_fsid | |
2dfeca9b DS |
374 | * |
375 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
376 | * The returned struct is not linked onto any lists and can be destroyed with | |
377 | * kfree() right away. | |
378 | */ | |
f7361d8c | 379 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) |
2208a378 ID |
380 | { |
381 | struct btrfs_fs_devices *fs_devs; | |
382 | ||
78f2c9e6 | 383 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
384 | if (!fs_devs) |
385 | return ERR_PTR(-ENOMEM); | |
386 | ||
387 | mutex_init(&fs_devs->device_list_mutex); | |
388 | ||
389 | INIT_LIST_HEAD(&fs_devs->devices); | |
390 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 391 | INIT_LIST_HEAD(&fs_devs->fs_list); |
944d3f9f | 392 | INIT_LIST_HEAD(&fs_devs->seed_list); |
2208a378 | 393 | |
19c4c49c AJ |
394 | if (fsid) { |
395 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
f7361d8c | 396 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); |
19c4c49c | 397 | } |
7239ff4b | 398 | |
2208a378 ID |
399 | return fs_devs; |
400 | } | |
401 | ||
f2db4d5c | 402 | static void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 403 | { |
bbbf7243 | 404 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 405 | rcu_string_free(device->name); |
611ccc58 | 406 | extent_io_tree_release(&device->alloc_state); |
5b316468 | 407 | btrfs_destroy_dev_zone_info(device); |
48dae9cf DS |
408 | kfree(device); |
409 | } | |
410 | ||
e4404d6e YZ |
411 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
412 | { | |
413 | struct btrfs_device *device; | |
5f58d783 | 414 | |
e4404d6e YZ |
415 | WARN_ON(fs_devices->opened); |
416 | while (!list_empty(&fs_devices->devices)) { | |
417 | device = list_entry(fs_devices->devices.next, | |
418 | struct btrfs_device, dev_list); | |
419 | list_del(&device->dev_list); | |
a425f9d4 | 420 | btrfs_free_device(device); |
e4404d6e YZ |
421 | } |
422 | kfree(fs_devices); | |
423 | } | |
424 | ||
ffc5a379 | 425 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
426 | { |
427 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 428 | |
2b82032c YZ |
429 | while (!list_empty(&fs_uuids)) { |
430 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
431 | struct btrfs_fs_devices, fs_list); |
432 | list_del(&fs_devices->fs_list); | |
e4404d6e | 433 | free_fs_devices(fs_devices); |
8a4b83cc | 434 | } |
8a4b83cc CM |
435 | } |
436 | ||
1a898345 AJ |
437 | static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices, |
438 | const u8 *fsid, const u8 *metadata_fsid) | |
439 | { | |
440 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0) | |
441 | return false; | |
442 | ||
443 | if (!metadata_fsid) | |
444 | return true; | |
445 | ||
446 | if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0) | |
447 | return false; | |
448 | ||
449 | return true; | |
450 | } | |
451 | ||
7239ff4b NB |
452 | static noinline struct btrfs_fs_devices *find_fsid( |
453 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 454 | { |
8a4b83cc CM |
455 | struct btrfs_fs_devices *fs_devices; |
456 | ||
7239ff4b NB |
457 | ASSERT(fsid); |
458 | ||
7a62d0f0 | 459 | /* Handle non-split brain cases */ |
c4babc5e | 460 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
1a898345 AJ |
461 | if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid)) |
462 | return fs_devices; | |
8a4b83cc CM |
463 | } |
464 | return NULL; | |
465 | } | |
466 | ||
beaf8ab3 | 467 | static int |
05bdb996 | 468 | btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder, |
9ae061cf | 469 | int flush, struct file **bdev_file, |
8f32380d | 470 | struct btrfs_super_block **disk_super) |
beaf8ab3 | 471 | { |
86ec15d0 | 472 | struct block_device *bdev; |
beaf8ab3 SB |
473 | int ret; |
474 | ||
9ae061cf | 475 | *bdev_file = bdev_file_open_by_path(device_path, flags, holder, NULL); |
beaf8ab3 | 476 | |
9ae061cf CB |
477 | if (IS_ERR(*bdev_file)) { |
478 | ret = PTR_ERR(*bdev_file); | |
beaf8ab3 SB |
479 | goto error; |
480 | } | |
9ae061cf | 481 | bdev = file_bdev(*bdev_file); |
beaf8ab3 SB |
482 | |
483 | if (flush) | |
86ec15d0 | 484 | sync_blockdev(bdev); |
b85c4298 | 485 | if (holder) { |
ead083ae | 486 | ret = set_blocksize(*bdev_file, BTRFS_BDEV_BLOCKSIZE); |
b85c4298 AV |
487 | if (ret) { |
488 | fput(*bdev_file); | |
489 | goto error; | |
490 | } | |
beaf8ab3 | 491 | } |
86ec15d0 JK |
492 | invalidate_bdev(bdev); |
493 | *disk_super = btrfs_read_dev_super(bdev); | |
8f32380d JT |
494 | if (IS_ERR(*disk_super)) { |
495 | ret = PTR_ERR(*disk_super); | |
9ae061cf | 496 | fput(*bdev_file); |
beaf8ab3 SB |
497 | goto error; |
498 | } | |
499 | ||
500 | return 0; | |
501 | ||
502 | error: | |
b85c4298 | 503 | *disk_super = NULL; |
9ae061cf | 504 | *bdev_file = NULL; |
beaf8ab3 SB |
505 | return ret; |
506 | } | |
507 | ||
43dd529a DS |
508 | /* |
509 | * Search and remove all stale devices (which are not mounted). When both | |
510 | * inputs are NULL, it will search and release all stale devices. | |
16cab91a | 511 | * |
43dd529a DS |
512 | * @devt: Optional. When provided will it release all unmounted devices |
513 | * matching this devt only. | |
16cab91a | 514 | * @skip_device: Optional. Will skip this device when searching for the stale |
43dd529a | 515 | * devices. |
16cab91a AJ |
516 | * |
517 | * Return: 0 for success or if @devt is 0. | |
518 | * -EBUSY if @devt is a mounted device. | |
519 | * -ENOENT if @devt does not match any device in the list. | |
d8367db3 | 520 | */ |
16cab91a | 521 | static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) |
4fde46f0 | 522 | { |
fa6d2ae5 AJ |
523 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
524 | struct btrfs_device *device, *tmp_device; | |
bc27d6f0 AJ |
525 | int ret; |
526 | bool freed = false; | |
70bc7088 | 527 | |
c1247069 AJ |
528 | lockdep_assert_held(&uuid_mutex); |
529 | ||
bc27d6f0 AJ |
530 | /* Return good status if there is no instance of devt. */ |
531 | ret = 0; | |
fa6d2ae5 | 532 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 533 | |
70bc7088 | 534 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
535 | list_for_each_entry_safe(device, tmp_device, |
536 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 537 | if (skip_device && skip_device == device) |
d8367db3 | 538 | continue; |
330a5bf4 | 539 | if (devt && devt != device->devt) |
38cf665d | 540 | continue; |
70bc7088 | 541 | if (fs_devices->opened) { |
bc27d6f0 | 542 | if (devt) |
70bc7088 AJ |
543 | ret = -EBUSY; |
544 | break; | |
545 | } | |
4fde46f0 | 546 | |
4fde46f0 | 547 | /* delete the stale device */ |
7bcb8164 AJ |
548 | fs_devices->num_devices--; |
549 | list_del(&device->dev_list); | |
550 | btrfs_free_device(device); | |
551 | ||
bc27d6f0 | 552 | freed = true; |
7bcb8164 AJ |
553 | } |
554 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 555 | |
7bcb8164 AJ |
556 | if (fs_devices->num_devices == 0) { |
557 | btrfs_sysfs_remove_fsid(fs_devices); | |
558 | list_del(&fs_devices->fs_list); | |
559 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
560 | } |
561 | } | |
70bc7088 | 562 | |
bc27d6f0 AJ |
563 | /* If there is at least one freed device return 0. */ |
564 | if (freed) | |
565 | return 0; | |
566 | ||
70bc7088 | 567 | return ret; |
4fde46f0 AJ |
568 | } |
569 | ||
69d427f3 | 570 | static struct btrfs_fs_devices *find_fsid_by_device( |
a5b8a5f9 AJ |
571 | struct btrfs_super_block *disk_super, |
572 | dev_t devt, bool *same_fsid_diff_dev) | |
69d427f3 AJ |
573 | { |
574 | struct btrfs_fs_devices *fsid_fs_devices; | |
a5b8a5f9 | 575 | struct btrfs_fs_devices *devt_fs_devices; |
69d427f3 AJ |
576 | const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
577 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
a5b8a5f9 | 578 | bool found_by_devt = false; |
69d427f3 AJ |
579 | |
580 | /* Find the fs_device by the usual method, if found use it. */ | |
581 | fsid_fs_devices = find_fsid(disk_super->fsid, | |
582 | has_metadata_uuid ? disk_super->metadata_uuid : NULL); | |
583 | ||
a5b8a5f9 AJ |
584 | /* The temp_fsid feature is supported only with single device filesystem. */ |
585 | if (btrfs_super_num_devices(disk_super) != 1) | |
586 | return fsid_fs_devices; | |
587 | ||
c47b02c1 AJ |
588 | /* |
589 | * A seed device is an integral component of the sprout device, which | |
590 | * functions as a multi-device filesystem. So, temp-fsid feature is | |
591 | * not supported. | |
592 | */ | |
593 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) | |
594 | return fsid_fs_devices; | |
595 | ||
a5b8a5f9 AJ |
596 | /* Try to find a fs_devices by matching devt. */ |
597 | list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) { | |
598 | struct btrfs_device *device; | |
599 | ||
600 | list_for_each_entry(device, &devt_fs_devices->devices, dev_list) { | |
601 | if (device->devt == devt) { | |
602 | found_by_devt = true; | |
603 | break; | |
604 | } | |
605 | } | |
606 | if (found_by_devt) | |
607 | break; | |
608 | } | |
609 | ||
610 | if (found_by_devt) { | |
611 | /* Existing device. */ | |
612 | if (fsid_fs_devices == NULL) { | |
613 | if (devt_fs_devices->opened == 0) { | |
614 | /* Stale device. */ | |
615 | return NULL; | |
616 | } else { | |
617 | /* temp_fsid is mounting a subvol. */ | |
618 | return devt_fs_devices; | |
619 | } | |
620 | } else { | |
621 | /* Regular or temp_fsid device mounting a subvol. */ | |
622 | return devt_fs_devices; | |
623 | } | |
624 | } else { | |
625 | /* New device. */ | |
626 | if (fsid_fs_devices == NULL) { | |
627 | return NULL; | |
628 | } else { | |
629 | /* sb::fsid is already used create a new temp_fsid. */ | |
630 | *same_fsid_diff_dev = true; | |
631 | return NULL; | |
632 | } | |
633 | } | |
634 | ||
635 | /* Not reached. */ | |
69d427f3 AJ |
636 | } |
637 | ||
18c850fd JB |
638 | /* |
639 | * This is only used on mount, and we are protected from competing things | |
640 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
641 | * fs_devices->device_list_mutex here. | |
642 | */ | |
0fb08bcc | 643 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 644 | struct btrfs_device *device, blk_mode_t flags, |
0fb08bcc AJ |
645 | void *holder) |
646 | { | |
9ae061cf | 647 | struct file *bdev_file; |
0fb08bcc AJ |
648 | struct btrfs_super_block *disk_super; |
649 | u64 devid; | |
650 | int ret; | |
651 | ||
652 | if (device->bdev) | |
653 | return -EINVAL; | |
654 | if (!device->name) | |
655 | return -EINVAL; | |
656 | ||
657 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
9ae061cf | 658 | &bdev_file, &disk_super); |
0fb08bcc AJ |
659 | if (ret) |
660 | return ret; | |
661 | ||
0fb08bcc AJ |
662 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
663 | if (devid != device->devid) | |
8f32380d | 664 | goto error_free_page; |
0fb08bcc AJ |
665 | |
666 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 667 | goto error_free_page; |
0fb08bcc AJ |
668 | |
669 | device->generation = btrfs_super_generation(disk_super); | |
670 | ||
671 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
672 | if (btrfs_super_incompat_flags(disk_super) & |
673 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
674 | pr_err( | |
675 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 676 | goto error_free_page; |
7239ff4b NB |
677 | } |
678 | ||
ebbede42 | 679 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 680 | fs_devices->seeding = true; |
0fb08bcc | 681 | } else { |
9ae061cf | 682 | if (bdev_read_only(file_bdev(bdev_file))) |
ebbede42 AJ |
683 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
684 | else | |
685 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
686 | } |
687 | ||
9ae061cf | 688 | if (!bdev_nonrot(file_bdev(bdev_file))) |
7f0432d0 | 689 | fs_devices->rotating = true; |
0fb08bcc | 690 | |
9ae061cf | 691 | if (bdev_max_discard_sectors(file_bdev(bdev_file))) |
63a7cb13 DS |
692 | fs_devices->discardable = true; |
693 | ||
9ae061cf CB |
694 | device->bdev_file = bdev_file; |
695 | device->bdev = file_bdev(bdev_file); | |
e12c9621 | 696 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc | 697 | |
9f7eb840 AJ |
698 | if (device->devt != device->bdev->bd_dev) { |
699 | btrfs_warn(NULL, | |
700 | "device %s maj:min changed from %d:%d to %d:%d", | |
701 | device->name->str, MAJOR(device->devt), | |
702 | MINOR(device->devt), MAJOR(device->bdev->bd_dev), | |
703 | MINOR(device->bdev->bd_dev)); | |
704 | ||
705 | device->devt = device->bdev->bd_dev; | |
706 | } | |
707 | ||
0fb08bcc | 708 | fs_devices->open_devices++; |
ebbede42 AJ |
709 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
710 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 711 | fs_devices->rw_devices++; |
b1b8e386 | 712 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 713 | } |
8f32380d | 714 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
715 | |
716 | return 0; | |
717 | ||
8f32380d JT |
718 | error_free_page: |
719 | btrfs_release_disk_super(disk_super); | |
9ae061cf | 720 | fput(bdev_file); |
0fb08bcc AJ |
721 | |
722 | return -EINVAL; | |
723 | } | |
724 | ||
4844c366 AJ |
725 | u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb) |
726 | { | |
727 | bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) & | |
728 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
729 | ||
730 | return has_metadata_uuid ? sb->metadata_uuid : sb->fsid; | |
731 | } | |
732 | ||
60999ca4 DS |
733 | /* |
734 | * Add new device to list of registered devices | |
735 | * | |
736 | * Returns: | |
e124ece5 AJ |
737 | * device pointer which was just added or updated when successful |
738 | * error pointer when failed | |
60999ca4 | 739 | */ |
e124ece5 | 740 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
741 | struct btrfs_super_block *disk_super, |
742 | bool *new_device_added) | |
8a4b83cc CM |
743 | { |
744 | struct btrfs_device *device; | |
7a62d0f0 | 745 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 746 | struct rcu_string *name; |
8a4b83cc | 747 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 748 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
4889bc05 AJ |
749 | dev_t path_devt; |
750 | int error; | |
a5b8a5f9 | 751 | bool same_fsid_diff_dev = false; |
7239ff4b NB |
752 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
753 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
754 | ||
5966930d | 755 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { |
197a9ece AJ |
756 | btrfs_err(NULL, |
757 | "device %s has incomplete metadata_uuid change, please use btrfstune to complete", | |
758 | path); | |
759 | return ERR_PTR(-EAGAIN); | |
760 | } | |
7239ff4b | 761 | |
4889bc05 | 762 | error = lookup_bdev(path, &path_devt); |
ed02363f QW |
763 | if (error) { |
764 | btrfs_err(NULL, "failed to lookup block device for path %s: %d", | |
765 | path, error); | |
4889bc05 | 766 | return ERR_PTR(error); |
ed02363f | 767 | } |
4889bc05 | 768 | |
a5b8a5f9 | 769 | fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev); |
8a4b83cc | 770 | |
8a4b83cc | 771 | if (!fs_devices) { |
f7361d8c | 772 | fs_devices = alloc_fs_devices(disk_super->fsid); |
cd63ffbd FM |
773 | if (IS_ERR(fs_devices)) |
774 | return ERR_CAST(fs_devices); | |
775 | ||
f7361d8c AJ |
776 | if (has_metadata_uuid) |
777 | memcpy(fs_devices->metadata_uuid, | |
778 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); | |
779 | ||
a5b8a5f9 AJ |
780 | if (same_fsid_diff_dev) { |
781 | generate_random_uuid(fs_devices->fsid); | |
782 | fs_devices->temp_fsid = true; | |
0782303a AJ |
783 | pr_info("BTRFS: device %s (%d:%d) using temp-fsid %pU\n", |
784 | path, MAJOR(path_devt), MINOR(path_devt), | |
785 | fs_devices->fsid); | |
a5b8a5f9 | 786 | } |
92900e51 | 787 | |
9c6d173e | 788 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 789 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 790 | |
8a4b83cc CM |
791 | device = NULL; |
792 | } else { | |
562d7b15 JB |
793 | struct btrfs_dev_lookup_args args = { |
794 | .devid = devid, | |
795 | .uuid = disk_super->dev_item.uuid, | |
796 | }; | |
797 | ||
9c6d173e | 798 | mutex_lock(&fs_devices->device_list_mutex); |
562d7b15 | 799 | device = btrfs_find_device(fs_devices, &args); |
7a62d0f0 | 800 | |
5966930d | 801 | if (found_transid > fs_devices->latest_generation) { |
7a62d0f0 NB |
802 | memcpy(fs_devices->fsid, disk_super->fsid, |
803 | BTRFS_FSID_SIZE); | |
319baafc AJ |
804 | memcpy(fs_devices->metadata_uuid, |
805 | btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE); | |
7a62d0f0 | 806 | } |
8a4b83cc | 807 | } |
443f24fe | 808 | |
8a4b83cc | 809 | if (!device) { |
bb21e302 AJ |
810 | unsigned int nofs_flag; |
811 | ||
9c6d173e | 812 | if (fs_devices->opened) { |
ed02363f | 813 | btrfs_err(NULL, |
0782303a AJ |
814 | "device %s (%d:%d) belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)", |
815 | path, MAJOR(path_devt), MINOR(path_devt), | |
816 | fs_devices->fsid, current->comm, | |
7f9879eb | 817 | task_pid_nr(current)); |
9c6d173e | 818 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 819 | return ERR_PTR(-EBUSY); |
9c6d173e | 820 | } |
2b82032c | 821 | |
bb21e302 | 822 | nofs_flag = memalloc_nofs_save(); |
12bd2fc0 | 823 | device = btrfs_alloc_device(NULL, &devid, |
bb21e302 AJ |
824 | disk_super->dev_item.uuid, path); |
825 | memalloc_nofs_restore(nofs_flag); | |
12bd2fc0 | 826 | if (IS_ERR(device)) { |
9c6d173e | 827 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 828 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 829 | return device; |
8a4b83cc | 830 | } |
606686ee | 831 | |
4889bc05 | 832 | device->devt = path_devt; |
90519d66 | 833 | |
1f78160c | 834 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 835 | fs_devices->num_devices++; |
e5e9a520 | 836 | |
2b82032c | 837 | device->fs_devices = fs_devices; |
4306a974 | 838 | *new_device_added = true; |
327f18cc AJ |
839 | |
840 | if (disk_super->label[0]) | |
aa6c0df7 | 841 | pr_info( |
0782303a | 842 | "BTRFS: device label %s devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n", |
aa6c0df7 | 843 | disk_super->label, devid, found_transid, path, |
0782303a | 844 | MAJOR(path_devt), MINOR(path_devt), |
aa6c0df7 | 845 | current->comm, task_pid_nr(current)); |
327f18cc | 846 | else |
aa6c0df7 | 847 | pr_info( |
0782303a | 848 | "BTRFS: device fsid %pU devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n", |
aa6c0df7 | 849 | disk_super->fsid, devid, found_transid, path, |
0782303a | 850 | MAJOR(path_devt), MINOR(path_devt), |
aa6c0df7 | 851 | current->comm, task_pid_nr(current)); |
327f18cc | 852 | |
606686ee | 853 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
854 | /* |
855 | * When FS is already mounted. | |
856 | * 1. If you are here and if the device->name is NULL that | |
857 | * means this device was missing at time of FS mount. | |
858 | * 2. If you are here and if the device->name is different | |
859 | * from 'path' that means either | |
860 | * a. The same device disappeared and reappeared with | |
861 | * different name. or | |
862 | * b. The missing-disk-which-was-replaced, has | |
863 | * reappeared now. | |
864 | * | |
865 | * We must allow 1 and 2a above. But 2b would be a spurious | |
866 | * and unintentional. | |
867 | * | |
868 | * Further in case of 1 and 2a above, the disk at 'path' | |
869 | * would have missed some transaction when it was away and | |
870 | * in case of 2a the stale bdev has to be updated as well. | |
871 | * 2b must not be allowed at all time. | |
872 | */ | |
873 | ||
874 | /* | |
0f23ae74 CM |
875 | * For now, we do allow update to btrfs_fs_device through the |
876 | * btrfs dev scan cli after FS has been mounted. We're still | |
877 | * tracking a problem where systems fail mount by subvolume id | |
878 | * when we reject replacement on a mounted FS. | |
b96de000 | 879 | */ |
0f23ae74 | 880 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
881 | /* |
882 | * That is if the FS is _not_ mounted and if you | |
883 | * are here, that means there is more than one | |
884 | * disk with same uuid and devid.We keep the one | |
885 | * with larger generation number or the last-in if | |
886 | * generation are equal. | |
887 | */ | |
9c6d173e | 888 | mutex_unlock(&fs_devices->device_list_mutex); |
ed02363f QW |
889 | btrfs_err(NULL, |
890 | "device %s already registered with a higher generation, found %llu expect %llu", | |
891 | path, found_transid, device->generation); | |
e124ece5 | 892 | return ERR_PTR(-EEXIST); |
77bdae4d | 893 | } |
b96de000 | 894 | |
a9261d41 AJ |
895 | /* |
896 | * We are going to replace the device path for a given devid, | |
897 | * make sure it's the same device if the device is mounted | |
79c9234b DM |
898 | * |
899 | * NOTE: the device->fs_info may not be reliable here so pass | |
900 | * in a NULL to message helpers instead. This avoids a possible | |
901 | * use-after-free when the fs_info and fs_info->sb are already | |
902 | * torn down. | |
a9261d41 AJ |
903 | */ |
904 | if (device->bdev) { | |
4889bc05 | 905 | if (device->devt != path_devt) { |
a9261d41 | 906 | mutex_unlock(&fs_devices->device_list_mutex); |
0697d9a6 | 907 | btrfs_warn_in_rcu(NULL, |
79dae17d AJ |
908 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
909 | path, devid, found_transid, | |
910 | current->comm, | |
911 | task_pid_nr(current)); | |
a9261d41 AJ |
912 | return ERR_PTR(-EEXIST); |
913 | } | |
79c9234b | 914 | btrfs_info_in_rcu(NULL, |
79dae17d | 915 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
cb3e217b | 916 | devid, btrfs_dev_name(device), |
79dae17d AJ |
917 | path, current->comm, |
918 | task_pid_nr(current)); | |
a9261d41 AJ |
919 | } |
920 | ||
606686ee | 921 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
922 | if (!name) { |
923 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 924 | return ERR_PTR(-ENOMEM); |
9c6d173e | 925 | } |
606686ee JB |
926 | rcu_string_free(device->name); |
927 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 928 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 929 | fs_devices->missing_devices--; |
e6e674bd | 930 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 931 | } |
4889bc05 | 932 | device->devt = path_devt; |
8a4b83cc CM |
933 | } |
934 | ||
77bdae4d AJ |
935 | /* |
936 | * Unmount does not free the btrfs_device struct but would zero | |
937 | * generation along with most of the other members. So just update | |
938 | * it back. We need it to pick the disk with largest generation | |
939 | * (as above). | |
940 | */ | |
d1a63002 | 941 | if (!fs_devices->opened) { |
77bdae4d | 942 | device->generation = found_transid; |
d1a63002 NB |
943 | fs_devices->latest_generation = max_t(u64, found_transid, |
944 | fs_devices->latest_generation); | |
945 | } | |
77bdae4d | 946 | |
f2788d2f AJ |
947 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
948 | ||
9c6d173e | 949 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 950 | return device; |
8a4b83cc CM |
951 | } |
952 | ||
e4404d6e YZ |
953 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
954 | { | |
955 | struct btrfs_fs_devices *fs_devices; | |
956 | struct btrfs_device *device; | |
957 | struct btrfs_device *orig_dev; | |
d2979aa2 | 958 | int ret = 0; |
e4404d6e | 959 | |
c1247069 AJ |
960 | lockdep_assert_held(&uuid_mutex); |
961 | ||
f7361d8c | 962 | fs_devices = alloc_fs_devices(orig->fsid); |
2208a378 ID |
963 | if (IS_ERR(fs_devices)) |
964 | return fs_devices; | |
e4404d6e | 965 | |
02db0844 | 966 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
967 | |
968 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
bb21e302 AJ |
969 | const char *dev_path = NULL; |
970 | ||
971 | /* | |
972 | * This is ok to do without RCU read locked because we hold the | |
973 | * uuid mutex so nothing we touch in here is going to disappear. | |
974 | */ | |
975 | if (orig_dev->name) | |
976 | dev_path = orig_dev->name->str; | |
606686ee | 977 | |
12bd2fc0 | 978 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
bb21e302 | 979 | orig_dev->uuid, dev_path); |
d2979aa2 AJ |
980 | if (IS_ERR(device)) { |
981 | ret = PTR_ERR(device); | |
e4404d6e | 982 | goto error; |
d2979aa2 | 983 | } |
e4404d6e | 984 | |
21e61ec6 JT |
985 | if (orig_dev->zone_info) { |
986 | struct btrfs_zoned_device_info *zone_info; | |
987 | ||
988 | zone_info = btrfs_clone_dev_zone_info(orig_dev); | |
989 | if (!zone_info) { | |
990 | btrfs_free_device(device); | |
991 | ret = -ENOMEM; | |
992 | goto error; | |
993 | } | |
994 | device->zone_info = zone_info; | |
995 | } | |
996 | ||
e4404d6e YZ |
997 | list_add(&device->dev_list, &fs_devices->devices); |
998 | device->fs_devices = fs_devices; | |
999 | fs_devices->num_devices++; | |
1000 | } | |
1001 | return fs_devices; | |
1002 | error: | |
1003 | free_fs_devices(fs_devices); | |
d2979aa2 | 1004 | return ERR_PTR(ret); |
e4404d6e YZ |
1005 | } |
1006 | ||
3712ccb7 | 1007 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
bacce86a | 1008 | struct btrfs_device **latest_dev) |
dfe25020 | 1009 | { |
c6e30871 | 1010 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1011 | |
46224705 | 1012 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1013 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1014 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1015 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1016 | &device->dev_state) && |
998a0671 AJ |
1017 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1018 | &device->dev_state) && | |
3712ccb7 NB |
1019 | (!*latest_dev || |
1020 | device->generation > (*latest_dev)->generation)) { | |
1021 | *latest_dev = device; | |
a6b0d5c8 | 1022 | } |
2b82032c | 1023 | continue; |
a6b0d5c8 | 1024 | } |
2b82032c | 1025 | |
cf89af14 AJ |
1026 | /* |
1027 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, | |
1028 | * in btrfs_init_dev_replace() so just continue. | |
1029 | */ | |
1030 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) | |
1031 | continue; | |
1032 | ||
9ae061cf CB |
1033 | if (device->bdev_file) { |
1034 | fput(device->bdev_file); | |
2b82032c | 1035 | device->bdev = NULL; |
9ae061cf | 1036 | device->bdev_file = NULL; |
2b82032c YZ |
1037 | fs_devices->open_devices--; |
1038 | } | |
ebbede42 | 1039 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1040 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1041 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
b2a61667 | 1042 | fs_devices->rw_devices--; |
2b82032c | 1043 | } |
e4404d6e YZ |
1044 | list_del_init(&device->dev_list); |
1045 | fs_devices->num_devices--; | |
a425f9d4 | 1046 | btrfs_free_device(device); |
dfe25020 | 1047 | } |
2b82032c | 1048 | |
3712ccb7 NB |
1049 | } |
1050 | ||
1051 | /* | |
1052 | * After we have read the system tree and know devids belonging to this | |
1053 | * filesystem, remove the device which does not belong there. | |
1054 | */ | |
bacce86a | 1055 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
3712ccb7 NB |
1056 | { |
1057 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1058 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1059 | |
1060 | mutex_lock(&uuid_mutex); | |
bacce86a | 1061 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
944d3f9f NB |
1062 | |
1063 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
bacce86a | 1064 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
2b82032c | 1065 | |
d24fa5c1 | 1066 | fs_devices->latest_dev = latest_dev; |
a6b0d5c8 | 1067 | |
dfe25020 | 1068 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1069 | } |
a0af469b | 1070 | |
14238819 AJ |
1071 | static void btrfs_close_bdev(struct btrfs_device *device) |
1072 | { | |
08ffcae8 DS |
1073 | if (!device->bdev) |
1074 | return; | |
1075 | ||
ebbede42 | 1076 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1077 | sync_blockdev(device->bdev); |
1078 | invalidate_bdev(device->bdev); | |
1079 | } | |
1080 | ||
9ae061cf | 1081 | fput(device->bdev_file); |
14238819 AJ |
1082 | } |
1083 | ||
959b1c04 | 1084 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1085 | { |
1086 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1087 | |
ebbede42 | 1088 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1089 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1090 | list_del_init(&device->dev_alloc_list); | |
1091 | fs_devices->rw_devices--; | |
1092 | } | |
1093 | ||
0d977e0e DCZX |
1094 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
1095 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); | |
1096 | ||
5d03dbeb LZ |
1097 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
1098 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); | |
f448341a | 1099 | fs_devices->missing_devices--; |
5d03dbeb | 1100 | } |
f448341a | 1101 | |
959b1c04 | 1102 | btrfs_close_bdev(device); |
321f69f8 | 1103 | if (device->bdev) { |
3fff3975 | 1104 | fs_devices->open_devices--; |
321f69f8 | 1105 | device->bdev = NULL; |
f448341a | 1106 | } |
321f69f8 | 1107 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
5b316468 | 1108 | btrfs_destroy_dev_zone_info(device); |
f448341a | 1109 | |
321f69f8 JT |
1110 | device->fs_info = NULL; |
1111 | atomic_set(&device->dev_stats_ccnt, 0); | |
1112 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1113 | |
6b225baa FM |
1114 | /* |
1115 | * Reset the flush error record. We might have a transient flush error | |
1116 | * in this mount, and if so we aborted the current transaction and set | |
1117 | * the fs to an error state, guaranteeing no super blocks can be further | |
1118 | * committed. However that error might be transient and if we unmount the | |
1119 | * filesystem and mount it again, we should allow the mount to succeed | |
1120 | * (btrfs_check_rw_degradable() should not fail) - if after mounting the | |
1121 | * filesystem again we still get flush errors, then we will again abort | |
1122 | * any transaction and set the error state, guaranteeing no commits of | |
1123 | * unsafe super blocks. | |
1124 | */ | |
1125 | device->last_flush_error = 0; | |
1126 | ||
321f69f8 | 1127 | /* Verify the device is back in a pristine state */ |
1f16033c AJ |
1128 | WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); |
1129 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1130 | WARN_ON(!list_empty(&device->dev_alloc_list)); | |
1131 | WARN_ON(!list_empty(&device->post_commit_list)); | |
f448341a AJ |
1132 | } |
1133 | ||
54eed6ae | 1134 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1135 | { |
2037a093 | 1136 | struct btrfs_device *device, *tmp; |
e4404d6e | 1137 | |
425c6ed6 JB |
1138 | lockdep_assert_held(&uuid_mutex); |
1139 | ||
2b82032c | 1140 | if (--fs_devices->opened > 0) |
54eed6ae | 1141 | return; |
8a4b83cc | 1142 | |
425c6ed6 | 1143 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1144 | btrfs_close_one_device(device); |
c9513edb | 1145 | |
e4404d6e YZ |
1146 | WARN_ON(fs_devices->open_devices); |
1147 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1148 | fs_devices->opened = 0; |
0395d84f | 1149 | fs_devices->seeding = false; |
c4989c2f | 1150 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1151 | } |
1152 | ||
54eed6ae | 1153 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1154 | { |
944d3f9f NB |
1155 | LIST_HEAD(list); |
1156 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1157 | |
1158 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1159 | close_fs_devices(fs_devices); |
5f58d783 | 1160 | if (!fs_devices->opened) { |
944d3f9f | 1161 | list_splice_init(&fs_devices->seed_list, &list); |
e4404d6e | 1162 | |
5f58d783 AJ |
1163 | /* |
1164 | * If the struct btrfs_fs_devices is not assembled with any | |
1165 | * other device, it can be re-initialized during the next mount | |
1166 | * without the needing device-scan step. Therefore, it can be | |
1167 | * fully freed. | |
1168 | */ | |
1169 | if (fs_devices->num_devices == 1) { | |
1170 | list_del(&fs_devices->fs_list); | |
1171 | free_fs_devices(fs_devices); | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | ||
944d3f9f | 1176 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
0226e0eb | 1177 | close_fs_devices(fs_devices); |
944d3f9f | 1178 | list_del(&fs_devices->seed_list); |
e4404d6e YZ |
1179 | free_fs_devices(fs_devices); |
1180 | } | |
425c6ed6 | 1181 | mutex_unlock(&uuid_mutex); |
2b82032c YZ |
1182 | } |
1183 | ||
897fb573 | 1184 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 1185 | blk_mode_t flags, void *holder) |
8a4b83cc | 1186 | { |
8a4b83cc | 1187 | struct btrfs_device *device; |
443f24fe | 1188 | struct btrfs_device *latest_dev = NULL; |
96c2e067 | 1189 | struct btrfs_device *tmp_device; |
2f1aeab9 | 1190 | int ret = 0; |
8a4b83cc | 1191 | |
96c2e067 AJ |
1192 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
1193 | dev_list) { | |
2f1aeab9 | 1194 | int ret2; |
a0af469b | 1195 | |
2f1aeab9 AJ |
1196 | ret2 = btrfs_open_one_device(fs_devices, device, flags, holder); |
1197 | if (ret2 == 0 && | |
96c2e067 | 1198 | (!latest_dev || device->generation > latest_dev->generation)) { |
9f050db4 | 1199 | latest_dev = device; |
2f1aeab9 | 1200 | } else if (ret2 == -ENODATA) { |
96c2e067 AJ |
1201 | fs_devices->num_devices--; |
1202 | list_del(&device->dev_list); | |
1203 | btrfs_free_device(device); | |
1204 | } | |
2f1aeab9 AJ |
1205 | if (ret == 0 && ret2 != 0) |
1206 | ret = ret2; | |
8a4b83cc | 1207 | } |
2f1aeab9 AJ |
1208 | |
1209 | if (fs_devices->open_devices == 0) { | |
1210 | if (ret) | |
1211 | return ret; | |
1ed802c9 | 1212 | return -EINVAL; |
2f1aeab9 | 1213 | } |
1ed802c9 | 1214 | |
2b82032c | 1215 | fs_devices->opened = 1; |
d24fa5c1 | 1216 | fs_devices->latest_dev = latest_dev; |
2b82032c | 1217 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1218 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
33fd2f71 | 1219 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
1ed802c9 AJ |
1220 | |
1221 | return 0; | |
2b82032c YZ |
1222 | } |
1223 | ||
4f0f586b ST |
1224 | static int devid_cmp(void *priv, const struct list_head *a, |
1225 | const struct list_head *b) | |
f8e10cd3 | 1226 | { |
214cc184 | 1227 | const struct btrfs_device *dev1, *dev2; |
f8e10cd3 AJ |
1228 | |
1229 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1230 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1231 | ||
1232 | if (dev1->devid < dev2->devid) | |
1233 | return -1; | |
1234 | else if (dev1->devid > dev2->devid) | |
1235 | return 1; | |
1236 | return 0; | |
1237 | } | |
1238 | ||
2b82032c | 1239 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 1240 | blk_mode_t flags, void *holder) |
2b82032c YZ |
1241 | { |
1242 | int ret; | |
1243 | ||
f5194e34 | 1244 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1245 | /* |
1246 | * The device_list_mutex cannot be taken here in case opening the | |
a8698707 | 1247 | * underlying device takes further locks like open_mutex. |
18c850fd JB |
1248 | * |
1249 | * We also don't need the lock here as this is called during mount and | |
1250 | * exclusion is provided by uuid_mutex | |
1251 | */ | |
f5194e34 | 1252 | |
2b82032c | 1253 | if (fs_devices->opened) { |
e4404d6e YZ |
1254 | fs_devices->opened++; |
1255 | ret = 0; | |
2b82032c | 1256 | } else { |
f8e10cd3 | 1257 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1258 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1259 | } |
542c5908 | 1260 | |
8a4b83cc CM |
1261 | return ret; |
1262 | } | |
1263 | ||
8f32380d | 1264 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1265 | { |
8f32380d JT |
1266 | struct page *page = virt_to_page(super); |
1267 | ||
6cf86a00 AJ |
1268 | put_page(page); |
1269 | } | |
1270 | ||
b335eab8 | 1271 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
12659251 | 1272 | u64 bytenr, u64 bytenr_orig) |
6cf86a00 | 1273 | { |
b335eab8 NB |
1274 | struct btrfs_super_block *disk_super; |
1275 | struct page *page; | |
6cf86a00 AJ |
1276 | void *p; |
1277 | pgoff_t index; | |
1278 | ||
1279 | /* make sure our super fits in the device */ | |
cda00eba | 1280 | if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) |
b335eab8 | 1281 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1282 | |
1283 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1284 | if (sizeof(*disk_super) > PAGE_SIZE) |
1285 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1286 | |
1287 | /* make sure our super doesn't straddle pages on disk */ | |
1288 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1289 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1290 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1291 | |
1292 | /* pull in the page with our super */ | |
224941e8 | 1293 | page = read_cache_page_gfp(bdev->bd_mapping, index, GFP_KERNEL); |
6cf86a00 | 1294 | |
b335eab8 NB |
1295 | if (IS_ERR(page)) |
1296 | return ERR_CAST(page); | |
6cf86a00 | 1297 | |
b335eab8 | 1298 | p = page_address(page); |
6cf86a00 AJ |
1299 | |
1300 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1301 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1302 | |
12659251 | 1303 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
b335eab8 | 1304 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
8f32380d | 1305 | btrfs_release_disk_super(p); |
b335eab8 | 1306 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1307 | } |
1308 | ||
b335eab8 NB |
1309 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1310 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1311 | |
b335eab8 | 1312 | return disk_super; |
6cf86a00 AJ |
1313 | } |
1314 | ||
16cab91a | 1315 | int btrfs_forget_devices(dev_t devt) |
228a73ab AJ |
1316 | { |
1317 | int ret; | |
1318 | ||
1319 | mutex_lock(&uuid_mutex); | |
16cab91a | 1320 | ret = btrfs_free_stale_devices(devt, NULL); |
228a73ab AJ |
1321 | mutex_unlock(&uuid_mutex); |
1322 | ||
1323 | return ret; | |
1324 | } | |
1325 | ||
d565fffa AJ |
1326 | static bool btrfs_skip_registration(struct btrfs_super_block *disk_super, |
1327 | const char *path, dev_t devt, | |
1328 | bool mount_arg_dev) | |
1329 | { | |
1330 | struct btrfs_fs_devices *fs_devices; | |
1331 | ||
1332 | /* | |
1333 | * Do not skip device registration for mounted devices with matching | |
1334 | * maj:min but different paths. Booting without initrd relies on | |
1335 | * /dev/root initially, later replaced with the actual root device. | |
1336 | * A successful scan ensures grub2-probe selects the correct device. | |
1337 | */ | |
1338 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
1339 | struct btrfs_device *device; | |
1340 | ||
1341 | mutex_lock(&fs_devices->device_list_mutex); | |
1342 | ||
1343 | if (!fs_devices->opened) { | |
1344 | mutex_unlock(&fs_devices->device_list_mutex); | |
1345 | continue; | |
1346 | } | |
1347 | ||
1348 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
1349 | if (device->bdev && (device->bdev->bd_dev == devt) && | |
1350 | strcmp(device->name->str, path) != 0) { | |
1351 | mutex_unlock(&fs_devices->device_list_mutex); | |
1352 | ||
1353 | /* Do not skip registration. */ | |
1354 | return false; | |
1355 | } | |
1356 | } | |
1357 | mutex_unlock(&fs_devices->device_list_mutex); | |
1358 | } | |
1359 | ||
1360 | if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 && | |
1361 | !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)) | |
1362 | return true; | |
1363 | ||
1364 | return false; | |
1365 | } | |
1366 | ||
6f60cbd3 DS |
1367 | /* |
1368 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1369 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
bc27d6f0 AJ |
1370 | * is read via pagecache. |
1371 | * | |
1372 | * With @mount_arg_dev it's a scan during mount time that will always register | |
1373 | * the device or return an error. Multi-device and seeding devices are registered | |
1374 | * in both cases. | |
6f60cbd3 | 1375 | */ |
bc27d6f0 AJ |
1376 | struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags, |
1377 | bool mount_arg_dev) | |
8a4b83cc CM |
1378 | { |
1379 | struct btrfs_super_block *disk_super; | |
4306a974 | 1380 | bool new_device_added = false; |
36350e95 | 1381 | struct btrfs_device *device = NULL; |
9ae061cf | 1382 | struct file *bdev_file; |
12659251 | 1383 | u64 bytenr, bytenr_orig; |
d565fffa | 1384 | dev_t devt; |
12659251 | 1385 | int ret; |
8a4b83cc | 1386 | |
899f9307 DS |
1387 | lockdep_assert_held(&uuid_mutex); |
1388 | ||
6f60cbd3 DS |
1389 | /* |
1390 | * we would like to check all the supers, but that would make | |
1391 | * a btrfs mount succeed after a mkfs from a different FS. | |
1392 | * So, we need to add a special mount option to scan for | |
1393 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1394 | */ | |
6f60cbd3 | 1395 | |
50d281fc | 1396 | /* |
2ef78928 CH |
1397 | * Avoid an exclusive open here, as the systemd-udev may initiate the |
1398 | * device scan which may race with the user's mount or mkfs command, | |
1399 | * resulting in failure. | |
1400 | * Since the device scan is solely for reading purposes, there is no | |
1401 | * need for an exclusive open. Additionally, the devices are read again | |
50d281fc AJ |
1402 | * during the mount process. It is ok to get some inconsistent |
1403 | * values temporarily, as the device paths of the fsid are the only | |
1404 | * required information for assembling the volume. | |
1405 | */ | |
9ae061cf CB |
1406 | bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL); |
1407 | if (IS_ERR(bdev_file)) | |
1408 | return ERR_CAST(bdev_file); | |
6f60cbd3 | 1409 | |
12659251 | 1410 | bytenr_orig = btrfs_sb_offset(0); |
9ae061cf | 1411 | ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr); |
4989d4a0 SK |
1412 | if (ret) { |
1413 | device = ERR_PTR(ret); | |
1414 | goto error_bdev_put; | |
1415 | } | |
12659251 | 1416 | |
9ae061cf | 1417 | disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr, |
86ec15d0 | 1418 | bytenr_orig); |
b335eab8 NB |
1419 | if (IS_ERR(disk_super)) { |
1420 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1421 | goto error_bdev_put; |
05a5c55d | 1422 | } |
6f60cbd3 | 1423 | |
d565fffa AJ |
1424 | devt = file_bdev(bdev_file)->bd_dev; |
1425 | if (btrfs_skip_registration(disk_super, path, devt, mount_arg_dev)) { | |
1426 | pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n", | |
1427 | path, MAJOR(devt), MINOR(devt)); | |
bc27d6f0 | 1428 | |
d565fffa | 1429 | btrfs_free_stale_devices(devt, NULL); |
bc27d6f0 | 1430 | |
bc27d6f0 AJ |
1431 | device = NULL; |
1432 | goto free_disk_super; | |
1433 | } | |
1434 | ||
4306a974 | 1435 | device = device_list_add(path, disk_super, &new_device_added); |
4889bc05 AJ |
1436 | if (!IS_ERR(device) && new_device_added) |
1437 | btrfs_free_stale_devices(device->devt, device); | |
6f60cbd3 | 1438 | |
bc27d6f0 | 1439 | free_disk_super: |
8f32380d | 1440 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1441 | |
1442 | error_bdev_put: | |
9ae061cf | 1443 | fput(bdev_file); |
b6ed73bc | 1444 | |
36350e95 | 1445 | return device; |
8a4b83cc | 1446 | } |
0b86a832 | 1447 | |
1c11b63e JM |
1448 | /* |
1449 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1450 | * such is found, record the end of it in *start | |
1451 | */ | |
1c11b63e JM |
1452 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1453 | u64 len) | |
6df9a95e | 1454 | { |
1c11b63e | 1455 | u64 physical_start, physical_end; |
6df9a95e | 1456 | |
1c11b63e | 1457 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1458 | |
e5860f82 FM |
1459 | if (find_first_extent_bit(&device->alloc_state, *start, |
1460 | &physical_start, &physical_end, | |
1461 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1462 | |
1c11b63e JM |
1463 | if (in_range(physical_start, *start, len) || |
1464 | in_range(*start, physical_start, | |
ae6bd7f9 | 1465 | physical_end + 1 - physical_start)) { |
1c11b63e JM |
1466 | *start = physical_end + 1; |
1467 | return true; | |
6df9a95e JB |
1468 | } |
1469 | } | |
1c11b63e | 1470 | return false; |
6df9a95e JB |
1471 | } |
1472 | ||
ed8947bc | 1473 | static u64 dev_extent_search_start(struct btrfs_device *device) |
3b4ffa40 NA |
1474 | { |
1475 | switch (device->fs_devices->chunk_alloc_policy) { | |
1476 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
ed8947bc | 1477 | return BTRFS_DEVICE_RANGE_RESERVED; |
1cd6121f NA |
1478 | case BTRFS_CHUNK_ALLOC_ZONED: |
1479 | /* | |
1480 | * We don't care about the starting region like regular | |
1481 | * allocator, because we anyway use/reserve the first two zones | |
1482 | * for superblock logging. | |
1483 | */ | |
ed8947bc | 1484 | return 0; |
3b4ffa40 NA |
1485 | default: |
1486 | BUG(); | |
1487 | } | |
1488 | } | |
1489 | ||
1cd6121f NA |
1490 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
1491 | u64 *hole_start, u64 *hole_size, | |
1492 | u64 num_bytes) | |
1493 | { | |
1494 | u64 zone_size = device->zone_info->zone_size; | |
1495 | u64 pos; | |
1496 | int ret; | |
1497 | bool changed = false; | |
1498 | ||
1499 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); | |
1500 | ||
1501 | while (*hole_size > 0) { | |
1502 | pos = btrfs_find_allocatable_zones(device, *hole_start, | |
1503 | *hole_start + *hole_size, | |
1504 | num_bytes); | |
1505 | if (pos != *hole_start) { | |
1506 | *hole_size = *hole_start + *hole_size - pos; | |
1507 | *hole_start = pos; | |
1508 | changed = true; | |
1509 | if (*hole_size < num_bytes) | |
1510 | break; | |
1511 | } | |
1512 | ||
1513 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); | |
1514 | ||
1515 | /* Range is ensured to be empty */ | |
1516 | if (!ret) | |
1517 | return changed; | |
1518 | ||
1519 | /* Given hole range was invalid (outside of device) */ | |
1520 | if (ret == -ERANGE) { | |
1521 | *hole_start += *hole_size; | |
d6f67afb | 1522 | *hole_size = 0; |
7000babd | 1523 | return true; |
1cd6121f NA |
1524 | } |
1525 | ||
1526 | *hole_start += zone_size; | |
1527 | *hole_size -= zone_size; | |
1528 | changed = true; | |
1529 | } | |
1530 | ||
1531 | return changed; | |
1532 | } | |
1533 | ||
43dd529a DS |
1534 | /* |
1535 | * Check if specified hole is suitable for allocation. | |
1536 | * | |
3b4ffa40 NA |
1537 | * @device: the device which we have the hole |
1538 | * @hole_start: starting position of the hole | |
1539 | * @hole_size: the size of the hole | |
1540 | * @num_bytes: the size of the free space that we need | |
1541 | * | |
1cd6121f | 1542 | * This function may modify @hole_start and @hole_size to reflect the suitable |
3b4ffa40 NA |
1543 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
1544 | */ | |
1545 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1546 | u64 *hole_size, u64 num_bytes) | |
1547 | { | |
1548 | bool changed = false; | |
1549 | u64 hole_end = *hole_start + *hole_size; | |
1550 | ||
1cd6121f NA |
1551 | for (;;) { |
1552 | /* | |
1553 | * Check before we set max_hole_start, otherwise we could end up | |
1554 | * sending back this offset anyway. | |
1555 | */ | |
1556 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1557 | if (hole_end >= *hole_start) | |
1558 | *hole_size = hole_end - *hole_start; | |
1559 | else | |
1560 | *hole_size = 0; | |
1561 | changed = true; | |
1562 | } | |
1563 | ||
1564 | switch (device->fs_devices->chunk_alloc_policy) { | |
1565 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1566 | /* No extra check */ | |
1567 | break; | |
1568 | case BTRFS_CHUNK_ALLOC_ZONED: | |
1569 | if (dev_extent_hole_check_zoned(device, hole_start, | |
1570 | hole_size, num_bytes)) { | |
1571 | changed = true; | |
1572 | /* | |
1573 | * The changed hole can contain pending extent. | |
1574 | * Loop again to check that. | |
1575 | */ | |
1576 | continue; | |
1577 | } | |
1578 | break; | |
1579 | default: | |
1580 | BUG(); | |
1581 | } | |
3b4ffa40 | 1582 | |
3b4ffa40 | 1583 | break; |
3b4ffa40 NA |
1584 | } |
1585 | ||
1586 | return changed; | |
1587 | } | |
6df9a95e | 1588 | |
0b86a832 | 1589 | /* |
43dd529a DS |
1590 | * Find free space in the specified device. |
1591 | * | |
499f377f JM |
1592 | * @device: the device which we search the free space in |
1593 | * @num_bytes: the size of the free space that we need | |
1594 | * @search_start: the position from which to begin the search | |
1595 | * @start: store the start of the free space. | |
1596 | * @len: the size of the free space. that we find, or the size | |
1597 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1598 | * |
43dd529a DS |
1599 | * This does a pretty simple search, the expectation is that it is called very |
1600 | * infrequently and that a given device has a small number of extents. | |
7bfc837d MX |
1601 | * |
1602 | * @start is used to store the start of the free space if we find. But if we | |
1603 | * don't find suitable free space, it will be used to store the start position | |
1604 | * of the max free space. | |
1605 | * | |
1606 | * @len is used to store the size of the free space that we find. | |
1607 | * But if we don't find suitable free space, it is used to store the size of | |
1608 | * the max free space. | |
135da976 QW |
1609 | * |
1610 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1611 | * check to ensure dev extents are not double allocated. | |
1612 | * This makes the function safe to allocate dev extents but may not report | |
1613 | * correct usable device space, as device extent freed in current transaction | |
1a9fd417 | 1614 | * is not reported as available. |
0b86a832 | 1615 | */ |
ed8947bc FM |
1616 | static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
1617 | u64 *start, u64 *len) | |
0b86a832 | 1618 | { |
0b246afa JM |
1619 | struct btrfs_fs_info *fs_info = device->fs_info; |
1620 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1621 | struct btrfs_key key; |
7bfc837d | 1622 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1623 | struct btrfs_path *path; |
ed8947bc | 1624 | u64 search_start; |
7bfc837d MX |
1625 | u64 hole_size; |
1626 | u64 max_hole_start; | |
20218dfb | 1627 | u64 max_hole_size = 0; |
7bfc837d | 1628 | u64 extent_end; |
0b86a832 CM |
1629 | u64 search_end = device->total_bytes; |
1630 | int ret; | |
7bfc837d | 1631 | int slot; |
0b86a832 | 1632 | struct extent_buffer *l; |
8cdc7c5b | 1633 | |
ed8947bc | 1634 | search_start = dev_extent_search_start(device); |
20218dfb | 1635 | max_hole_start = search_start; |
0b86a832 | 1636 | |
1cd6121f NA |
1637 | WARN_ON(device->zone_info && |
1638 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); | |
1639 | ||
6df9a95e | 1640 | path = btrfs_alloc_path(); |
20218dfb JB |
1641 | if (!path) { |
1642 | ret = -ENOMEM; | |
1643 | goto out; | |
1644 | } | |
f2ab7618 | 1645 | again: |
401e29c1 AJ |
1646 | if (search_start >= search_end || |
1647 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1648 | ret = -ENOSPC; |
6df9a95e | 1649 | goto out; |
7bfc837d MX |
1650 | } |
1651 | ||
e4058b54 | 1652 | path->reada = READA_FORWARD; |
6df9a95e JB |
1653 | path->search_commit_root = 1; |
1654 | path->skip_locking = 1; | |
7bfc837d | 1655 | |
0b86a832 CM |
1656 | key.objectid = device->devid; |
1657 | key.offset = search_start; | |
1658 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1659 | |
0ff40a91 | 1660 | ret = btrfs_search_backwards(root, &key, path); |
0b86a832 | 1661 | if (ret < 0) |
7bfc837d | 1662 | goto out; |
7bfc837d | 1663 | |
3c538de0 | 1664 | while (search_start < search_end) { |
0b86a832 CM |
1665 | l = path->nodes[0]; |
1666 | slot = path->slots[0]; | |
1667 | if (slot >= btrfs_header_nritems(l)) { | |
1668 | ret = btrfs_next_leaf(root, path); | |
1669 | if (ret == 0) | |
1670 | continue; | |
1671 | if (ret < 0) | |
7bfc837d MX |
1672 | goto out; |
1673 | ||
1674 | break; | |
0b86a832 CM |
1675 | } |
1676 | btrfs_item_key_to_cpu(l, &key, slot); | |
1677 | ||
1678 | if (key.objectid < device->devid) | |
1679 | goto next; | |
1680 | ||
1681 | if (key.objectid > device->devid) | |
7bfc837d | 1682 | break; |
0b86a832 | 1683 | |
962a298f | 1684 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1685 | goto next; |
9779b72f | 1686 | |
3c538de0 JB |
1687 | if (key.offset > search_end) |
1688 | break; | |
1689 | ||
7bfc837d MX |
1690 | if (key.offset > search_start) { |
1691 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1692 | dev_extent_hole_check(device, &search_start, &hole_size, |
1693 | num_bytes); | |
6df9a95e | 1694 | |
7bfc837d MX |
1695 | if (hole_size > max_hole_size) { |
1696 | max_hole_start = search_start; | |
1697 | max_hole_size = hole_size; | |
1698 | } | |
9779b72f | 1699 | |
7bfc837d MX |
1700 | /* |
1701 | * If this free space is greater than which we need, | |
1702 | * it must be the max free space that we have found | |
1703 | * until now, so max_hole_start must point to the start | |
1704 | * of this free space and the length of this free space | |
1705 | * is stored in max_hole_size. Thus, we return | |
1706 | * max_hole_start and max_hole_size and go back to the | |
1707 | * caller. | |
1708 | */ | |
1709 | if (hole_size >= num_bytes) { | |
1710 | ret = 0; | |
1711 | goto out; | |
0b86a832 CM |
1712 | } |
1713 | } | |
0b86a832 | 1714 | |
0b86a832 | 1715 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1716 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1717 | dev_extent); | |
1718 | if (extent_end > search_start) | |
1719 | search_start = extent_end; | |
0b86a832 CM |
1720 | next: |
1721 | path->slots[0]++; | |
1722 | cond_resched(); | |
1723 | } | |
0b86a832 | 1724 | |
38c01b96 | 1725 | /* |
1726 | * At this point, search_start should be the end of | |
1727 | * allocated dev extents, and when shrinking the device, | |
1728 | * search_end may be smaller than search_start. | |
1729 | */ | |
f2ab7618 | 1730 | if (search_end > search_start) { |
38c01b96 | 1731 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1732 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1733 | num_bytes)) { | |
f2ab7618 ZL |
1734 | btrfs_release_path(path); |
1735 | goto again; | |
1736 | } | |
0b86a832 | 1737 | |
f2ab7618 ZL |
1738 | if (hole_size > max_hole_size) { |
1739 | max_hole_start = search_start; | |
1740 | max_hole_size = hole_size; | |
1741 | } | |
6df9a95e JB |
1742 | } |
1743 | ||
7bfc837d | 1744 | /* See above. */ |
f2ab7618 | 1745 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1746 | ret = -ENOSPC; |
1747 | else | |
1748 | ret = 0; | |
1749 | ||
3c538de0 | 1750 | ASSERT(max_hole_start + max_hole_size <= search_end); |
7bfc837d | 1751 | out: |
2b82032c | 1752 | btrfs_free_path(path); |
7bfc837d | 1753 | *start = max_hole_start; |
b2117a39 | 1754 | if (len) |
7bfc837d | 1755 | *len = max_hole_size; |
0b86a832 CM |
1756 | return ret; |
1757 | } | |
1758 | ||
b2950863 | 1759 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1760 | struct btrfs_device *device, |
2196d6e8 | 1761 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1762 | { |
0b246afa JM |
1763 | struct btrfs_fs_info *fs_info = device->fs_info; |
1764 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1765 | int ret; |
1766 | struct btrfs_path *path; | |
8f18cf13 | 1767 | struct btrfs_key key; |
a061fc8d CM |
1768 | struct btrfs_key found_key; |
1769 | struct extent_buffer *leaf = NULL; | |
1770 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1771 | |
1772 | path = btrfs_alloc_path(); | |
1773 | if (!path) | |
1774 | return -ENOMEM; | |
1775 | ||
1776 | key.objectid = device->devid; | |
1777 | key.offset = start; | |
1778 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1779 | again: |
8f18cf13 | 1780 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1781 | if (ret > 0) { |
1782 | ret = btrfs_previous_item(root, path, key.objectid, | |
1783 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1784 | if (ret) |
1785 | goto out; | |
a061fc8d CM |
1786 | leaf = path->nodes[0]; |
1787 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1788 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1789 | struct btrfs_dev_extent); | |
1790 | BUG_ON(found_key.offset > start || found_key.offset + | |
1791 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1792 | key = found_key; |
1793 | btrfs_release_path(path); | |
1794 | goto again; | |
a061fc8d CM |
1795 | } else if (ret == 0) { |
1796 | leaf = path->nodes[0]; | |
1797 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1798 | struct btrfs_dev_extent); | |
79787eaa | 1799 | } else { |
79787eaa | 1800 | goto out; |
a061fc8d | 1801 | } |
8f18cf13 | 1802 | |
2196d6e8 MX |
1803 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1804 | ||
8f18cf13 | 1805 | ret = btrfs_del_item(trans, root, path); |
79bd3712 | 1806 | if (ret == 0) |
3204d33c | 1807 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
b0b802d7 | 1808 | out: |
8f18cf13 CM |
1809 | btrfs_free_path(path); |
1810 | return ret; | |
1811 | } | |
1812 | ||
6df9a95e | 1813 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1814 | { |
6df9a95e JB |
1815 | struct rb_node *n; |
1816 | u64 ret = 0; | |
0b86a832 | 1817 | |
7dc66abb FM |
1818 | read_lock(&fs_info->mapping_tree_lock); |
1819 | n = rb_last(&fs_info->mapping_tree.rb_root); | |
6df9a95e | 1820 | if (n) { |
7dc66abb FM |
1821 | struct btrfs_chunk_map *map; |
1822 | ||
1823 | map = rb_entry(n, struct btrfs_chunk_map, rb_node); | |
1824 | ret = map->start + map->chunk_len; | |
0b86a832 | 1825 | } |
7dc66abb | 1826 | read_unlock(&fs_info->mapping_tree_lock); |
6df9a95e | 1827 | |
0b86a832 CM |
1828 | return ret; |
1829 | } | |
1830 | ||
53f10659 ID |
1831 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1832 | u64 *devid_ret) | |
0b86a832 CM |
1833 | { |
1834 | int ret; | |
1835 | struct btrfs_key key; | |
1836 | struct btrfs_key found_key; | |
2b82032c YZ |
1837 | struct btrfs_path *path; |
1838 | ||
2b82032c YZ |
1839 | path = btrfs_alloc_path(); |
1840 | if (!path) | |
1841 | return -ENOMEM; | |
0b86a832 CM |
1842 | |
1843 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1844 | key.type = BTRFS_DEV_ITEM_KEY; | |
1845 | key.offset = (u64)-1; | |
1846 | ||
53f10659 | 1847 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1848 | if (ret < 0) |
1849 | goto error; | |
1850 | ||
a06dee4d AJ |
1851 | if (ret == 0) { |
1852 | /* Corruption */ | |
1853 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1854 | ret = -EUCLEAN; | |
1855 | goto error; | |
1856 | } | |
0b86a832 | 1857 | |
53f10659 ID |
1858 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1859 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1860 | BTRFS_DEV_ITEM_KEY); |
1861 | if (ret) { | |
53f10659 | 1862 | *devid_ret = 1; |
0b86a832 CM |
1863 | } else { |
1864 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1865 | path->slots[0]); | |
53f10659 | 1866 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1867 | } |
1868 | ret = 0; | |
1869 | error: | |
2b82032c | 1870 | btrfs_free_path(path); |
0b86a832 CM |
1871 | return ret; |
1872 | } | |
1873 | ||
1874 | /* | |
1875 | * the device information is stored in the chunk root | |
1876 | * the btrfs_device struct should be fully filled in | |
1877 | */ | |
c74a0b02 | 1878 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1879 | struct btrfs_device *device) |
0b86a832 CM |
1880 | { |
1881 | int ret; | |
1882 | struct btrfs_path *path; | |
1883 | struct btrfs_dev_item *dev_item; | |
1884 | struct extent_buffer *leaf; | |
1885 | struct btrfs_key key; | |
1886 | unsigned long ptr; | |
0b86a832 | 1887 | |
0b86a832 CM |
1888 | path = btrfs_alloc_path(); |
1889 | if (!path) | |
1890 | return -ENOMEM; | |
1891 | ||
0b86a832 CM |
1892 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1893 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1894 | key.offset = device->devid; |
0b86a832 | 1895 | |
2bb2e00e | 1896 | btrfs_reserve_chunk_metadata(trans, true); |
8e87e856 NB |
1897 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1898 | &key, sizeof(*dev_item)); | |
2bb2e00e | 1899 | btrfs_trans_release_chunk_metadata(trans); |
0b86a832 CM |
1900 | if (ret) |
1901 | goto out; | |
1902 | ||
1903 | leaf = path->nodes[0]; | |
1904 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1905 | ||
1906 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1907 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1908 | btrfs_set_device_type(leaf, dev_item, device->type); |
1909 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1910 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1911 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1912 | btrfs_set_device_total_bytes(leaf, dev_item, |
1913 | btrfs_device_get_disk_total_bytes(device)); | |
1914 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1915 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1916 | btrfs_set_device_group(leaf, dev_item, 0); |
1917 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1918 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1919 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1920 | |
410ba3a2 | 1921 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1922 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1923 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1924 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1925 | ptr, BTRFS_FSID_SIZE); | |
50564b65 | 1926 | btrfs_mark_buffer_dirty(trans, leaf); |
0b86a832 | 1927 | |
2b82032c | 1928 | ret = 0; |
0b86a832 CM |
1929 | out: |
1930 | btrfs_free_path(path); | |
1931 | return ret; | |
1932 | } | |
8f18cf13 | 1933 | |
5a1972bd QW |
1934 | /* |
1935 | * Function to update ctime/mtime for a given device path. | |
1936 | * Mainly used for ctime/mtime based probe like libblkid. | |
54fde91f JB |
1937 | * |
1938 | * We don't care about errors here, this is just to be kind to userspace. | |
5a1972bd | 1939 | */ |
54fde91f | 1940 | static void update_dev_time(const char *device_path) |
5a1972bd | 1941 | { |
54fde91f | 1942 | struct path path; |
54fde91f | 1943 | int ret; |
5a1972bd | 1944 | |
54fde91f JB |
1945 | ret = kern_path(device_path, LOOKUP_FOLLOW, &path); |
1946 | if (ret) | |
5a1972bd | 1947 | return; |
8f96a5bf | 1948 | |
913e9928 | 1949 | inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION); |
54fde91f | 1950 | path_put(&path); |
5a1972bd QW |
1951 | } |
1952 | ||
bbac5869 QW |
1953 | static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, |
1954 | struct btrfs_device *device) | |
a061fc8d | 1955 | { |
f331a952 | 1956 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1957 | int ret; |
1958 | struct btrfs_path *path; | |
a061fc8d | 1959 | struct btrfs_key key; |
a061fc8d | 1960 | |
a061fc8d CM |
1961 | path = btrfs_alloc_path(); |
1962 | if (!path) | |
1963 | return -ENOMEM; | |
1964 | ||
a061fc8d CM |
1965 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1966 | key.type = BTRFS_DEV_ITEM_KEY; | |
1967 | key.offset = device->devid; | |
1968 | ||
2bb2e00e | 1969 | btrfs_reserve_chunk_metadata(trans, false); |
a061fc8d | 1970 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
2bb2e00e | 1971 | btrfs_trans_release_chunk_metadata(trans); |
5e9f2ad5 NB |
1972 | if (ret) { |
1973 | if (ret > 0) | |
1974 | ret = -ENOENT; | |
a061fc8d CM |
1975 | goto out; |
1976 | } | |
1977 | ||
1978 | ret = btrfs_del_item(trans, root, path); | |
a061fc8d CM |
1979 | out: |
1980 | btrfs_free_path(path); | |
a061fc8d CM |
1981 | return ret; |
1982 | } | |
1983 | ||
3cc31a0d DS |
1984 | /* |
1985 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1986 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1987 | * replace. | |
1988 | */ | |
1989 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1990 | u64 num_devices) | |
a061fc8d | 1991 | { |
a061fc8d | 1992 | u64 all_avail; |
de98ced9 | 1993 | unsigned seq; |
418775a2 | 1994 | int i; |
a061fc8d | 1995 | |
de98ced9 | 1996 | do { |
bd45ffbc | 1997 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1998 | |
bd45ffbc AJ |
1999 | all_avail = fs_info->avail_data_alloc_bits | |
2000 | fs_info->avail_system_alloc_bits | | |
2001 | fs_info->avail_metadata_alloc_bits; | |
2002 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 2003 | |
418775a2 | 2004 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 2005 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 2006 | continue; |
a061fc8d | 2007 | |
efc222f8 AJ |
2008 | if (num_devices < btrfs_raid_array[i].devs_min) |
2009 | return btrfs_raid_array[i].mindev_error; | |
53b381b3 DW |
2010 | } |
2011 | ||
bd45ffbc | 2012 | return 0; |
f1fa7f26 AJ |
2013 | } |
2014 | ||
c9162bdf OS |
2015 | static struct btrfs_device * btrfs_find_next_active_device( |
2016 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 2017 | { |
2b82032c | 2018 | struct btrfs_device *next_device; |
88acff64 AJ |
2019 | |
2020 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
2021 | if (next_device != device && | |
e6e674bd AJ |
2022 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
2023 | && next_device->bdev) | |
88acff64 AJ |
2024 | return next_device; |
2025 | } | |
2026 | ||
2027 | return NULL; | |
2028 | } | |
2029 | ||
2030 | /* | |
d24fa5c1 | 2031 | * Helper function to check if the given device is part of s_bdev / latest_dev |
88acff64 AJ |
2032 | * and replace it with the provided or the next active device, in the context |
2033 | * where this function called, there should be always be another device (or | |
2034 | * this_dev) which is active. | |
2035 | */ | |
b105e927 | 2036 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
e493e8f9 | 2037 | struct btrfs_device *next_device) |
88acff64 | 2038 | { |
d6507cf1 | 2039 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 | 2040 | |
e493e8f9 | 2041 | if (!next_device) |
88acff64 | 2042 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
e493e8f9 | 2043 | device); |
88acff64 AJ |
2044 | ASSERT(next_device); |
2045 | ||
2046 | if (fs_info->sb->s_bdev && | |
2047 | (fs_info->sb->s_bdev == device->bdev)) | |
2048 | fs_info->sb->s_bdev = next_device->bdev; | |
2049 | ||
d24fa5c1 AJ |
2050 | if (fs_info->fs_devices->latest_dev->bdev == device->bdev) |
2051 | fs_info->fs_devices->latest_dev = next_device; | |
88acff64 AJ |
2052 | } |
2053 | ||
1da73967 AJ |
2054 | /* |
2055 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2056 | * currently replaced. | |
2057 | */ | |
2058 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2059 | { | |
2060 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2061 | ||
cb5583dd | 2062 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2063 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2064 | ASSERT(num_devices > 1); | |
2065 | num_devices--; | |
2066 | } | |
cb5583dd | 2067 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2068 | |
2069 | return num_devices; | |
2070 | } | |
2071 | ||
0e0078f7 CH |
2072 | static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, |
2073 | struct block_device *bdev, int copy_num) | |
2074 | { | |
2075 | struct btrfs_super_block *disk_super; | |
26ecf243 CH |
2076 | const size_t len = sizeof(disk_super->magic); |
2077 | const u64 bytenr = btrfs_sb_offset(copy_num); | |
0e0078f7 CH |
2078 | int ret; |
2079 | ||
26ecf243 | 2080 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); |
0e0078f7 CH |
2081 | if (IS_ERR(disk_super)) |
2082 | return; | |
2083 | ||
26ecf243 CH |
2084 | memset(&disk_super->magic, 0, len); |
2085 | folio_mark_dirty(virt_to_folio(disk_super)); | |
2086 | btrfs_release_disk_super(disk_super); | |
2087 | ||
2088 | ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); | |
0e0078f7 CH |
2089 | if (ret) |
2090 | btrfs_warn(fs_info, "error clearing superblock number %d (%d)", | |
2091 | copy_num, ret); | |
0e0078f7 CH |
2092 | } |
2093 | ||
1cdeac6d | 2094 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device) |
6fbceb9f | 2095 | { |
6fbceb9f | 2096 | int copy_num; |
1cdeac6d | 2097 | struct block_device *bdev = device->bdev; |
6fbceb9f JT |
2098 | |
2099 | if (!bdev) | |
2100 | return; | |
2101 | ||
2102 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
0e0078f7 | 2103 | if (bdev_is_zoned(bdev)) |
12659251 | 2104 | btrfs_reset_sb_log_zones(bdev, copy_num); |
0e0078f7 CH |
2105 | else |
2106 | btrfs_scratch_superblock(fs_info, bdev, copy_num); | |
6fbceb9f JT |
2107 | } |
2108 | ||
2109 | /* Notify udev that device has changed */ | |
2110 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2111 | ||
2112 | /* Update ctime/mtime for device path for libblkid */ | |
1cdeac6d | 2113 | update_dev_time(device->name->str); |
6fbceb9f JT |
2114 | } |
2115 | ||
1a15eb72 JB |
2116 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, |
2117 | struct btrfs_dev_lookup_args *args, | |
9ae061cf | 2118 | struct file **bdev_file) |
f1fa7f26 | 2119 | { |
bbac5869 | 2120 | struct btrfs_trans_handle *trans; |
f1fa7f26 | 2121 | struct btrfs_device *device; |
1f78160c | 2122 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2123 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2124 | u64 num_devices; |
a061fc8d CM |
2125 | int ret = 0; |
2126 | ||
914a519b JB |
2127 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2128 | btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); | |
2129 | return -EINVAL; | |
2130 | } | |
2131 | ||
8ef9dc0f JB |
2132 | /* |
2133 | * The device list in fs_devices is accessed without locks (neither | |
2134 | * uuid_mutex nor device_list_mutex) as it won't change on a mounted | |
2135 | * filesystem and another device rm cannot run. | |
2136 | */ | |
1da73967 | 2137 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2138 | |
0b246afa | 2139 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2140 | if (ret) |
bbac5869 | 2141 | return ret; |
a061fc8d | 2142 | |
1a15eb72 JB |
2143 | device = btrfs_find_device(fs_info->fs_devices, args); |
2144 | if (!device) { | |
2145 | if (args->missing) | |
a27a94c2 NB |
2146 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
2147 | else | |
1a15eb72 | 2148 | ret = -ENOENT; |
bbac5869 | 2149 | return ret; |
a27a94c2 | 2150 | } |
dfe25020 | 2151 | |
eede2bf3 OS |
2152 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2153 | btrfs_warn_in_rcu(fs_info, | |
2154 | "cannot remove device %s (devid %llu) due to active swapfile", | |
cb3e217b | 2155 | btrfs_dev_name(device), device->devid); |
bbac5869 | 2156 | return -ETXTBSY; |
eede2bf3 OS |
2157 | } |
2158 | ||
bbac5869 QW |
2159 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
2160 | return BTRFS_ERROR_DEV_TGT_REPLACE; | |
63a212ab | 2161 | |
ebbede42 | 2162 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
bbac5869 QW |
2163 | fs_info->fs_devices->rw_devices == 1) |
2164 | return BTRFS_ERROR_DEV_ONLY_WRITABLE; | |
2b82032c | 2165 | |
ebbede42 | 2166 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2167 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2168 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2169 | device->fs_devices->rw_devices--; |
34441361 | 2170 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2171 | } |
a061fc8d CM |
2172 | |
2173 | ret = btrfs_shrink_device(device, 0); | |
2174 | if (ret) | |
9b3517e9 | 2175 | goto error_undo; |
a061fc8d | 2176 | |
bbac5869 QW |
2177 | trans = btrfs_start_transaction(fs_info->chunk_root, 0); |
2178 | if (IS_ERR(trans)) { | |
2179 | ret = PTR_ERR(trans); | |
9b3517e9 | 2180 | goto error_undo; |
bbac5869 QW |
2181 | } |
2182 | ||
2183 | ret = btrfs_rm_dev_item(trans, device); | |
2184 | if (ret) { | |
2185 | /* Any error in dev item removal is critical */ | |
2186 | btrfs_crit(fs_info, | |
2187 | "failed to remove device item for devid %llu: %d", | |
2188 | device->devid, ret); | |
2189 | btrfs_abort_transaction(trans, ret); | |
2190 | btrfs_end_transaction(trans); | |
2191 | return ret; | |
2192 | } | |
a061fc8d | 2193 | |
e12c9621 | 2194 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2195 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2196 | |
2197 | /* | |
2198 | * the device list mutex makes sure that we don't change | |
2199 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2200 | * the device supers. Whoever is writing all supers, should |
2201 | * lock the device list mutex before getting the number of | |
2202 | * devices in the super block (super_copy). Conversely, | |
2203 | * whoever updates the number of devices in the super block | |
2204 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2205 | */ |
1f78160c | 2206 | |
41a52a0f AJ |
2207 | /* |
2208 | * In normal cases the cur_devices == fs_devices. But in case | |
2209 | * of deleting a seed device, the cur_devices should point to | |
9675ea8c | 2210 | * its own fs_devices listed under the fs_devices->seed_list. |
41a52a0f | 2211 | */ |
1f78160c | 2212 | cur_devices = device->fs_devices; |
b5185197 | 2213 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2214 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2215 | |
41a52a0f AJ |
2216 | cur_devices->num_devices--; |
2217 | cur_devices->total_devices--; | |
b4993e64 AJ |
2218 | /* Update total_devices of the parent fs_devices if it's seed */ |
2219 | if (cur_devices != fs_devices) | |
2220 | fs_devices->total_devices--; | |
2b82032c | 2221 | |
e6e674bd | 2222 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2223 | cur_devices->missing_devices--; |
cd02dca5 | 2224 | |
d6507cf1 | 2225 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2226 | |
9ae061cf | 2227 | if (device->bdev_file) { |
41a52a0f | 2228 | cur_devices->open_devices--; |
0bfaa9c5 | 2229 | /* remove sysfs entry */ |
53f8a74c | 2230 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2231 | } |
99994cde | 2232 | |
0b246afa JM |
2233 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2234 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2235 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2236 | |
cea67ab9 | 2237 | /* |
3fa421de JB |
2238 | * At this point, the device is zero sized and detached from the |
2239 | * devices list. All that's left is to zero out the old supers and | |
2240 | * free the device. | |
2241 | * | |
2242 | * We cannot call btrfs_close_bdev() here because we're holding the sb | |
9ae061cf CB |
2243 | * write lock, and fput() on the block device will pull in the |
2244 | * ->open_mutex on the block device and it's dependencies. Instead | |
2245 | * just flush the device and let the caller do the final bdev_release. | |
cea67ab9 | 2246 | */ |
3fa421de | 2247 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
1cdeac6d | 2248 | btrfs_scratch_superblocks(fs_info, device); |
3fa421de JB |
2249 | if (device->bdev) { |
2250 | sync_blockdev(device->bdev); | |
2251 | invalidate_bdev(device->bdev); | |
2252 | } | |
2253 | } | |
cea67ab9 | 2254 | |
9ae061cf | 2255 | *bdev_file = device->bdev_file; |
8e75fd89 NB |
2256 | synchronize_rcu(); |
2257 | btrfs_free_device(device); | |
cea67ab9 | 2258 | |
8b41393f JB |
2259 | /* |
2260 | * This can happen if cur_devices is the private seed devices list. We | |
2261 | * cannot call close_fs_devices() here because it expects the uuid_mutex | |
2262 | * to be held, but in fact we don't need that for the private | |
2263 | * seed_devices, we can simply decrement cur_devices->opened and then | |
2264 | * remove it from our list and free the fs_devices. | |
2265 | */ | |
8e906945 | 2266 | if (cur_devices->num_devices == 0) { |
944d3f9f | 2267 | list_del_init(&cur_devices->seed_list); |
8b41393f JB |
2268 | ASSERT(cur_devices->opened == 1); |
2269 | cur_devices->opened--; | |
1f78160c | 2270 | free_fs_devices(cur_devices); |
2b82032c YZ |
2271 | } |
2272 | ||
bbac5869 QW |
2273 | ret = btrfs_commit_transaction(trans); |
2274 | ||
a061fc8d | 2275 | return ret; |
24fc572f | 2276 | |
9b3517e9 | 2277 | error_undo: |
ebbede42 | 2278 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2279 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2280 | list_add(&device->dev_alloc_list, |
b5185197 | 2281 | &fs_devices->alloc_list); |
c3929c36 | 2282 | device->fs_devices->rw_devices++; |
34441361 | 2283 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2284 | } |
bbac5869 | 2285 | return ret; |
a061fc8d CM |
2286 | } |
2287 | ||
68a9db5f | 2288 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2289 | { |
d51908ce AJ |
2290 | struct btrfs_fs_devices *fs_devices; |
2291 | ||
68a9db5f | 2292 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2293 | |
25e8e911 AJ |
2294 | /* |
2295 | * in case of fs with no seed, srcdev->fs_devices will point | |
2296 | * to fs_devices of fs_info. However when the dev being replaced is | |
2297 | * a seed dev it will point to the seed's local fs_devices. In short | |
2298 | * srcdev will have its correct fs_devices in both the cases. | |
2299 | */ | |
2300 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2301 | |
e93c89c1 | 2302 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2303 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2304 | fs_devices->num_devices--; |
e6e674bd | 2305 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2306 | fs_devices->missing_devices--; |
e93c89c1 | 2307 | |
ebbede42 | 2308 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2309 | fs_devices->rw_devices--; |
1357272f | 2310 | |
82372bc8 | 2311 | if (srcdev->bdev) |
d51908ce | 2312 | fs_devices->open_devices--; |
084b6e7c QW |
2313 | } |
2314 | ||
65237ee3 | 2315 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2316 | { |
2317 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2318 | |
a466c85e JB |
2319 | mutex_lock(&uuid_mutex); |
2320 | ||
14238819 | 2321 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2322 | synchronize_rcu(); |
2323 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2324 | |
94d5f0c2 AJ |
2325 | /* if this is no devs we rather delete the fs_devices */ |
2326 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2327 | /* |
2328 | * On a mounted FS, num_devices can't be zero unless it's a | |
2329 | * seed. In case of a seed device being replaced, the replace | |
2330 | * target added to the sprout FS, so there will be no more | |
2331 | * device left under the seed FS. | |
2332 | */ | |
2333 | ASSERT(fs_devices->seeding); | |
2334 | ||
944d3f9f | 2335 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2336 | close_fs_devices(fs_devices); |
8bef8401 | 2337 | free_fs_devices(fs_devices); |
94d5f0c2 | 2338 | } |
a466c85e | 2339 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2340 | } |
2341 | ||
4f5ad7bd | 2342 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2343 | { |
4f5ad7bd | 2344 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2345 | |
d9a071f0 | 2346 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2347 | |
53f8a74c | 2348 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2349 | |
779bf3fe | 2350 | if (tgtdev->bdev) |
d9a071f0 | 2351 | fs_devices->open_devices--; |
779bf3fe | 2352 | |
d9a071f0 | 2353 | fs_devices->num_devices--; |
e93c89c1 | 2354 | |
d6507cf1 | 2355 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2356 | |
e93c89c1 | 2357 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2358 | |
d9a071f0 | 2359 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe | 2360 | |
1cdeac6d | 2361 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev); |
14238819 AJ |
2362 | |
2363 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2364 | synchronize_rcu(); |
2365 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2366 | } |
2367 | ||
43dd529a DS |
2368 | /* |
2369 | * Populate args from device at path. | |
faa775c4 JB |
2370 | * |
2371 | * @fs_info: the filesystem | |
2372 | * @args: the args to populate | |
2373 | * @path: the path to the device | |
2374 | * | |
2375 | * This will read the super block of the device at @path and populate @args with | |
2376 | * the devid, fsid, and uuid. This is meant to be used for ioctls that need to | |
2377 | * lookup a device to operate on, but need to do it before we take any locks. | |
2378 | * This properly handles the special case of "missing" that a user may pass in, | |
2379 | * and does some basic sanity checks. The caller must make sure that @path is | |
2380 | * properly NUL terminated before calling in, and must call | |
2381 | * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and | |
2382 | * uuid buffers. | |
2383 | * | |
2384 | * Return: 0 for success, -errno for failure | |
2385 | */ | |
2386 | int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, | |
2387 | struct btrfs_dev_lookup_args *args, | |
2388 | const char *path) | |
7ba15b7d | 2389 | { |
7ba15b7d | 2390 | struct btrfs_super_block *disk_super; |
9ae061cf | 2391 | struct file *bdev_file; |
faa775c4 | 2392 | int ret; |
7ba15b7d | 2393 | |
faa775c4 JB |
2394 | if (!path || !path[0]) |
2395 | return -EINVAL; | |
2396 | if (!strcmp(path, "missing")) { | |
2397 | args->missing = true; | |
2398 | return 0; | |
2399 | } | |
8f32380d | 2400 | |
faa775c4 JB |
2401 | args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); |
2402 | args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); | |
2403 | if (!args->uuid || !args->fsid) { | |
2404 | btrfs_put_dev_args_from_path(args); | |
2405 | return -ENOMEM; | |
2406 | } | |
8f32380d | 2407 | |
05bdb996 | 2408 | ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0, |
9ae061cf | 2409 | &bdev_file, &disk_super); |
9ea0106a ZF |
2410 | if (ret) { |
2411 | btrfs_put_dev_args_from_path(args); | |
faa775c4 | 2412 | return ret; |
9ea0106a ZF |
2413 | } |
2414 | ||
faa775c4 JB |
2415 | args->devid = btrfs_stack_device_id(&disk_super->dev_item); |
2416 | memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); | |
7239ff4b | 2417 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
faa775c4 | 2418 | memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); |
7239ff4b | 2419 | else |
faa775c4 | 2420 | memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); |
8f32380d | 2421 | btrfs_release_disk_super(disk_super); |
9ae061cf | 2422 | fput(bdev_file); |
faa775c4 | 2423 | return 0; |
7ba15b7d SB |
2424 | } |
2425 | ||
5c5c0df0 | 2426 | /* |
faa775c4 JB |
2427 | * Only use this jointly with btrfs_get_dev_args_from_path() because we will |
2428 | * allocate our ->uuid and ->fsid pointers, everybody else uses local variables | |
2429 | * that don't need to be freed. | |
5c5c0df0 | 2430 | */ |
faa775c4 JB |
2431 | void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) |
2432 | { | |
2433 | kfree(args->uuid); | |
2434 | kfree(args->fsid); | |
2435 | args->uuid = NULL; | |
2436 | args->fsid = NULL; | |
2437 | } | |
2438 | ||
a27a94c2 | 2439 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2440 | struct btrfs_fs_info *fs_info, u64 devid, |
2441 | const char *device_path) | |
24e0474b | 2442 | { |
562d7b15 | 2443 | BTRFS_DEV_LOOKUP_ARGS(args); |
a27a94c2 | 2444 | struct btrfs_device *device; |
faa775c4 | 2445 | int ret; |
24e0474b | 2446 | |
5c5c0df0 | 2447 | if (devid) { |
562d7b15 JB |
2448 | args.devid = devid; |
2449 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
a27a94c2 NB |
2450 | if (!device) |
2451 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2452 | return device; |
2453 | } | |
2454 | ||
faa775c4 JB |
2455 | ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); |
2456 | if (ret) | |
2457 | return ERR_PTR(ret); | |
2458 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
2459 | btrfs_put_dev_args_from_path(&args); | |
2460 | if (!device) | |
6e927ceb | 2461 | return ERR_PTR(-ENOENT); |
faa775c4 | 2462 | return device; |
24e0474b AJ |
2463 | } |
2464 | ||
849eae5e | 2465 | static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2466 | { |
0b246afa | 2467 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2468 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2469 | struct btrfs_fs_devices *seed_devices; |
2b82032c | 2470 | |
a32bf9a3 | 2471 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2472 | if (!fs_devices->seeding) |
849eae5e | 2473 | return ERR_PTR(-EINVAL); |
2b82032c | 2474 | |
427c8fdd NB |
2475 | /* |
2476 | * Private copy of the seed devices, anchored at | |
2477 | * fs_info->fs_devices->seed_list | |
2478 | */ | |
f7361d8c | 2479 | seed_devices = alloc_fs_devices(NULL); |
2208a378 | 2480 | if (IS_ERR(seed_devices)) |
849eae5e | 2481 | return seed_devices; |
2b82032c | 2482 | |
427c8fdd NB |
2483 | /* |
2484 | * It's necessary to retain a copy of the original seed fs_devices in | |
2485 | * fs_uuids so that filesystems which have been seeded can successfully | |
2486 | * reference the seed device from open_seed_devices. This also supports | |
2487 | * multiple fs seed. | |
2488 | */ | |
e4404d6e YZ |
2489 | old_devices = clone_fs_devices(fs_devices); |
2490 | if (IS_ERR(old_devices)) { | |
2491 | kfree(seed_devices); | |
849eae5e | 2492 | return old_devices; |
2b82032c | 2493 | } |
e4404d6e | 2494 | |
c4babc5e | 2495 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2496 | |
e4404d6e YZ |
2497 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2498 | seed_devices->opened = 1; | |
2499 | INIT_LIST_HEAD(&seed_devices->devices); | |
2500 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2501 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2502 | |
849eae5e AJ |
2503 | return seed_devices; |
2504 | } | |
2505 | ||
2506 | /* | |
2507 | * Splice seed devices into the sprout fs_devices. | |
2508 | * Generate a new fsid for the sprouted read-write filesystem. | |
2509 | */ | |
2510 | static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, | |
2511 | struct btrfs_fs_devices *seed_devices) | |
2512 | { | |
2513 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2514 | struct btrfs_super_block *disk_super = fs_info->super_copy; | |
2515 | struct btrfs_device *device; | |
2516 | u64 super_flags; | |
2517 | ||
2518 | /* | |
2519 | * We are updating the fsid, the thread leading to device_list_add() | |
2520 | * could race, so uuid_mutex is needed. | |
2521 | */ | |
2522 | lockdep_assert_held(&uuid_mutex); | |
2523 | ||
2524 | /* | |
2525 | * The threads listed below may traverse dev_list but can do that without | |
2526 | * device_list_mutex: | |
2527 | * - All device ops and balance - as we are in btrfs_exclop_start. | |
2528 | * - Various dev_list readers - are using RCU. | |
2529 | * - btrfs_ioctl_fitrim() - is using RCU. | |
2530 | * | |
2531 | * For-read threads as below are using device_list_mutex: | |
2532 | * - Readonly scrub btrfs_scrub_dev() | |
2533 | * - Readonly scrub btrfs_scrub_progress() | |
2534 | * - btrfs_get_dev_stats() | |
2535 | */ | |
2536 | lockdep_assert_held(&fs_devices->device_list_mutex); | |
2537 | ||
1f78160c XG |
2538 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2539 | synchronize_rcu); | |
2196d6e8 MX |
2540 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2541 | device->fs_devices = seed_devices; | |
c9513edb | 2542 | |
0395d84f | 2543 | fs_devices->seeding = false; |
2b82032c YZ |
2544 | fs_devices->num_devices = 0; |
2545 | fs_devices->open_devices = 0; | |
69611ac8 | 2546 | fs_devices->missing_devices = 0; |
7f0432d0 | 2547 | fs_devices->rotating = false; |
944d3f9f | 2548 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2549 | |
2550 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2551 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2552 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
f7171750 | 2553 | |
2b82032c YZ |
2554 | super_flags = btrfs_super_flags(disk_super) & |
2555 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2556 | btrfs_set_super_flags(disk_super, super_flags); | |
2b82032c YZ |
2557 | } |
2558 | ||
2559 | /* | |
01327610 | 2560 | * Store the expected generation for seed devices in device items. |
2b82032c | 2561 | */ |
5c466629 | 2562 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2563 | { |
562d7b15 | 2564 | BTRFS_DEV_LOOKUP_ARGS(args); |
5c466629 | 2565 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2566 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2567 | struct btrfs_path *path; |
2568 | struct extent_buffer *leaf; | |
2569 | struct btrfs_dev_item *dev_item; | |
2570 | struct btrfs_device *device; | |
2571 | struct btrfs_key key; | |
44880fdc | 2572 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c | 2573 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2b82032c YZ |
2574 | int ret; |
2575 | ||
2576 | path = btrfs_alloc_path(); | |
2577 | if (!path) | |
2578 | return -ENOMEM; | |
2579 | ||
2b82032c YZ |
2580 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2581 | key.offset = 0; | |
2582 | key.type = BTRFS_DEV_ITEM_KEY; | |
2583 | ||
2584 | while (1) { | |
2bb2e00e | 2585 | btrfs_reserve_chunk_metadata(trans, false); |
2b82032c | 2586 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
2bb2e00e | 2587 | btrfs_trans_release_chunk_metadata(trans); |
2b82032c YZ |
2588 | if (ret < 0) |
2589 | goto error; | |
2590 | ||
2591 | leaf = path->nodes[0]; | |
2592 | next_slot: | |
2593 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2594 | ret = btrfs_next_leaf(root, path); | |
2595 | if (ret > 0) | |
2596 | break; | |
2597 | if (ret < 0) | |
2598 | goto error; | |
2599 | leaf = path->nodes[0]; | |
2600 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2601 | btrfs_release_path(path); |
2b82032c YZ |
2602 | continue; |
2603 | } | |
2604 | ||
2605 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2606 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2607 | key.type != BTRFS_DEV_ITEM_KEY) | |
2608 | break; | |
2609 | ||
2610 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2611 | struct btrfs_dev_item); | |
562d7b15 | 2612 | args.devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 2613 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2614 | BTRFS_UUID_SIZE); |
1473b24e | 2615 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2616 | BTRFS_FSID_SIZE); |
562d7b15 JB |
2617 | args.uuid = dev_uuid; |
2618 | args.fsid = fs_uuid; | |
2619 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
79787eaa | 2620 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2621 | |
2622 | if (device->fs_devices->seeding) { | |
2623 | btrfs_set_device_generation(leaf, dev_item, | |
2624 | device->generation); | |
50564b65 | 2625 | btrfs_mark_buffer_dirty(trans, leaf); |
2b82032c YZ |
2626 | } |
2627 | ||
2628 | path->slots[0]++; | |
2629 | goto next_slot; | |
2630 | } | |
2631 | ret = 0; | |
2632 | error: | |
2633 | btrfs_free_path(path); | |
2634 | return ret; | |
2635 | } | |
2636 | ||
da353f6b | 2637 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2638 | { |
5112febb | 2639 | struct btrfs_root *root = fs_info->dev_root; |
788f20eb CM |
2640 | struct btrfs_trans_handle *trans; |
2641 | struct btrfs_device *device; | |
9ae061cf | 2642 | struct file *bdev_file; |
0b246afa | 2643 | struct super_block *sb = fs_info->sb; |
5da54bc1 | 2644 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8ba7d5f5 | 2645 | struct btrfs_fs_devices *seed_devices = NULL; |
39379faa NA |
2646 | u64 orig_super_total_bytes; |
2647 | u64 orig_super_num_devices; | |
788f20eb | 2648 | int ret = 0; |
fd880809 | 2649 | bool seeding_dev = false; |
44cab9ba | 2650 | bool locked = false; |
788f20eb | 2651 | |
5da54bc1 | 2652 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2653 | return -EROFS; |
788f20eb | 2654 | |
9ae061cf | 2655 | bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE, |
86ec15d0 | 2656 | fs_info->bdev_holder, NULL); |
9ae061cf CB |
2657 | if (IS_ERR(bdev_file)) |
2658 | return PTR_ERR(bdev_file); | |
a2135011 | 2659 | |
9ae061cf | 2660 | if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) { |
b70f5097 NA |
2661 | ret = -EINVAL; |
2662 | goto error; | |
2663 | } | |
2664 | ||
5da54bc1 | 2665 | if (fs_devices->seeding) { |
fd880809 | 2666 | seeding_dev = true; |
2b82032c YZ |
2667 | down_write(&sb->s_umount); |
2668 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2669 | locked = true; |
2b82032c YZ |
2670 | } |
2671 | ||
9ae061cf | 2672 | sync_blockdev(file_bdev(bdev_file)); |
a2135011 | 2673 | |
f4cfa9bd NB |
2674 | rcu_read_lock(); |
2675 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
9ae061cf | 2676 | if (device->bdev == file_bdev(bdev_file)) { |
788f20eb | 2677 | ret = -EEXIST; |
f4cfa9bd | 2678 | rcu_read_unlock(); |
2b82032c | 2679 | goto error; |
788f20eb CM |
2680 | } |
2681 | } | |
f4cfa9bd | 2682 | rcu_read_unlock(); |
788f20eb | 2683 | |
bb21e302 | 2684 | device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); |
12bd2fc0 | 2685 | if (IS_ERR(device)) { |
788f20eb | 2686 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2687 | ret = PTR_ERR(device); |
2b82032c | 2688 | goto error; |
788f20eb CM |
2689 | } |
2690 | ||
5b316468 | 2691 | device->fs_info = fs_info; |
9ae061cf CB |
2692 | device->bdev_file = bdev_file; |
2693 | device->bdev = file_bdev(bdev_file); | |
4889bc05 AJ |
2694 | ret = lookup_bdev(device_path, &device->devt); |
2695 | if (ret) | |
2696 | goto error_free_device; | |
5b316468 | 2697 | |
16beac87 | 2698 | ret = btrfs_get_dev_zone_info(device, false); |
5b316468 NA |
2699 | if (ret) |
2700 | goto error_free_device; | |
2701 | ||
a22285a6 | 2702 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2703 | if (IS_ERR(trans)) { |
98d5dc13 | 2704 | ret = PTR_ERR(trans); |
5b316468 | 2705 | goto error_free_zone; |
98d5dc13 TI |
2706 | } |
2707 | ||
ebbede42 | 2708 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2709 | device->generation = trans->transid; |
0b246afa JM |
2710 | device->io_width = fs_info->sectorsize; |
2711 | device->io_align = fs_info->sectorsize; | |
2712 | device->sector_size = fs_info->sectorsize; | |
cda00eba | 2713 | device->total_bytes = |
86ec15d0 | 2714 | round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize); |
2cc3c559 | 2715 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2716 | device->commit_total_bytes = device->total_bytes; |
e12c9621 | 2717 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2718 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
27087f37 | 2719 | device->dev_stats_valid = 1; |
ead083ae | 2720 | set_blocksize(device->bdev_file, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2721 | |
2b82032c | 2722 | if (seeding_dev) { |
a0a1db70 | 2723 | btrfs_clear_sb_rdonly(sb); |
849eae5e AJ |
2724 | |
2725 | /* GFP_KERNEL allocation must not be under device_list_mutex */ | |
2726 | seed_devices = btrfs_init_sprout(fs_info); | |
2727 | if (IS_ERR(seed_devices)) { | |
2728 | ret = PTR_ERR(seed_devices); | |
d31c32f6 AJ |
2729 | btrfs_abort_transaction(trans, ret); |
2730 | goto error_trans; | |
2731 | } | |
849eae5e AJ |
2732 | } |
2733 | ||
2734 | mutex_lock(&fs_devices->device_list_mutex); | |
2735 | if (seeding_dev) { | |
2736 | btrfs_setup_sprout(fs_info, seed_devices); | |
b7cb29e6 AJ |
2737 | btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, |
2738 | device); | |
2b82032c | 2739 | } |
788f20eb | 2740 | |
5da54bc1 | 2741 | device->fs_devices = fs_devices; |
e5e9a520 | 2742 | |
34441361 | 2743 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2744 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2745 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2746 | fs_devices->num_devices++; | |
2747 | fs_devices->open_devices++; | |
2748 | fs_devices->rw_devices++; | |
2749 | fs_devices->total_devices++; | |
2750 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2751 | |
a5ed45f8 | 2752 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2753 | |
86ec15d0 | 2754 | if (!bdev_nonrot(device->bdev)) |
7f0432d0 | 2755 | fs_devices->rotating = true; |
c289811c | 2756 | |
39379faa | 2757 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2758 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2759 | round_down(orig_super_total_bytes + device->total_bytes, |
2760 | fs_info->sectorsize)); | |
788f20eb | 2761 | |
39379faa NA |
2762 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2763 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2764 | orig_super_num_devices + 1); | |
0d39376a | 2765 | |
2196d6e8 MX |
2766 | /* |
2767 | * we've got more storage, clear any full flags on the space | |
2768 | * infos | |
2769 | */ | |
0b246afa | 2770 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2771 | |
34441361 | 2772 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2773 | |
2774 | /* Add sysfs device entry */ | |
cd36da2e | 2775 | btrfs_sysfs_add_device(device); |
ca10845a | 2776 | |
5da54bc1 | 2777 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2778 | |
2b82032c | 2779 | if (seeding_dev) { |
34441361 | 2780 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2781 | ret = init_first_rw_device(trans); |
34441361 | 2782 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2783 | if (ret) { |
66642832 | 2784 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2785 | goto error_sysfs; |
005d6427 | 2786 | } |
2196d6e8 MX |
2787 | } |
2788 | ||
8e87e856 | 2789 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2790 | if (ret) { |
66642832 | 2791 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2792 | goto error_sysfs; |
2196d6e8 MX |
2793 | } |
2794 | ||
2795 | if (seeding_dev) { | |
5c466629 | 2796 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2797 | if (ret) { |
66642832 | 2798 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2799 | goto error_sysfs; |
005d6427 | 2800 | } |
b2373f25 | 2801 | |
8e560081 NB |
2802 | /* |
2803 | * fs_devices now represents the newly sprouted filesystem and | |
849eae5e | 2804 | * its fsid has been changed by btrfs_sprout_splice(). |
8e560081 NB |
2805 | */ |
2806 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2807 | } |
2808 | ||
3a45bb20 | 2809 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2810 | |
2b82032c YZ |
2811 | if (seeding_dev) { |
2812 | mutex_unlock(&uuid_mutex); | |
2813 | up_write(&sb->s_umount); | |
44cab9ba | 2814 | locked = false; |
788f20eb | 2815 | |
79787eaa JM |
2816 | if (ret) /* transaction commit */ |
2817 | return ret; | |
2818 | ||
2ff7e61e | 2819 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2820 | if (ret < 0) |
0b246afa | 2821 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2822 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2823 | trans = btrfs_attach_transaction(root); |
2824 | if (IS_ERR(trans)) { | |
2825 | if (PTR_ERR(trans) == -ENOENT) | |
2826 | return 0; | |
7132a262 AJ |
2827 | ret = PTR_ERR(trans); |
2828 | trans = NULL; | |
2829 | goto error_sysfs; | |
671415b7 | 2830 | } |
3a45bb20 | 2831 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2832 | } |
c9e9f97b | 2833 | |
7f551d96 AJ |
2834 | /* |
2835 | * Now that we have written a new super block to this device, check all | |
2836 | * other fs_devices list if device_path alienates any other scanned | |
2837 | * device. | |
2838 | * We can ignore the return value as it typically returns -EINVAL and | |
2839 | * only succeeds if the device was an alien. | |
2840 | */ | |
4889bc05 | 2841 | btrfs_forget_devices(device->devt); |
7f551d96 AJ |
2842 | |
2843 | /* Update ctime/mtime for blkid or udev */ | |
54fde91f | 2844 | update_dev_time(device_path); |
7f551d96 | 2845 | |
2b82032c | 2846 | return ret; |
79787eaa | 2847 | |
d31c32f6 | 2848 | error_sysfs: |
53f8a74c | 2849 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2850 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2851 | mutex_lock(&fs_info->chunk_mutex); | |
2852 | list_del_rcu(&device->dev_list); | |
2853 | list_del(&device->dev_alloc_list); | |
2854 | fs_info->fs_devices->num_devices--; | |
2855 | fs_info->fs_devices->open_devices--; | |
2856 | fs_info->fs_devices->rw_devices--; | |
2857 | fs_info->fs_devices->total_devices--; | |
2858 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2859 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2860 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2861 | orig_super_total_bytes); | |
2862 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2863 | orig_super_num_devices); | |
2864 | mutex_unlock(&fs_info->chunk_mutex); | |
2865 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2866 | error_trans: |
0af2c4bf | 2867 | if (seeding_dev) |
a0a1db70 | 2868 | btrfs_set_sb_rdonly(sb); |
7132a262 AJ |
2869 | if (trans) |
2870 | btrfs_end_transaction(trans); | |
5b316468 NA |
2871 | error_free_zone: |
2872 | btrfs_destroy_dev_zone_info(device); | |
5c4cf6c9 | 2873 | error_free_device: |
a425f9d4 | 2874 | btrfs_free_device(device); |
2b82032c | 2875 | error: |
9ae061cf | 2876 | fput(bdev_file); |
44cab9ba | 2877 | if (locked) { |
2b82032c YZ |
2878 | mutex_unlock(&uuid_mutex); |
2879 | up_write(&sb->s_umount); | |
2880 | } | |
c9e9f97b | 2881 | return ret; |
788f20eb CM |
2882 | } |
2883 | ||
d397712b CM |
2884 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2885 | struct btrfs_device *device) | |
0b86a832 CM |
2886 | { |
2887 | int ret; | |
2888 | struct btrfs_path *path; | |
0b246afa | 2889 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2890 | struct btrfs_dev_item *dev_item; |
2891 | struct extent_buffer *leaf; | |
2892 | struct btrfs_key key; | |
2893 | ||
0b86a832 CM |
2894 | path = btrfs_alloc_path(); |
2895 | if (!path) | |
2896 | return -ENOMEM; | |
2897 | ||
2898 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2899 | key.type = BTRFS_DEV_ITEM_KEY; | |
2900 | key.offset = device->devid; | |
2901 | ||
2902 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2903 | if (ret < 0) | |
2904 | goto out; | |
2905 | ||
2906 | if (ret > 0) { | |
2907 | ret = -ENOENT; | |
2908 | goto out; | |
2909 | } | |
2910 | ||
2911 | leaf = path->nodes[0]; | |
2912 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2913 | ||
2914 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2915 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2916 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2917 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2918 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2919 | btrfs_set_device_total_bytes(leaf, dev_item, |
2920 | btrfs_device_get_disk_total_bytes(device)); | |
2921 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2922 | btrfs_device_get_bytes_used(device)); | |
50564b65 | 2923 | btrfs_mark_buffer_dirty(trans, leaf); |
0b86a832 CM |
2924 | |
2925 | out: | |
2926 | btrfs_free_path(path); | |
2927 | return ret; | |
2928 | } | |
2929 | ||
2196d6e8 | 2930 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2931 | struct btrfs_device *device, u64 new_size) |
2932 | { | |
0b246afa JM |
2933 | struct btrfs_fs_info *fs_info = device->fs_info; |
2934 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2935 | u64 old_total; |
2936 | u64 diff; | |
2bb2e00e | 2937 | int ret; |
8f18cf13 | 2938 | |
ebbede42 | 2939 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2940 | return -EACCES; |
2196d6e8 | 2941 | |
7dfb8be1 NB |
2942 | new_size = round_down(new_size, fs_info->sectorsize); |
2943 | ||
34441361 | 2944 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2945 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2946 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2947 | |
63a212ab | 2948 | if (new_size <= device->total_bytes || |
401e29c1 | 2949 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2950 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2951 | return -EINVAL; |
2196d6e8 | 2952 | } |
2b82032c | 2953 | |
7dfb8be1 NB |
2954 | btrfs_set_super_total_bytes(super_copy, |
2955 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c | 2956 | device->fs_devices->total_rw_bytes += diff; |
6f2d3c01 | 2957 | atomic64_add(diff, &fs_info->free_chunk_space); |
2b82032c | 2958 | |
7cc8e58d MX |
2959 | btrfs_device_set_total_bytes(device, new_size); |
2960 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2961 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2962 | if (list_empty(&device->post_commit_list)) |
2963 | list_add_tail(&device->post_commit_list, | |
2964 | &trans->transaction->dev_update_list); | |
34441361 | 2965 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2966 | |
2bb2e00e FM |
2967 | btrfs_reserve_chunk_metadata(trans, false); |
2968 | ret = btrfs_update_device(trans, device); | |
2969 | btrfs_trans_release_chunk_metadata(trans); | |
2970 | ||
2971 | return ret; | |
8f18cf13 CM |
2972 | } |
2973 | ||
f4208794 | 2974 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2975 | { |
f4208794 | 2976 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2977 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2978 | int ret; |
2979 | struct btrfs_path *path; | |
2980 | struct btrfs_key key; | |
2981 | ||
8f18cf13 CM |
2982 | path = btrfs_alloc_path(); |
2983 | if (!path) | |
2984 | return -ENOMEM; | |
2985 | ||
408fbf19 | 2986 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2987 | key.offset = chunk_offset; |
2988 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2989 | ||
2990 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2991 | if (ret < 0) |
2992 | goto out; | |
2993 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2994 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2995 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2996 | ret = -ENOENT; |
2997 | goto out; | |
2998 | } | |
8f18cf13 CM |
2999 | |
3000 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 3001 | if (ret < 0) |
0b246afa JM |
3002 | btrfs_handle_fs_error(fs_info, ret, |
3003 | "Failed to delete chunk item."); | |
79787eaa | 3004 | out: |
8f18cf13 | 3005 | btrfs_free_path(path); |
65a246c5 | 3006 | return ret; |
8f18cf13 CM |
3007 | } |
3008 | ||
408fbf19 | 3009 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 3010 | { |
0b246afa | 3011 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
3012 | struct btrfs_disk_key *disk_key; |
3013 | struct btrfs_chunk *chunk; | |
3014 | u8 *ptr; | |
3015 | int ret = 0; | |
3016 | u32 num_stripes; | |
3017 | u32 array_size; | |
3018 | u32 len = 0; | |
3019 | u32 cur; | |
3020 | struct btrfs_key key; | |
3021 | ||
79bd3712 | 3022 | lockdep_assert_held(&fs_info->chunk_mutex); |
8f18cf13 CM |
3023 | array_size = btrfs_super_sys_array_size(super_copy); |
3024 | ||
3025 | ptr = super_copy->sys_chunk_array; | |
3026 | cur = 0; | |
3027 | ||
3028 | while (cur < array_size) { | |
3029 | disk_key = (struct btrfs_disk_key *)ptr; | |
3030 | btrfs_disk_key_to_cpu(&key, disk_key); | |
3031 | ||
3032 | len = sizeof(*disk_key); | |
3033 | ||
3034 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
3035 | chunk = (struct btrfs_chunk *)(ptr + len); | |
3036 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
3037 | len += btrfs_chunk_item_size(num_stripes); | |
3038 | } else { | |
3039 | ret = -EIO; | |
3040 | break; | |
3041 | } | |
408fbf19 | 3042 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
3043 | key.offset == chunk_offset) { |
3044 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
3045 | array_size -= len; | |
3046 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
3047 | } else { | |
3048 | ptr += len; | |
3049 | cur += len; | |
3050 | } | |
3051 | } | |
3052 | return ret; | |
3053 | } | |
3054 | ||
7dc66abb FM |
3055 | struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info, |
3056 | u64 logical, u64 length) | |
3057 | { | |
3058 | struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node; | |
3059 | struct rb_node *prev = NULL; | |
3060 | struct rb_node *orig_prev; | |
3061 | struct btrfs_chunk_map *map; | |
3062 | struct btrfs_chunk_map *prev_map = NULL; | |
3063 | ||
3064 | while (node) { | |
3065 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); | |
3066 | prev = node; | |
3067 | prev_map = map; | |
3068 | ||
3069 | if (logical < map->start) { | |
3070 | node = node->rb_left; | |
3071 | } else if (logical >= map->start + map->chunk_len) { | |
3072 | node = node->rb_right; | |
3073 | } else { | |
3074 | refcount_inc(&map->refs); | |
3075 | return map; | |
3076 | } | |
3077 | } | |
3078 | ||
3079 | if (!prev) | |
3080 | return NULL; | |
3081 | ||
3082 | orig_prev = prev; | |
3083 | while (prev && logical >= prev_map->start + prev_map->chunk_len) { | |
3084 | prev = rb_next(prev); | |
3085 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); | |
3086 | } | |
3087 | ||
3088 | if (!prev) { | |
3089 | prev = orig_prev; | |
3090 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); | |
3091 | while (prev && logical < prev_map->start) { | |
3092 | prev = rb_prev(prev); | |
3093 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); | |
3094 | } | |
3095 | } | |
3096 | ||
3097 | if (prev) { | |
3098 | u64 end = logical + length; | |
3099 | ||
3100 | /* | |
3101 | * Caller can pass a U64_MAX length when it wants to get any | |
3102 | * chunk starting at an offset of 'logical' or higher, so deal | |
3103 | * with underflow by resetting the end offset to U64_MAX. | |
3104 | */ | |
3105 | if (end < logical) | |
3106 | end = U64_MAX; | |
3107 | ||
3108 | if (end > prev_map->start && | |
3109 | logical < prev_map->start + prev_map->chunk_len) { | |
3110 | refcount_inc(&prev_map->refs); | |
3111 | return prev_map; | |
3112 | } | |
3113 | } | |
3114 | ||
3115 | return NULL; | |
3116 | } | |
3117 | ||
3118 | struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info, | |
3119 | u64 logical, u64 length) | |
3120 | { | |
3121 | struct btrfs_chunk_map *map; | |
3122 | ||
3123 | read_lock(&fs_info->mapping_tree_lock); | |
3124 | map = btrfs_find_chunk_map_nolock(fs_info, logical, length); | |
3125 | read_unlock(&fs_info->mapping_tree_lock); | |
3126 | ||
3127 | return map; | |
3128 | } | |
3129 | ||
60ca842e | 3130 | /* |
9580503b DS |
3131 | * Find the mapping containing the given logical extent. |
3132 | * | |
60ca842e OS |
3133 | * @logical: Logical block offset in bytes. |
3134 | * @length: Length of extent in bytes. | |
3135 | * | |
3136 | * Return: Chunk mapping or ERR_PTR. | |
3137 | */ | |
7dc66abb FM |
3138 | struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, |
3139 | u64 logical, u64 length) | |
592d92ee | 3140 | { |
7dc66abb | 3141 | struct btrfs_chunk_map *map; |
592d92ee | 3142 | |
7dc66abb | 3143 | map = btrfs_find_chunk_map(fs_info, logical, length); |
592d92ee | 3144 | |
7dc66abb | 3145 | if (unlikely(!map)) { |
7d410d5e FM |
3146 | btrfs_crit(fs_info, |
3147 | "unable to find chunk map for logical %llu length %llu", | |
592d92ee LB |
3148 | logical, length); |
3149 | return ERR_PTR(-EINVAL); | |
3150 | } | |
3151 | ||
7dc66abb | 3152 | if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) { |
592d92ee | 3153 | btrfs_crit(fs_info, |
7d410d5e | 3154 | "found a bad chunk map, wanted %llu-%llu, found %llu-%llu", |
7dc66abb FM |
3155 | logical, logical + length, map->start, |
3156 | map->start + map->chunk_len); | |
3157 | btrfs_free_chunk_map(map); | |
592d92ee LB |
3158 | return ERR_PTR(-EINVAL); |
3159 | } | |
3160 | ||
7dc66abb FM |
3161 | /* Callers are responsible for dropping the reference. */ |
3162 | return map; | |
592d92ee LB |
3163 | } |
3164 | ||
79bd3712 | 3165 | static int remove_chunk_item(struct btrfs_trans_handle *trans, |
7dc66abb | 3166 | struct btrfs_chunk_map *map, u64 chunk_offset) |
79bd3712 FM |
3167 | { |
3168 | int i; | |
3169 | ||
3170 | /* | |
3171 | * Removing chunk items and updating the device items in the chunks btree | |
3172 | * requires holding the chunk_mutex. | |
3173 | * See the comment at btrfs_chunk_alloc() for the details. | |
3174 | */ | |
3175 | lockdep_assert_held(&trans->fs_info->chunk_mutex); | |
3176 | ||
3177 | for (i = 0; i < map->num_stripes; i++) { | |
3178 | int ret; | |
3179 | ||
3180 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
3181 | if (ret) | |
3182 | return ret; | |
3183 | } | |
3184 | ||
3185 | return btrfs_free_chunk(trans, chunk_offset); | |
3186 | } | |
3187 | ||
97aff912 | 3188 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3189 | { |
97aff912 | 3190 | struct btrfs_fs_info *fs_info = trans->fs_info; |
7dc66abb | 3191 | struct btrfs_chunk_map *map; |
2196d6e8 | 3192 | u64 dev_extent_len = 0; |
47ab2a6c | 3193 | int i, ret = 0; |
0b246afa | 3194 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3195 | |
7dc66abb FM |
3196 | map = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
3197 | if (IS_ERR(map)) { | |
47ab2a6c JB |
3198 | /* |
3199 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3200 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3201 | * do anything we still error out. |
3202 | */ | |
3203 | ASSERT(0); | |
7dc66abb | 3204 | return PTR_ERR(map); |
47ab2a6c | 3205 | } |
8f18cf13 | 3206 | |
57ba4cb8 | 3207 | /* |
79bd3712 FM |
3208 | * First delete the device extent items from the devices btree. |
3209 | * We take the device_list_mutex to avoid racing with the finishing phase | |
3210 | * of a device replace operation. See the comment below before acquiring | |
3211 | * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex | |
3212 | * because that can result in a deadlock when deleting the device extent | |
3213 | * items from the devices btree - COWing an extent buffer from the btree | |
3214 | * may result in allocating a new metadata chunk, which would attempt to | |
3215 | * lock again fs_info->chunk_mutex. | |
57ba4cb8 FM |
3216 | */ |
3217 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3218 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3219 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3220 | ret = btrfs_free_dev_extent(trans, device, |
3221 | map->stripes[i].physical, | |
3222 | &dev_extent_len); | |
47ab2a6c | 3223 | if (ret) { |
57ba4cb8 | 3224 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3225 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3226 | goto out; |
3227 | } | |
a061fc8d | 3228 | |
2196d6e8 | 3229 | if (device->bytes_used > 0) { |
34441361 | 3230 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3231 | btrfs_device_set_bytes_used(device, |
3232 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3233 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3234 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3235 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3236 | } |
79bd3712 FM |
3237 | } |
3238 | mutex_unlock(&fs_devices->device_list_mutex); | |
a061fc8d | 3239 | |
79bd3712 FM |
3240 | /* |
3241 | * We acquire fs_info->chunk_mutex for 2 reasons: | |
3242 | * | |
3243 | * 1) Just like with the first phase of the chunk allocation, we must | |
3244 | * reserve system space, do all chunk btree updates and deletions, and | |
3245 | * update the system chunk array in the superblock while holding this | |
3246 | * mutex. This is for similar reasons as explained on the comment at | |
3247 | * the top of btrfs_chunk_alloc(); | |
3248 | * | |
3249 | * 2) Prevent races with the final phase of a device replace operation | |
3250 | * that replaces the device object associated with the map's stripes, | |
3251 | * because the device object's id can change at any time during that | |
3252 | * final phase of the device replace operation | |
3253 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
3254 | * replaced device and then see it with an ID of | |
3255 | * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating | |
3256 | * the device item, which does not exists on the chunk btree. | |
3257 | * The finishing phase of device replace acquires both the | |
3258 | * device_list_mutex and the chunk_mutex, in that order, so we are | |
3259 | * safe by just acquiring the chunk_mutex. | |
3260 | */ | |
3261 | trans->removing_chunk = true; | |
3262 | mutex_lock(&fs_info->chunk_mutex); | |
3263 | ||
3264 | check_system_chunk(trans, map->type); | |
3265 | ||
3266 | ret = remove_chunk_item(trans, map, chunk_offset); | |
3267 | /* | |
3268 | * Normally we should not get -ENOSPC since we reserved space before | |
3269 | * through the call to check_system_chunk(). | |
3270 | * | |
3271 | * Despite our system space_info having enough free space, we may not | |
3272 | * be able to allocate extents from its block groups, because all have | |
3273 | * an incompatible profile, which will force us to allocate a new system | |
3274 | * block group with the right profile, or right after we called | |
3275 | * check_system_space() above, a scrub turned the only system block group | |
3276 | * with enough free space into RO mode. | |
3277 | * This is explained with more detail at do_chunk_alloc(). | |
3278 | * | |
3279 | * So if we get -ENOSPC, allocate a new system chunk and retry once. | |
3280 | */ | |
3281 | if (ret == -ENOSPC) { | |
3282 | const u64 sys_flags = btrfs_system_alloc_profile(fs_info); | |
3283 | struct btrfs_block_group *sys_bg; | |
3284 | ||
f6f39f7a | 3285 | sys_bg = btrfs_create_chunk(trans, sys_flags); |
79bd3712 FM |
3286 | if (IS_ERR(sys_bg)) { |
3287 | ret = PTR_ERR(sys_bg); | |
3288 | btrfs_abort_transaction(trans, ret); | |
3289 | goto out; | |
3290 | } | |
3291 | ||
3292 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
64bc6c2a | 3293 | if (ret) { |
64bc6c2a NB |
3294 | btrfs_abort_transaction(trans, ret); |
3295 | goto out; | |
dfe25020 | 3296 | } |
57ba4cb8 | 3297 | |
79bd3712 FM |
3298 | ret = remove_chunk_item(trans, map, chunk_offset); |
3299 | if (ret) { | |
3300 | btrfs_abort_transaction(trans, ret); | |
3301 | goto out; | |
3302 | } | |
3303 | } else if (ret) { | |
66642832 | 3304 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3305 | goto out; |
3306 | } | |
8f18cf13 | 3307 | |
7dc66abb | 3308 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len); |
1abe9b8a | 3309 | |
8f18cf13 | 3310 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3311 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3312 | if (ret) { |
66642832 | 3313 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3314 | goto out; |
3315 | } | |
8f18cf13 CM |
3316 | } |
3317 | ||
79bd3712 FM |
3318 | mutex_unlock(&fs_info->chunk_mutex); |
3319 | trans->removing_chunk = false; | |
3320 | ||
3321 | /* | |
3322 | * We are done with chunk btree updates and deletions, so release the | |
3323 | * system space we previously reserved (with check_system_chunk()). | |
3324 | */ | |
3325 | btrfs_trans_release_chunk_metadata(trans); | |
3326 | ||
7dc66abb | 3327 | ret = btrfs_remove_block_group(trans, map); |
47ab2a6c | 3328 | if (ret) { |
66642832 | 3329 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3330 | goto out; |
3331 | } | |
2b82032c | 3332 | |
47ab2a6c | 3333 | out: |
79bd3712 FM |
3334 | if (trans->removing_chunk) { |
3335 | mutex_unlock(&fs_info->chunk_mutex); | |
3336 | trans->removing_chunk = false; | |
3337 | } | |
2b82032c | 3338 | /* once for us */ |
7dc66abb | 3339 | btrfs_free_chunk_map(map); |
47ab2a6c JB |
3340 | return ret; |
3341 | } | |
2b82032c | 3342 | |
18bb8bbf | 3343 | int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3344 | { |
5b4aacef | 3345 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3346 | struct btrfs_trans_handle *trans; |
b0643e59 | 3347 | struct btrfs_block_group *block_group; |
01e86008 | 3348 | u64 length; |
47ab2a6c | 3349 | int ret; |
2b82032c | 3350 | |
4b349253 JB |
3351 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
3352 | btrfs_err(fs_info, | |
3353 | "relocate: not supported on extent tree v2 yet"); | |
3354 | return -EINVAL; | |
3355 | } | |
3356 | ||
67c5e7d4 FM |
3357 | /* |
3358 | * Prevent races with automatic removal of unused block groups. | |
3359 | * After we relocate and before we remove the chunk with offset | |
3360 | * chunk_offset, automatic removal of the block group can kick in, | |
3361 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3362 | * | |
3363 | * Make sure to acquire this mutex before doing a tree search (dev | |
3364 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3365 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3366 | * we release the path used to search the chunk/dev tree and before | |
3367 | * the current task acquires this mutex and calls us. | |
3368 | */ | |
f3372065 | 3369 | lockdep_assert_held(&fs_info->reclaim_bgs_lock); |
67c5e7d4 | 3370 | |
47ab2a6c | 3371 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3372 | btrfs_scrub_pause(fs_info); |
0b246afa | 3373 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3374 | btrfs_scrub_continue(fs_info); |
2d82a40a FM |
3375 | if (ret) { |
3376 | /* | |
3377 | * If we had a transaction abort, stop all running scrubs. | |
3378 | * See transaction.c:cleanup_transaction() why we do it here. | |
3379 | */ | |
3380 | if (BTRFS_FS_ERROR(fs_info)) | |
3381 | btrfs_scrub_cancel(fs_info); | |
47ab2a6c | 3382 | return ret; |
2d82a40a | 3383 | } |
47ab2a6c | 3384 | |
b0643e59 DZ |
3385 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3386 | if (!block_group) | |
3387 | return -ENOENT; | |
3388 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
01e86008 | 3389 | length = block_group->length; |
b0643e59 DZ |
3390 | btrfs_put_block_group(block_group); |
3391 | ||
01e86008 JT |
3392 | /* |
3393 | * On a zoned file system, discard the whole block group, this will | |
3394 | * trigger a REQ_OP_ZONE_RESET operation on the device zone. If | |
3395 | * resetting the zone fails, don't treat it as a fatal problem from the | |
3396 | * filesystem's point of view. | |
3397 | */ | |
3398 | if (btrfs_is_zoned(fs_info)) { | |
3399 | ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); | |
3400 | if (ret) | |
3401 | btrfs_info(fs_info, | |
3402 | "failed to reset zone %llu after relocation", | |
3403 | chunk_offset); | |
3404 | } | |
3405 | ||
19c4d2f9 CM |
3406 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3407 | chunk_offset); | |
3408 | if (IS_ERR(trans)) { | |
3409 | ret = PTR_ERR(trans); | |
3410 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3411 | return ret; | |
3412 | } | |
3413 | ||
47ab2a6c | 3414 | /* |
19c4d2f9 CM |
3415 | * step two, delete the device extents and the |
3416 | * chunk tree entries | |
47ab2a6c | 3417 | */ |
97aff912 | 3418 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3419 | btrfs_end_transaction(trans); |
19c4d2f9 | 3420 | return ret; |
2b82032c YZ |
3421 | } |
3422 | ||
2ff7e61e | 3423 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3424 | { |
0b246afa | 3425 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3426 | struct btrfs_path *path; |
3427 | struct extent_buffer *leaf; | |
3428 | struct btrfs_chunk *chunk; | |
3429 | struct btrfs_key key; | |
3430 | struct btrfs_key found_key; | |
2b82032c | 3431 | u64 chunk_type; |
ba1bf481 JB |
3432 | bool retried = false; |
3433 | int failed = 0; | |
2b82032c YZ |
3434 | int ret; |
3435 | ||
3436 | path = btrfs_alloc_path(); | |
3437 | if (!path) | |
3438 | return -ENOMEM; | |
3439 | ||
ba1bf481 | 3440 | again: |
2b82032c YZ |
3441 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3442 | key.offset = (u64)-1; | |
3443 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3444 | ||
3445 | while (1) { | |
f3372065 | 3446 | mutex_lock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3447 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3448 | if (ret < 0) { |
f3372065 | 3449 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3450 | goto error; |
67c5e7d4 | 3451 | } |
7411055d DS |
3452 | if (ret == 0) { |
3453 | /* | |
3454 | * On the first search we would find chunk tree with | |
3455 | * offset -1, which is not possible. On subsequent | |
3456 | * loops this would find an existing item on an invalid | |
3457 | * offset (one less than the previous one, wrong | |
3458 | * alignment and size). | |
3459 | */ | |
3460 | ret = -EUCLEAN; | |
9af503d9 | 3461 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
7411055d DS |
3462 | goto error; |
3463 | } | |
2b82032c YZ |
3464 | |
3465 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3466 | key.type); | |
67c5e7d4 | 3467 | if (ret) |
f3372065 | 3468 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c YZ |
3469 | if (ret < 0) |
3470 | goto error; | |
3471 | if (ret > 0) | |
3472 | break; | |
1a40e23b | 3473 | |
2b82032c YZ |
3474 | leaf = path->nodes[0]; |
3475 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3476 | |
2b82032c YZ |
3477 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3478 | struct btrfs_chunk); | |
3479 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3480 | btrfs_release_path(path); |
8f18cf13 | 3481 | |
2b82032c | 3482 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3483 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3484 | if (ret == -ENOSPC) |
3485 | failed++; | |
14586651 HS |
3486 | else |
3487 | BUG_ON(ret); | |
2b82032c | 3488 | } |
f3372065 | 3489 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 3490 | |
2b82032c YZ |
3491 | if (found_key.offset == 0) |
3492 | break; | |
3493 | key.offset = found_key.offset - 1; | |
3494 | } | |
3495 | ret = 0; | |
ba1bf481 JB |
3496 | if (failed && !retried) { |
3497 | failed = 0; | |
3498 | retried = true; | |
3499 | goto again; | |
fae7f21c | 3500 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3501 | ret = -ENOSPC; |
3502 | } | |
2b82032c YZ |
3503 | error: |
3504 | btrfs_free_path(path); | |
3505 | return ret; | |
8f18cf13 CM |
3506 | } |
3507 | ||
a6f93c71 LB |
3508 | /* |
3509 | * return 1 : allocate a data chunk successfully, | |
3510 | * return <0: errors during allocating a data chunk, | |
3511 | * return 0 : no need to allocate a data chunk. | |
3512 | */ | |
3513 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3514 | u64 chunk_offset) | |
3515 | { | |
32da5386 | 3516 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3517 | u64 bytes_used; |
3518 | u64 chunk_type; | |
3519 | ||
3520 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3521 | ASSERT(cache); | |
3522 | chunk_type = cache->flags; | |
3523 | btrfs_put_block_group(cache); | |
3524 | ||
5ae21692 JT |
3525 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3526 | return 0; | |
3527 | ||
3528 | spin_lock(&fs_info->data_sinfo->lock); | |
3529 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3530 | spin_unlock(&fs_info->data_sinfo->lock); | |
3531 | ||
3532 | if (!bytes_used) { | |
3533 | struct btrfs_trans_handle *trans; | |
3534 | int ret; | |
3535 | ||
3536 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3537 | if (IS_ERR(trans)) | |
3538 | return PTR_ERR(trans); | |
3539 | ||
3540 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3541 | btrfs_end_transaction(trans); | |
3542 | if (ret < 0) | |
3543 | return ret; | |
3544 | return 1; | |
a6f93c71 | 3545 | } |
5ae21692 | 3546 | |
a6f93c71 LB |
3547 | return 0; |
3548 | } | |
3549 | ||
e6052347 DS |
3550 | static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu, |
3551 | const struct btrfs_disk_balance_args *disk) | |
3552 | { | |
3553 | memset(cpu, 0, sizeof(*cpu)); | |
3554 | ||
3555 | cpu->profiles = le64_to_cpu(disk->profiles); | |
3556 | cpu->usage = le64_to_cpu(disk->usage); | |
3557 | cpu->devid = le64_to_cpu(disk->devid); | |
3558 | cpu->pstart = le64_to_cpu(disk->pstart); | |
3559 | cpu->pend = le64_to_cpu(disk->pend); | |
3560 | cpu->vstart = le64_to_cpu(disk->vstart); | |
3561 | cpu->vend = le64_to_cpu(disk->vend); | |
3562 | cpu->target = le64_to_cpu(disk->target); | |
3563 | cpu->flags = le64_to_cpu(disk->flags); | |
3564 | cpu->limit = le64_to_cpu(disk->limit); | |
3565 | cpu->stripes_min = le32_to_cpu(disk->stripes_min); | |
3566 | cpu->stripes_max = le32_to_cpu(disk->stripes_max); | |
3567 | } | |
3568 | ||
3569 | static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk, | |
3570 | const struct btrfs_balance_args *cpu) | |
3571 | { | |
3572 | memset(disk, 0, sizeof(*disk)); | |
3573 | ||
3574 | disk->profiles = cpu_to_le64(cpu->profiles); | |
3575 | disk->usage = cpu_to_le64(cpu->usage); | |
3576 | disk->devid = cpu_to_le64(cpu->devid); | |
3577 | disk->pstart = cpu_to_le64(cpu->pstart); | |
3578 | disk->pend = cpu_to_le64(cpu->pend); | |
3579 | disk->vstart = cpu_to_le64(cpu->vstart); | |
3580 | disk->vend = cpu_to_le64(cpu->vend); | |
3581 | disk->target = cpu_to_le64(cpu->target); | |
3582 | disk->flags = cpu_to_le64(cpu->flags); | |
3583 | disk->limit = cpu_to_le64(cpu->limit); | |
3584 | disk->stripes_min = cpu_to_le32(cpu->stripes_min); | |
3585 | disk->stripes_max = cpu_to_le32(cpu->stripes_max); | |
3586 | } | |
3587 | ||
6bccf3ab | 3588 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3589 | struct btrfs_balance_control *bctl) |
3590 | { | |
6bccf3ab | 3591 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3592 | struct btrfs_trans_handle *trans; |
3593 | struct btrfs_balance_item *item; | |
3594 | struct btrfs_disk_balance_args disk_bargs; | |
3595 | struct btrfs_path *path; | |
3596 | struct extent_buffer *leaf; | |
3597 | struct btrfs_key key; | |
3598 | int ret, err; | |
3599 | ||
3600 | path = btrfs_alloc_path(); | |
3601 | if (!path) | |
3602 | return -ENOMEM; | |
3603 | ||
3604 | trans = btrfs_start_transaction(root, 0); | |
3605 | if (IS_ERR(trans)) { | |
3606 | btrfs_free_path(path); | |
3607 | return PTR_ERR(trans); | |
3608 | } | |
3609 | ||
3610 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3611 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3612 | key.offset = 0; |
3613 | ||
3614 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3615 | sizeof(*item)); | |
3616 | if (ret) | |
3617 | goto out; | |
3618 | ||
3619 | leaf = path->nodes[0]; | |
3620 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3621 | ||
b159fa28 | 3622 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3623 | |
3624 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3625 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3626 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3627 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3628 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3629 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3630 | ||
3631 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3632 | ||
50564b65 | 3633 | btrfs_mark_buffer_dirty(trans, leaf); |
0940ebf6 ID |
3634 | out: |
3635 | btrfs_free_path(path); | |
3a45bb20 | 3636 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3637 | if (err && !ret) |
3638 | ret = err; | |
3639 | return ret; | |
3640 | } | |
3641 | ||
6bccf3ab | 3642 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3643 | { |
6bccf3ab | 3644 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3645 | struct btrfs_trans_handle *trans; |
3646 | struct btrfs_path *path; | |
3647 | struct btrfs_key key; | |
3648 | int ret, err; | |
3649 | ||
3650 | path = btrfs_alloc_path(); | |
3651 | if (!path) | |
3652 | return -ENOMEM; | |
3653 | ||
3502a8c0 | 3654 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3655 | if (IS_ERR(trans)) { |
3656 | btrfs_free_path(path); | |
3657 | return PTR_ERR(trans); | |
3658 | } | |
3659 | ||
3660 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3661 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3662 | key.offset = 0; |
3663 | ||
3664 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3665 | if (ret < 0) | |
3666 | goto out; | |
3667 | if (ret > 0) { | |
3668 | ret = -ENOENT; | |
3669 | goto out; | |
3670 | } | |
3671 | ||
3672 | ret = btrfs_del_item(trans, root, path); | |
3673 | out: | |
3674 | btrfs_free_path(path); | |
3a45bb20 | 3675 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3676 | if (err && !ret) |
3677 | ret = err; | |
3678 | return ret; | |
3679 | } | |
3680 | ||
59641015 ID |
3681 | /* |
3682 | * This is a heuristic used to reduce the number of chunks balanced on | |
3683 | * resume after balance was interrupted. | |
3684 | */ | |
3685 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3686 | { | |
3687 | /* | |
3688 | * Turn on soft mode for chunk types that were being converted. | |
3689 | */ | |
3690 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3691 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3692 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3693 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3694 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3695 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3696 | ||
3697 | /* | |
3698 | * Turn on usage filter if is not already used. The idea is | |
3699 | * that chunks that we have already balanced should be | |
3700 | * reasonably full. Don't do it for chunks that are being | |
3701 | * converted - that will keep us from relocating unconverted | |
3702 | * (albeit full) chunks. | |
3703 | */ | |
3704 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3705 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3706 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3707 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3708 | bctl->data.usage = 90; | |
3709 | } | |
3710 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3711 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3712 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3713 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3714 | bctl->sys.usage = 90; | |
3715 | } | |
3716 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3717 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3718 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3719 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3720 | bctl->meta.usage = 90; | |
3721 | } | |
3722 | } | |
3723 | ||
149196a2 DS |
3724 | /* |
3725 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3726 | */ | |
3727 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3728 | { |
3729 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3730 | int ret; |
c9e9f97b | 3731 | |
53e4d8c2 | 3732 | ASSERT(fs_info->balance_ctl); |
c9e9f97b ID |
3733 | |
3734 | spin_lock(&fs_info->balance_lock); | |
3735 | fs_info->balance_ctl = NULL; | |
3736 | spin_unlock(&fs_info->balance_lock); | |
3737 | ||
3738 | kfree(bctl); | |
149196a2 DS |
3739 | ret = del_balance_item(fs_info); |
3740 | if (ret) | |
3741 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3742 | } |
3743 | ||
ed25e9b2 ID |
3744 | /* |
3745 | * Balance filters. Return 1 if chunk should be filtered out | |
3746 | * (should not be balanced). | |
3747 | */ | |
899c81ea | 3748 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3749 | struct btrfs_balance_args *bargs) |
3750 | { | |
899c81ea ID |
3751 | chunk_type = chunk_to_extended(chunk_type) & |
3752 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3753 | |
899c81ea | 3754 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3755 | return 0; |
3756 | ||
3757 | return 1; | |
3758 | } | |
3759 | ||
dba72cb3 | 3760 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3761 | struct btrfs_balance_args *bargs) |
bc309467 | 3762 | { |
32da5386 | 3763 | struct btrfs_block_group *cache; |
bc309467 DS |
3764 | u64 chunk_used; |
3765 | u64 user_thresh_min; | |
3766 | u64 user_thresh_max; | |
3767 | int ret = 1; | |
3768 | ||
3769 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3770 | chunk_used = cache->used; |
bc309467 DS |
3771 | |
3772 | if (bargs->usage_min == 0) | |
3773 | user_thresh_min = 0; | |
3774 | else | |
428c8e03 | 3775 | user_thresh_min = mult_perc(cache->length, bargs->usage_min); |
bc309467 DS |
3776 | |
3777 | if (bargs->usage_max == 0) | |
3778 | user_thresh_max = 1; | |
3779 | else if (bargs->usage_max > 100) | |
b3470b5d | 3780 | user_thresh_max = cache->length; |
bc309467 | 3781 | else |
428c8e03 | 3782 | user_thresh_max = mult_perc(cache->length, bargs->usage_max); |
bc309467 DS |
3783 | |
3784 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3785 | ret = 0; | |
3786 | ||
3787 | btrfs_put_block_group(cache); | |
3788 | return ret; | |
3789 | } | |
3790 | ||
dba72cb3 | 3791 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3792 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3793 | { |
32da5386 | 3794 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3795 | u64 chunk_used, user_thresh; |
3796 | int ret = 1; | |
3797 | ||
3798 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3799 | chunk_used = cache->used; |
5ce5b3c0 | 3800 | |
bc309467 | 3801 | if (bargs->usage_min == 0) |
3e39cea6 | 3802 | user_thresh = 1; |
a105bb88 | 3803 | else if (bargs->usage > 100) |
b3470b5d | 3804 | user_thresh = cache->length; |
a105bb88 | 3805 | else |
428c8e03 | 3806 | user_thresh = mult_perc(cache->length, bargs->usage); |
a105bb88 | 3807 | |
5ce5b3c0 ID |
3808 | if (chunk_used < user_thresh) |
3809 | ret = 0; | |
3810 | ||
3811 | btrfs_put_block_group(cache); | |
3812 | return ret; | |
3813 | } | |
3814 | ||
409d404b ID |
3815 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3816 | struct btrfs_chunk *chunk, | |
3817 | struct btrfs_balance_args *bargs) | |
3818 | { | |
3819 | struct btrfs_stripe *stripe; | |
3820 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3821 | int i; | |
3822 | ||
3823 | for (i = 0; i < num_stripes; i++) { | |
3824 | stripe = btrfs_stripe_nr(chunk, i); | |
3825 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3826 | return 0; | |
3827 | } | |
3828 | ||
3829 | return 1; | |
3830 | } | |
3831 | ||
946c9256 DS |
3832 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3833 | { | |
3834 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3835 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3836 | const int nparity = btrfs_raid_array[index].nparity; | |
3837 | ||
d58ede8d | 3838 | return (num_stripes - nparity) / ncopies; |
946c9256 DS |
3839 | } |
3840 | ||
94e60d5a ID |
3841 | /* [pstart, pend) */ |
3842 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3843 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3844 | struct btrfs_balance_args *bargs) |
3845 | { | |
3846 | struct btrfs_stripe *stripe; | |
3847 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3848 | u64 stripe_offset; | |
3849 | u64 stripe_length; | |
946c9256 | 3850 | u64 type; |
94e60d5a ID |
3851 | int factor; |
3852 | int i; | |
3853 | ||
3854 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3855 | return 0; | |
3856 | ||
946c9256 DS |
3857 | type = btrfs_chunk_type(leaf, chunk); |
3858 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3859 | |
3860 | for (i = 0; i < num_stripes; i++) { | |
3861 | stripe = btrfs_stripe_nr(chunk, i); | |
3862 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3863 | continue; | |
3864 | ||
3865 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3866 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3867 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3868 | |
3869 | if (stripe_offset < bargs->pend && | |
3870 | stripe_offset + stripe_length > bargs->pstart) | |
3871 | return 0; | |
3872 | } | |
3873 | ||
3874 | return 1; | |
3875 | } | |
3876 | ||
ea67176a ID |
3877 | /* [vstart, vend) */ |
3878 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3879 | struct btrfs_chunk *chunk, | |
3880 | u64 chunk_offset, | |
3881 | struct btrfs_balance_args *bargs) | |
3882 | { | |
3883 | if (chunk_offset < bargs->vend && | |
3884 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3885 | /* at least part of the chunk is inside this vrange */ | |
3886 | return 0; | |
3887 | ||
3888 | return 1; | |
3889 | } | |
3890 | ||
dee32d0a GAP |
3891 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3892 | struct btrfs_chunk *chunk, | |
3893 | struct btrfs_balance_args *bargs) | |
3894 | { | |
3895 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3896 | ||
3897 | if (bargs->stripes_min <= num_stripes | |
3898 | && num_stripes <= bargs->stripes_max) | |
3899 | return 0; | |
3900 | ||
3901 | return 1; | |
3902 | } | |
3903 | ||
899c81ea | 3904 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3905 | struct btrfs_balance_args *bargs) |
3906 | { | |
3907 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3908 | return 0; | |
3909 | ||
899c81ea ID |
3910 | chunk_type = chunk_to_extended(chunk_type) & |
3911 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3912 | |
899c81ea | 3913 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3914 | return 1; |
3915 | ||
3916 | return 0; | |
3917 | } | |
3918 | ||
6ec0896c | 3919 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3920 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3921 | { | |
6ec0896c | 3922 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3923 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3924 | struct btrfs_balance_args *bargs = NULL; |
3925 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3926 | ||
3927 | /* type filter */ | |
3928 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3929 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3930 | return 0; | |
3931 | } | |
3932 | ||
3933 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3934 | bargs = &bctl->data; | |
3935 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3936 | bargs = &bctl->sys; | |
3937 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3938 | bargs = &bctl->meta; | |
3939 | ||
ed25e9b2 ID |
3940 | /* profiles filter */ |
3941 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3942 | chunk_profiles_filter(chunk_type, bargs)) { | |
3943 | return 0; | |
5ce5b3c0 ID |
3944 | } |
3945 | ||
3946 | /* usage filter */ | |
3947 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3948 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3949 | return 0; |
bc309467 | 3950 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3951 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3952 | return 0; |
409d404b ID |
3953 | } |
3954 | ||
3955 | /* devid filter */ | |
3956 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3957 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3958 | return 0; | |
94e60d5a ID |
3959 | } |
3960 | ||
3961 | /* drange filter, makes sense only with devid filter */ | |
3962 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3963 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3964 | return 0; |
ea67176a ID |
3965 | } |
3966 | ||
3967 | /* vrange filter */ | |
3968 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3969 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3970 | return 0; | |
ed25e9b2 ID |
3971 | } |
3972 | ||
dee32d0a GAP |
3973 | /* stripes filter */ |
3974 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3975 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3976 | return 0; | |
3977 | } | |
3978 | ||
cfa4c961 ID |
3979 | /* soft profile changing mode */ |
3980 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3981 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3982 | return 0; | |
3983 | } | |
3984 | ||
7d824b6f DS |
3985 | /* |
3986 | * limited by count, must be the last filter | |
3987 | */ | |
3988 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3989 | if (bargs->limit == 0) | |
3990 | return 0; | |
3991 | else | |
3992 | bargs->limit--; | |
12907fc7 DS |
3993 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3994 | /* | |
3995 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3996 | * determined here because we do not have the global information |
12907fc7 DS |
3997 | * about the count of all chunks that satisfy the filters. |
3998 | */ | |
3999 | if (bargs->limit_max == 0) | |
4000 | return 0; | |
4001 | else | |
4002 | bargs->limit_max--; | |
7d824b6f DS |
4003 | } |
4004 | ||
f43ffb60 ID |
4005 | return 1; |
4006 | } | |
4007 | ||
c9e9f97b | 4008 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 4009 | { |
19a39dce | 4010 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 4011 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 4012 | u64 chunk_type; |
f43ffb60 | 4013 | struct btrfs_chunk *chunk; |
5a488b9d | 4014 | struct btrfs_path *path = NULL; |
ec44a35c | 4015 | struct btrfs_key key; |
ec44a35c | 4016 | struct btrfs_key found_key; |
f43ffb60 ID |
4017 | struct extent_buffer *leaf; |
4018 | int slot; | |
c9e9f97b ID |
4019 | int ret; |
4020 | int enospc_errors = 0; | |
19a39dce | 4021 | bool counting = true; |
12907fc7 | 4022 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
4023 | u64 limit_data = bctl->data.limit; |
4024 | u64 limit_meta = bctl->meta.limit; | |
4025 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
4026 | u32 count_data = 0; |
4027 | u32 count_meta = 0; | |
4028 | u32 count_sys = 0; | |
2c9fe835 | 4029 | int chunk_reserved = 0; |
ec44a35c | 4030 | |
ec44a35c | 4031 | path = btrfs_alloc_path(); |
17e9f796 MF |
4032 | if (!path) { |
4033 | ret = -ENOMEM; | |
4034 | goto error; | |
4035 | } | |
19a39dce ID |
4036 | |
4037 | /* zero out stat counters */ | |
4038 | spin_lock(&fs_info->balance_lock); | |
4039 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
4040 | spin_unlock(&fs_info->balance_lock); | |
4041 | again: | |
7d824b6f | 4042 | if (!counting) { |
12907fc7 DS |
4043 | /* |
4044 | * The single value limit and min/max limits use the same bytes | |
4045 | * in the | |
4046 | */ | |
7d824b6f DS |
4047 | bctl->data.limit = limit_data; |
4048 | bctl->meta.limit = limit_meta; | |
4049 | bctl->sys.limit = limit_sys; | |
4050 | } | |
ec44a35c CM |
4051 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4052 | key.offset = (u64)-1; | |
4053 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4054 | ||
d397712b | 4055 | while (1) { |
19a39dce | 4056 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 4057 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
4058 | ret = -ECANCELED; |
4059 | goto error; | |
4060 | } | |
4061 | ||
f3372065 | 4062 | mutex_lock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 4063 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 4064 | if (ret < 0) { |
f3372065 | 4065 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 4066 | goto error; |
67c5e7d4 | 4067 | } |
ec44a35c CM |
4068 | |
4069 | /* | |
4070 | * this shouldn't happen, it means the last relocate | |
4071 | * failed | |
4072 | */ | |
4073 | if (ret == 0) | |
c9e9f97b | 4074 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
4075 | |
4076 | ret = btrfs_previous_item(chunk_root, path, 0, | |
4077 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 4078 | if (ret) { |
f3372065 | 4079 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
c9e9f97b | 4080 | ret = 0; |
ec44a35c | 4081 | break; |
c9e9f97b | 4082 | } |
7d9eb12c | 4083 | |
f43ffb60 ID |
4084 | leaf = path->nodes[0]; |
4085 | slot = path->slots[0]; | |
4086 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 4087 | |
67c5e7d4 | 4088 | if (found_key.objectid != key.objectid) { |
f3372065 | 4089 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 4090 | break; |
67c5e7d4 | 4091 | } |
7d9eb12c | 4092 | |
f43ffb60 | 4093 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 4094 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 4095 | |
19a39dce ID |
4096 | if (!counting) { |
4097 | spin_lock(&fs_info->balance_lock); | |
4098 | bctl->stat.considered++; | |
4099 | spin_unlock(&fs_info->balance_lock); | |
4100 | } | |
4101 | ||
6ec0896c | 4102 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 4103 | |
b3b4aa74 | 4104 | btrfs_release_path(path); |
67c5e7d4 | 4105 | if (!ret) { |
f3372065 | 4106 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f43ffb60 | 4107 | goto loop; |
67c5e7d4 | 4108 | } |
f43ffb60 | 4109 | |
19a39dce | 4110 | if (counting) { |
f3372065 | 4111 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
4112 | spin_lock(&fs_info->balance_lock); |
4113 | bctl->stat.expected++; | |
4114 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
4115 | |
4116 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
4117 | count_data++; | |
4118 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
4119 | count_sys++; | |
4120 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
4121 | count_meta++; | |
4122 | ||
4123 | goto loop; | |
4124 | } | |
4125 | ||
4126 | /* | |
4127 | * Apply limit_min filter, no need to check if the LIMITS | |
4128 | * filter is used, limit_min is 0 by default | |
4129 | */ | |
4130 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
4131 | count_data < bctl->data.limit_min) | |
4132 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
4133 | count_meta < bctl->meta.limit_min) | |
4134 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
4135 | count_sys < bctl->sys.limit_min)) { | |
f3372065 | 4136 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
4137 | goto loop; |
4138 | } | |
4139 | ||
a6f93c71 LB |
4140 | if (!chunk_reserved) { |
4141 | /* | |
4142 | * We may be relocating the only data chunk we have, | |
4143 | * which could potentially end up with losing data's | |
4144 | * raid profile, so lets allocate an empty one in | |
4145 | * advance. | |
4146 | */ | |
4147 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
4148 | found_key.offset); | |
2c9fe835 | 4149 | if (ret < 0) { |
f3372065 | 4150 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2c9fe835 | 4151 | goto error; |
a6f93c71 LB |
4152 | } else if (ret == 1) { |
4153 | chunk_reserved = 1; | |
2c9fe835 | 4154 | } |
2c9fe835 ZL |
4155 | } |
4156 | ||
5b4aacef | 4157 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
f3372065 | 4158 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce | 4159 | if (ret == -ENOSPC) { |
c9e9f97b | 4160 | enospc_errors++; |
eede2bf3 OS |
4161 | } else if (ret == -ETXTBSY) { |
4162 | btrfs_info(fs_info, | |
4163 | "skipping relocation of block group %llu due to active swapfile", | |
4164 | found_key.offset); | |
4165 | ret = 0; | |
4166 | } else if (ret) { | |
4167 | goto error; | |
19a39dce ID |
4168 | } else { |
4169 | spin_lock(&fs_info->balance_lock); | |
4170 | bctl->stat.completed++; | |
4171 | spin_unlock(&fs_info->balance_lock); | |
4172 | } | |
f43ffb60 | 4173 | loop: |
795a3321 ID |
4174 | if (found_key.offset == 0) |
4175 | break; | |
ba1bf481 | 4176 | key.offset = found_key.offset - 1; |
ec44a35c | 4177 | } |
c9e9f97b | 4178 | |
19a39dce ID |
4179 | if (counting) { |
4180 | btrfs_release_path(path); | |
4181 | counting = false; | |
4182 | goto again; | |
4183 | } | |
ec44a35c CM |
4184 | error: |
4185 | btrfs_free_path(path); | |
c9e9f97b | 4186 | if (enospc_errors) { |
efe120a0 | 4187 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 4188 | enospc_errors); |
c9e9f97b ID |
4189 | if (!ret) |
4190 | ret = -ENOSPC; | |
4191 | } | |
4192 | ||
ec44a35c CM |
4193 | return ret; |
4194 | } | |
4195 | ||
43dd529a DS |
4196 | /* |
4197 | * See if a given profile is valid and reduced. | |
4198 | * | |
4199 | * @flags: profile to validate | |
4200 | * @extended: if true @flags is treated as an extended profile | |
0c460c0d ID |
4201 | */ |
4202 | static int alloc_profile_is_valid(u64 flags, int extended) | |
4203 | { | |
4204 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
4205 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
4206 | ||
4207 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4208 | ||
4209 | /* 1) check that all other bits are zeroed */ | |
4210 | if (flags & ~mask) | |
4211 | return 0; | |
4212 | ||
4213 | /* 2) see if profile is reduced */ | |
4214 | if (flags == 0) | |
4215 | return !extended; /* "0" is valid for usual profiles */ | |
4216 | ||
c1499166 | 4217 | return has_single_bit_set(flags); |
0c460c0d ID |
4218 | } |
4219 | ||
5ba366c3 DS |
4220 | /* |
4221 | * Validate target profile against allowed profiles and return true if it's OK. | |
4222 | * Otherwise print the error message and return false. | |
4223 | */ | |
4224 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
4225 | const struct btrfs_balance_args *bargs, | |
4226 | u64 allowed, const char *type) | |
bdcd3c97 | 4227 | { |
5ba366c3 DS |
4228 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
4229 | return true; | |
4230 | ||
4231 | /* Profile is valid and does not have bits outside of the allowed set */ | |
4232 | if (alloc_profile_is_valid(bargs->target, 1) && | |
4233 | (bargs->target & ~allowed) == 0) | |
4234 | return true; | |
4235 | ||
4236 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
4237 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
4238 | return false; | |
bdcd3c97 AM |
4239 | } |
4240 | ||
56fc37d9 AJ |
4241 | /* |
4242 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
4243 | * @size_buf including the terminating null. The output may be trimmed if it | |
4244 | * does not fit into the provided buffer. | |
4245 | */ | |
4246 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
4247 | u32 size_buf) | |
4248 | { | |
4249 | int ret; | |
4250 | u32 size_bp = size_buf; | |
4251 | char *bp = buf; | |
4252 | u64 flags = bargs->flags; | |
4253 | char tmp_buf[128] = {'\0'}; | |
4254 | ||
4255 | if (!flags) | |
4256 | return; | |
4257 | ||
4258 | #define CHECK_APPEND_NOARG(a) \ | |
4259 | do { \ | |
4260 | ret = snprintf(bp, size_bp, (a)); \ | |
4261 | if (ret < 0 || ret >= size_bp) \ | |
4262 | goto out_overflow; \ | |
4263 | size_bp -= ret; \ | |
4264 | bp += ret; \ | |
4265 | } while (0) | |
4266 | ||
4267 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4268 | do { \ | |
4269 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4270 | if (ret < 0 || ret >= size_bp) \ | |
4271 | goto out_overflow; \ | |
4272 | size_bp -= ret; \ | |
4273 | bp += ret; \ | |
4274 | } while (0) | |
4275 | ||
4276 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
4277 | do { \ | |
4278 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
4279 | if (ret < 0 || ret >= size_bp) \ | |
4280 | goto out_overflow; \ | |
4281 | size_bp -= ret; \ | |
4282 | bp += ret; \ | |
4283 | } while (0) | |
4284 | ||
158da513 DS |
4285 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
4286 | CHECK_APPEND_1ARG("convert=%s,", | |
4287 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
4288 | |
4289 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
4290 | CHECK_APPEND_NOARG("soft,"); | |
4291 | ||
4292 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
4293 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
4294 | sizeof(tmp_buf)); | |
4295 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
4296 | } | |
4297 | ||
4298 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
4299 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
4300 | ||
4301 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
4302 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
4303 | bargs->usage_min, bargs->usage_max); | |
4304 | ||
4305 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
4306 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
4307 | ||
4308 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4309 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4310 | bargs->pstart, bargs->pend); | |
4311 | ||
4312 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4313 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4314 | bargs->vstart, bargs->vend); | |
4315 | ||
4316 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4317 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4318 | ||
4319 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4320 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4321 | bargs->limit_min, bargs->limit_max); | |
4322 | ||
4323 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4324 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4325 | bargs->stripes_min, bargs->stripes_max); | |
4326 | ||
4327 | #undef CHECK_APPEND_2ARG | |
4328 | #undef CHECK_APPEND_1ARG | |
4329 | #undef CHECK_APPEND_NOARG | |
4330 | ||
4331 | out_overflow: | |
4332 | ||
4333 | if (size_bp < size_buf) | |
4334 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4335 | else | |
4336 | buf[0] = '\0'; | |
4337 | } | |
4338 | ||
4339 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4340 | { | |
4341 | u32 size_buf = 1024; | |
4342 | char tmp_buf[192] = {'\0'}; | |
4343 | char *buf; | |
4344 | char *bp; | |
4345 | u32 size_bp = size_buf; | |
4346 | int ret; | |
4347 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4348 | ||
4349 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4350 | if (!buf) | |
4351 | return; | |
4352 | ||
4353 | bp = buf; | |
4354 | ||
4355 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4356 | do { \ | |
4357 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4358 | if (ret < 0 || ret >= size_bp) \ | |
4359 | goto out_overflow; \ | |
4360 | size_bp -= ret; \ | |
4361 | bp += ret; \ | |
4362 | } while (0) | |
4363 | ||
4364 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4365 | CHECK_APPEND_1ARG("%s", "-f "); | |
4366 | ||
4367 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4368 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4369 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4370 | } | |
4371 | ||
4372 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4373 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4374 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4375 | } | |
4376 | ||
4377 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4378 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4379 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4380 | } | |
4381 | ||
4382 | #undef CHECK_APPEND_1ARG | |
4383 | ||
4384 | out_overflow: | |
4385 | ||
4386 | if (size_bp < size_buf) | |
4387 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4388 | btrfs_info(fs_info, "balance: %s %s", | |
4389 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4390 | "resume" : "start", buf); | |
4391 | ||
4392 | kfree(buf); | |
4393 | } | |
4394 | ||
c9e9f97b | 4395 | /* |
dccdb07b | 4396 | * Should be called with balance mutexe held |
c9e9f97b | 4397 | */ |
6fcf6e2b DS |
4398 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4399 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4400 | struct btrfs_ioctl_balance_args *bargs) |
4401 | { | |
14506127 | 4402 | u64 meta_target, data_target; |
f43ffb60 | 4403 | u64 allowed; |
e4837f8f | 4404 | int mixed = 0; |
c9e9f97b | 4405 | int ret; |
8dabb742 | 4406 | u64 num_devices; |
de98ced9 | 4407 | unsigned seq; |
e62869be | 4408 | bool reducing_redundancy; |
b19c98f2 | 4409 | bool paused = false; |
081db89b | 4410 | int i; |
c9e9f97b | 4411 | |
837d5b6e | 4412 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4413 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4414 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4415 | ret = -EINVAL; |
4416 | goto out; | |
4417 | } | |
4418 | ||
e4837f8f ID |
4419 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4420 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4421 | mixed = 1; | |
4422 | ||
f43ffb60 ID |
4423 | /* |
4424 | * In case of mixed groups both data and meta should be picked, | |
4425 | * and identical options should be given for both of them. | |
4426 | */ | |
e4837f8f ID |
4427 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4428 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4429 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4430 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4431 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4432 | btrfs_err(fs_info, |
6dac13f8 | 4433 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4434 | ret = -EINVAL; |
4435 | goto out; | |
4436 | } | |
4437 | } | |
4438 | ||
b35cf1f0 JB |
4439 | /* |
4440 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4441 | * are exclusive |
b35cf1f0 JB |
4442 | */ |
4443 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4444 | |
4445 | /* | |
4446 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4447 | * special bit for it, to make it easier to distinguish. Thus we need | |
4448 | * to set it manually, or balance would refuse the profile. | |
4449 | */ | |
4450 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4451 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4452 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4453 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4454 | |
5ba366c3 DS |
4455 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4456 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4457 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4458 | ret = -EINVAL; |
4459 | goto out; | |
4460 | } | |
4461 | ||
6079e12c DS |
4462 | /* |
4463 | * Allow to reduce metadata or system integrity only if force set for | |
4464 | * profiles with redundancy (copies, parity) | |
4465 | */ | |
4466 | allowed = 0; | |
4467 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4468 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4469 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4470 | allowed |= btrfs_raid_array[i].bg_flag; | |
4471 | } | |
de98ced9 MX |
4472 | do { |
4473 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4474 | ||
4475 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4476 | (fs_info->avail_system_alloc_bits & allowed) && | |
4477 | !(bctl->sys.target & allowed)) || | |
4478 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4479 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4480 | !(bctl->meta.target & allowed))) |
e62869be | 4481 | reducing_redundancy = true; |
5a8067c0 | 4482 | else |
e62869be | 4483 | reducing_redundancy = false; |
5a8067c0 FM |
4484 | |
4485 | /* if we're not converting, the target field is uninitialized */ | |
4486 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4487 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4488 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4489 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4490 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4491 | |
e62869be | 4492 | if (reducing_redundancy) { |
5a8067c0 FM |
4493 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4494 | btrfs_info(fs_info, | |
e62869be | 4495 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4496 | } else { |
4497 | btrfs_err(fs_info, | |
e62869be | 4498 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4499 | ret = -EINVAL; |
4500 | goto out; | |
4501 | } | |
4502 | } | |
4503 | ||
14506127 AB |
4504 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4505 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4506 | btrfs_warn(fs_info, |
6dac13f8 | 4507 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4508 | btrfs_bg_type_to_raid_name(meta_target), |
4509 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4510 | } |
4511 | ||
6bccf3ab | 4512 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4513 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4514 | goto out; |
4515 | ||
59641015 ID |
4516 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4517 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4518 | BUG_ON(fs_info->balance_ctl); |
4519 | spin_lock(&fs_info->balance_lock); | |
4520 | fs_info->balance_ctl = bctl; | |
4521 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4522 | } else { |
4523 | BUG_ON(ret != -EEXIST); | |
4524 | spin_lock(&fs_info->balance_lock); | |
4525 | update_balance_args(bctl); | |
4526 | spin_unlock(&fs_info->balance_lock); | |
4527 | } | |
c9e9f97b | 4528 | |
3009a62f DS |
4529 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4530 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4531 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4532 | mutex_unlock(&fs_info->balance_mutex); |
4533 | ||
4534 | ret = __btrfs_balance(fs_info); | |
4535 | ||
4536 | mutex_lock(&fs_info->balance_mutex); | |
efc0e69c | 4537 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { |
7333bd02 | 4538 | btrfs_info(fs_info, "balance: paused"); |
efc0e69c | 4539 | btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); |
b19c98f2 | 4540 | paused = true; |
efc0e69c | 4541 | } |
44d354ab QW |
4542 | /* |
4543 | * Balance can be canceled by: | |
4544 | * | |
4545 | * - Regular cancel request | |
4546 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4547 | * | |
4548 | * - Fatal signal to "btrfs" process | |
4549 | * Either the signal caught by wait_reserve_ticket() and callers | |
4550 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4551 | * got -ECANCELED. | |
4552 | * Either way, in this case balance_cancel_req = 0, and | |
4553 | * ret == -EINTR or ret == -ECANCELED. | |
4554 | * | |
4555 | * So here we only check the return value to catch canceled balance. | |
4556 | */ | |
4557 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4558 | btrfs_info(fs_info, "balance: canceled"); |
4559 | else | |
4560 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4561 | ||
3009a62f | 4562 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4563 | |
4564 | if (bargs) { | |
4565 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4566 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4567 | } |
4568 | ||
b19c98f2 JB |
4569 | /* We didn't pause, we can clean everything up. */ |
4570 | if (!paused) { | |
149196a2 | 4571 | reset_balance_state(fs_info); |
c3e1f96c | 4572 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4573 | } |
4574 | ||
837d5b6e | 4575 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4576 | |
4577 | return ret; | |
4578 | out: | |
59641015 | 4579 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4580 | reset_balance_state(fs_info); |
a17c95df | 4581 | else |
59641015 | 4582 | kfree(bctl); |
c3e1f96c | 4583 | btrfs_exclop_finish(fs_info); |
a17c95df | 4584 | |
59641015 ID |
4585 | return ret; |
4586 | } | |
4587 | ||
4588 | static int balance_kthread(void *data) | |
4589 | { | |
2b6ba629 | 4590 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4591 | int ret = 0; |
59641015 | 4592 | |
a690e5f2 | 4593 | sb_start_write(fs_info->sb); |
59641015 | 4594 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4595 | if (fs_info->balance_ctl) |
6fcf6e2b | 4596 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4597 | mutex_unlock(&fs_info->balance_mutex); |
a690e5f2 | 4598 | sb_end_write(fs_info->sb); |
2b6ba629 | 4599 | |
59641015 ID |
4600 | return ret; |
4601 | } | |
4602 | ||
2b6ba629 ID |
4603 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4604 | { | |
4605 | struct task_struct *tsk; | |
4606 | ||
1354e1a1 | 4607 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4608 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4609 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4610 | return 0; |
4611 | } | |
1354e1a1 | 4612 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4613 | |
3cdde224 | 4614 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4615 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4616 | return 0; |
4617 | } | |
4618 | ||
efc0e69c NB |
4619 | spin_lock(&fs_info->super_lock); |
4620 | ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); | |
4621 | fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; | |
4622 | spin_unlock(&fs_info->super_lock); | |
02ee654d AJ |
4623 | /* |
4624 | * A ro->rw remount sequence should continue with the paused balance | |
4625 | * regardless of who pauses it, system or the user as of now, so set | |
4626 | * the resume flag. | |
4627 | */ | |
4628 | spin_lock(&fs_info->balance_lock); | |
4629 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4630 | spin_unlock(&fs_info->balance_lock); | |
4631 | ||
2b6ba629 | 4632 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4633 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4634 | } |
4635 | ||
68310a5e | 4636 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4637 | { |
59641015 ID |
4638 | struct btrfs_balance_control *bctl; |
4639 | struct btrfs_balance_item *item; | |
4640 | struct btrfs_disk_balance_args disk_bargs; | |
4641 | struct btrfs_path *path; | |
4642 | struct extent_buffer *leaf; | |
4643 | struct btrfs_key key; | |
4644 | int ret; | |
4645 | ||
4646 | path = btrfs_alloc_path(); | |
4647 | if (!path) | |
4648 | return -ENOMEM; | |
4649 | ||
59641015 | 4650 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4651 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4652 | key.offset = 0; |
4653 | ||
68310a5e | 4654 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4655 | if (ret < 0) |
68310a5e | 4656 | goto out; |
59641015 ID |
4657 | if (ret > 0) { /* ret = -ENOENT; */ |
4658 | ret = 0; | |
68310a5e ID |
4659 | goto out; |
4660 | } | |
4661 | ||
4662 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4663 | if (!bctl) { | |
4664 | ret = -ENOMEM; | |
4665 | goto out; | |
59641015 ID |
4666 | } |
4667 | ||
4668 | leaf = path->nodes[0]; | |
4669 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4670 | ||
68310a5e ID |
4671 | bctl->flags = btrfs_balance_flags(leaf, item); |
4672 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4673 | |
4674 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4675 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4676 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4677 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4678 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4679 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4680 | ||
eee95e3f DS |
4681 | /* |
4682 | * This should never happen, as the paused balance state is recovered | |
4683 | * during mount without any chance of other exclusive ops to collide. | |
4684 | * | |
4685 | * This gives the exclusive op status to balance and keeps in paused | |
4686 | * state until user intervention (cancel or umount). If the ownership | |
4687 | * cannot be assigned, show a message but do not fail. The balance | |
4688 | * is in a paused state and must have fs_info::balance_ctl properly | |
4689 | * set up. | |
4690 | */ | |
efc0e69c | 4691 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) |
eee95e3f | 4692 | btrfs_warn(fs_info, |
6dac13f8 | 4693 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4694 | |
fb286100 JB |
4695 | btrfs_release_path(path); |
4696 | ||
68310a5e | 4697 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4698 | BUG_ON(fs_info->balance_ctl); |
4699 | spin_lock(&fs_info->balance_lock); | |
4700 | fs_info->balance_ctl = bctl; | |
4701 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4702 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4703 | out: |
4704 | btrfs_free_path(path); | |
ec44a35c CM |
4705 | return ret; |
4706 | } | |
4707 | ||
837d5b6e ID |
4708 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4709 | { | |
4710 | int ret = 0; | |
4711 | ||
4712 | mutex_lock(&fs_info->balance_mutex); | |
4713 | if (!fs_info->balance_ctl) { | |
4714 | mutex_unlock(&fs_info->balance_mutex); | |
4715 | return -ENOTCONN; | |
4716 | } | |
4717 | ||
3009a62f | 4718 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4719 | atomic_inc(&fs_info->balance_pause_req); |
4720 | mutex_unlock(&fs_info->balance_mutex); | |
4721 | ||
4722 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4723 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4724 | |
4725 | mutex_lock(&fs_info->balance_mutex); | |
4726 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4727 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4728 | atomic_dec(&fs_info->balance_pause_req); |
4729 | } else { | |
4730 | ret = -ENOTCONN; | |
4731 | } | |
4732 | ||
4733 | mutex_unlock(&fs_info->balance_mutex); | |
4734 | return ret; | |
4735 | } | |
4736 | ||
a7e99c69 ID |
4737 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4738 | { | |
4739 | mutex_lock(&fs_info->balance_mutex); | |
4740 | if (!fs_info->balance_ctl) { | |
4741 | mutex_unlock(&fs_info->balance_mutex); | |
4742 | return -ENOTCONN; | |
4743 | } | |
4744 | ||
cf7d20f4 DS |
4745 | /* |
4746 | * A paused balance with the item stored on disk can be resumed at | |
4747 | * mount time if the mount is read-write. Otherwise it's still paused | |
4748 | * and we must not allow cancelling as it deletes the item. | |
4749 | */ | |
4750 | if (sb_rdonly(fs_info->sb)) { | |
4751 | mutex_unlock(&fs_info->balance_mutex); | |
4752 | return -EROFS; | |
4753 | } | |
4754 | ||
a7e99c69 ID |
4755 | atomic_inc(&fs_info->balance_cancel_req); |
4756 | /* | |
4757 | * if we are running just wait and return, balance item is | |
4758 | * deleted in btrfs_balance in this case | |
4759 | */ | |
3009a62f | 4760 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4761 | mutex_unlock(&fs_info->balance_mutex); |
4762 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4763 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4764 | mutex_lock(&fs_info->balance_mutex); |
4765 | } else { | |
a7e99c69 | 4766 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4767 | /* |
4768 | * Lock released to allow other waiters to continue, we'll | |
4769 | * reexamine the status again. | |
4770 | */ | |
a7e99c69 ID |
4771 | mutex_lock(&fs_info->balance_mutex); |
4772 | ||
a17c95df | 4773 | if (fs_info->balance_ctl) { |
149196a2 | 4774 | reset_balance_state(fs_info); |
c3e1f96c | 4775 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4776 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4777 | } |
a7e99c69 ID |
4778 | } |
4779 | ||
29eefa6d | 4780 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4781 | atomic_dec(&fs_info->balance_cancel_req); |
4782 | mutex_unlock(&fs_info->balance_mutex); | |
4783 | return 0; | |
4784 | } | |
4785 | ||
97f4dd09 | 4786 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4787 | { |
4788 | struct btrfs_fs_info *fs_info = data; | |
4789 | struct btrfs_root *root = fs_info->tree_root; | |
4790 | struct btrfs_key key; | |
803b2f54 SB |
4791 | struct btrfs_path *path = NULL; |
4792 | int ret = 0; | |
4793 | struct extent_buffer *eb; | |
4794 | int slot; | |
4795 | struct btrfs_root_item root_item; | |
4796 | u32 item_size; | |
f45388f3 | 4797 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4798 | bool closing = false; |
803b2f54 SB |
4799 | |
4800 | path = btrfs_alloc_path(); | |
4801 | if (!path) { | |
4802 | ret = -ENOMEM; | |
4803 | goto out; | |
4804 | } | |
4805 | ||
4806 | key.objectid = 0; | |
4807 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4808 | key.offset = 0; | |
4809 | ||
803b2f54 | 4810 | while (1) { |
c94bec2c JB |
4811 | if (btrfs_fs_closing(fs_info)) { |
4812 | closing = true; | |
4813 | break; | |
4814 | } | |
7c829b72 AJ |
4815 | ret = btrfs_search_forward(root, &key, path, |
4816 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4817 | if (ret) { |
4818 | if (ret > 0) | |
4819 | ret = 0; | |
4820 | break; | |
4821 | } | |
4822 | ||
4823 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4824 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4825 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4826 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4827 | goto skip; | |
4828 | ||
4829 | eb = path->nodes[0]; | |
4830 | slot = path->slots[0]; | |
3212fa14 | 4831 | item_size = btrfs_item_size(eb, slot); |
803b2f54 SB |
4832 | if (item_size < sizeof(root_item)) |
4833 | goto skip; | |
4834 | ||
803b2f54 SB |
4835 | read_extent_buffer(eb, &root_item, |
4836 | btrfs_item_ptr_offset(eb, slot), | |
4837 | (int)sizeof(root_item)); | |
4838 | if (btrfs_root_refs(&root_item) == 0) | |
4839 | goto skip; | |
f45388f3 FDBM |
4840 | |
4841 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4842 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4843 | if (trans) | |
4844 | goto update_tree; | |
4845 | ||
4846 | btrfs_release_path(path); | |
803b2f54 SB |
4847 | /* |
4848 | * 1 - subvol uuid item | |
4849 | * 1 - received_subvol uuid item | |
4850 | */ | |
4851 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4852 | if (IS_ERR(trans)) { | |
4853 | ret = PTR_ERR(trans); | |
4854 | break; | |
4855 | } | |
f45388f3 FDBM |
4856 | continue; |
4857 | } else { | |
4858 | goto skip; | |
4859 | } | |
4860 | update_tree: | |
9771a5cf | 4861 | btrfs_release_path(path); |
f45388f3 | 4862 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4863 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4864 | BTRFS_UUID_KEY_SUBVOL, |
4865 | key.objectid); | |
4866 | if (ret < 0) { | |
efe120a0 | 4867 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4868 | ret); |
803b2f54 SB |
4869 | break; |
4870 | } | |
4871 | } | |
4872 | ||
4873 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4874 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4875 | root_item.received_uuid, |
4876 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4877 | key.objectid); | |
4878 | if (ret < 0) { | |
efe120a0 | 4879 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4880 | ret); |
803b2f54 SB |
4881 | break; |
4882 | } | |
4883 | } | |
4884 | ||
f45388f3 | 4885 | skip: |
9771a5cf | 4886 | btrfs_release_path(path); |
803b2f54 | 4887 | if (trans) { |
3a45bb20 | 4888 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4889 | trans = NULL; |
803b2f54 SB |
4890 | if (ret) |
4891 | break; | |
4892 | } | |
4893 | ||
803b2f54 SB |
4894 | if (key.offset < (u64)-1) { |
4895 | key.offset++; | |
4896 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4897 | key.offset = 0; | |
4898 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4899 | } else if (key.objectid < (u64)-1) { | |
4900 | key.offset = 0; | |
4901 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4902 | key.objectid++; | |
4903 | } else { | |
4904 | break; | |
4905 | } | |
4906 | cond_resched(); | |
4907 | } | |
4908 | ||
4909 | out: | |
4910 | btrfs_free_path(path); | |
f45388f3 | 4911 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4912 | btrfs_end_transaction(trans); |
803b2f54 | 4913 | if (ret) |
efe120a0 | 4914 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4915 | else if (!closing) |
afcdd129 | 4916 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4917 | up(&fs_info->uuid_tree_rescan_sem); |
4918 | return 0; | |
4919 | } | |
4920 | ||
f7a81ea4 SB |
4921 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4922 | { | |
4923 | struct btrfs_trans_handle *trans; | |
4924 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4925 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4926 | struct task_struct *task; |
4927 | int ret; | |
f7a81ea4 SB |
4928 | |
4929 | /* | |
4930 | * 1 - root node | |
4931 | * 1 - root item | |
4932 | */ | |
4933 | trans = btrfs_start_transaction(tree_root, 2); | |
4934 | if (IS_ERR(trans)) | |
4935 | return PTR_ERR(trans); | |
4936 | ||
9b7a2440 | 4937 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4938 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4939 | ret = PTR_ERR(uuid_root); |
66642832 | 4940 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4941 | btrfs_end_transaction(trans); |
6d13f549 | 4942 | return ret; |
f7a81ea4 SB |
4943 | } |
4944 | ||
4945 | fs_info->uuid_root = uuid_root; | |
4946 | ||
3a45bb20 | 4947 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4948 | if (ret) |
4949 | return ret; | |
4950 | ||
4951 | down(&fs_info->uuid_tree_rescan_sem); | |
4952 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4953 | if (IS_ERR(task)) { | |
70f80175 | 4954 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4955 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4956 | up(&fs_info->uuid_tree_rescan_sem); |
4957 | return PTR_ERR(task); | |
4958 | } | |
4959 | ||
4960 | return 0; | |
f7a81ea4 | 4961 | } |
803b2f54 | 4962 | |
8f18cf13 CM |
4963 | /* |
4964 | * shrinking a device means finding all of the device extents past | |
4965 | * the new size, and then following the back refs to the chunks. | |
4966 | * The chunk relocation code actually frees the device extent | |
4967 | */ | |
4968 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4969 | { | |
0b246afa JM |
4970 | struct btrfs_fs_info *fs_info = device->fs_info; |
4971 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4972 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4973 | struct btrfs_dev_extent *dev_extent = NULL; |
4974 | struct btrfs_path *path; | |
4975 | u64 length; | |
8f18cf13 CM |
4976 | u64 chunk_offset; |
4977 | int ret; | |
4978 | int slot; | |
ba1bf481 JB |
4979 | int failed = 0; |
4980 | bool retried = false; | |
8f18cf13 CM |
4981 | struct extent_buffer *l; |
4982 | struct btrfs_key key; | |
0b246afa | 4983 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4984 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4985 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4986 | u64 diff; |
61d0d0d2 | 4987 | u64 start; |
e9fd2c05 | 4988 | u64 free_diff = 0; |
7dfb8be1 NB |
4989 | |
4990 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4991 | start = new_size; |
0e4324a4 | 4992 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4993 | |
401e29c1 | 4994 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4995 | return -EINVAL; |
4996 | ||
8f18cf13 CM |
4997 | path = btrfs_alloc_path(); |
4998 | if (!path) | |
4999 | return -ENOMEM; | |
5000 | ||
0338dff6 | 5001 | path->reada = READA_BACK; |
8f18cf13 | 5002 | |
61d0d0d2 NB |
5003 | trans = btrfs_start_transaction(root, 0); |
5004 | if (IS_ERR(trans)) { | |
5005 | btrfs_free_path(path); | |
5006 | return PTR_ERR(trans); | |
5007 | } | |
5008 | ||
34441361 | 5009 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 5010 | |
7cc8e58d | 5011 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 5012 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 5013 | device->fs_devices->total_rw_bytes -= diff; |
e9fd2c05 JB |
5014 | |
5015 | /* | |
5016 | * The new free_chunk_space is new_size - used, so we have to | |
5017 | * subtract the delta of the old free_chunk_space which included | |
5018 | * old_size - used. If used > new_size then just subtract this | |
5019 | * entire device's free space. | |
5020 | */ | |
5021 | if (device->bytes_used < new_size) | |
5022 | free_diff = (old_size - device->bytes_used) - | |
5023 | (new_size - device->bytes_used); | |
5024 | else | |
5025 | free_diff = old_size - device->bytes_used; | |
5026 | atomic64_sub(free_diff, &fs_info->free_chunk_space); | |
2bf64758 | 5027 | } |
61d0d0d2 NB |
5028 | |
5029 | /* | |
5030 | * Once the device's size has been set to the new size, ensure all | |
5031 | * in-memory chunks are synced to disk so that the loop below sees them | |
5032 | * and relocates them accordingly. | |
5033 | */ | |
1c11b63e | 5034 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
5035 | mutex_unlock(&fs_info->chunk_mutex); |
5036 | ret = btrfs_commit_transaction(trans); | |
5037 | if (ret) | |
5038 | goto done; | |
5039 | } else { | |
5040 | mutex_unlock(&fs_info->chunk_mutex); | |
5041 | btrfs_end_transaction(trans); | |
5042 | } | |
8f18cf13 | 5043 | |
ba1bf481 | 5044 | again: |
8f18cf13 CM |
5045 | key.objectid = device->devid; |
5046 | key.offset = (u64)-1; | |
5047 | key.type = BTRFS_DEV_EXTENT_KEY; | |
5048 | ||
213e64da | 5049 | do { |
f3372065 | 5050 | mutex_lock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 5051 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 5052 | if (ret < 0) { |
f3372065 | 5053 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 5054 | goto done; |
67c5e7d4 | 5055 | } |
8f18cf13 CM |
5056 | |
5057 | ret = btrfs_previous_item(root, path, 0, key.type); | |
8f18cf13 | 5058 | if (ret) { |
f3372065 | 5059 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
7056bf69 NB |
5060 | if (ret < 0) |
5061 | goto done; | |
8f18cf13 | 5062 | ret = 0; |
b3b4aa74 | 5063 | btrfs_release_path(path); |
bf1fb512 | 5064 | break; |
8f18cf13 CM |
5065 | } |
5066 | ||
5067 | l = path->nodes[0]; | |
5068 | slot = path->slots[0]; | |
5069 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
5070 | ||
ba1bf481 | 5071 | if (key.objectid != device->devid) { |
f3372065 | 5072 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 5073 | btrfs_release_path(path); |
bf1fb512 | 5074 | break; |
ba1bf481 | 5075 | } |
8f18cf13 CM |
5076 | |
5077 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
5078 | length = btrfs_dev_extent_length(l, dev_extent); | |
5079 | ||
ba1bf481 | 5080 | if (key.offset + length <= new_size) { |
f3372065 | 5081 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 5082 | btrfs_release_path(path); |
d6397bae | 5083 | break; |
ba1bf481 | 5084 | } |
8f18cf13 | 5085 | |
8f18cf13 | 5086 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 5087 | btrfs_release_path(path); |
8f18cf13 | 5088 | |
a6f93c71 LB |
5089 | /* |
5090 | * We may be relocating the only data chunk we have, | |
5091 | * which could potentially end up with losing data's | |
5092 | * raid profile, so lets allocate an empty one in | |
5093 | * advance. | |
5094 | */ | |
5095 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
5096 | if (ret < 0) { | |
f3372065 | 5097 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
a6f93c71 LB |
5098 | goto done; |
5099 | } | |
5100 | ||
0b246afa | 5101 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
f3372065 | 5102 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
eede2bf3 | 5103 | if (ret == -ENOSPC) { |
ba1bf481 | 5104 | failed++; |
eede2bf3 OS |
5105 | } else if (ret) { |
5106 | if (ret == -ETXTBSY) { | |
5107 | btrfs_warn(fs_info, | |
5108 | "could not shrink block group %llu due to active swapfile", | |
5109 | chunk_offset); | |
5110 | } | |
5111 | goto done; | |
5112 | } | |
213e64da | 5113 | } while (key.offset-- > 0); |
ba1bf481 JB |
5114 | |
5115 | if (failed && !retried) { | |
5116 | failed = 0; | |
5117 | retried = true; | |
5118 | goto again; | |
5119 | } else if (failed && retried) { | |
5120 | ret = -ENOSPC; | |
ba1bf481 | 5121 | goto done; |
8f18cf13 CM |
5122 | } |
5123 | ||
d6397bae | 5124 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 5125 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
5126 | if (IS_ERR(trans)) { |
5127 | ret = PTR_ERR(trans); | |
5128 | goto done; | |
5129 | } | |
5130 | ||
34441361 | 5131 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
5132 | /* Clear all state bits beyond the shrunk device size */ |
5133 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
5134 | CHUNK_STATE_MASK); | |
5135 | ||
7cc8e58d | 5136 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
5137 | if (list_empty(&device->post_commit_list)) |
5138 | list_add_tail(&device->post_commit_list, | |
5139 | &trans->transaction->dev_update_list); | |
d6397bae | 5140 | |
d6397bae | 5141 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
5142 | btrfs_set_super_total_bytes(super_copy, |
5143 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 5144 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 5145 | |
2bb2e00e | 5146 | btrfs_reserve_chunk_metadata(trans, false); |
2196d6e8 MX |
5147 | /* Now btrfs_update_device() will change the on-disk size. */ |
5148 | ret = btrfs_update_device(trans, device); | |
2bb2e00e | 5149 | btrfs_trans_release_chunk_metadata(trans); |
801660b0 AJ |
5150 | if (ret < 0) { |
5151 | btrfs_abort_transaction(trans, ret); | |
5152 | btrfs_end_transaction(trans); | |
5153 | } else { | |
5154 | ret = btrfs_commit_transaction(trans); | |
5155 | } | |
8f18cf13 CM |
5156 | done: |
5157 | btrfs_free_path(path); | |
53e489bc | 5158 | if (ret) { |
34441361 | 5159 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 5160 | btrfs_device_set_total_bytes(device, old_size); |
e9fd2c05 | 5161 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
53e489bc | 5162 | device->fs_devices->total_rw_bytes += diff; |
e9fd2c05 JB |
5163 | atomic64_add(free_diff, &fs_info->free_chunk_space); |
5164 | } | |
34441361 | 5165 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 5166 | } |
8f18cf13 CM |
5167 | return ret; |
5168 | } | |
5169 | ||
2ff7e61e | 5170 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
5171 | struct btrfs_key *key, |
5172 | struct btrfs_chunk *chunk, int item_size) | |
5173 | { | |
0b246afa | 5174 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
5175 | struct btrfs_disk_key disk_key; |
5176 | u32 array_size; | |
5177 | u8 *ptr; | |
5178 | ||
79bd3712 FM |
5179 | lockdep_assert_held(&fs_info->chunk_mutex); |
5180 | ||
0b86a832 | 5181 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 5182 | if (array_size + item_size + sizeof(disk_key) |
79bd3712 | 5183 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) |
0b86a832 CM |
5184 | return -EFBIG; |
5185 | ||
5186 | ptr = super_copy->sys_chunk_array + array_size; | |
5187 | btrfs_cpu_key_to_disk(&disk_key, key); | |
5188 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
5189 | ptr += sizeof(disk_key); | |
5190 | memcpy(ptr, chunk, item_size); | |
5191 | item_size += sizeof(disk_key); | |
5192 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 | 5193 | |
0b86a832 CM |
5194 | return 0; |
5195 | } | |
5196 | ||
73c5de00 AJ |
5197 | /* |
5198 | * sort the devices in descending order by max_avail, total_avail | |
5199 | */ | |
5200 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 5201 | { |
73c5de00 AJ |
5202 | const struct btrfs_device_info *di_a = a; |
5203 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 5204 | |
73c5de00 | 5205 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 5206 | return -1; |
73c5de00 | 5207 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 5208 | return 1; |
73c5de00 AJ |
5209 | if (di_a->total_avail > di_b->total_avail) |
5210 | return -1; | |
5211 | if (di_a->total_avail < di_b->total_avail) | |
5212 | return 1; | |
5213 | return 0; | |
b2117a39 | 5214 | } |
0b86a832 | 5215 | |
53b381b3 DW |
5216 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5217 | { | |
ffe2d203 | 5218 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
5219 | return; |
5220 | ||
ceda0864 | 5221 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
5222 | } |
5223 | ||
cfbb825c DS |
5224 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5225 | { | |
5226 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
5227 | return; | |
5228 | ||
5229 | btrfs_set_fs_incompat(info, RAID1C34); | |
5230 | } | |
5231 | ||
4f2bafe8 | 5232 | /* |
f6f39f7a | 5233 | * Structure used internally for btrfs_create_chunk() function. |
4f2bafe8 NA |
5234 | * Wraps needed parameters. |
5235 | */ | |
5236 | struct alloc_chunk_ctl { | |
5237 | u64 start; | |
5238 | u64 type; | |
5239 | /* Total number of stripes to allocate */ | |
5240 | int num_stripes; | |
5241 | /* sub_stripes info for map */ | |
5242 | int sub_stripes; | |
5243 | /* Stripes per device */ | |
5244 | int dev_stripes; | |
5245 | /* Maximum number of devices to use */ | |
5246 | int devs_max; | |
5247 | /* Minimum number of devices to use */ | |
5248 | int devs_min; | |
5249 | /* ndevs has to be a multiple of this */ | |
5250 | int devs_increment; | |
5251 | /* Number of copies */ | |
5252 | int ncopies; | |
5253 | /* Number of stripes worth of bytes to store parity information */ | |
5254 | int nparity; | |
5255 | u64 max_stripe_size; | |
5256 | u64 max_chunk_size; | |
6aafb303 | 5257 | u64 dev_extent_min; |
4f2bafe8 NA |
5258 | u64 stripe_size; |
5259 | u64 chunk_size; | |
5260 | int ndevs; | |
5261 | }; | |
5262 | ||
27c314d5 NA |
5263 | static void init_alloc_chunk_ctl_policy_regular( |
5264 | struct btrfs_fs_devices *fs_devices, | |
5265 | struct alloc_chunk_ctl *ctl) | |
5266 | { | |
f6fca391 | 5267 | struct btrfs_space_info *space_info; |
27c314d5 | 5268 | |
f6fca391 SR |
5269 | space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); |
5270 | ASSERT(space_info); | |
5271 | ||
5272 | ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); | |
8a540e99 | 5273 | ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G); |
f6fca391 SR |
5274 | |
5275 | if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) | |
5276 | ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); | |
27c314d5 NA |
5277 | |
5278 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5279 | ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), |
27c314d5 | 5280 | ctl->max_chunk_size); |
cb091225 | 5281 | ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes); |
27c314d5 NA |
5282 | } |
5283 | ||
1cd6121f NA |
5284 | static void init_alloc_chunk_ctl_policy_zoned( |
5285 | struct btrfs_fs_devices *fs_devices, | |
5286 | struct alloc_chunk_ctl *ctl) | |
5287 | { | |
5288 | u64 zone_size = fs_devices->fs_info->zone_size; | |
5289 | u64 limit; | |
5290 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; | |
5291 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; | |
5292 | u64 min_chunk_size = min_data_stripes * zone_size; | |
5293 | u64 type = ctl->type; | |
5294 | ||
5295 | ctl->max_stripe_size = zone_size; | |
5296 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
5297 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, | |
5298 | zone_size); | |
5299 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
5300 | ctl->max_chunk_size = ctl->max_stripe_size; | |
5301 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
5302 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
5303 | ctl->devs_max = min_t(int, ctl->devs_max, | |
5304 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
bb05b298 AB |
5305 | } else { |
5306 | BUG(); | |
1cd6121f NA |
5307 | } |
5308 | ||
5309 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5310 | limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), |
1cd6121f NA |
5311 | zone_size), |
5312 | min_chunk_size); | |
5313 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); | |
5314 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; | |
5315 | } | |
5316 | ||
27c314d5 NA |
5317 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
5318 | struct alloc_chunk_ctl *ctl) | |
5319 | { | |
5320 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
5321 | ||
5322 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
5323 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5324 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
5325 | if (!ctl->devs_max) | |
5326 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
5327 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
5328 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
5329 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
5330 | ctl->nparity = btrfs_raid_array[index].nparity; | |
5331 | ctl->ndevs = 0; | |
5332 | ||
5333 | switch (fs_devices->chunk_alloc_policy) { | |
5334 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5335 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
5336 | break; | |
1cd6121f NA |
5337 | case BTRFS_CHUNK_ALLOC_ZONED: |
5338 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); | |
5339 | break; | |
27c314d5 NA |
5340 | default: |
5341 | BUG(); | |
5342 | } | |
5343 | } | |
5344 | ||
560156cb NA |
5345 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
5346 | struct alloc_chunk_ctl *ctl, | |
5347 | struct btrfs_device_info *devices_info) | |
b2117a39 | 5348 | { |
560156cb | 5349 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 5350 | struct btrfs_device *device; |
73c5de00 | 5351 | u64 total_avail; |
560156cb | 5352 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 5353 | int ret; |
560156cb NA |
5354 | int ndevs = 0; |
5355 | u64 max_avail; | |
5356 | u64 dev_offset; | |
0cad8a11 | 5357 | |
9f680ce0 | 5358 | /* |
73c5de00 AJ |
5359 | * in the first pass through the devices list, we gather information |
5360 | * about the available holes on each device. | |
9f680ce0 | 5361 | */ |
ebcc9301 | 5362 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 5363 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5364 | WARN(1, KERN_ERR |
efe120a0 | 5365 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5366 | continue; |
5367 | } | |
b2117a39 | 5368 | |
e12c9621 AJ |
5369 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5370 | &device->dev_state) || | |
401e29c1 | 5371 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5372 | continue; |
b2117a39 | 5373 | |
73c5de00 AJ |
5374 | if (device->total_bytes > device->bytes_used) |
5375 | total_avail = device->total_bytes - device->bytes_used; | |
5376 | else | |
5377 | total_avail = 0; | |
38c01b96 | 5378 | |
5379 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 5380 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 5381 | continue; |
b2117a39 | 5382 | |
560156cb NA |
5383 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
5384 | &max_avail); | |
73c5de00 | 5385 | if (ret && ret != -ENOSPC) |
560156cb | 5386 | return ret; |
b2117a39 | 5387 | |
73c5de00 | 5388 | if (ret == 0) |
560156cb | 5389 | max_avail = dev_extent_want; |
b2117a39 | 5390 | |
6aafb303 | 5391 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
5392 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
5393 | btrfs_debug(info, | |
560156cb | 5394 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 5395 | __func__, device->devid, max_avail, |
6aafb303 | 5396 | ctl->dev_extent_min); |
73c5de00 | 5397 | continue; |
4117f207 | 5398 | } |
b2117a39 | 5399 | |
063d006f ES |
5400 | if (ndevs == fs_devices->rw_devices) { |
5401 | WARN(1, "%s: found more than %llu devices\n", | |
5402 | __func__, fs_devices->rw_devices); | |
5403 | break; | |
5404 | } | |
73c5de00 AJ |
5405 | devices_info[ndevs].dev_offset = dev_offset; |
5406 | devices_info[ndevs].max_avail = max_avail; | |
5407 | devices_info[ndevs].total_avail = total_avail; | |
5408 | devices_info[ndevs].dev = device; | |
5409 | ++ndevs; | |
5410 | } | |
560156cb | 5411 | ctl->ndevs = ndevs; |
b2117a39 | 5412 | |
73c5de00 AJ |
5413 | /* |
5414 | * now sort the devices by hole size / available space | |
5415 | */ | |
560156cb | 5416 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 5417 | btrfs_cmp_device_info, NULL); |
b2117a39 | 5418 | |
560156cb NA |
5419 | return 0; |
5420 | } | |
5421 | ||
5badf512 NA |
5422 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
5423 | struct btrfs_device_info *devices_info) | |
5424 | { | |
5425 | /* Number of stripes that count for block group size */ | |
5426 | int data_stripes; | |
5427 | ||
5428 | /* | |
5429 | * The primary goal is to maximize the number of stripes, so use as | |
5430 | * many devices as possible, even if the stripes are not maximum sized. | |
5431 | * | |
5432 | * The DUP profile stores more than one stripe per device, the | |
5433 | * max_avail is the total size so we have to adjust. | |
5434 | */ | |
5435 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5436 | ctl->dev_stripes); | |
5437 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5438 | ||
5439 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5440 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5441 | ||
5442 | /* | |
5443 | * Use the number of data stripes to figure out how big this chunk is | |
5444 | * really going to be in terms of logical address space, and compare | |
5445 | * that answer with the max chunk size. If it's higher, we try to | |
5446 | * reduce stripe_size. | |
5447 | */ | |
5448 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5449 | /* | |
5450 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5451 | * then use it, unless it ends up being even bigger than the | |
5452 | * previous value we had already. | |
5453 | */ | |
5454 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5455 | data_stripes), SZ_16M), | |
5456 | ctl->stripe_size); | |
5457 | } | |
5458 | ||
5da431b7 QW |
5459 | /* Stripe size should not go beyond 1G. */ |
5460 | ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); | |
5461 | ||
5badf512 NA |
5462 | /* Align to BTRFS_STRIPE_LEN */ |
5463 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5464 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5465 | ||
5466 | return 0; | |
5467 | } | |
5468 | ||
1cd6121f NA |
5469 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
5470 | struct btrfs_device_info *devices_info) | |
5471 | { | |
5472 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; | |
5473 | /* Number of stripes that count for block group size */ | |
5474 | int data_stripes; | |
5475 | ||
5476 | /* | |
5477 | * It should hold because: | |
5478 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes | |
5479 | */ | |
5480 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); | |
5481 | ||
5482 | ctl->stripe_size = zone_size; | |
5483 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5484 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5485 | ||
5486 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ | |
5487 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5488 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, | |
5489 | ctl->stripe_size) + ctl->nparity, | |
5490 | ctl->dev_stripes); | |
5491 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5492 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5493 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); | |
5494 | } | |
5495 | ||
5496 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5497 | ||
5498 | return 0; | |
5499 | } | |
5500 | ||
5badf512 NA |
5501 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
5502 | struct alloc_chunk_ctl *ctl, | |
5503 | struct btrfs_device_info *devices_info) | |
5504 | { | |
5505 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5506 | ||
5507 | /* | |
5508 | * Round down to number of usable stripes, devs_increment can be any | |
5509 | * number so we can't use round_down() that requires power of 2, while | |
5510 | * rounddown is safe. | |
5511 | */ | |
5512 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5513 | ||
5514 | if (ctl->ndevs < ctl->devs_min) { | |
5515 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5516 | btrfs_debug(info, | |
5517 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5518 | __func__, ctl->ndevs, ctl->devs_min); | |
5519 | } | |
5520 | return -ENOSPC; | |
5521 | } | |
5522 | ||
5523 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5524 | ||
5525 | switch (fs_devices->chunk_alloc_policy) { | |
5526 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5527 | return decide_stripe_size_regular(ctl, devices_info); | |
1cd6121f NA |
5528 | case BTRFS_CHUNK_ALLOC_ZONED: |
5529 | return decide_stripe_size_zoned(ctl, devices_info); | |
5badf512 NA |
5530 | default: |
5531 | BUG(); | |
5532 | } | |
5533 | } | |
5534 | ||
7dc66abb FM |
5535 | static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits) |
5536 | { | |
5537 | for (int i = 0; i < map->num_stripes; i++) { | |
5538 | struct btrfs_io_stripe *stripe = &map->stripes[i]; | |
5539 | struct btrfs_device *device = stripe->dev; | |
5540 | ||
5541 | set_extent_bit(&device->alloc_state, stripe->physical, | |
5542 | stripe->physical + map->stripe_size - 1, | |
5543 | bits | EXTENT_NOWAIT, NULL); | |
5544 | } | |
5545 | } | |
5546 | ||
5547 | static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits) | |
5548 | { | |
5549 | for (int i = 0; i < map->num_stripes; i++) { | |
5550 | struct btrfs_io_stripe *stripe = &map->stripes[i]; | |
5551 | struct btrfs_device *device = stripe->dev; | |
5552 | ||
5553 | __clear_extent_bit(&device->alloc_state, stripe->physical, | |
5554 | stripe->physical + map->stripe_size - 1, | |
5555 | bits | EXTENT_NOWAIT, | |
5556 | NULL, NULL); | |
5557 | } | |
5558 | } | |
5559 | ||
5560 | void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map) | |
5561 | { | |
5562 | write_lock(&fs_info->mapping_tree_lock); | |
5563 | rb_erase_cached(&map->rb_node, &fs_info->mapping_tree); | |
5564 | RB_CLEAR_NODE(&map->rb_node); | |
5565 | chunk_map_device_clear_bits(map, CHUNK_ALLOCATED); | |
5566 | write_unlock(&fs_info->mapping_tree_lock); | |
5567 | ||
5568 | /* Once for the tree reference. */ | |
5569 | btrfs_free_chunk_map(map); | |
5570 | } | |
5571 | ||
5572 | EXPORT_FOR_TESTS | |
5573 | int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map) | |
5574 | { | |
5575 | struct rb_node **p; | |
5576 | struct rb_node *parent = NULL; | |
5577 | bool leftmost = true; | |
5578 | ||
5579 | write_lock(&fs_info->mapping_tree_lock); | |
5580 | p = &fs_info->mapping_tree.rb_root.rb_node; | |
5581 | while (*p) { | |
5582 | struct btrfs_chunk_map *entry; | |
5583 | ||
5584 | parent = *p; | |
5585 | entry = rb_entry(parent, struct btrfs_chunk_map, rb_node); | |
5586 | ||
5587 | if (map->start < entry->start) { | |
5588 | p = &(*p)->rb_left; | |
5589 | } else if (map->start > entry->start) { | |
5590 | p = &(*p)->rb_right; | |
5591 | leftmost = false; | |
5592 | } else { | |
5593 | write_unlock(&fs_info->mapping_tree_lock); | |
5594 | return -EEXIST; | |
5595 | } | |
5596 | } | |
5597 | rb_link_node(&map->rb_node, parent, p); | |
5598 | rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost); | |
5599 | chunk_map_device_set_bits(map, CHUNK_ALLOCATED); | |
5600 | chunk_map_device_clear_bits(map, CHUNK_TRIMMED); | |
5601 | write_unlock(&fs_info->mapping_tree_lock); | |
5602 | ||
5603 | return 0; | |
5604 | } | |
5605 | ||
5606 | EXPORT_FOR_TESTS | |
5607 | struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp) | |
5608 | { | |
5609 | struct btrfs_chunk_map *map; | |
5610 | ||
5611 | map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp); | |
5612 | if (!map) | |
5613 | return NULL; | |
5614 | ||
5615 | refcount_set(&map->refs, 1); | |
5616 | RB_CLEAR_NODE(&map->rb_node); | |
5617 | ||
5618 | return map; | |
5619 | } | |
5620 | ||
79bd3712 | 5621 | static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, |
dce580ca NA |
5622 | struct alloc_chunk_ctl *ctl, |
5623 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5624 | { |
5625 | struct btrfs_fs_info *info = trans->fs_info; | |
7dc66abb | 5626 | struct btrfs_chunk_map *map; |
79bd3712 | 5627 | struct btrfs_block_group *block_group; |
dce580ca NA |
5628 | u64 start = ctl->start; |
5629 | u64 type = ctl->type; | |
560156cb NA |
5630 | int ret; |
5631 | int i; | |
5632 | int j; | |
5633 | ||
7dc66abb | 5634 | map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS); |
dce580ca | 5635 | if (!map) |
79bd3712 | 5636 | return ERR_PTR(-ENOMEM); |
7dc66abb FM |
5637 | |
5638 | map->start = start; | |
5639 | map->chunk_len = ctl->chunk_size; | |
5640 | map->stripe_size = ctl->stripe_size; | |
5641 | map->type = type; | |
5642 | map->io_align = BTRFS_STRIPE_LEN; | |
5643 | map->io_width = BTRFS_STRIPE_LEN; | |
5644 | map->sub_stripes = ctl->sub_stripes; | |
dce580ca | 5645 | map->num_stripes = ctl->num_stripes; |
560156cb | 5646 | |
dce580ca NA |
5647 | for (i = 0; i < ctl->ndevs; ++i) { |
5648 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5649 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5650 | map->stripes[s].dev = devices_info[i].dev; |
5651 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5652 | j * ctl->stripe_size; |
6324fbf3 | 5653 | } |
6324fbf3 | 5654 | } |
0b86a832 | 5655 | |
dce580ca | 5656 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5657 | |
7dc66abb | 5658 | ret = btrfs_add_chunk_map(info, map); |
0f5d42b2 | 5659 | if (ret) { |
7dc66abb | 5660 | btrfs_free_chunk_map(map); |
79bd3712 | 5661 | return ERR_PTR(ret); |
0f5d42b2 | 5662 | } |
1efb72a3 | 5663 | |
5758d1bd | 5664 | block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); |
7dc66abb FM |
5665 | if (IS_ERR(block_group)) { |
5666 | btrfs_remove_chunk_map(info, map); | |
5667 | return block_group; | |
5668 | } | |
2b82032c | 5669 | |
7dc66abb | 5670 | for (int i = 0; i < map->num_stripes; i++) { |
bbbf7243 NB |
5671 | struct btrfs_device *dev = map->stripes[i].dev; |
5672 | ||
4f2bafe8 | 5673 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5674 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5675 | if (list_empty(&dev->post_commit_list)) |
5676 | list_add_tail(&dev->post_commit_list, | |
5677 | &trans->transaction->dev_update_list); | |
5678 | } | |
43530c46 | 5679 | |
dce580ca | 5680 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5681 | &info->free_chunk_space); |
1c116187 | 5682 | |
0b246afa | 5683 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5684 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5685 | |
79bd3712 | 5686 | return block_group; |
dce580ca NA |
5687 | } |
5688 | ||
f6f39f7a | 5689 | struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, |
79bd3712 | 5690 | u64 type) |
dce580ca NA |
5691 | { |
5692 | struct btrfs_fs_info *info = trans->fs_info; | |
5693 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5694 | struct btrfs_device_info *devices_info = NULL; | |
5695 | struct alloc_chunk_ctl ctl; | |
79bd3712 | 5696 | struct btrfs_block_group *block_group; |
dce580ca NA |
5697 | int ret; |
5698 | ||
11c67b1a NB |
5699 | lockdep_assert_held(&info->chunk_mutex); |
5700 | ||
dce580ca NA |
5701 | if (!alloc_profile_is_valid(type, 0)) { |
5702 | ASSERT(0); | |
79bd3712 | 5703 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5704 | } |
5705 | ||
5706 | if (list_empty(&fs_devices->alloc_list)) { | |
5707 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5708 | btrfs_debug(info, "%s: no writable device", __func__); | |
79bd3712 | 5709 | return ERR_PTR(-ENOSPC); |
dce580ca NA |
5710 | } |
5711 | ||
5712 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5713 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5714 | ASSERT(0); | |
79bd3712 | 5715 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5716 | } |
5717 | ||
11c67b1a | 5718 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5719 | ctl.type = type; |
5720 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5721 | ||
5722 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5723 | GFP_NOFS); | |
5724 | if (!devices_info) | |
79bd3712 | 5725 | return ERR_PTR(-ENOMEM); |
dce580ca NA |
5726 | |
5727 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5728 | if (ret < 0) { |
5729 | block_group = ERR_PTR(ret); | |
dce580ca | 5730 | goto out; |
79bd3712 | 5731 | } |
dce580ca NA |
5732 | |
5733 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5734 | if (ret < 0) { |
5735 | block_group = ERR_PTR(ret); | |
dce580ca | 5736 | goto out; |
79bd3712 | 5737 | } |
dce580ca | 5738 | |
79bd3712 | 5739 | block_group = create_chunk(trans, &ctl, devices_info); |
dce580ca NA |
5740 | |
5741 | out: | |
b2117a39 | 5742 | kfree(devices_info); |
79bd3712 | 5743 | return block_group; |
2b82032c YZ |
5744 | } |
5745 | ||
79bd3712 FM |
5746 | /* |
5747 | * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the | |
5748 | * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system | |
5749 | * chunks. | |
5750 | * | |
5751 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
5752 | * phases. | |
5753 | */ | |
5754 | int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, | |
5755 | struct btrfs_block_group *bg) | |
5756 | { | |
5757 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
79bd3712 FM |
5758 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
5759 | struct btrfs_key key; | |
5760 | struct btrfs_chunk *chunk; | |
5761 | struct btrfs_stripe *stripe; | |
7dc66abb | 5762 | struct btrfs_chunk_map *map; |
79bd3712 FM |
5763 | size_t item_size; |
5764 | int i; | |
5765 | int ret; | |
5766 | ||
5767 | /* | |
5768 | * We take the chunk_mutex for 2 reasons: | |
5769 | * | |
5770 | * 1) Updates and insertions in the chunk btree must be done while holding | |
5771 | * the chunk_mutex, as well as updating the system chunk array in the | |
5772 | * superblock. See the comment on top of btrfs_chunk_alloc() for the | |
5773 | * details; | |
5774 | * | |
5775 | * 2) To prevent races with the final phase of a device replace operation | |
5776 | * that replaces the device object associated with the map's stripes, | |
5777 | * because the device object's id can change at any time during that | |
5778 | * final phase of the device replace operation | |
5779 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
5780 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
5781 | * which would cause a failure when updating the device item, which does | |
5782 | * not exists, or persisting a stripe of the chunk item with such ID. | |
5783 | * Here we can't use the device_list_mutex because our caller already | |
5784 | * has locked the chunk_mutex, and the final phase of device replace | |
5785 | * acquires both mutexes - first the device_list_mutex and then the | |
5786 | * chunk_mutex. Using any of those two mutexes protects us from a | |
5787 | * concurrent device replace. | |
5788 | */ | |
5789 | lockdep_assert_held(&fs_info->chunk_mutex); | |
5790 | ||
7dc66abb FM |
5791 | map = btrfs_get_chunk_map(fs_info, bg->start, bg->length); |
5792 | if (IS_ERR(map)) { | |
5793 | ret = PTR_ERR(map); | |
79bd3712 FM |
5794 | btrfs_abort_transaction(trans, ret); |
5795 | return ret; | |
5796 | } | |
5797 | ||
79bd3712 FM |
5798 | item_size = btrfs_chunk_item_size(map->num_stripes); |
5799 | ||
5800 | chunk = kzalloc(item_size, GFP_NOFS); | |
5801 | if (!chunk) { | |
5802 | ret = -ENOMEM; | |
5803 | btrfs_abort_transaction(trans, ret); | |
50460e37 | 5804 | goto out; |
2b82032c YZ |
5805 | } |
5806 | ||
79bd3712 FM |
5807 | for (i = 0; i < map->num_stripes; i++) { |
5808 | struct btrfs_device *device = map->stripes[i].dev; | |
5809 | ||
5810 | ret = btrfs_update_device(trans, device); | |
5811 | if (ret) | |
5812 | goto out; | |
5813 | } | |
5814 | ||
2b82032c | 5815 | stripe = &chunk->stripe; |
6df9a95e | 5816 | for (i = 0; i < map->num_stripes; i++) { |
79bd3712 FM |
5817 | struct btrfs_device *device = map->stripes[i].dev; |
5818 | const u64 dev_offset = map->stripes[i].physical; | |
0b86a832 | 5819 | |
e17cade2 CM |
5820 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5821 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5822 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5823 | stripe++; |
0b86a832 CM |
5824 | } |
5825 | ||
79bd3712 | 5826 | btrfs_set_stack_chunk_length(chunk, bg->length); |
fd51eb2f | 5827 | btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); |
a97699d1 | 5828 | btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); |
2b82032c YZ |
5829 | btrfs_set_stack_chunk_type(chunk, map->type); |
5830 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
a97699d1 QW |
5831 | btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); |
5832 | btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); | |
0b246afa | 5833 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5834 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5835 | |
2b82032c YZ |
5836 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5837 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
79bd3712 | 5838 | key.offset = bg->start; |
0b86a832 | 5839 | |
2b82032c | 5840 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
79bd3712 FM |
5841 | if (ret) |
5842 | goto out; | |
5843 | ||
3349b57f | 5844 | set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); |
79bd3712 FM |
5845 | |
5846 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2ff7e61e | 5847 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
79bd3712 FM |
5848 | if (ret) |
5849 | goto out; | |
8f18cf13 | 5850 | } |
1abe9b8a | 5851 | |
6df9a95e | 5852 | out: |
0b86a832 | 5853 | kfree(chunk); |
7dc66abb | 5854 | btrfs_free_chunk_map(map); |
4ed1d16e | 5855 | return ret; |
2b82032c | 5856 | } |
0b86a832 | 5857 | |
6f8e0fc7 | 5858 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5859 | { |
6f8e0fc7 | 5860 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5861 | u64 alloc_profile; |
79bd3712 FM |
5862 | struct btrfs_block_group *meta_bg; |
5863 | struct btrfs_block_group *sys_bg; | |
5864 | ||
5865 | /* | |
5866 | * When adding a new device for sprouting, the seed device is read-only | |
5867 | * so we must first allocate a metadata and a system chunk. But before | |
5868 | * adding the block group items to the extent, device and chunk btrees, | |
5869 | * we must first: | |
5870 | * | |
5871 | * 1) Create both chunks without doing any changes to the btrees, as | |
5872 | * otherwise we would get -ENOSPC since the block groups from the | |
5873 | * seed device are read-only; | |
5874 | * | |
5875 | * 2) Add the device item for the new sprout device - finishing the setup | |
5876 | * of a new block group requires updating the device item in the chunk | |
5877 | * btree, so it must exist when we attempt to do it. The previous step | |
5878 | * ensures this does not fail with -ENOSPC. | |
5879 | * | |
5880 | * After that we can add the block group items to their btrees: | |
5881 | * update existing device item in the chunk btree, add a new block group | |
5882 | * item to the extent btree, add a new chunk item to the chunk btree and | |
5883 | * finally add the new device extent items to the devices btree. | |
5884 | */ | |
2b82032c | 5885 | |
1b86826d | 5886 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
f6f39f7a | 5887 | meta_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5888 | if (IS_ERR(meta_bg)) |
5889 | return PTR_ERR(meta_bg); | |
2b82032c | 5890 | |
1b86826d | 5891 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
f6f39f7a | 5892 | sys_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5893 | if (IS_ERR(sys_bg)) |
5894 | return PTR_ERR(sys_bg); | |
5895 | ||
5896 | return 0; | |
2b82032c YZ |
5897 | } |
5898 | ||
7dc66abb | 5899 | static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map) |
d20983b4 | 5900 | { |
fc9a2ac7 | 5901 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5902 | |
fc9a2ac7 | 5903 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5904 | } |
5905 | ||
a09f23c3 | 5906 | bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c | 5907 | { |
7dc66abb | 5908 | struct btrfs_chunk_map *map; |
d20983b4 | 5909 | int miss_ndevs = 0; |
2b82032c | 5910 | int i; |
a09f23c3 | 5911 | bool ret = true; |
2b82032c | 5912 | |
7dc66abb FM |
5913 | map = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
5914 | if (IS_ERR(map)) | |
a09f23c3 | 5915 | return false; |
2b82032c | 5916 | |
2b82032c | 5917 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5918 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5919 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5920 | miss_ndevs++; |
5921 | continue; | |
5922 | } | |
ebbede42 AJ |
5923 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5924 | &map->stripes[i].dev->dev_state)) { | |
a09f23c3 | 5925 | ret = false; |
d20983b4 | 5926 | goto end; |
2b82032c YZ |
5927 | } |
5928 | } | |
d20983b4 MX |
5929 | |
5930 | /* | |
a09f23c3 AJ |
5931 | * If the number of missing devices is larger than max errors, we can |
5932 | * not write the data into that chunk successfully. | |
d20983b4 MX |
5933 | */ |
5934 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
a09f23c3 | 5935 | ret = false; |
d20983b4 | 5936 | end: |
7dc66abb | 5937 | btrfs_free_chunk_map(map); |
a09f23c3 | 5938 | return ret; |
0b86a832 CM |
5939 | } |
5940 | ||
7dc66abb | 5941 | void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info) |
0b86a832 | 5942 | { |
7dc66abb FM |
5943 | write_lock(&fs_info->mapping_tree_lock); |
5944 | while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) { | |
5945 | struct btrfs_chunk_map *map; | |
5946 | struct rb_node *node; | |
0b86a832 | 5947 | |
7dc66abb FM |
5948 | node = rb_first_cached(&fs_info->mapping_tree); |
5949 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); | |
5950 | rb_erase_cached(&map->rb_node, &fs_info->mapping_tree); | |
5951 | RB_CLEAR_NODE(&map->rb_node); | |
5952 | chunk_map_device_clear_bits(map, CHUNK_ALLOCATED); | |
5953 | /* Once for the tree ref. */ | |
5954 | btrfs_free_chunk_map(map); | |
5955 | cond_resched_rwlock_write(&fs_info->mapping_tree_lock); | |
0b86a832 | 5956 | } |
7dc66abb | 5957 | write_unlock(&fs_info->mapping_tree_lock); |
0b86a832 CM |
5958 | } |
5959 | ||
5d964051 | 5960 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 5961 | { |
7dc66abb | 5962 | struct btrfs_chunk_map *map; |
6d322b48 QW |
5963 | enum btrfs_raid_types index; |
5964 | int ret = 1; | |
f188591e | 5965 | |
7dc66abb FM |
5966 | map = btrfs_get_chunk_map(fs_info, logical, len); |
5967 | if (IS_ERR(map)) | |
592d92ee LB |
5968 | /* |
5969 | * We could return errors for these cases, but that could get | |
5970 | * ugly and we'd probably do the same thing which is just not do | |
5971 | * anything else and exit, so return 1 so the callers don't try | |
5972 | * to use other copies. | |
5973 | */ | |
fb7669b5 | 5974 | return 1; |
fb7669b5 | 5975 | |
6d322b48 QW |
5976 | index = btrfs_bg_flags_to_raid_index(map->type); |
5977 | ||
5978 | /* Non-RAID56, use their ncopies from btrfs_raid_array. */ | |
5979 | if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
5980 | ret = btrfs_raid_array[index].ncopies; | |
53b381b3 DW |
5981 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5982 | ret = 2; | |
5983 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5984 | /* |
5985 | * There could be two corrupted data stripes, we need | |
5986 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5987 | * |
8810f751 LB |
5988 | * Fail a stripe at a time on every retry except the |
5989 | * stripe under reconstruction. | |
5990 | */ | |
5991 | ret = map->num_stripes; | |
7dc66abb | 5992 | btrfs_free_chunk_map(map); |
f188591e CM |
5993 | return ret; |
5994 | } | |
5995 | ||
2ff7e61e | 5996 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5997 | u64 logical) |
5998 | { | |
7dc66abb | 5999 | struct btrfs_chunk_map *map; |
0b246afa | 6000 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 6001 | |
b036f479 QW |
6002 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
6003 | return len; | |
6004 | ||
7dc66abb | 6005 | map = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 6006 | |
7dc66abb | 6007 | if (!WARN_ON(IS_ERR(map))) { |
69f03f13 | 6008 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
cb091225 | 6009 | len = btrfs_stripe_nr_to_offset(nr_data_stripes(map)); |
7dc66abb | 6010 | btrfs_free_chunk_map(map); |
69f03f13 | 6011 | } |
53b381b3 DW |
6012 | return len; |
6013 | } | |
6014 | ||
e4ff5fb5 | 6015 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 | 6016 | { |
7dc66abb | 6017 | struct btrfs_chunk_map *map; |
53b381b3 DW |
6018 | int ret = 0; |
6019 | ||
b036f479 QW |
6020 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
6021 | return 0; | |
6022 | ||
7dc66abb | 6023 | map = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 6024 | |
7dc66abb | 6025 | if (!WARN_ON(IS_ERR(map))) { |
69f03f13 NB |
6026 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
6027 | ret = 1; | |
7dc66abb | 6028 | btrfs_free_chunk_map(map); |
69f03f13 | 6029 | } |
53b381b3 DW |
6030 | return ret; |
6031 | } | |
6032 | ||
30d9861f | 6033 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
7dc66abb | 6034 | struct btrfs_chunk_map *map, int first, |
8ba0ae78 | 6035 | int dev_replace_is_ongoing) |
dfe25020 | 6036 | { |
4dc4a3be | 6037 | const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy); |
dfe25020 | 6038 | int i; |
99f92a7c | 6039 | int num_stripes; |
8ba0ae78 | 6040 | int preferred_mirror; |
30d9861f SB |
6041 | int tolerance; |
6042 | struct btrfs_device *srcdev; | |
6043 | ||
99f92a7c | 6044 | ASSERT((map->type & |
c7369b3f | 6045 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
6046 | |
6047 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
6048 | num_stripes = map->sub_stripes; | |
6049 | else | |
6050 | num_stripes = map->num_stripes; | |
6051 | ||
4dc4a3be | 6052 | switch (policy) { |
33fd2f71 AJ |
6053 | default: |
6054 | /* Shouldn't happen, just warn and use pid instead of failing */ | |
4dc4a3be NA |
6055 | btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid", |
6056 | policy); | |
6057 | WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID); | |
33fd2f71 AJ |
6058 | fallthrough; |
6059 | case BTRFS_READ_POLICY_PID: | |
6060 | preferred_mirror = first + (current->pid % num_stripes); | |
6061 | break; | |
6062 | } | |
8ba0ae78 | 6063 | |
30d9861f SB |
6064 | if (dev_replace_is_ongoing && |
6065 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
6066 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
6067 | srcdev = fs_info->dev_replace.srcdev; | |
6068 | else | |
6069 | srcdev = NULL; | |
6070 | ||
6071 | /* | |
6072 | * try to avoid the drive that is the source drive for a | |
6073 | * dev-replace procedure, only choose it if no other non-missing | |
6074 | * mirror is available | |
6075 | */ | |
6076 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
6077 | if (map->stripes[preferred_mirror].dev->bdev && |
6078 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
6079 | return preferred_mirror; | |
99f92a7c | 6080 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
6081 | if (map->stripes[i].dev->bdev && |
6082 | (tolerance || map->stripes[i].dev != srcdev)) | |
6083 | return i; | |
6084 | } | |
dfe25020 | 6085 | } |
30d9861f | 6086 | |
dfe25020 CM |
6087 | /* we couldn't find one that doesn't fail. Just return something |
6088 | * and the io error handling code will clean up eventually | |
6089 | */ | |
8ba0ae78 | 6090 | return preferred_mirror; |
dfe25020 CM |
6091 | } |
6092 | ||
731ccf15 | 6093 | static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, |
02c372e1 | 6094 | u64 logical, |
1faf3885 | 6095 | u16 total_stripes) |
6e9606d2 | 6096 | { |
4ced85f8 QW |
6097 | struct btrfs_io_context *bioc; |
6098 | ||
6099 | bioc = kzalloc( | |
4c664611 QW |
6100 | /* The size of btrfs_io_context */ |
6101 | sizeof(struct btrfs_io_context) + | |
6102 | /* Plus the variable array for the stripes */ | |
18d758a2 | 6103 | sizeof(struct btrfs_io_stripe) * (total_stripes), |
9f0eac07 L |
6104 | GFP_NOFS); |
6105 | ||
6106 | if (!bioc) | |
6107 | return NULL; | |
6e9606d2 | 6108 | |
4c664611 | 6109 | refcount_set(&bioc->refs, 1); |
6e9606d2 | 6110 | |
731ccf15 | 6111 | bioc->fs_info = fs_info; |
1faf3885 | 6112 | bioc->replace_stripe_src = -1; |
18d758a2 | 6113 | bioc->full_stripe_logical = (u64)-1; |
02c372e1 | 6114 | bioc->logical = logical; |
608769a4 | 6115 | |
4c664611 | 6116 | return bioc; |
6e9606d2 ZL |
6117 | } |
6118 | ||
4c664611 | 6119 | void btrfs_get_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 6120 | { |
4c664611 QW |
6121 | WARN_ON(!refcount_read(&bioc->refs)); |
6122 | refcount_inc(&bioc->refs); | |
6e9606d2 ZL |
6123 | } |
6124 | ||
4c664611 | 6125 | void btrfs_put_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 6126 | { |
4c664611 | 6127 | if (!bioc) |
6e9606d2 | 6128 | return; |
4c664611 QW |
6129 | if (refcount_dec_and_test(&bioc->refs)) |
6130 | kfree(bioc); | |
6e9606d2 ZL |
6131 | } |
6132 | ||
0b3d4cd3 LB |
6133 | /* |
6134 | * Please note that, discard won't be sent to target device of device | |
6135 | * replace. | |
6136 | */ | |
a4012f06 CH |
6137 | struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, |
6138 | u64 logical, u64 *length_ret, | |
6139 | u32 *num_stripes) | |
0b3d4cd3 | 6140 | { |
7dc66abb | 6141 | struct btrfs_chunk_map *map; |
a4012f06 | 6142 | struct btrfs_discard_stripe *stripes; |
6b7faadd | 6143 | u64 length = *length_ret; |
0b3d4cd3 | 6144 | u64 offset; |
6ded22c1 QW |
6145 | u32 stripe_nr; |
6146 | u32 stripe_nr_end; | |
6147 | u32 stripe_cnt; | |
0b3d4cd3 | 6148 | u64 stripe_end_offset; |
0b3d4cd3 | 6149 | u64 stripe_offset; |
0b3d4cd3 LB |
6150 | u32 stripe_index; |
6151 | u32 factor = 0; | |
6152 | u32 sub_stripes = 0; | |
6ded22c1 | 6153 | u32 stripes_per_dev = 0; |
0b3d4cd3 LB |
6154 | u32 remaining_stripes = 0; |
6155 | u32 last_stripe = 0; | |
a4012f06 | 6156 | int ret; |
0b3d4cd3 LB |
6157 | int i; |
6158 | ||
7dc66abb FM |
6159 | map = btrfs_get_chunk_map(fs_info, logical, length); |
6160 | if (IS_ERR(map)) | |
6161 | return ERR_CAST(map); | |
a4012f06 | 6162 | |
0b3d4cd3 LB |
6163 | /* we don't discard raid56 yet */ |
6164 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
6165 | ret = -EOPNOTSUPP; | |
a4012f06 | 6166 | goto out_free_map; |
a97699d1 | 6167 | } |
0b3d4cd3 | 6168 | |
7dc66abb FM |
6169 | offset = logical - map->start; |
6170 | length = min_t(u64, map->start + map->chunk_len - logical, length); | |
6b7faadd | 6171 | *length_ret = length; |
0b3d4cd3 | 6172 | |
0b3d4cd3 LB |
6173 | /* |
6174 | * stripe_nr counts the total number of stripes we have to stride | |
6175 | * to get to this block | |
6176 | */ | |
a97699d1 | 6177 | stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; |
0b3d4cd3 LB |
6178 | |
6179 | /* stripe_offset is the offset of this block in its stripe */ | |
cb091225 | 6180 | stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr); |
0b3d4cd3 | 6181 | |
a97699d1 QW |
6182 | stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> |
6183 | BTRFS_STRIPE_LEN_SHIFT; | |
0b3d4cd3 | 6184 | stripe_cnt = stripe_nr_end - stripe_nr; |
cb091225 | 6185 | stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) - |
0b3d4cd3 LB |
6186 | (offset + length); |
6187 | /* | |
6188 | * after this, stripe_nr is the number of stripes on this | |
6189 | * device we have to walk to find the data, and stripe_index is | |
6190 | * the number of our device in the stripe array | |
6191 | */ | |
a4012f06 | 6192 | *num_stripes = 1; |
0b3d4cd3 LB |
6193 | stripe_index = 0; |
6194 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
6195 | BTRFS_BLOCK_GROUP_RAID10)) { | |
6196 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
6197 | sub_stripes = 1; | |
6198 | else | |
6199 | sub_stripes = map->sub_stripes; | |
6200 | ||
6201 | factor = map->num_stripes / sub_stripes; | |
a4012f06 | 6202 | *num_stripes = min_t(u64, map->num_stripes, |
0b3d4cd3 | 6203 | sub_stripes * stripe_cnt); |
6ded22c1 QW |
6204 | stripe_index = stripe_nr % factor; |
6205 | stripe_nr /= factor; | |
0b3d4cd3 | 6206 | stripe_index *= sub_stripes; |
6ded22c1 QW |
6207 | |
6208 | remaining_stripes = stripe_cnt % factor; | |
6209 | stripes_per_dev = stripe_cnt / factor; | |
6210 | last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; | |
c7369b3f | 6211 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 | 6212 | BTRFS_BLOCK_GROUP_DUP)) { |
a4012f06 | 6213 | *num_stripes = map->num_stripes; |
0b3d4cd3 | 6214 | } else { |
6ded22c1 QW |
6215 | stripe_index = stripe_nr % map->num_stripes; |
6216 | stripe_nr /= map->num_stripes; | |
0b3d4cd3 LB |
6217 | } |
6218 | ||
a4012f06 CH |
6219 | stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); |
6220 | if (!stripes) { | |
0b3d4cd3 | 6221 | ret = -ENOMEM; |
a4012f06 | 6222 | goto out_free_map; |
0b3d4cd3 LB |
6223 | } |
6224 | ||
a4012f06 CH |
6225 | for (i = 0; i < *num_stripes; i++) { |
6226 | stripes[i].physical = | |
0b3d4cd3 | 6227 | map->stripes[stripe_index].physical + |
cb091225 | 6228 | stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr); |
a4012f06 | 6229 | stripes[i].dev = map->stripes[stripe_index].dev; |
0b3d4cd3 LB |
6230 | |
6231 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
6232 | BTRFS_BLOCK_GROUP_RAID10)) { | |
cb091225 | 6233 | stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev); |
0b3d4cd3 LB |
6234 | |
6235 | if (i / sub_stripes < remaining_stripes) | |
a97699d1 | 6236 | stripes[i].length += BTRFS_STRIPE_LEN; |
0b3d4cd3 LB |
6237 | |
6238 | /* | |
6239 | * Special for the first stripe and | |
6240 | * the last stripe: | |
6241 | * | |
6242 | * |-------|...|-------| | |
6243 | * |----------| | |
6244 | * off end_off | |
6245 | */ | |
6246 | if (i < sub_stripes) | |
a4012f06 | 6247 | stripes[i].length -= stripe_offset; |
0b3d4cd3 LB |
6248 | |
6249 | if (stripe_index >= last_stripe && | |
6250 | stripe_index <= (last_stripe + | |
6251 | sub_stripes - 1)) | |
a4012f06 | 6252 | stripes[i].length -= stripe_end_offset; |
0b3d4cd3 LB |
6253 | |
6254 | if (i == sub_stripes - 1) | |
6255 | stripe_offset = 0; | |
6256 | } else { | |
a4012f06 | 6257 | stripes[i].length = length; |
0b3d4cd3 LB |
6258 | } |
6259 | ||
6260 | stripe_index++; | |
6261 | if (stripe_index == map->num_stripes) { | |
6262 | stripe_index = 0; | |
6263 | stripe_nr++; | |
6264 | } | |
6265 | } | |
6266 | ||
7dc66abb | 6267 | btrfs_free_chunk_map(map); |
a4012f06 CH |
6268 | return stripes; |
6269 | out_free_map: | |
7dc66abb | 6270 | btrfs_free_chunk_map(map); |
a4012f06 | 6271 | return ERR_PTR(ret); |
0b3d4cd3 LB |
6272 | } |
6273 | ||
6143c23c NA |
6274 | static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) |
6275 | { | |
6276 | struct btrfs_block_group *cache; | |
6277 | bool ret; | |
6278 | ||
de17addc | 6279 | /* Non zoned filesystem does not use "to_copy" flag */ |
6143c23c NA |
6280 | if (!btrfs_is_zoned(fs_info)) |
6281 | return false; | |
6282 | ||
6283 | cache = btrfs_lookup_block_group(fs_info, logical); | |
6284 | ||
3349b57f | 6285 | ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); |
6143c23c NA |
6286 | |
6287 | btrfs_put_block_group(cache); | |
6288 | return ret; | |
6289 | } | |
6290 | ||
73c0f228 | 6291 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
be5c7edb | 6292 | struct btrfs_io_context *bioc, |
73c0f228 | 6293 | struct btrfs_dev_replace *dev_replace, |
6143c23c | 6294 | u64 logical, |
73c0f228 LB |
6295 | int *num_stripes_ret, int *max_errors_ret) |
6296 | { | |
73c0f228 | 6297 | u64 srcdev_devid = dev_replace->srcdev->devid; |
1faf3885 QW |
6298 | /* |
6299 | * At this stage, num_stripes is still the real number of stripes, | |
6300 | * excluding the duplicated stripes. | |
6301 | */ | |
73c0f228 | 6302 | int num_stripes = *num_stripes_ret; |
1faf3885 | 6303 | int nr_extra_stripes = 0; |
73c0f228 LB |
6304 | int max_errors = *max_errors_ret; |
6305 | int i; | |
6306 | ||
1faf3885 QW |
6307 | /* |
6308 | * A block group which has "to_copy" set will eventually be copied by | |
6309 | * the dev-replace process. We can avoid cloning IO here. | |
6310 | */ | |
6311 | if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) | |
6312 | return; | |
73c0f228 | 6313 | |
1faf3885 QW |
6314 | /* |
6315 | * Duplicate the write operations while the dev-replace procedure is | |
6316 | * running. Since the copying of the old disk to the new disk takes | |
6317 | * place at run time while the filesystem is mounted writable, the | |
6318 | * regular write operations to the old disk have to be duplicated to go | |
6319 | * to the new disk as well. | |
6320 | * | |
6321 | * Note that device->missing is handled by the caller, and that the | |
6322 | * write to the old disk is already set up in the stripes array. | |
6323 | */ | |
6324 | for (i = 0; i < num_stripes; i++) { | |
6325 | struct btrfs_io_stripe *old = &bioc->stripes[i]; | |
6326 | struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; | |
6143c23c | 6327 | |
1faf3885 QW |
6328 | if (old->dev->devid != srcdev_devid) |
6329 | continue; | |
73c0f228 | 6330 | |
1faf3885 QW |
6331 | new->physical = old->physical; |
6332 | new->dev = dev_replace->tgtdev; | |
6333 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
6334 | bioc->replace_stripe_src = i; | |
6335 | nr_extra_stripes++; | |
6336 | } | |
73c0f228 | 6337 | |
1faf3885 QW |
6338 | /* We can only have at most 2 extra nr_stripes (for DUP). */ |
6339 | ASSERT(nr_extra_stripes <= 2); | |
6340 | /* | |
6341 | * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for | |
6342 | * replace. | |
6343 | * If we have 2 extra stripes, only choose the one with smaller physical. | |
6344 | */ | |
6345 | if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { | |
6346 | struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; | |
6347 | struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; | |
73c0f228 | 6348 | |
1faf3885 QW |
6349 | /* Only DUP can have two extra stripes. */ |
6350 | ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); | |
6351 | ||
6352 | /* | |
6353 | * Swap the last stripe stripes and reduce @nr_extra_stripes. | |
6354 | * The extra stripe would still be there, but won't be accessed. | |
6355 | */ | |
6356 | if (first->physical > second->physical) { | |
6357 | swap(second->physical, first->physical); | |
6358 | swap(second->dev, first->dev); | |
6359 | nr_extra_stripes--; | |
73c0f228 LB |
6360 | } |
6361 | } | |
6362 | ||
1faf3885 QW |
6363 | *num_stripes_ret = num_stripes + nr_extra_stripes; |
6364 | *max_errors_ret = max_errors + nr_extra_stripes; | |
6365 | bioc->replace_nr_stripes = nr_extra_stripes; | |
73c0f228 LB |
6366 | } |
6367 | ||
e94dfb7a JT |
6368 | static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset, |
6369 | struct btrfs_io_geometry *io_geom) | |
5f141126 | 6370 | { |
cc353a8b | 6371 | /* |
f8a02dc6 CH |
6372 | * Stripe_nr is the stripe where this block falls. stripe_offset is |
6373 | * the offset of this block in its stripe. | |
cc353a8b | 6374 | */ |
e94dfb7a JT |
6375 | io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; |
6376 | io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; | |
6377 | ASSERT(io_geom->stripe_offset < U32_MAX); | |
5f141126 | 6378 | |
f8a02dc6 | 6379 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
cb091225 QW |
6380 | unsigned long full_stripe_len = |
6381 | btrfs_stripe_nr_to_offset(nr_data_stripes(map)); | |
5f141126 | 6382 | |
a97699d1 QW |
6383 | /* |
6384 | * For full stripe start, we use previously calculated | |
6385 | * @stripe_nr. Align it to nr_data_stripes, then multiply with | |
6386 | * STRIPE_LEN. | |
6387 | * | |
6388 | * By this we can avoid u64 division completely. And we have | |
6389 | * to go rounddown(), not round_down(), as nr_data_stripes is | |
6390 | * not ensured to be power of 2. | |
6391 | */ | |
e94dfb7a JT |
6392 | io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset( |
6393 | rounddown(io_geom->stripe_nr, nr_data_stripes(map))); | |
5f141126 | 6394 | |
e94dfb7a JT |
6395 | ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset); |
6396 | ASSERT(io_geom->raid56_full_stripe_start <= offset); | |
5f141126 | 6397 | /* |
f8a02dc6 CH |
6398 | * For writes to RAID56, allow to write a full stripe set, but |
6399 | * no straddling of stripe sets. | |
5f141126 | 6400 | */ |
e94dfb7a JT |
6401 | if (io_geom->op == BTRFS_MAP_WRITE) |
6402 | return full_stripe_len - (offset - io_geom->raid56_full_stripe_start); | |
5f141126 NB |
6403 | } |
6404 | ||
f8a02dc6 CH |
6405 | /* |
6406 | * For other RAID types and for RAID56 reads, allow a single stripe (on | |
6407 | * a single disk). | |
6408 | */ | |
6409 | if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) | |
e94dfb7a | 6410 | return BTRFS_STRIPE_LEN - io_geom->stripe_offset; |
f8a02dc6 | 6411 | return U64_MAX; |
5f141126 NB |
6412 | } |
6413 | ||
6edf6822 JT |
6414 | static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical, |
6415 | u64 *length, struct btrfs_io_stripe *dst, | |
6416 | struct btrfs_chunk_map *map, | |
6417 | struct btrfs_io_geometry *io_geom) | |
03793cbb | 6418 | { |
6edf6822 | 6419 | dst->dev = map->stripes[io_geom->stripe_index].dev; |
10e27980 | 6420 | |
6edf6822 JT |
6421 | if (io_geom->op == BTRFS_MAP_READ && |
6422 | btrfs_need_stripe_tree_update(fs_info, map->type)) | |
10e27980 | 6423 | return btrfs_get_raid_extent_offset(fs_info, logical, length, |
6edf6822 JT |
6424 | map->type, |
6425 | io_geom->stripe_index, dst); | |
10e27980 | 6426 | |
6edf6822 JT |
6427 | dst->physical = map->stripes[io_geom->stripe_index].physical + |
6428 | io_geom->stripe_offset + | |
6429 | btrfs_stripe_nr_to_offset(io_geom->stripe_nr); | |
10e27980 | 6430 | return 0; |
03793cbb CH |
6431 | } |
6432 | ||
02d05b64 JT |
6433 | static bool is_single_device_io(struct btrfs_fs_info *fs_info, |
6434 | const struct btrfs_io_stripe *smap, | |
6435 | const struct btrfs_chunk_map *map, | |
6436 | int num_alloc_stripes, | |
6437 | enum btrfs_map_op op, int mirror_num) | |
6438 | { | |
6439 | if (!smap) | |
6440 | return false; | |
6441 | ||
6442 | if (num_alloc_stripes != 1) | |
6443 | return false; | |
6444 | ||
6445 | if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ) | |
6446 | return false; | |
6447 | ||
6448 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) | |
6449 | return false; | |
6450 | ||
6451 | return true; | |
6452 | } | |
6453 | ||
30e8534b JT |
6454 | static void map_blocks_raid0(const struct btrfs_chunk_map *map, |
6455 | struct btrfs_io_geometry *io_geom) | |
6456 | { | |
6457 | io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes; | |
6458 | io_geom->stripe_nr /= map->num_stripes; | |
6459 | if (io_geom->op == BTRFS_MAP_READ) | |
6460 | io_geom->mirror_num = 1; | |
6461 | } | |
6462 | ||
5e36aba8 JT |
6463 | static void map_blocks_raid1(struct btrfs_fs_info *fs_info, |
6464 | struct btrfs_chunk_map *map, | |
6465 | struct btrfs_io_geometry *io_geom, | |
6466 | bool dev_replace_is_ongoing) | |
6467 | { | |
6468 | if (io_geom->op != BTRFS_MAP_READ) { | |
6469 | io_geom->num_stripes = map->num_stripes; | |
6470 | return; | |
6471 | } | |
6472 | ||
6473 | if (io_geom->mirror_num) { | |
6474 | io_geom->stripe_index = io_geom->mirror_num - 1; | |
6475 | return; | |
6476 | } | |
6477 | ||
6478 | io_geom->stripe_index = find_live_mirror(fs_info, map, 0, | |
6479 | dev_replace_is_ongoing); | |
6480 | io_geom->mirror_num = io_geom->stripe_index + 1; | |
6481 | } | |
6482 | ||
5aeb15c8 JT |
6483 | static void map_blocks_dup(const struct btrfs_chunk_map *map, |
6484 | struct btrfs_io_geometry *io_geom) | |
6485 | { | |
6486 | if (io_geom->op != BTRFS_MAP_READ) { | |
6487 | io_geom->num_stripes = map->num_stripes; | |
6488 | return; | |
6489 | } | |
6490 | ||
6491 | if (io_geom->mirror_num) { | |
6492 | io_geom->stripe_index = io_geom->mirror_num - 1; | |
6493 | return; | |
6494 | } | |
6495 | ||
6496 | io_geom->mirror_num = 1; | |
6497 | } | |
6498 | ||
8938f112 JT |
6499 | static void map_blocks_raid10(struct btrfs_fs_info *fs_info, |
6500 | struct btrfs_chunk_map *map, | |
6501 | struct btrfs_io_geometry *io_geom, | |
6502 | bool dev_replace_is_ongoing) | |
6503 | { | |
6504 | u32 factor = map->num_stripes / map->sub_stripes; | |
6505 | int old_stripe_index; | |
6506 | ||
6507 | io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes; | |
6508 | io_geom->stripe_nr /= factor; | |
6509 | ||
6510 | if (io_geom->op != BTRFS_MAP_READ) { | |
6511 | io_geom->num_stripes = map->sub_stripes; | |
6512 | return; | |
6513 | } | |
6514 | ||
6515 | if (io_geom->mirror_num) { | |
6516 | io_geom->stripe_index += io_geom->mirror_num - 1; | |
6517 | return; | |
6518 | } | |
6519 | ||
6520 | old_stripe_index = io_geom->stripe_index; | |
6521 | io_geom->stripe_index = find_live_mirror(fs_info, map, | |
6522 | io_geom->stripe_index, | |
6523 | dev_replace_is_ongoing); | |
6524 | io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1; | |
6525 | } | |
6526 | ||
089221d3 JT |
6527 | static void map_blocks_raid56_write(struct btrfs_chunk_map *map, |
6528 | struct btrfs_io_geometry *io_geom, | |
6529 | u64 logical, u64 *length) | |
6530 | { | |
6531 | int data_stripes = nr_data_stripes(map); | |
6532 | ||
6533 | /* | |
6534 | * Needs full stripe mapping. | |
6535 | * | |
6536 | * Push stripe_nr back to the start of the full stripe For those cases | |
6537 | * needing a full stripe, @stripe_nr is the full stripe number. | |
6538 | * | |
6539 | * Originally we go raid56_full_stripe_start / full_stripe_len, but | |
6540 | * that can be expensive. Here we just divide @stripe_nr with | |
6541 | * @data_stripes. | |
6542 | */ | |
6543 | io_geom->stripe_nr /= data_stripes; | |
6544 | ||
6545 | /* RAID[56] write or recovery. Return all stripes */ | |
6546 | io_geom->num_stripes = map->num_stripes; | |
6547 | io_geom->max_errors = btrfs_chunk_max_errors(map); | |
6548 | ||
6549 | /* Return the length to the full stripe end. */ | |
6550 | *length = min(logical + *length, | |
6551 | io_geom->raid56_full_stripe_start + map->start + | |
6552 | btrfs_stripe_nr_to_offset(data_stripes)) - | |
6553 | logical; | |
6554 | io_geom->stripe_index = 0; | |
6555 | io_geom->stripe_offset = 0; | |
6556 | } | |
6557 | ||
6558 | static void map_blocks_raid56_read(struct btrfs_chunk_map *map, | |
6559 | struct btrfs_io_geometry *io_geom) | |
6560 | { | |
6561 | int data_stripes = nr_data_stripes(map); | |
6562 | ||
6563 | ASSERT(io_geom->mirror_num <= 1); | |
6564 | /* Just grab the data stripe directly. */ | |
6565 | io_geom->stripe_index = io_geom->stripe_nr % data_stripes; | |
6566 | io_geom->stripe_nr /= data_stripes; | |
6567 | ||
6568 | /* We distribute the parity blocks across stripes. */ | |
6569 | io_geom->stripe_index = | |
6570 | (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes; | |
6571 | ||
6572 | if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1) | |
6573 | io_geom->mirror_num = 1; | |
6574 | } | |
6575 | ||
a16fb8c6 JT |
6576 | static void map_blocks_single(const struct btrfs_chunk_map *map, |
6577 | struct btrfs_io_geometry *io_geom) | |
6578 | { | |
6579 | io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes; | |
6580 | io_geom->stripe_nr /= map->num_stripes; | |
6581 | io_geom->mirror_num = io_geom->stripe_index + 1; | |
6582 | } | |
6583 | ||
ed3764f7 QW |
6584 | /* |
6585 | * Map one logical range to one or more physical ranges. | |
6586 | * | |
6587 | * @length: (Mandatory) mapped length of this run. | |
6588 | * One logical range can be split into different segments | |
6589 | * due to factors like zones and RAID0/5/6/10 stripe | |
6590 | * boundaries. | |
6591 | * | |
6592 | * @bioc_ret: (Mandatory) returned btrfs_io_context structure. | |
6593 | * which has one or more physical ranges (btrfs_io_stripe) | |
6594 | * recorded inside. | |
6595 | * Caller should call btrfs_put_bioc() to free it after use. | |
6596 | * | |
6597 | * @smap: (Optional) single physical range optimization. | |
6598 | * If the map request can be fulfilled by one single | |
6599 | * physical range, and this is parameter is not NULL, | |
6600 | * then @bioc_ret would be NULL, and @smap would be | |
6601 | * updated. | |
6602 | * | |
6603 | * @mirror_num_ret: (Mandatory) returned mirror number if the original | |
6604 | * value is 0. | |
6605 | * | |
6606 | * Mirror number 0 means to choose any live mirrors. | |
6607 | * | |
6608 | * For non-RAID56 profiles, non-zero mirror_num means | |
6609 | * the Nth mirror. (e.g. mirror_num 1 means the first | |
6610 | * copy). | |
6611 | * | |
6612 | * For RAID56 profile, mirror 1 means rebuild from P and | |
6613 | * the remaining data stripes. | |
6614 | * | |
6615 | * For RAID6 profile, mirror > 2 means mark another | |
6616 | * data/P stripe error and rebuild from the remaining | |
6617 | * stripes.. | |
ed3764f7 | 6618 | */ |
cd4efd21 CH |
6619 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
6620 | u64 logical, u64 *length, | |
6621 | struct btrfs_io_context **bioc_ret, | |
9fb2acc2 | 6622 | struct btrfs_io_stripe *smap, int *mirror_num_ret) |
0b86a832 | 6623 | { |
7dc66abb | 6624 | struct btrfs_chunk_map *map; |
fd747f2d | 6625 | struct btrfs_io_geometry io_geom = { 0 }; |
f8a02dc6 | 6626 | u64 map_offset; |
cea9e445 | 6627 | int i; |
de11cc12 | 6628 | int ret = 0; |
5f50fa91 | 6629 | int num_copies; |
4c664611 | 6630 | struct btrfs_io_context *bioc = NULL; |
472262f3 SB |
6631 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6632 | int dev_replace_is_ongoing = 0; | |
4ced85f8 | 6633 | u16 num_alloc_stripes; |
f8a02dc6 | 6634 | u64 max_len; |
89b798ad | 6635 | |
4c664611 | 6636 | ASSERT(bioc_ret); |
0b3d4cd3 | 6637 | |
fd747f2d JT |
6638 | io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); |
6639 | io_geom.num_stripes = 1; | |
6640 | io_geom.stripe_index = 0; | |
6641 | io_geom.op = op; | |
6642 | ||
5f50fa91 | 6643 | num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); |
fd747f2d | 6644 | if (io_geom.mirror_num > num_copies) |
5f50fa91 QW |
6645 | return -EINVAL; |
6646 | ||
7dc66abb FM |
6647 | map = btrfs_get_chunk_map(fs_info, logical, *length); |
6648 | if (IS_ERR(map)) | |
6649 | return PTR_ERR(map); | |
42034313 | 6650 | |
7dc66abb | 6651 | map_offset = logical - map->start; |
fd747f2d | 6652 | io_geom.raid56_full_stripe_start = (u64)-1; |
e94dfb7a | 6653 | max_len = btrfs_max_io_len(map, map_offset, &io_geom); |
7dc66abb | 6654 | *length = min_t(u64, map->chunk_len - map_offset, max_len); |
593060d7 | 6655 | |
cb5583dd | 6656 | down_read(&dev_replace->rwsem); |
472262f3 | 6657 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6658 | /* |
6659 | * Hold the semaphore for read during the whole operation, write is | |
6660 | * requested at commit time but must wait. | |
6661 | */ | |
472262f3 | 6662 | if (!dev_replace_is_ongoing) |
cb5583dd | 6663 | up_read(&dev_replace->rwsem); |
472262f3 | 6664 | |
b55b3077 JT |
6665 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
6666 | case BTRFS_BLOCK_GROUP_RAID0: | |
30e8534b | 6667 | map_blocks_raid0(map, &io_geom); |
b55b3077 JT |
6668 | break; |
6669 | case BTRFS_BLOCK_GROUP_RAID1: | |
6670 | case BTRFS_BLOCK_GROUP_RAID1C3: | |
6671 | case BTRFS_BLOCK_GROUP_RAID1C4: | |
5e36aba8 | 6672 | map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing); |
b55b3077 JT |
6673 | break; |
6674 | case BTRFS_BLOCK_GROUP_DUP: | |
5aeb15c8 | 6675 | map_blocks_dup(map, &io_geom); |
b55b3077 JT |
6676 | break; |
6677 | case BTRFS_BLOCK_GROUP_RAID10: | |
8938f112 | 6678 | map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing); |
b55b3077 JT |
6679 | break; |
6680 | case BTRFS_BLOCK_GROUP_RAID5: | |
6681 | case BTRFS_BLOCK_GROUP_RAID6: | |
089221d3 JT |
6682 | if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1) |
6683 | map_blocks_raid56_write(map, &io_geom, logical, length); | |
6684 | else | |
6685 | map_blocks_raid56_read(map, &io_geom); | |
b55b3077 JT |
6686 | break; |
6687 | default: | |
8790d502 | 6688 | /* |
6ded22c1 | 6689 | * After this, stripe_nr is the number of stripes on this |
47c5713f DS |
6690 | * device we have to walk to find the data, and stripe_index is |
6691 | * the number of our device in the stripe array | |
8790d502 | 6692 | */ |
a16fb8c6 | 6693 | map_blocks_single(map, &io_geom); |
b55b3077 | 6694 | break; |
8790d502 | 6695 | } |
fd747f2d | 6696 | if (io_geom.stripe_index >= map->num_stripes) { |
5d163e0e JM |
6697 | btrfs_crit(fs_info, |
6698 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
fd747f2d | 6699 | io_geom.stripe_index, map->num_stripes); |
e042d1ec JB |
6700 | ret = -EINVAL; |
6701 | goto out; | |
6702 | } | |
cea9e445 | 6703 | |
fd747f2d | 6704 | num_alloc_stripes = io_geom.num_stripes; |
1faf3885 QW |
6705 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
6706 | op != BTRFS_MAP_READ) | |
6707 | /* | |
6708 | * For replace case, we need to add extra stripes for extra | |
6709 | * duplicated stripes. | |
6710 | * | |
6711 | * For both WRITE and GET_READ_MIRRORS, we may have at most | |
6712 | * 2 more stripes (DUP types, otherwise 1). | |
6713 | */ | |
6714 | num_alloc_stripes += 2; | |
2c8cdd6e | 6715 | |
03793cbb CH |
6716 | /* |
6717 | * If this I/O maps to a single device, try to return the device and | |
6718 | * physical block information on the stack instead of allocating an | |
6719 | * I/O context structure. | |
6720 | */ | |
02d05b64 | 6721 | if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op, |
fd747f2d | 6722 | io_geom.mirror_num)) { |
6edf6822 | 6723 | ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom); |
4e7de35e | 6724 | if (mirror_num_ret) |
fd747f2d | 6725 | *mirror_num_ret = io_geom.mirror_num; |
03793cbb | 6726 | *bioc_ret = NULL; |
03793cbb CH |
6727 | goto out; |
6728 | } | |
6729 | ||
02c372e1 | 6730 | bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes); |
4c664611 | 6731 | if (!bioc) { |
de11cc12 LZ |
6732 | ret = -ENOMEM; |
6733 | goto out; | |
6734 | } | |
1faf3885 | 6735 | bioc->map_type = map->type; |
608769a4 | 6736 | |
18d758a2 QW |
6737 | /* |
6738 | * For RAID56 full map, we need to make sure the stripes[] follows the | |
6739 | * rule that data stripes are all ordered, then followed with P and Q | |
6740 | * (if we have). | |
6741 | * | |
6742 | * It's still mostly the same as other profiles, just with extra rotation. | |
6743 | */ | |
9fb2acc2 | 6744 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
fd747f2d | 6745 | (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) { |
18d758a2 QW |
6746 | /* |
6747 | * For RAID56 @stripe_nr is already the number of full stripes | |
6748 | * before us, which is also the rotation value (needs to modulo | |
6749 | * with num_stripes). | |
6750 | * | |
6751 | * In this case, we just add @stripe_nr with @i, then do the | |
6752 | * modulo, to reduce one modulo call. | |
6753 | */ | |
7dc66abb | 6754 | bioc->full_stripe_logical = map->start + |
d9d4ce9f JT |
6755 | btrfs_stripe_nr_to_offset(io_geom.stripe_nr * |
6756 | nr_data_stripes(map)); | |
fd747f2d | 6757 | for (int i = 0; i < io_geom.num_stripes; i++) { |
89f547c6 JT |
6758 | struct btrfs_io_stripe *dst = &bioc->stripes[i]; |
6759 | u32 stripe_index; | |
6760 | ||
6761 | stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes; | |
6762 | dst->dev = map->stripes[stripe_index].dev; | |
6763 | dst->physical = | |
6764 | map->stripes[stripe_index].physical + | |
6765 | io_geom.stripe_offset + | |
6766 | btrfs_stripe_nr_to_offset(io_geom.stripe_nr); | |
10e27980 | 6767 | } |
18d758a2 QW |
6768 | } else { |
6769 | /* | |
6770 | * For all other non-RAID56 profiles, just copy the target | |
6771 | * stripe into the bioc. | |
6772 | */ | |
fd747f2d | 6773 | for (i = 0; i < io_geom.num_stripes; i++) { |
6edf6822 JT |
6774 | ret = set_io_stripe(fs_info, logical, length, |
6775 | &bioc->stripes[i], map, &io_geom); | |
10e27980 JT |
6776 | if (ret < 0) |
6777 | break; | |
fd747f2d | 6778 | io_geom.stripe_index++; |
18d758a2 | 6779 | } |
593060d7 | 6780 | } |
de11cc12 | 6781 | |
10e27980 JT |
6782 | if (ret) { |
6783 | *bioc_ret = NULL; | |
6784 | btrfs_put_bioc(bioc); | |
6785 | goto out; | |
6786 | } | |
6787 | ||
8680e587 | 6788 | if (op != BTRFS_MAP_READ) |
fd747f2d | 6789 | io_geom.max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6790 | |
73c0f228 | 6791 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
8680e587 | 6792 | op != BTRFS_MAP_READ) { |
be5c7edb | 6793 | handle_ops_on_dev_replace(op, bioc, dev_replace, logical, |
fd747f2d | 6794 | &io_geom.num_stripes, &io_geom.max_errors); |
472262f3 SB |
6795 | } |
6796 | ||
4c664611 | 6797 | *bioc_ret = bioc; |
fd747f2d JT |
6798 | bioc->num_stripes = io_geom.num_stripes; |
6799 | bioc->max_errors = io_geom.max_errors; | |
6800 | bioc->mirror_num = io_geom.mirror_num; | |
ad6d620e | 6801 | |
cea9e445 | 6802 | out: |
73beece9 | 6803 | if (dev_replace_is_ongoing) { |
53176dde DS |
6804 | lockdep_assert_held(&dev_replace->rwsem); |
6805 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6806 | up_read(&dev_replace->rwsem); |
73beece9 | 6807 | } |
7dc66abb | 6808 | btrfs_free_chunk_map(map); |
de11cc12 | 6809 | return ret; |
0b86a832 CM |
6810 | } |
6811 | ||
562d7b15 JB |
6812 | static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, |
6813 | const struct btrfs_fs_devices *fs_devices) | |
6814 | { | |
6815 | if (args->fsid == NULL) | |
6816 | return true; | |
6817 | if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) | |
6818 | return true; | |
6819 | return false; | |
6820 | } | |
6821 | ||
6822 | static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, | |
6823 | const struct btrfs_device *device) | |
6824 | { | |
0fca385d LS |
6825 | if (args->missing) { |
6826 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && | |
6827 | !device->bdev) | |
6828 | return true; | |
6829 | return false; | |
6830 | } | |
562d7b15 | 6831 | |
0fca385d | 6832 | if (device->devid != args->devid) |
562d7b15 JB |
6833 | return false; |
6834 | if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) | |
6835 | return false; | |
0fca385d | 6836 | return true; |
562d7b15 JB |
6837 | } |
6838 | ||
09ba3bc9 AJ |
6839 | /* |
6840 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6841 | * return NULL. | |
6842 | * | |
6843 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6844 | * only devid is used. | |
09ba3bc9 | 6845 | */ |
562d7b15 JB |
6846 | struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, |
6847 | const struct btrfs_dev_lookup_args *args) | |
0b86a832 | 6848 | { |
2b82032c | 6849 | struct btrfs_device *device; |
944d3f9f NB |
6850 | struct btrfs_fs_devices *seed_devs; |
6851 | ||
562d7b15 | 6852 | if (dev_args_match_fs_devices(args, fs_devices)) { |
944d3f9f | 6853 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
562d7b15 | 6854 | if (dev_args_match_device(args, device)) |
944d3f9f NB |
6855 | return device; |
6856 | } | |
6857 | } | |
2b82032c | 6858 | |
944d3f9f | 6859 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
562d7b15 JB |
6860 | if (!dev_args_match_fs_devices(args, seed_devs)) |
6861 | continue; | |
6862 | list_for_each_entry(device, &seed_devs->devices, dev_list) { | |
6863 | if (dev_args_match_device(args, device)) | |
6864 | return device; | |
2b82032c | 6865 | } |
2b82032c | 6866 | } |
944d3f9f | 6867 | |
2b82032c | 6868 | return NULL; |
0b86a832 CM |
6869 | } |
6870 | ||
2ff7e61e | 6871 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6872 | u64 devid, u8 *dev_uuid) |
6873 | { | |
6874 | struct btrfs_device *device; | |
fccc0007 | 6875 | unsigned int nofs_flag; |
dfe25020 | 6876 | |
fccc0007 JB |
6877 | /* |
6878 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6879 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6880 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6881 | * places. | |
6882 | */ | |
bb21e302 | 6883 | |
fccc0007 | 6884 | nofs_flag = memalloc_nofs_save(); |
bb21e302 | 6885 | device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); |
fccc0007 | 6886 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6887 | if (IS_ERR(device)) |
adfb69af | 6888 | return device; |
12bd2fc0 ID |
6889 | |
6890 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6891 | device->fs_devices = fs_devices; |
dfe25020 | 6892 | fs_devices->num_devices++; |
12bd2fc0 | 6893 | |
e6e674bd | 6894 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6895 | fs_devices->missing_devices++; |
12bd2fc0 | 6896 | |
dfe25020 CM |
6897 | return device; |
6898 | } | |
6899 | ||
43dd529a DS |
6900 | /* |
6901 | * Allocate new device struct, set up devid and UUID. | |
6902 | * | |
12bd2fc0 ID |
6903 | * @fs_info: used only for generating a new devid, can be NULL if |
6904 | * devid is provided (i.e. @devid != NULL). | |
6905 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6906 | * is generated. | |
6907 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6908 | * is generated. | |
bb21e302 | 6909 | * @path: a pointer to device path if available, NULL otherwise. |
12bd2fc0 ID |
6910 | * |
6911 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6912 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6913 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6914 | */ |
6915 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
bb21e302 AJ |
6916 | const u64 *devid, const u8 *uuid, |
6917 | const char *path) | |
12bd2fc0 ID |
6918 | { |
6919 | struct btrfs_device *dev; | |
6920 | u64 tmp; | |
6921 | ||
fae7f21c | 6922 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6923 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6924 | |
fe4f46d4 DS |
6925 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
6926 | if (!dev) | |
6927 | return ERR_PTR(-ENOMEM); | |
6928 | ||
fe4f46d4 DS |
6929 | INIT_LIST_HEAD(&dev->dev_list); |
6930 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
6931 | INIT_LIST_HEAD(&dev->post_commit_list); | |
6932 | ||
fe4f46d4 DS |
6933 | atomic_set(&dev->dev_stats_ccnt, 0); |
6934 | btrfs_device_data_ordered_init(dev); | |
35da5a7e | 6935 | extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); |
12bd2fc0 ID |
6936 | |
6937 | if (devid) | |
6938 | tmp = *devid; | |
6939 | else { | |
6940 | int ret; | |
6941 | ||
6942 | ret = find_next_devid(fs_info, &tmp); | |
6943 | if (ret) { | |
a425f9d4 | 6944 | btrfs_free_device(dev); |
12bd2fc0 ID |
6945 | return ERR_PTR(ret); |
6946 | } | |
6947 | } | |
6948 | dev->devid = tmp; | |
6949 | ||
6950 | if (uuid) | |
6951 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6952 | else | |
6953 | generate_random_uuid(dev->uuid); | |
6954 | ||
bb21e302 AJ |
6955 | if (path) { |
6956 | struct rcu_string *name; | |
6957 | ||
6958 | name = rcu_string_strdup(path, GFP_KERNEL); | |
6959 | if (!name) { | |
6960 | btrfs_free_device(dev); | |
6961 | return ERR_PTR(-ENOMEM); | |
6962 | } | |
6963 | rcu_assign_pointer(dev->name, name); | |
6964 | } | |
6965 | ||
12bd2fc0 ID |
6966 | return dev; |
6967 | } | |
6968 | ||
5a2b8e60 | 6969 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6970 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6971 | { |
2b902dfc AJ |
6972 | if (error) |
6973 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6974 | devid, uuid); | |
6975 | else | |
6976 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6977 | devid, uuid); | |
5a2b8e60 AJ |
6978 | } |
6979 | ||
7dc66abb | 6980 | u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map) |
39e264a4 | 6981 | { |
bc88b486 | 6982 | const int data_stripes = calc_data_stripes(map->type, map->num_stripes); |
e4f6c6be | 6983 | |
7dc66abb | 6984 | return div_u64(map->chunk_len, data_stripes); |
39e264a4 NB |
6985 | } |
6986 | ||
e9306ad4 QW |
6987 | #if BITS_PER_LONG == 32 |
6988 | /* | |
6989 | * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE | |
6990 | * can't be accessed on 32bit systems. | |
6991 | * | |
6992 | * This function do mount time check to reject the fs if it already has | |
6993 | * metadata chunk beyond that limit. | |
6994 | */ | |
6995 | static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6996 | u64 logical, u64 length, u64 type) | |
6997 | { | |
6998 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6999 | return 0; | |
7000 | ||
7001 | if (logical + length < MAX_LFS_FILESIZE) | |
7002 | return 0; | |
7003 | ||
7004 | btrfs_err_32bit_limit(fs_info); | |
7005 | return -EOVERFLOW; | |
7006 | } | |
7007 | ||
7008 | /* | |
7009 | * This is to give early warning for any metadata chunk reaching | |
7010 | * BTRFS_32BIT_EARLY_WARN_THRESHOLD. | |
7011 | * Although we can still access the metadata, it's not going to be possible | |
7012 | * once the limit is reached. | |
7013 | */ | |
7014 | static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
7015 | u64 logical, u64 length, u64 type) | |
7016 | { | |
7017 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
7018 | return; | |
7019 | ||
7020 | if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
7021 | return; | |
7022 | ||
7023 | btrfs_warn_32bit_limit(fs_info); | |
7024 | } | |
7025 | #endif | |
7026 | ||
ff37c89f NB |
7027 | static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, |
7028 | u64 devid, u8 *uuid) | |
7029 | { | |
7030 | struct btrfs_device *dev; | |
7031 | ||
7032 | if (!btrfs_test_opt(fs_info, DEGRADED)) { | |
7033 | btrfs_report_missing_device(fs_info, devid, uuid, true); | |
7034 | return ERR_PTR(-ENOENT); | |
7035 | } | |
7036 | ||
7037 | dev = add_missing_dev(fs_info->fs_devices, devid, uuid); | |
7038 | if (IS_ERR(dev)) { | |
7039 | btrfs_err(fs_info, "failed to init missing device %llu: %ld", | |
7040 | devid, PTR_ERR(dev)); | |
7041 | return dev; | |
7042 | } | |
7043 | btrfs_report_missing_device(fs_info, devid, uuid, false); | |
7044 | ||
7045 | return dev; | |
7046 | } | |
7047 | ||
9690ac09 | 7048 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
7049 | struct btrfs_chunk *chunk) |
7050 | { | |
562d7b15 | 7051 | BTRFS_DEV_LOOKUP_ARGS(args); |
9690ac09 | 7052 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
7dc66abb | 7053 | struct btrfs_chunk_map *map; |
e06cd3dd LB |
7054 | u64 logical; |
7055 | u64 length; | |
e06cd3dd | 7056 | u64 devid; |
e9306ad4 | 7057 | u64 type; |
e06cd3dd | 7058 | u8 uuid[BTRFS_UUID_SIZE]; |
76a66ba1 | 7059 | int index; |
e06cd3dd LB |
7060 | int num_stripes; |
7061 | int ret; | |
7062 | int i; | |
7063 | ||
7064 | logical = key->offset; | |
7065 | length = btrfs_chunk_length(leaf, chunk); | |
e9306ad4 | 7066 | type = btrfs_chunk_type(leaf, chunk); |
76a66ba1 | 7067 | index = btrfs_bg_flags_to_raid_index(type); |
e06cd3dd LB |
7068 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
7069 | ||
e9306ad4 QW |
7070 | #if BITS_PER_LONG == 32 |
7071 | ret = check_32bit_meta_chunk(fs_info, logical, length, type); | |
7072 | if (ret < 0) | |
7073 | return ret; | |
7074 | warn_32bit_meta_chunk(fs_info, logical, length, type); | |
7075 | #endif | |
7076 | ||
075cb3c7 QW |
7077 | /* |
7078 | * Only need to verify chunk item if we're reading from sys chunk array, | |
7079 | * as chunk item in tree block is already verified by tree-checker. | |
7080 | */ | |
7081 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 7082 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
7083 | if (ret) |
7084 | return ret; | |
7085 | } | |
a061fc8d | 7086 | |
7dc66abb | 7087 | map = btrfs_find_chunk_map(fs_info, logical, 1); |
0b86a832 CM |
7088 | |
7089 | /* already mapped? */ | |
7dc66abb FM |
7090 | if (map && map->start <= logical && map->start + map->chunk_len > logical) { |
7091 | btrfs_free_chunk_map(map); | |
0b86a832 | 7092 | return 0; |
7dc66abb FM |
7093 | } else if (map) { |
7094 | btrfs_free_chunk_map(map); | |
0b86a832 | 7095 | } |
0b86a832 | 7096 | |
7dc66abb FM |
7097 | map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS); |
7098 | if (!map) | |
0b86a832 | 7099 | return -ENOMEM; |
0b86a832 | 7100 | |
7dc66abb FM |
7101 | map->start = logical; |
7102 | map->chunk_len = length; | |
593060d7 CM |
7103 | map->num_stripes = num_stripes; |
7104 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
7105 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
e9306ad4 | 7106 | map->type = type; |
76a66ba1 QW |
7107 | /* |
7108 | * We can't use the sub_stripes value, as for profiles other than | |
7109 | * RAID10, they may have 0 as sub_stripes for filesystems created by | |
7110 | * older mkfs (<v5.4). | |
7111 | * In that case, it can cause divide-by-zero errors later. | |
7112 | * Since currently sub_stripes is fixed for each profile, let's | |
7113 | * use the trusted value instead. | |
7114 | */ | |
7115 | map->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
cf90d884 | 7116 | map->verified_stripes = 0; |
7dc66abb | 7117 | map->stripe_size = btrfs_calc_stripe_length(map); |
593060d7 CM |
7118 | for (i = 0; i < num_stripes; i++) { |
7119 | map->stripes[i].physical = | |
7120 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
7121 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
562d7b15 | 7122 | args.devid = devid; |
a443755f CM |
7123 | read_extent_buffer(leaf, uuid, (unsigned long) |
7124 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
7125 | BTRFS_UUID_SIZE); | |
562d7b15 JB |
7126 | args.uuid = uuid; |
7127 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); | |
dfe25020 | 7128 | if (!map->stripes[i].dev) { |
ff37c89f NB |
7129 | map->stripes[i].dev = handle_missing_device(fs_info, |
7130 | devid, uuid); | |
adfb69af | 7131 | if (IS_ERR(map->stripes[i].dev)) { |
1742e1c9 | 7132 | ret = PTR_ERR(map->stripes[i].dev); |
7dc66abb | 7133 | btrfs_free_chunk_map(map); |
1742e1c9 | 7134 | return ret; |
dfe25020 CM |
7135 | } |
7136 | } | |
ff37c89f | 7137 | |
e12c9621 AJ |
7138 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
7139 | &(map->stripes[i].dev->dev_state)); | |
0b86a832 CM |
7140 | } |
7141 | ||
7dc66abb | 7142 | ret = btrfs_add_chunk_map(fs_info, map); |
64f64f43 QW |
7143 | if (ret < 0) { |
7144 | btrfs_err(fs_info, | |
7145 | "failed to add chunk map, start=%llu len=%llu: %d", | |
7dc66abb | 7146 | map->start, map->chunk_len, ret); |
64f64f43 | 7147 | } |
0b86a832 | 7148 | |
64f64f43 | 7149 | return ret; |
0b86a832 CM |
7150 | } |
7151 | ||
143bede5 | 7152 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
7153 | struct btrfs_dev_item *dev_item, |
7154 | struct btrfs_device *device) | |
7155 | { | |
7156 | unsigned long ptr; | |
0b86a832 CM |
7157 | |
7158 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
7159 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
7160 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 7161 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 7162 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 7163 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
7164 | device->type = btrfs_device_type(leaf, dev_item); |
7165 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
7166 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
7167 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 7168 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 7169 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 7170 | |
410ba3a2 | 7171 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 7172 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
7173 | } |
7174 | ||
2ff7e61e | 7175 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 7176 | u8 *fsid) |
2b82032c YZ |
7177 | { |
7178 | struct btrfs_fs_devices *fs_devices; | |
7179 | int ret; | |
7180 | ||
a32bf9a3 | 7181 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 7182 | ASSERT(fsid); |
2b82032c | 7183 | |
427c8fdd | 7184 | /* This will match only for multi-device seed fs */ |
944d3f9f | 7185 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 7186 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
7187 | return fs_devices; |
7188 | ||
2b82032c | 7189 | |
7239ff4b | 7190 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 7191 | if (!fs_devices) { |
0b246afa | 7192 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
7193 | return ERR_PTR(-ENOENT); |
7194 | ||
f7361d8c | 7195 | fs_devices = alloc_fs_devices(fsid); |
5f375835 MX |
7196 | if (IS_ERR(fs_devices)) |
7197 | return fs_devices; | |
7198 | ||
0395d84f | 7199 | fs_devices->seeding = true; |
5f375835 MX |
7200 | fs_devices->opened = 1; |
7201 | return fs_devices; | |
2b82032c | 7202 | } |
e4404d6e | 7203 | |
427c8fdd NB |
7204 | /* |
7205 | * Upon first call for a seed fs fsid, just create a private copy of the | |
7206 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
7207 | */ | |
e4404d6e | 7208 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
7209 | if (IS_ERR(fs_devices)) |
7210 | return fs_devices; | |
2b82032c | 7211 | |
05bdb996 | 7212 | ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder); |
48d28232 JL |
7213 | if (ret) { |
7214 | free_fs_devices(fs_devices); | |
c83b60c0 | 7215 | return ERR_PTR(ret); |
48d28232 | 7216 | } |
2b82032c YZ |
7217 | |
7218 | if (!fs_devices->seeding) { | |
0226e0eb | 7219 | close_fs_devices(fs_devices); |
e4404d6e | 7220 | free_fs_devices(fs_devices); |
c83b60c0 | 7221 | return ERR_PTR(-EINVAL); |
2b82032c YZ |
7222 | } |
7223 | ||
944d3f9f | 7224 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
c83b60c0 | 7225 | |
5f375835 | 7226 | return fs_devices; |
2b82032c YZ |
7227 | } |
7228 | ||
17850759 | 7229 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
7230 | struct btrfs_dev_item *dev_item) |
7231 | { | |
562d7b15 | 7232 | BTRFS_DEV_LOOKUP_ARGS(args); |
17850759 | 7233 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 7234 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
7235 | struct btrfs_device *device; |
7236 | u64 devid; | |
7237 | int ret; | |
44880fdc | 7238 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
7239 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
7240 | ||
c1867eb3 DS |
7241 | devid = btrfs_device_id(leaf, dev_item); |
7242 | args.devid = devid; | |
410ba3a2 | 7243 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 7244 | BTRFS_UUID_SIZE); |
1473b24e | 7245 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 7246 | BTRFS_FSID_SIZE); |
562d7b15 JB |
7247 | args.uuid = dev_uuid; |
7248 | args.fsid = fs_uuid; | |
2b82032c | 7249 | |
de37aa51 | 7250 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 7251 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
7252 | if (IS_ERR(fs_devices)) |
7253 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
7254 | } |
7255 | ||
562d7b15 | 7256 | device = btrfs_find_device(fs_info->fs_devices, &args); |
5f375835 | 7257 | if (!device) { |
c5502451 | 7258 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
7259 | btrfs_report_missing_device(fs_info, devid, |
7260 | dev_uuid, true); | |
45dbdbc9 | 7261 | return -ENOENT; |
c5502451 | 7262 | } |
2b82032c | 7263 | |
2ff7e61e | 7264 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
7265 | if (IS_ERR(device)) { |
7266 | btrfs_err(fs_info, | |
7267 | "failed to add missing dev %llu: %ld", | |
7268 | devid, PTR_ERR(device)); | |
7269 | return PTR_ERR(device); | |
7270 | } | |
2b902dfc | 7271 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 7272 | } else { |
c5502451 | 7273 | if (!device->bdev) { |
2b902dfc AJ |
7274 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
7275 | btrfs_report_missing_device(fs_info, | |
7276 | devid, dev_uuid, true); | |
45dbdbc9 | 7277 | return -ENOENT; |
2b902dfc AJ |
7278 | } |
7279 | btrfs_report_missing_device(fs_info, devid, | |
7280 | dev_uuid, false); | |
c5502451 | 7281 | } |
5f375835 | 7282 | |
e6e674bd AJ |
7283 | if (!device->bdev && |
7284 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7285 | /* |
7286 | * this happens when a device that was properly setup | |
7287 | * in the device info lists suddenly goes bad. | |
7288 | * device->bdev is NULL, and so we have to set | |
7289 | * device->missing to one here | |
7290 | */ | |
5f375835 | 7291 | device->fs_devices->missing_devices++; |
e6e674bd | 7292 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7293 | } |
5f375835 MX |
7294 | |
7295 | /* Move the device to its own fs_devices */ | |
7296 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7297 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7298 | &device->dev_state)); | |
5f375835 MX |
7299 | |
7300 | list_move(&device->dev_list, &fs_devices->devices); | |
7301 | device->fs_devices->num_devices--; | |
7302 | fs_devices->num_devices++; | |
7303 | ||
7304 | device->fs_devices->missing_devices--; | |
7305 | fs_devices->missing_devices++; | |
7306 | ||
7307 | device->fs_devices = fs_devices; | |
7308 | } | |
2b82032c YZ |
7309 | } |
7310 | ||
0b246afa | 7311 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7312 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7313 | if (device->generation != |
7314 | btrfs_device_generation(leaf, dev_item)) | |
7315 | return -EINVAL; | |
6324fbf3 | 7316 | } |
0b86a832 CM |
7317 | |
7318 | fill_device_from_item(leaf, dev_item, device); | |
3a160a93 | 7319 | if (device->bdev) { |
cda00eba | 7320 | u64 max_total_bytes = bdev_nr_bytes(device->bdev); |
3a160a93 AJ |
7321 | |
7322 | if (device->total_bytes > max_total_bytes) { | |
7323 | btrfs_err(fs_info, | |
7324 | "device total_bytes should be at most %llu but found %llu", | |
7325 | max_total_bytes, device->total_bytes); | |
7326 | return -EINVAL; | |
7327 | } | |
7328 | } | |
e12c9621 | 7329 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7330 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7331 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7332 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7333 | atomic64_add(device->total_bytes - device->bytes_used, |
7334 | &fs_info->free_chunk_space); | |
2bf64758 | 7335 | } |
0b86a832 | 7336 | ret = 0; |
0b86a832 CM |
7337 | return ret; |
7338 | } | |
7339 | ||
6bccf3ab | 7340 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7341 | { |
ab8d0fc4 | 7342 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7343 | struct extent_buffer *sb; |
0b86a832 | 7344 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7345 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7346 | u8 *array_ptr; |
7347 | unsigned long sb_array_offset; | |
84eed90f | 7348 | int ret = 0; |
0b86a832 CM |
7349 | u32 num_stripes; |
7350 | u32 array_size; | |
7351 | u32 len = 0; | |
1ffb22cf | 7352 | u32 cur_offset; |
e06cd3dd | 7353 | u64 type; |
84eed90f | 7354 | struct btrfs_key key; |
0b86a832 | 7355 | |
0b246afa | 7356 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
e959d3c1 | 7357 | |
a83fffb7 | 7358 | /* |
e959d3c1 QW |
7359 | * We allocated a dummy extent, just to use extent buffer accessors. |
7360 | * There will be unused space after BTRFS_SUPER_INFO_SIZE, but | |
7361 | * that's fine, we will not go beyond system chunk array anyway. | |
a83fffb7 | 7362 | */ |
e959d3c1 QW |
7363 | sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); |
7364 | if (!sb) | |
7365 | return -ENOMEM; | |
4db8c528 | 7366 | set_extent_buffer_uptodate(sb); |
4008c04a | 7367 | |
a061fc8d | 7368 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7369 | array_size = btrfs_super_sys_array_size(super_copy); |
7370 | ||
1ffb22cf DS |
7371 | array_ptr = super_copy->sys_chunk_array; |
7372 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7373 | cur_offset = 0; | |
0b86a832 | 7374 | |
1ffb22cf DS |
7375 | while (cur_offset < array_size) { |
7376 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7377 | len = sizeof(*disk_key); |
7378 | if (cur_offset + len > array_size) | |
7379 | goto out_short_read; | |
7380 | ||
0b86a832 CM |
7381 | btrfs_disk_key_to_cpu(&key, disk_key); |
7382 | ||
1ffb22cf DS |
7383 | array_ptr += len; |
7384 | sb_array_offset += len; | |
7385 | cur_offset += len; | |
0b86a832 | 7386 | |
32ab3d1b JT |
7387 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
7388 | btrfs_err(fs_info, | |
7389 | "unexpected item type %u in sys_array at offset %u", | |
7390 | (u32)key.type, cur_offset); | |
7391 | ret = -EIO; | |
7392 | break; | |
7393 | } | |
f5cdedd7 | 7394 | |
32ab3d1b JT |
7395 | chunk = (struct btrfs_chunk *)sb_array_offset; |
7396 | /* | |
7397 | * At least one btrfs_chunk with one stripe must be present, | |
7398 | * exact stripe count check comes afterwards | |
7399 | */ | |
7400 | len = btrfs_chunk_item_size(1); | |
7401 | if (cur_offset + len > array_size) | |
7402 | goto out_short_read; | |
e06cd3dd | 7403 | |
32ab3d1b JT |
7404 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
7405 | if (!num_stripes) { | |
7406 | btrfs_err(fs_info, | |
7407 | "invalid number of stripes %u in sys_array at offset %u", | |
7408 | num_stripes, cur_offset); | |
7409 | ret = -EIO; | |
7410 | break; | |
7411 | } | |
e3540eab | 7412 | |
32ab3d1b JT |
7413 | type = btrfs_chunk_type(sb, chunk); |
7414 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7415 | btrfs_err(fs_info, |
32ab3d1b JT |
7416 | "invalid chunk type %llu in sys_array at offset %u", |
7417 | type, cur_offset); | |
84eed90f CM |
7418 | ret = -EIO; |
7419 | break; | |
0b86a832 | 7420 | } |
32ab3d1b JT |
7421 | |
7422 | len = btrfs_chunk_item_size(num_stripes); | |
7423 | if (cur_offset + len > array_size) | |
7424 | goto out_short_read; | |
7425 | ||
7426 | ret = read_one_chunk(&key, sb, chunk); | |
7427 | if (ret) | |
7428 | break; | |
7429 | ||
1ffb22cf DS |
7430 | array_ptr += len; |
7431 | sb_array_offset += len; | |
7432 | cur_offset += len; | |
0b86a832 | 7433 | } |
d865177a | 7434 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7435 | free_extent_buffer_stale(sb); |
84eed90f | 7436 | return ret; |
e3540eab DS |
7437 | |
7438 | out_short_read: | |
ab8d0fc4 | 7439 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7440 | len, cur_offset); |
d865177a | 7441 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7442 | free_extent_buffer_stale(sb); |
e3540eab | 7443 | return -EIO; |
0b86a832 CM |
7444 | } |
7445 | ||
21634a19 QW |
7446 | /* |
7447 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7448 | * | |
6528b99d AJ |
7449 | * If the @failing_dev is specified, it's accounted as missing. |
7450 | * | |
21634a19 QW |
7451 | * Return true if all chunks meet the minimal RW mount requirements. |
7452 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7453 | */ | |
6528b99d AJ |
7454 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7455 | struct btrfs_device *failing_dev) | |
21634a19 | 7456 | { |
7dc66abb FM |
7457 | struct btrfs_chunk_map *map; |
7458 | u64 next_start; | |
21634a19 QW |
7459 | bool ret = true; |
7460 | ||
7dc66abb | 7461 | map = btrfs_find_chunk_map(fs_info, 0, U64_MAX); |
21634a19 | 7462 | /* No chunk at all? Return false anyway */ |
7dc66abb | 7463 | if (!map) { |
21634a19 QW |
7464 | ret = false; |
7465 | goto out; | |
7466 | } | |
7dc66abb | 7467 | while (map) { |
21634a19 QW |
7468 | int missing = 0; |
7469 | int max_tolerated; | |
7470 | int i; | |
7471 | ||
21634a19 QW |
7472 | max_tolerated = |
7473 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7474 | map->type); | |
7475 | for (i = 0; i < map->num_stripes; i++) { | |
7476 | struct btrfs_device *dev = map->stripes[i].dev; | |
7477 | ||
e6e674bd AJ |
7478 | if (!dev || !dev->bdev || |
7479 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7480 | dev->last_flush_error) |
7481 | missing++; | |
6528b99d AJ |
7482 | else if (failing_dev && failing_dev == dev) |
7483 | missing++; | |
21634a19 QW |
7484 | } |
7485 | if (missing > max_tolerated) { | |
6528b99d AJ |
7486 | if (!failing_dev) |
7487 | btrfs_warn(fs_info, | |
52042d8e | 7488 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
7dc66abb FM |
7489 | map->start, missing, max_tolerated); |
7490 | btrfs_free_chunk_map(map); | |
21634a19 QW |
7491 | ret = false; |
7492 | goto out; | |
7493 | } | |
7dc66abb FM |
7494 | next_start = map->start + map->chunk_len; |
7495 | btrfs_free_chunk_map(map); | |
21634a19 | 7496 | |
7dc66abb | 7497 | map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start); |
21634a19 QW |
7498 | } |
7499 | out: | |
7500 | return ret; | |
7501 | } | |
7502 | ||
d85327b1 DS |
7503 | static void readahead_tree_node_children(struct extent_buffer *node) |
7504 | { | |
7505 | int i; | |
7506 | const int nr_items = btrfs_header_nritems(node); | |
7507 | ||
bfb484d9 JB |
7508 | for (i = 0; i < nr_items; i++) |
7509 | btrfs_readahead_node_child(node, i); | |
d85327b1 DS |
7510 | } |
7511 | ||
5b4aacef | 7512 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7513 | { |
5b4aacef | 7514 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7515 | struct btrfs_path *path; |
7516 | struct extent_buffer *leaf; | |
7517 | struct btrfs_key key; | |
7518 | struct btrfs_key found_key; | |
7519 | int ret; | |
7520 | int slot; | |
43cb1478 | 7521 | int iter_ret = 0; |
99e3ecfc | 7522 | u64 total_dev = 0; |
d85327b1 | 7523 | u64 last_ra_node = 0; |
0b86a832 | 7524 | |
0b86a832 CM |
7525 | path = btrfs_alloc_path(); |
7526 | if (!path) | |
7527 | return -ENOMEM; | |
7528 | ||
3dd0f7a3 AJ |
7529 | /* |
7530 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7531 | * otherwise we don't need it. | |
7532 | */ | |
b367e47f | 7533 | mutex_lock(&uuid_mutex); |
b367e47f | 7534 | |
48cfa61b BB |
7535 | /* |
7536 | * It is possible for mount and umount to race in such a way that | |
7537 | * we execute this code path, but open_fs_devices failed to clear | |
7538 | * total_rw_bytes. We certainly want it cleared before reading the | |
7539 | * device items, so clear it here. | |
7540 | */ | |
7541 | fs_info->fs_devices->total_rw_bytes = 0; | |
7542 | ||
4d9380e0 FM |
7543 | /* |
7544 | * Lockdep complains about possible circular locking dependency between | |
7545 | * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores | |
7546 | * used for freeze procection of a fs (struct super_block.s_writers), | |
7547 | * which we take when starting a transaction, and extent buffers of the | |
7548 | * chunk tree if we call read_one_dev() while holding a lock on an | |
7549 | * extent buffer of the chunk tree. Since we are mounting the filesystem | |
7550 | * and at this point there can't be any concurrent task modifying the | |
7551 | * chunk tree, to keep it simple, just skip locking on the chunk tree. | |
7552 | */ | |
7553 | ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); | |
7554 | path->skip_locking = 1; | |
7555 | ||
395927a9 FDBM |
7556 | /* |
7557 | * Read all device items, and then all the chunk items. All | |
7558 | * device items are found before any chunk item (their object id | |
7559 | * is smaller than the lowest possible object id for a chunk | |
7560 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7561 | */ |
7562 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7563 | key.offset = 0; | |
7564 | key.type = 0; | |
43cb1478 GN |
7565 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
7566 | struct extent_buffer *node = path->nodes[1]; | |
d85327b1 | 7567 | |
0b86a832 CM |
7568 | leaf = path->nodes[0]; |
7569 | slot = path->slots[0]; | |
43cb1478 | 7570 | |
d85327b1 DS |
7571 | if (node) { |
7572 | if (last_ra_node != node->start) { | |
7573 | readahead_tree_node_children(node); | |
7574 | last_ra_node = node->start; | |
7575 | } | |
7576 | } | |
395927a9 FDBM |
7577 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7578 | struct btrfs_dev_item *dev_item; | |
7579 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7580 | struct btrfs_dev_item); |
17850759 | 7581 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7582 | if (ret) |
7583 | goto error; | |
99e3ecfc | 7584 | total_dev++; |
0b86a832 CM |
7585 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7586 | struct btrfs_chunk *chunk; | |
79bd3712 FM |
7587 | |
7588 | /* | |
7589 | * We are only called at mount time, so no need to take | |
7590 | * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, | |
7591 | * we always lock first fs_info->chunk_mutex before | |
7592 | * acquiring any locks on the chunk tree. This is a | |
7593 | * requirement for chunk allocation, see the comment on | |
7594 | * top of btrfs_chunk_alloc() for details. | |
7595 | */ | |
0b86a832 | 7596 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
9690ac09 | 7597 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7598 | if (ret) |
7599 | goto error; | |
0b86a832 | 7600 | } |
43cb1478 GN |
7601 | } |
7602 | /* Catch error found during iteration */ | |
7603 | if (iter_ret < 0) { | |
7604 | ret = iter_ret; | |
7605 | goto error; | |
0b86a832 | 7606 | } |
99e3ecfc LB |
7607 | |
7608 | /* | |
7609 | * After loading chunk tree, we've got all device information, | |
7610 | * do another round of validation checks. | |
7611 | */ | |
0b246afa | 7612 | if (total_dev != fs_info->fs_devices->total_devices) { |
d201238c QW |
7613 | btrfs_warn(fs_info, |
7614 | "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", | |
0b246afa | 7615 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc | 7616 | total_dev); |
d201238c QW |
7617 | fs_info->fs_devices->total_devices = total_dev; |
7618 | btrfs_set_super_num_devices(fs_info->super_copy, total_dev); | |
99e3ecfc | 7619 | } |
0b246afa JM |
7620 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7621 | fs_info->fs_devices->total_rw_bytes) { | |
7622 | btrfs_err(fs_info, | |
99e3ecfc | 7623 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7624 | btrfs_super_total_bytes(fs_info->super_copy), |
7625 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7626 | ret = -EINVAL; |
7627 | goto error; | |
7628 | } | |
0b86a832 CM |
7629 | ret = 0; |
7630 | error: | |
b367e47f LZ |
7631 | mutex_unlock(&uuid_mutex); |
7632 | ||
2b82032c | 7633 | btrfs_free_path(path); |
0b86a832 CM |
7634 | return ret; |
7635 | } | |
442a4f63 | 7636 | |
a8d1b164 | 7637 | int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
cb517eab | 7638 | { |
944d3f9f | 7639 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab | 7640 | struct btrfs_device *device; |
a8d1b164 | 7641 | int ret = 0; |
cb517eab | 7642 | |
944d3f9f NB |
7643 | fs_devices->fs_info = fs_info; |
7644 | ||
7645 | mutex_lock(&fs_devices->device_list_mutex); | |
7646 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7647 | device->fs_info = fs_info; | |
944d3f9f NB |
7648 | |
7649 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
a8d1b164 | 7650 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
fb456252 | 7651 | device->fs_info = fs_info; |
a8d1b164 JT |
7652 | ret = btrfs_get_dev_zone_info(device, false); |
7653 | if (ret) | |
7654 | break; | |
7655 | } | |
29cc83f6 | 7656 | |
944d3f9f | 7657 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7658 | } |
e17125b5 | 7659 | mutex_unlock(&fs_devices->device_list_mutex); |
a8d1b164 JT |
7660 | |
7661 | return ret; | |
cb517eab MX |
7662 | } |
7663 | ||
1dc990df DS |
7664 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7665 | const struct btrfs_dev_stats_item *ptr, | |
7666 | int index) | |
7667 | { | |
7668 | u64 val; | |
7669 | ||
7670 | read_extent_buffer(eb, &val, | |
7671 | offsetof(struct btrfs_dev_stats_item, values) + | |
7672 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7673 | sizeof(val)); | |
7674 | return val; | |
7675 | } | |
7676 | ||
7677 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7678 | struct btrfs_dev_stats_item *ptr, | |
7679 | int index, u64 val) | |
7680 | { | |
7681 | write_extent_buffer(eb, &val, | |
7682 | offsetof(struct btrfs_dev_stats_item, values) + | |
7683 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7684 | sizeof(val)); | |
7685 | } | |
7686 | ||
92e26df4 JB |
7687 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
7688 | struct btrfs_path *path) | |
733f4fbb | 7689 | { |
124604eb | 7690 | struct btrfs_dev_stats_item *ptr; |
733f4fbb | 7691 | struct extent_buffer *eb; |
124604eb JB |
7692 | struct btrfs_key key; |
7693 | int item_size; | |
7694 | int i, ret, slot; | |
7695 | ||
82d62d06 JB |
7696 | if (!device->fs_info->dev_root) |
7697 | return 0; | |
7698 | ||
124604eb JB |
7699 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7700 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
7701 | key.offset = device->devid; | |
7702 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); | |
7703 | if (ret) { | |
7704 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7705 | btrfs_dev_stat_set(device, i, 0); | |
7706 | device->dev_stats_valid = 1; | |
7707 | btrfs_release_path(path); | |
92e26df4 | 7708 | return ret < 0 ? ret : 0; |
124604eb JB |
7709 | } |
7710 | slot = path->slots[0]; | |
7711 | eb = path->nodes[0]; | |
3212fa14 | 7712 | item_size = btrfs_item_size(eb, slot); |
124604eb JB |
7713 | |
7714 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); | |
7715 | ||
7716 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7717 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7718 | btrfs_dev_stat_set(device, i, | |
7719 | btrfs_dev_stats_value(eb, ptr, i)); | |
7720 | else | |
7721 | btrfs_dev_stat_set(device, i, 0); | |
7722 | } | |
7723 | ||
7724 | device->dev_stats_valid = 1; | |
7725 | btrfs_dev_stat_print_on_load(device); | |
7726 | btrfs_release_path(path); | |
92e26df4 JB |
7727 | |
7728 | return 0; | |
124604eb JB |
7729 | } |
7730 | ||
7731 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7732 | { | |
7733 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
733f4fbb SB |
7734 | struct btrfs_device *device; |
7735 | struct btrfs_path *path = NULL; | |
92e26df4 | 7736 | int ret = 0; |
733f4fbb SB |
7737 | |
7738 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7739 | if (!path) |
7740 | return -ENOMEM; | |
733f4fbb SB |
7741 | |
7742 | mutex_lock(&fs_devices->device_list_mutex); | |
92e26df4 JB |
7743 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
7744 | ret = btrfs_device_init_dev_stats(device, path); | |
7745 | if (ret) | |
7746 | goto out; | |
7747 | } | |
124604eb | 7748 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
92e26df4 JB |
7749 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
7750 | ret = btrfs_device_init_dev_stats(device, path); | |
7751 | if (ret) | |
7752 | goto out; | |
7753 | } | |
733f4fbb | 7754 | } |
92e26df4 | 7755 | out: |
733f4fbb SB |
7756 | mutex_unlock(&fs_devices->device_list_mutex); |
7757 | ||
733f4fbb | 7758 | btrfs_free_path(path); |
92e26df4 | 7759 | return ret; |
733f4fbb SB |
7760 | } |
7761 | ||
7762 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7763 | struct btrfs_device *device) |
7764 | { | |
5495f195 | 7765 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7766 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7767 | struct btrfs_path *path; |
7768 | struct btrfs_key key; | |
7769 | struct extent_buffer *eb; | |
7770 | struct btrfs_dev_stats_item *ptr; | |
7771 | int ret; | |
7772 | int i; | |
7773 | ||
242e2956 DS |
7774 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7775 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7776 | key.offset = device->devid; |
7777 | ||
7778 | path = btrfs_alloc_path(); | |
fa252992 DS |
7779 | if (!path) |
7780 | return -ENOMEM; | |
733f4fbb SB |
7781 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7782 | if (ret < 0) { | |
0b246afa | 7783 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7784 | "error %d while searching for dev_stats item for device %s", |
cb3e217b | 7785 | ret, btrfs_dev_name(device)); |
733f4fbb SB |
7786 | goto out; |
7787 | } | |
7788 | ||
7789 | if (ret == 0 && | |
3212fa14 | 7790 | btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { |
733f4fbb SB |
7791 | /* need to delete old one and insert a new one */ |
7792 | ret = btrfs_del_item(trans, dev_root, path); | |
7793 | if (ret != 0) { | |
0b246afa | 7794 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7795 | "delete too small dev_stats item for device %s failed %d", |
cb3e217b | 7796 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7797 | goto out; |
7798 | } | |
7799 | ret = 1; | |
7800 | } | |
7801 | ||
7802 | if (ret == 1) { | |
7803 | /* need to insert a new item */ | |
7804 | btrfs_release_path(path); | |
7805 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7806 | &key, sizeof(*ptr)); | |
7807 | if (ret < 0) { | |
0b246afa | 7808 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7809 | "insert dev_stats item for device %s failed %d", |
cb3e217b | 7810 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7811 | goto out; |
7812 | } | |
7813 | } | |
7814 | ||
7815 | eb = path->nodes[0]; | |
7816 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7817 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7818 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7819 | btrfs_dev_stat_read(device, i)); | |
50564b65 | 7820 | btrfs_mark_buffer_dirty(trans, eb); |
733f4fbb SB |
7821 | |
7822 | out: | |
7823 | btrfs_free_path(path); | |
7824 | return ret; | |
7825 | } | |
7826 | ||
7827 | /* | |
7828 | * called from commit_transaction. Writes all changed device stats to disk. | |
7829 | */ | |
196c9d8d | 7830 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7831 | { |
196c9d8d | 7832 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7833 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7834 | struct btrfs_device *device; | |
addc3fa7 | 7835 | int stats_cnt; |
733f4fbb SB |
7836 | int ret = 0; |
7837 | ||
7838 | mutex_lock(&fs_devices->device_list_mutex); | |
7839 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7840 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7841 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7842 | continue; |
7843 | ||
9deae968 NB |
7844 | |
7845 | /* | |
7846 | * There is a LOAD-LOAD control dependency between the value of | |
7847 | * dev_stats_ccnt and updating the on-disk values which requires | |
7848 | * reading the in-memory counters. Such control dependencies | |
7849 | * require explicit read memory barriers. | |
7850 | * | |
7851 | * This memory barriers pairs with smp_mb__before_atomic in | |
7852 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7853 | * barrier implied by atomic_xchg in | |
7854 | * btrfs_dev_stats_read_and_reset | |
7855 | */ | |
7856 | smp_rmb(); | |
7857 | ||
5495f195 | 7858 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7859 | if (!ret) |
addc3fa7 | 7860 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7861 | } |
7862 | mutex_unlock(&fs_devices->device_list_mutex); | |
7863 | ||
7864 | return ret; | |
7865 | } | |
7866 | ||
442a4f63 SB |
7867 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7868 | { | |
7869 | btrfs_dev_stat_inc(dev, index); | |
442a4f63 | 7870 | |
733f4fbb SB |
7871 | if (!dev->dev_stats_valid) |
7872 | return; | |
fb456252 | 7873 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7874 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7875 | btrfs_dev_name(dev), |
442a4f63 SB |
7876 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7877 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7878 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7879 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7880 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7881 | } |
c11d2c23 | 7882 | |
733f4fbb SB |
7883 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7884 | { | |
a98cdb85 SB |
7885 | int i; |
7886 | ||
7887 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7888 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7889 | break; | |
7890 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7891 | return; /* all values == 0, suppress message */ | |
7892 | ||
fb456252 | 7893 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7894 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7895 | btrfs_dev_name(dev), |
733f4fbb SB |
7896 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7897 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7898 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7899 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7900 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7901 | } | |
7902 | ||
2ff7e61e | 7903 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7904 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 | 7905 | { |
562d7b15 | 7906 | BTRFS_DEV_LOOKUP_ARGS(args); |
c11d2c23 | 7907 | struct btrfs_device *dev; |
0b246afa | 7908 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7909 | int i; |
7910 | ||
7911 | mutex_lock(&fs_devices->device_list_mutex); | |
562d7b15 JB |
7912 | args.devid = stats->devid; |
7913 | dev = btrfs_find_device(fs_info->fs_devices, &args); | |
c11d2c23 SB |
7914 | mutex_unlock(&fs_devices->device_list_mutex); |
7915 | ||
7916 | if (!dev) { | |
0b246afa | 7917 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7918 | return -ENODEV; |
733f4fbb | 7919 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7920 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7921 | return -ENODEV; |
b27f7c0c | 7922 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7923 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7924 | if (stats->nr_items > i) | |
7925 | stats->values[i] = | |
7926 | btrfs_dev_stat_read_and_reset(dev, i); | |
7927 | else | |
4e411a7d | 7928 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7929 | } |
a69976bc AJ |
7930 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7931 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7932 | } else { |
7933 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7934 | if (stats->nr_items > i) | |
7935 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7936 | } | |
7937 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7938 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7939 | return 0; | |
7940 | } | |
a8a6dab7 | 7941 | |
935e5cc9 | 7942 | /* |
bbbf7243 NB |
7943 | * Update the size and bytes used for each device where it changed. This is |
7944 | * delayed since we would otherwise get errors while writing out the | |
7945 | * superblocks. | |
7946 | * | |
7947 | * Must be invoked during transaction commit. | |
935e5cc9 | 7948 | */ |
bbbf7243 | 7949 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7950 | { |
935e5cc9 MX |
7951 | struct btrfs_device *curr, *next; |
7952 | ||
bbbf7243 | 7953 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7954 | |
bbbf7243 | 7955 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7956 | return; |
7957 | ||
bbbf7243 NB |
7958 | /* |
7959 | * We don't need the device_list_mutex here. This list is owned by the | |
7960 | * transaction and the transaction must complete before the device is | |
7961 | * released. | |
7962 | */ | |
7963 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7964 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7965 | post_commit_list) { | |
7966 | list_del_init(&curr->post_commit_list); | |
7967 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7968 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7969 | } |
bbbf7243 | 7970 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7971 | } |
5a13f430 | 7972 | |
46df06b8 DS |
7973 | /* |
7974 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7975 | */ | |
7976 | int btrfs_bg_type_to_factor(u64 flags) | |
7977 | { | |
44b28ada DS |
7978 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7979 | ||
7980 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7981 | } |
cf90d884 QW |
7982 | |
7983 | ||
cf90d884 QW |
7984 | |
7985 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7986 | u64 chunk_offset, u64 devid, | |
7987 | u64 physical_offset, u64 physical_len) | |
7988 | { | |
562d7b15 | 7989 | struct btrfs_dev_lookup_args args = { .devid = devid }; |
7dc66abb | 7990 | struct btrfs_chunk_map *map; |
05a37c48 | 7991 | struct btrfs_device *dev; |
cf90d884 QW |
7992 | u64 stripe_len; |
7993 | bool found = false; | |
7994 | int ret = 0; | |
7995 | int i; | |
7996 | ||
7dc66abb FM |
7997 | map = btrfs_find_chunk_map(fs_info, chunk_offset, 1); |
7998 | if (!map) { | |
cf90d884 QW |
7999 | btrfs_err(fs_info, |
8000 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
8001 | physical_offset, devid); | |
8002 | ret = -EUCLEAN; | |
8003 | goto out; | |
8004 | } | |
8005 | ||
7dc66abb | 8006 | stripe_len = btrfs_calc_stripe_length(map); |
cf90d884 QW |
8007 | if (physical_len != stripe_len) { |
8008 | btrfs_err(fs_info, | |
8009 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7dc66abb | 8010 | physical_offset, devid, map->start, physical_len, |
cf90d884 QW |
8011 | stripe_len); |
8012 | ret = -EUCLEAN; | |
8013 | goto out; | |
8014 | } | |
8015 | ||
3613249a QW |
8016 | /* |
8017 | * Very old mkfs.btrfs (before v4.1) will not respect the reserved | |
8018 | * space. Although kernel can handle it without problem, better to warn | |
8019 | * the users. | |
8020 | */ | |
8021 | if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) | |
8022 | btrfs_warn(fs_info, | |
8023 | "devid %llu physical %llu len %llu inside the reserved space", | |
8024 | devid, physical_offset, physical_len); | |
8025 | ||
cf90d884 QW |
8026 | for (i = 0; i < map->num_stripes; i++) { |
8027 | if (map->stripes[i].dev->devid == devid && | |
8028 | map->stripes[i].physical == physical_offset) { | |
8029 | found = true; | |
8030 | if (map->verified_stripes >= map->num_stripes) { | |
8031 | btrfs_err(fs_info, | |
8032 | "too many dev extents for chunk %llu found", | |
7dc66abb | 8033 | map->start); |
cf90d884 QW |
8034 | ret = -EUCLEAN; |
8035 | goto out; | |
8036 | } | |
8037 | map->verified_stripes++; | |
8038 | break; | |
8039 | } | |
8040 | } | |
8041 | if (!found) { | |
8042 | btrfs_err(fs_info, | |
8043 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
8044 | physical_offset, devid); | |
8045 | ret = -EUCLEAN; | |
8046 | } | |
05a37c48 | 8047 | |
1a9fd417 | 8048 | /* Make sure no dev extent is beyond device boundary */ |
562d7b15 | 8049 | dev = btrfs_find_device(fs_info->fs_devices, &args); |
05a37c48 QW |
8050 | if (!dev) { |
8051 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
8052 | ret = -EUCLEAN; | |
8053 | goto out; | |
8054 | } | |
1b3922a8 | 8055 | |
05a37c48 QW |
8056 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
8057 | btrfs_err(fs_info, | |
8058 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
8059 | devid, physical_offset, physical_len, | |
8060 | dev->disk_total_bytes); | |
8061 | ret = -EUCLEAN; | |
8062 | goto out; | |
8063 | } | |
381a696e NA |
8064 | |
8065 | if (dev->zone_info) { | |
8066 | u64 zone_size = dev->zone_info->zone_size; | |
8067 | ||
8068 | if (!IS_ALIGNED(physical_offset, zone_size) || | |
8069 | !IS_ALIGNED(physical_len, zone_size)) { | |
8070 | btrfs_err(fs_info, | |
8071 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", | |
8072 | devid, physical_offset, physical_len); | |
8073 | ret = -EUCLEAN; | |
8074 | goto out; | |
8075 | } | |
8076 | } | |
8077 | ||
cf90d884 | 8078 | out: |
7dc66abb | 8079 | btrfs_free_chunk_map(map); |
cf90d884 QW |
8080 | return ret; |
8081 | } | |
8082 | ||
8083 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
8084 | { | |
cf90d884 QW |
8085 | struct rb_node *node; |
8086 | int ret = 0; | |
8087 | ||
7dc66abb FM |
8088 | read_lock(&fs_info->mapping_tree_lock); |
8089 | for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) { | |
8090 | struct btrfs_chunk_map *map; | |
8091 | ||
8092 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); | |
8093 | if (map->num_stripes != map->verified_stripes) { | |
cf90d884 QW |
8094 | btrfs_err(fs_info, |
8095 | "chunk %llu has missing dev extent, have %d expect %d", | |
7dc66abb | 8096 | map->start, map->verified_stripes, map->num_stripes); |
cf90d884 QW |
8097 | ret = -EUCLEAN; |
8098 | goto out; | |
8099 | } | |
8100 | } | |
8101 | out: | |
7dc66abb | 8102 | read_unlock(&fs_info->mapping_tree_lock); |
cf90d884 QW |
8103 | return ret; |
8104 | } | |
8105 | ||
8106 | /* | |
8107 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
8108 | * later chunk allocation/free would cause unexpected behavior. | |
8109 | * | |
8110 | * NOTE: This will iterate through the whole device tree, which should be of | |
8111 | * the same size level as the chunk tree. This slightly increases mount time. | |
8112 | */ | |
8113 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
8114 | { | |
8115 | struct btrfs_path *path; | |
8116 | struct btrfs_root *root = fs_info->dev_root; | |
8117 | struct btrfs_key key; | |
5eb19381 QW |
8118 | u64 prev_devid = 0; |
8119 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
8120 | int ret = 0; |
8121 | ||
42437a63 JB |
8122 | /* |
8123 | * We don't have a dev_root because we mounted with ignorebadroots and | |
8124 | * failed to load the root, so we want to skip the verification in this | |
8125 | * case for sure. | |
8126 | * | |
8127 | * However if the dev root is fine, but the tree itself is corrupted | |
8128 | * we'd still fail to mount. This verification is only to make sure | |
8129 | * writes can happen safely, so instead just bypass this check | |
8130 | * completely in the case of IGNOREBADROOTS. | |
8131 | */ | |
8132 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) | |
8133 | return 0; | |
8134 | ||
cf90d884 QW |
8135 | key.objectid = 1; |
8136 | key.type = BTRFS_DEV_EXTENT_KEY; | |
8137 | key.offset = 0; | |
8138 | ||
8139 | path = btrfs_alloc_path(); | |
8140 | if (!path) | |
8141 | return -ENOMEM; | |
8142 | ||
8143 | path->reada = READA_FORWARD; | |
8144 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
8145 | if (ret < 0) | |
8146 | goto out; | |
8147 | ||
8148 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 8149 | ret = btrfs_next_leaf(root, path); |
cf90d884 QW |
8150 | if (ret < 0) |
8151 | goto out; | |
8152 | /* No dev extents at all? Not good */ | |
8153 | if (ret > 0) { | |
8154 | ret = -EUCLEAN; | |
8155 | goto out; | |
8156 | } | |
8157 | } | |
8158 | while (1) { | |
8159 | struct extent_buffer *leaf = path->nodes[0]; | |
8160 | struct btrfs_dev_extent *dext; | |
8161 | int slot = path->slots[0]; | |
8162 | u64 chunk_offset; | |
8163 | u64 physical_offset; | |
8164 | u64 physical_len; | |
8165 | u64 devid; | |
8166 | ||
8167 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
8168 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
8169 | break; | |
8170 | devid = key.objectid; | |
8171 | physical_offset = key.offset; | |
8172 | ||
8173 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
8174 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
8175 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
8176 | ||
5eb19381 QW |
8177 | /* Check if this dev extent overlaps with the previous one */ |
8178 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
8179 | btrfs_err(fs_info, | |
8180 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
8181 | devid, physical_offset, prev_dev_ext_end); | |
8182 | ret = -EUCLEAN; | |
8183 | goto out; | |
8184 | } | |
8185 | ||
cf90d884 QW |
8186 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
8187 | physical_offset, physical_len); | |
8188 | if (ret < 0) | |
8189 | goto out; | |
5eb19381 QW |
8190 | prev_devid = devid; |
8191 | prev_dev_ext_end = physical_offset + physical_len; | |
8192 | ||
cf90d884 QW |
8193 | ret = btrfs_next_item(root, path); |
8194 | if (ret < 0) | |
8195 | goto out; | |
8196 | if (ret > 0) { | |
8197 | ret = 0; | |
8198 | break; | |
8199 | } | |
8200 | } | |
8201 | ||
8202 | /* Ensure all chunks have corresponding dev extents */ | |
8203 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
8204 | out: | |
8205 | btrfs_free_path(path); | |
8206 | return ret; | |
8207 | } | |
eede2bf3 OS |
8208 | |
8209 | /* | |
8210 | * Check whether the given block group or device is pinned by any inode being | |
8211 | * used as a swapfile. | |
8212 | */ | |
8213 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
8214 | { | |
8215 | struct btrfs_swapfile_pin *sp; | |
8216 | struct rb_node *node; | |
8217 | ||
8218 | spin_lock(&fs_info->swapfile_pins_lock); | |
8219 | node = fs_info->swapfile_pins.rb_node; | |
8220 | while (node) { | |
8221 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
8222 | if (ptr < sp->ptr) | |
8223 | node = node->rb_left; | |
8224 | else if (ptr > sp->ptr) | |
8225 | node = node->rb_right; | |
8226 | else | |
8227 | break; | |
8228 | } | |
8229 | spin_unlock(&fs_info->swapfile_pins_lock); | |
8230 | return node != NULL; | |
8231 | } | |
f7ef5287 NA |
8232 | |
8233 | static int relocating_repair_kthread(void *data) | |
8234 | { | |
0d031dc4 | 8235 | struct btrfs_block_group *cache = data; |
f7ef5287 NA |
8236 | struct btrfs_fs_info *fs_info = cache->fs_info; |
8237 | u64 target; | |
8238 | int ret = 0; | |
8239 | ||
8240 | target = cache->start; | |
8241 | btrfs_put_block_group(cache); | |
8242 | ||
ca5e4ea0 | 8243 | sb_start_write(fs_info->sb); |
f7ef5287 NA |
8244 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { |
8245 | btrfs_info(fs_info, | |
8246 | "zoned: skip relocating block group %llu to repair: EBUSY", | |
8247 | target); | |
ca5e4ea0 | 8248 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
8249 | return -EBUSY; |
8250 | } | |
8251 | ||
f3372065 | 8252 | mutex_lock(&fs_info->reclaim_bgs_lock); |
f7ef5287 NA |
8253 | |
8254 | /* Ensure block group still exists */ | |
8255 | cache = btrfs_lookup_block_group(fs_info, target); | |
8256 | if (!cache) | |
8257 | goto out; | |
8258 | ||
3349b57f | 8259 | if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) |
f7ef5287 NA |
8260 | goto out; |
8261 | ||
8262 | ret = btrfs_may_alloc_data_chunk(fs_info, target); | |
8263 | if (ret < 0) | |
8264 | goto out; | |
8265 | ||
8266 | btrfs_info(fs_info, | |
8267 | "zoned: relocating block group %llu to repair IO failure", | |
8268 | target); | |
8269 | ret = btrfs_relocate_chunk(fs_info, target); | |
8270 | ||
8271 | out: | |
8272 | if (cache) | |
8273 | btrfs_put_block_group(cache); | |
f3372065 | 8274 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f7ef5287 | 8275 | btrfs_exclop_finish(fs_info); |
ca5e4ea0 | 8276 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
8277 | |
8278 | return ret; | |
8279 | } | |
8280 | ||
554aed7d | 8281 | bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) |
f7ef5287 NA |
8282 | { |
8283 | struct btrfs_block_group *cache; | |
8284 | ||
554aed7d JT |
8285 | if (!btrfs_is_zoned(fs_info)) |
8286 | return false; | |
8287 | ||
f7ef5287 NA |
8288 | /* Do not attempt to repair in degraded state */ |
8289 | if (btrfs_test_opt(fs_info, DEGRADED)) | |
554aed7d | 8290 | return true; |
f7ef5287 NA |
8291 | |
8292 | cache = btrfs_lookup_block_group(fs_info, logical); | |
8293 | if (!cache) | |
554aed7d | 8294 | return true; |
f7ef5287 | 8295 | |
3349b57f | 8296 | if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { |
f7ef5287 | 8297 | btrfs_put_block_group(cache); |
554aed7d | 8298 | return true; |
f7ef5287 | 8299 | } |
f7ef5287 NA |
8300 | |
8301 | kthread_run(relocating_repair_kthread, cache, | |
8302 | "btrfs-relocating-repair"); | |
8303 | ||
554aed7d | 8304 | return true; |
f7ef5287 | 8305 | } |
4886ff7b QW |
8306 | |
8307 | static void map_raid56_repair_block(struct btrfs_io_context *bioc, | |
8308 | struct btrfs_io_stripe *smap, | |
8309 | u64 logical) | |
8310 | { | |
8311 | int data_stripes = nr_bioc_data_stripes(bioc); | |
8312 | int i; | |
8313 | ||
8314 | for (i = 0; i < data_stripes; i++) { | |
8315 | u64 stripe_start = bioc->full_stripe_logical + | |
cb091225 | 8316 | btrfs_stripe_nr_to_offset(i); |
4886ff7b QW |
8317 | |
8318 | if (logical >= stripe_start && | |
8319 | logical < stripe_start + BTRFS_STRIPE_LEN) | |
8320 | break; | |
8321 | } | |
8322 | ASSERT(i < data_stripes); | |
8323 | smap->dev = bioc->stripes[i].dev; | |
8324 | smap->physical = bioc->stripes[i].physical + | |
8325 | ((logical - bioc->full_stripe_logical) & | |
8326 | BTRFS_STRIPE_LEN_MASK); | |
8327 | } | |
8328 | ||
8329 | /* | |
8330 | * Map a repair write into a single device. | |
8331 | * | |
8332 | * A repair write is triggered by read time repair or scrub, which would only | |
8333 | * update the contents of a single device. | |
8334 | * Not update any other mirrors nor go through RMW path. | |
8335 | * | |
8336 | * Callers should ensure: | |
8337 | * | |
8338 | * - Call btrfs_bio_counter_inc_blocked() first | |
8339 | * - The range does not cross stripe boundary | |
8340 | * - Has a valid @mirror_num passed in. | |
8341 | */ | |
8342 | int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, | |
8343 | struct btrfs_io_stripe *smap, u64 logical, | |
8344 | u32 length, int mirror_num) | |
8345 | { | |
8346 | struct btrfs_io_context *bioc = NULL; | |
8347 | u64 map_length = length; | |
8348 | int mirror_ret = mirror_num; | |
8349 | int ret; | |
8350 | ||
8351 | ASSERT(mirror_num > 0); | |
8352 | ||
cd4efd21 | 8353 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, |
9fb2acc2 | 8354 | &bioc, smap, &mirror_ret); |
4886ff7b QW |
8355 | if (ret < 0) |
8356 | return ret; | |
8357 | ||
8358 | /* The map range should not cross stripe boundary. */ | |
8359 | ASSERT(map_length >= length); | |
8360 | ||
8361 | /* Already mapped to single stripe. */ | |
8362 | if (!bioc) | |
8363 | goto out; | |
8364 | ||
8365 | /* Map the RAID56 multi-stripe writes to a single one. */ | |
8366 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
8367 | map_raid56_repair_block(bioc, smap, logical); | |
8368 | goto out; | |
8369 | } | |
8370 | ||
8371 | ASSERT(mirror_num <= bioc->num_stripes); | |
8372 | smap->dev = bioc->stripes[mirror_num - 1].dev; | |
8373 | smap->physical = bioc->stripes[mirror_num - 1].physical; | |
8374 | out: | |
8375 | btrfs_put_bioc(bioc); | |
8376 | ASSERT(smap->dev); | |
8377 | return 0; | |
8378 | } |