Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 | 6 | #include <linux/sched.h> |
fccc0007 | 7 | #include <linux/sched/mm.h> |
0b86a832 | 8 | #include <linux/bio.h> |
5a0e3ad6 | 9 | #include <linux/slab.h> |
f2d8d74d | 10 | #include <linux/blkdev.h> |
442a4f63 | 11 | #include <linux/ratelimit.h> |
59641015 | 12 | #include <linux/kthread.h> |
53b381b3 | 13 | #include <linux/raid/pq.h> |
803b2f54 | 14 | #include <linux/semaphore.h> |
8da4b8c4 | 15 | #include <linux/uuid.h> |
f8e10cd3 | 16 | #include <linux/list_sort.h> |
784352fe | 17 | #include "misc.h" |
0b86a832 CM |
18 | #include "ctree.h" |
19 | #include "extent_map.h" | |
20 | #include "disk-io.h" | |
21 | #include "transaction.h" | |
22 | #include "print-tree.h" | |
23 | #include "volumes.h" | |
53b381b3 | 24 | #include "raid56.h" |
8b712842 | 25 | #include "async-thread.h" |
21adbd5c | 26 | #include "check-integrity.h" |
606686ee | 27 | #include "rcu-string.h" |
8dabb742 | 28 | #include "dev-replace.h" |
99994cde | 29 | #include "sysfs.h" |
82fc28fb | 30 | #include "tree-checker.h" |
8719aaae | 31 | #include "space-info.h" |
aac0023c | 32 | #include "block-group.h" |
b0643e59 | 33 | #include "discard.h" |
5b316468 | 34 | #include "zoned.h" |
0b86a832 | 35 | |
af902047 ZL |
36 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
37 | [BTRFS_RAID_RAID10] = { | |
38 | .sub_stripes = 2, | |
39 | .dev_stripes = 1, | |
40 | .devs_max = 0, /* 0 == as many as possible */ | |
b2f78e88 | 41 | .devs_min = 2, |
8789f4fe | 42 | .tolerated_failures = 1, |
af902047 ZL |
43 | .devs_increment = 2, |
44 | .ncopies = 2, | |
b50836ed | 45 | .nparity = 0, |
ed23467b | 46 | .raid_name = "raid10", |
41a6e891 | 47 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 48 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
49 | }, |
50 | [BTRFS_RAID_RAID1] = { | |
51 | .sub_stripes = 1, | |
52 | .dev_stripes = 1, | |
53 | .devs_max = 2, | |
54 | .devs_min = 2, | |
8789f4fe | 55 | .tolerated_failures = 1, |
af902047 ZL |
56 | .devs_increment = 2, |
57 | .ncopies = 2, | |
b50836ed | 58 | .nparity = 0, |
ed23467b | 59 | .raid_name = "raid1", |
41a6e891 | 60 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 61 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 62 | }, |
47e6f742 DS |
63 | [BTRFS_RAID_RAID1C3] = { |
64 | .sub_stripes = 1, | |
65 | .dev_stripes = 1, | |
cf93e15e | 66 | .devs_max = 3, |
47e6f742 DS |
67 | .devs_min = 3, |
68 | .tolerated_failures = 2, | |
69 | .devs_increment = 3, | |
70 | .ncopies = 3, | |
db26a024 | 71 | .nparity = 0, |
47e6f742 DS |
72 | .raid_name = "raid1c3", |
73 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
74 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
75 | }, | |
8d6fac00 DS |
76 | [BTRFS_RAID_RAID1C4] = { |
77 | .sub_stripes = 1, | |
78 | .dev_stripes = 1, | |
cf93e15e | 79 | .devs_max = 4, |
8d6fac00 DS |
80 | .devs_min = 4, |
81 | .tolerated_failures = 3, | |
82 | .devs_increment = 4, | |
83 | .ncopies = 4, | |
db26a024 | 84 | .nparity = 0, |
8d6fac00 DS |
85 | .raid_name = "raid1c4", |
86 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
87 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
88 | }, | |
af902047 ZL |
89 | [BTRFS_RAID_DUP] = { |
90 | .sub_stripes = 1, | |
91 | .dev_stripes = 2, | |
92 | .devs_max = 1, | |
93 | .devs_min = 1, | |
8789f4fe | 94 | .tolerated_failures = 0, |
af902047 ZL |
95 | .devs_increment = 1, |
96 | .ncopies = 2, | |
b50836ed | 97 | .nparity = 0, |
ed23467b | 98 | .raid_name = "dup", |
41a6e891 | 99 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 100 | .mindev_error = 0, |
af902047 ZL |
101 | }, |
102 | [BTRFS_RAID_RAID0] = { | |
103 | .sub_stripes = 1, | |
104 | .dev_stripes = 1, | |
105 | .devs_max = 0, | |
b2f78e88 | 106 | .devs_min = 1, |
8789f4fe | 107 | .tolerated_failures = 0, |
af902047 ZL |
108 | .devs_increment = 1, |
109 | .ncopies = 1, | |
b50836ed | 110 | .nparity = 0, |
ed23467b | 111 | .raid_name = "raid0", |
41a6e891 | 112 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 113 | .mindev_error = 0, |
af902047 ZL |
114 | }, |
115 | [BTRFS_RAID_SINGLE] = { | |
116 | .sub_stripes = 1, | |
117 | .dev_stripes = 1, | |
118 | .devs_max = 1, | |
119 | .devs_min = 1, | |
8789f4fe | 120 | .tolerated_failures = 0, |
af902047 ZL |
121 | .devs_increment = 1, |
122 | .ncopies = 1, | |
b50836ed | 123 | .nparity = 0, |
ed23467b | 124 | .raid_name = "single", |
41a6e891 | 125 | .bg_flag = 0, |
f9fbcaa2 | 126 | .mindev_error = 0, |
af902047 ZL |
127 | }, |
128 | [BTRFS_RAID_RAID5] = { | |
129 | .sub_stripes = 1, | |
130 | .dev_stripes = 1, | |
131 | .devs_max = 0, | |
132 | .devs_min = 2, | |
8789f4fe | 133 | .tolerated_failures = 1, |
af902047 | 134 | .devs_increment = 1, |
da612e31 | 135 | .ncopies = 1, |
b50836ed | 136 | .nparity = 1, |
ed23467b | 137 | .raid_name = "raid5", |
41a6e891 | 138 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 139 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
140 | }, |
141 | [BTRFS_RAID_RAID6] = { | |
142 | .sub_stripes = 1, | |
143 | .dev_stripes = 1, | |
144 | .devs_max = 0, | |
145 | .devs_min = 3, | |
8789f4fe | 146 | .tolerated_failures = 2, |
af902047 | 147 | .devs_increment = 1, |
da612e31 | 148 | .ncopies = 1, |
b50836ed | 149 | .nparity = 2, |
ed23467b | 150 | .raid_name = "raid6", |
41a6e891 | 151 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 152 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
153 | }, |
154 | }; | |
155 | ||
500a44c9 DS |
156 | /* |
157 | * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which | |
158 | * can be used as index to access btrfs_raid_array[]. | |
159 | */ | |
160 | enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) | |
161 | { | |
162 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
163 | return BTRFS_RAID_RAID10; | |
164 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) | |
165 | return BTRFS_RAID_RAID1; | |
166 | else if (flags & BTRFS_BLOCK_GROUP_RAID1C3) | |
167 | return BTRFS_RAID_RAID1C3; | |
168 | else if (flags & BTRFS_BLOCK_GROUP_RAID1C4) | |
169 | return BTRFS_RAID_RAID1C4; | |
170 | else if (flags & BTRFS_BLOCK_GROUP_DUP) | |
171 | return BTRFS_RAID_DUP; | |
172 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
173 | return BTRFS_RAID_RAID0; | |
174 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) | |
175 | return BTRFS_RAID_RAID5; | |
176 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) | |
177 | return BTRFS_RAID_RAID6; | |
178 | ||
179 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ | |
180 | } | |
181 | ||
158da513 | 182 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 183 | { |
158da513 DS |
184 | const int index = btrfs_bg_flags_to_raid_index(flags); |
185 | ||
186 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
187 | return NULL; |
188 | ||
158da513 | 189 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
190 | } |
191 | ||
f89e09cf AJ |
192 | /* |
193 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
194 | * bytes including terminating null byte. | |
195 | */ | |
196 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
197 | { | |
198 | int i; | |
199 | int ret; | |
200 | char *bp = buf; | |
201 | u64 flags = bg_flags; | |
202 | u32 size_bp = size_buf; | |
203 | ||
204 | if (!flags) { | |
205 | strcpy(bp, "NONE"); | |
206 | return; | |
207 | } | |
208 | ||
209 | #define DESCRIBE_FLAG(flag, desc) \ | |
210 | do { \ | |
211 | if (flags & (flag)) { \ | |
212 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
213 | if (ret < 0 || ret >= size_bp) \ | |
214 | goto out_overflow; \ | |
215 | size_bp -= ret; \ | |
216 | bp += ret; \ | |
217 | flags &= ~(flag); \ | |
218 | } \ | |
219 | } while (0) | |
220 | ||
221 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
222 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
223 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
224 | ||
225 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
226 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
227 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
228 | btrfs_raid_array[i].raid_name); | |
229 | #undef DESCRIBE_FLAG | |
230 | ||
231 | if (flags) { | |
232 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
233 | size_bp -= ret; | |
234 | } | |
235 | ||
236 | if (size_bp < size_buf) | |
237 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
238 | ||
239 | /* | |
240 | * The text is trimmed, it's up to the caller to provide sufficiently | |
241 | * large buffer | |
242 | */ | |
243 | out_overflow:; | |
244 | } | |
245 | ||
6f8e0fc7 | 246 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 247 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
48a3b636 | 248 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 249 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
5ab56090 LB |
250 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
251 | enum btrfs_map_op op, | |
252 | u64 logical, u64 *length, | |
253 | struct btrfs_bio **bbio_ret, | |
254 | int mirror_num, int need_raid_map); | |
2b82032c | 255 | |
9c6b1c4d DS |
256 | /* |
257 | * Device locking | |
258 | * ============== | |
259 | * | |
260 | * There are several mutexes that protect manipulation of devices and low-level | |
261 | * structures like chunks but not block groups, extents or files | |
262 | * | |
263 | * uuid_mutex (global lock) | |
264 | * ------------------------ | |
265 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
266 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
267 | * device) or requested by the device= mount option | |
268 | * | |
269 | * the mutex can be very coarse and can cover long-running operations | |
270 | * | |
271 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 272 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
273 | * |
274 | * global::fs_devs - add, remove, updates to the global list | |
275 | * | |
18c850fd JB |
276 | * does not protect: manipulation of the fs_devices::devices list in general |
277 | * but in mount context it could be used to exclude list modifications by eg. | |
278 | * scan ioctl | |
9c6b1c4d DS |
279 | * |
280 | * btrfs_device::name - renames (write side), read is RCU | |
281 | * | |
282 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
283 | * ------------------------------------------------ | |
284 | * protects updates to fs_devices::devices, ie. adding and deleting | |
285 | * | |
286 | * simple list traversal with read-only actions can be done with RCU protection | |
287 | * | |
288 | * may be used to exclude some operations from running concurrently without any | |
289 | * modifications to the list (see write_all_supers) | |
290 | * | |
18c850fd JB |
291 | * Is not required at mount and close times, because our device list is |
292 | * protected by the uuid_mutex at that point. | |
293 | * | |
9c6b1c4d DS |
294 | * balance_mutex |
295 | * ------------- | |
296 | * protects balance structures (status, state) and context accessed from | |
297 | * several places (internally, ioctl) | |
298 | * | |
299 | * chunk_mutex | |
300 | * ----------- | |
301 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
302 | * device is added/removed. Additionally it also protects post_commit_list of |
303 | * individual devices, since they can be added to the transaction's | |
304 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
305 | * |
306 | * cleaner_mutex | |
307 | * ------------- | |
308 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
309 | * cleaning together with relocation or delayed iputs | |
310 | * | |
311 | * | |
312 | * Lock nesting | |
313 | * ============ | |
314 | * | |
315 | * uuid_mutex | |
ae3e715f AJ |
316 | * device_list_mutex |
317 | * chunk_mutex | |
318 | * balance_mutex | |
89595e80 AJ |
319 | * |
320 | * | |
c3e1f96c GR |
321 | * Exclusive operations |
322 | * ==================== | |
89595e80 AJ |
323 | * |
324 | * Maintains the exclusivity of the following operations that apply to the | |
325 | * whole filesystem and cannot run in parallel. | |
326 | * | |
327 | * - Balance (*) | |
328 | * - Device add | |
329 | * - Device remove | |
330 | * - Device replace (*) | |
331 | * - Resize | |
332 | * | |
333 | * The device operations (as above) can be in one of the following states: | |
334 | * | |
335 | * - Running state | |
336 | * - Paused state | |
337 | * - Completed state | |
338 | * | |
339 | * Only device operations marked with (*) can go into the Paused state for the | |
340 | * following reasons: | |
341 | * | |
342 | * - ioctl (only Balance can be Paused through ioctl) | |
343 | * - filesystem remounted as read-only | |
344 | * - filesystem unmounted and mounted as read-only | |
345 | * - system power-cycle and filesystem mounted as read-only | |
346 | * - filesystem or device errors leading to forced read-only | |
347 | * | |
c3e1f96c GR |
348 | * The status of exclusive operation is set and cleared atomically. |
349 | * During the course of Paused state, fs_info::exclusive_operation remains set. | |
89595e80 AJ |
350 | * A device operation in Paused or Running state can be canceled or resumed |
351 | * either by ioctl (Balance only) or when remounted as read-write. | |
c3e1f96c | 352 | * The exclusive status is cleared when the device operation is canceled or |
89595e80 | 353 | * completed. |
9c6b1c4d DS |
354 | */ |
355 | ||
67a2c45e | 356 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 357 | static LIST_HEAD(fs_uuids); |
4143cb8b | 358 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
359 | { |
360 | return &fs_uuids; | |
361 | } | |
8a4b83cc | 362 | |
2dfeca9b DS |
363 | /* |
364 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
365 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
366 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
367 | * |
368 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
369 | * The returned struct is not linked onto any lists and can be destroyed with | |
370 | * kfree() right away. | |
371 | */ | |
7239ff4b NB |
372 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
373 | const u8 *metadata_fsid) | |
2208a378 ID |
374 | { |
375 | struct btrfs_fs_devices *fs_devs; | |
376 | ||
78f2c9e6 | 377 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
378 | if (!fs_devs) |
379 | return ERR_PTR(-ENOMEM); | |
380 | ||
381 | mutex_init(&fs_devs->device_list_mutex); | |
382 | ||
383 | INIT_LIST_HEAD(&fs_devs->devices); | |
384 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 385 | INIT_LIST_HEAD(&fs_devs->fs_list); |
944d3f9f | 386 | INIT_LIST_HEAD(&fs_devs->seed_list); |
2208a378 ID |
387 | if (fsid) |
388 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 389 | |
7239ff4b NB |
390 | if (metadata_fsid) |
391 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
392 | else if (fsid) | |
393 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
394 | ||
2208a378 ID |
395 | return fs_devs; |
396 | } | |
397 | ||
a425f9d4 | 398 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 399 | { |
bbbf7243 | 400 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 401 | rcu_string_free(device->name); |
1c11b63e | 402 | extent_io_tree_release(&device->alloc_state); |
48dae9cf | 403 | bio_put(device->flush_bio); |
5b316468 | 404 | btrfs_destroy_dev_zone_info(device); |
48dae9cf DS |
405 | kfree(device); |
406 | } | |
407 | ||
e4404d6e YZ |
408 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
409 | { | |
410 | struct btrfs_device *device; | |
411 | WARN_ON(fs_devices->opened); | |
412 | while (!list_empty(&fs_devices->devices)) { | |
413 | device = list_entry(fs_devices->devices.next, | |
414 | struct btrfs_device, dev_list); | |
415 | list_del(&device->dev_list); | |
a425f9d4 | 416 | btrfs_free_device(device); |
e4404d6e YZ |
417 | } |
418 | kfree(fs_devices); | |
419 | } | |
420 | ||
ffc5a379 | 421 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
422 | { |
423 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 424 | |
2b82032c YZ |
425 | while (!list_empty(&fs_uuids)) { |
426 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
427 | struct btrfs_fs_devices, fs_list); |
428 | list_del(&fs_devices->fs_list); | |
e4404d6e | 429 | free_fs_devices(fs_devices); |
8a4b83cc | 430 | } |
8a4b83cc CM |
431 | } |
432 | ||
7239ff4b NB |
433 | static noinline struct btrfs_fs_devices *find_fsid( |
434 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 435 | { |
8a4b83cc CM |
436 | struct btrfs_fs_devices *fs_devices; |
437 | ||
7239ff4b NB |
438 | ASSERT(fsid); |
439 | ||
7a62d0f0 | 440 | /* Handle non-split brain cases */ |
c4babc5e | 441 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
442 | if (metadata_fsid) { |
443 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
444 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
445 | BTRFS_FSID_SIZE) == 0) | |
446 | return fs_devices; | |
447 | } else { | |
448 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
449 | return fs_devices; | |
450 | } | |
8a4b83cc CM |
451 | } |
452 | return NULL; | |
453 | } | |
454 | ||
c6730a0e SY |
455 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
456 | struct btrfs_super_block *disk_super) | |
457 | { | |
458 | ||
459 | struct btrfs_fs_devices *fs_devices; | |
460 | ||
461 | /* | |
462 | * Handle scanned device having completed its fsid change but | |
463 | * belonging to a fs_devices that was created by first scanning | |
464 | * a device which didn't have its fsid/metadata_uuid changed | |
465 | * at all and the CHANGING_FSID_V2 flag set. | |
466 | */ | |
467 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
468 | if (fs_devices->fsid_change && | |
469 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
470 | BTRFS_FSID_SIZE) == 0 && | |
471 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
472 | BTRFS_FSID_SIZE) == 0) { | |
473 | return fs_devices; | |
474 | } | |
475 | } | |
476 | /* | |
477 | * Handle scanned device having completed its fsid change but | |
478 | * belonging to a fs_devices that was created by a device that | |
479 | * has an outdated pair of fsid/metadata_uuid and | |
480 | * CHANGING_FSID_V2 flag set. | |
481 | */ | |
482 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
483 | if (fs_devices->fsid_change && | |
484 | memcmp(fs_devices->metadata_uuid, | |
485 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
486 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
487 | BTRFS_FSID_SIZE) == 0) { | |
488 | return fs_devices; | |
489 | } | |
490 | } | |
491 | ||
492 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
493 | } | |
494 | ||
495 | ||
beaf8ab3 SB |
496 | static int |
497 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
498 | int flush, struct block_device **bdev, | |
8f32380d | 499 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
500 | { |
501 | int ret; | |
502 | ||
503 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
504 | ||
505 | if (IS_ERR(*bdev)) { | |
506 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
507 | goto error; |
508 | } | |
509 | ||
510 | if (flush) | |
511 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
9f6d2510 | 512 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
513 | if (ret) { |
514 | blkdev_put(*bdev, flags); | |
515 | goto error; | |
516 | } | |
517 | invalidate_bdev(*bdev); | |
8f32380d JT |
518 | *disk_super = btrfs_read_dev_super(*bdev); |
519 | if (IS_ERR(*disk_super)) { | |
520 | ret = PTR_ERR(*disk_super); | |
beaf8ab3 SB |
521 | blkdev_put(*bdev, flags); |
522 | goto error; | |
523 | } | |
524 | ||
525 | return 0; | |
526 | ||
527 | error: | |
528 | *bdev = NULL; | |
beaf8ab3 SB |
529 | return ret; |
530 | } | |
531 | ||
70bc7088 AJ |
532 | static bool device_path_matched(const char *path, struct btrfs_device *device) |
533 | { | |
534 | int found; | |
535 | ||
536 | rcu_read_lock(); | |
537 | found = strcmp(rcu_str_deref(device->name), path); | |
538 | rcu_read_unlock(); | |
539 | ||
540 | return found == 0; | |
541 | } | |
542 | ||
d8367db3 AJ |
543 | /* |
544 | * Search and remove all stale (devices which are not mounted) devices. | |
545 | * When both inputs are NULL, it will search and release all stale devices. | |
546 | * path: Optional. When provided will it release all unmounted devices | |
547 | * matching this path only. | |
548 | * skip_dev: Optional. Will skip this device when searching for the stale | |
549 | * devices. | |
70bc7088 AJ |
550 | * Return: 0 for success or if @path is NULL. |
551 | * -EBUSY if @path is a mounted device. | |
552 | * -ENOENT if @path does not match any device in the list. | |
d8367db3 | 553 | */ |
70bc7088 | 554 | static int btrfs_free_stale_devices(const char *path, |
fa6d2ae5 | 555 | struct btrfs_device *skip_device) |
4fde46f0 | 556 | { |
fa6d2ae5 AJ |
557 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
558 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
559 | int ret = 0; |
560 | ||
561 | if (path) | |
562 | ret = -ENOENT; | |
4fde46f0 | 563 | |
fa6d2ae5 | 564 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 565 | |
70bc7088 | 566 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
567 | list_for_each_entry_safe(device, tmp_device, |
568 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 569 | if (skip_device && skip_device == device) |
d8367db3 | 570 | continue; |
fa6d2ae5 | 571 | if (path && !device->name) |
4fde46f0 | 572 | continue; |
70bc7088 | 573 | if (path && !device_path_matched(path, device)) |
38cf665d | 574 | continue; |
70bc7088 AJ |
575 | if (fs_devices->opened) { |
576 | /* for an already deleted device return 0 */ | |
577 | if (path && ret != 0) | |
578 | ret = -EBUSY; | |
579 | break; | |
580 | } | |
4fde46f0 | 581 | |
4fde46f0 | 582 | /* delete the stale device */ |
7bcb8164 AJ |
583 | fs_devices->num_devices--; |
584 | list_del(&device->dev_list); | |
585 | btrfs_free_device(device); | |
586 | ||
70bc7088 | 587 | ret = 0; |
7bcb8164 AJ |
588 | } |
589 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 590 | |
7bcb8164 AJ |
591 | if (fs_devices->num_devices == 0) { |
592 | btrfs_sysfs_remove_fsid(fs_devices); | |
593 | list_del(&fs_devices->fs_list); | |
594 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
595 | } |
596 | } | |
70bc7088 AJ |
597 | |
598 | return ret; | |
4fde46f0 AJ |
599 | } |
600 | ||
18c850fd JB |
601 | /* |
602 | * This is only used on mount, and we are protected from competing things | |
603 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
604 | * fs_devices->device_list_mutex here. | |
605 | */ | |
0fb08bcc AJ |
606 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
607 | struct btrfs_device *device, fmode_t flags, | |
608 | void *holder) | |
609 | { | |
610 | struct request_queue *q; | |
611 | struct block_device *bdev; | |
0fb08bcc AJ |
612 | struct btrfs_super_block *disk_super; |
613 | u64 devid; | |
614 | int ret; | |
615 | ||
616 | if (device->bdev) | |
617 | return -EINVAL; | |
618 | if (!device->name) | |
619 | return -EINVAL; | |
620 | ||
621 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 622 | &bdev, &disk_super); |
0fb08bcc AJ |
623 | if (ret) |
624 | return ret; | |
625 | ||
0fb08bcc AJ |
626 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
627 | if (devid != device->devid) | |
8f32380d | 628 | goto error_free_page; |
0fb08bcc AJ |
629 | |
630 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 631 | goto error_free_page; |
0fb08bcc AJ |
632 | |
633 | device->generation = btrfs_super_generation(disk_super); | |
634 | ||
635 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
636 | if (btrfs_super_incompat_flags(disk_super) & |
637 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
638 | pr_err( | |
639 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 640 | goto error_free_page; |
7239ff4b NB |
641 | } |
642 | ||
ebbede42 | 643 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 644 | fs_devices->seeding = true; |
0fb08bcc | 645 | } else { |
ebbede42 AJ |
646 | if (bdev_read_only(bdev)) |
647 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
648 | else | |
649 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
650 | } |
651 | ||
652 | q = bdev_get_queue(bdev); | |
0fb08bcc | 653 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 654 | fs_devices->rotating = true; |
0fb08bcc AJ |
655 | |
656 | device->bdev = bdev; | |
e12c9621 | 657 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
658 | device->mode = flags; |
659 | ||
660 | fs_devices->open_devices++; | |
ebbede42 AJ |
661 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
662 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 663 | fs_devices->rw_devices++; |
b1b8e386 | 664 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 665 | } |
8f32380d | 666 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
667 | |
668 | return 0; | |
669 | ||
8f32380d JT |
670 | error_free_page: |
671 | btrfs_release_disk_super(disk_super); | |
0fb08bcc AJ |
672 | blkdev_put(bdev, flags); |
673 | ||
674 | return -EINVAL; | |
675 | } | |
676 | ||
7a62d0f0 NB |
677 | /* |
678 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
679 | * being created with a disk that has already completed its fsid change. Such |
680 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
681 | * Handle both cases here. | |
7a62d0f0 NB |
682 | */ |
683 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
684 | struct btrfs_super_block *disk_super) | |
685 | { | |
686 | struct btrfs_fs_devices *fs_devices; | |
687 | ||
688 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
689 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
690 | BTRFS_FSID_SIZE) != 0 && | |
691 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
692 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
693 | return fs_devices; | |
694 | } | |
695 | } | |
696 | ||
c0d81c7c | 697 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
698 | } |
699 | ||
cc5de4e7 NB |
700 | |
701 | static struct btrfs_fs_devices *find_fsid_changed( | |
702 | struct btrfs_super_block *disk_super) | |
703 | { | |
704 | struct btrfs_fs_devices *fs_devices; | |
705 | ||
706 | /* | |
707 | * Handles the case where scanned device is part of an fs that had | |
1a9fd417 | 708 | * multiple successful changes of FSID but currently device didn't |
05840710 NB |
709 | * observe it. Meaning our fsid will be different than theirs. We need |
710 | * to handle two subcases : | |
711 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
712 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
713 | * are equal). | |
cc5de4e7 NB |
714 | */ |
715 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 716 | /* Changed UUIDs */ |
cc5de4e7 NB |
717 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
718 | BTRFS_FSID_SIZE) != 0 && | |
719 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
720 | BTRFS_FSID_SIZE) == 0 && | |
721 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
722 | BTRFS_FSID_SIZE) != 0) |
723 | return fs_devices; | |
724 | ||
725 | /* Unchanged UUIDs */ | |
726 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
727 | BTRFS_FSID_SIZE) == 0 && | |
728 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
729 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 730 | return fs_devices; |
cc5de4e7 NB |
731 | } |
732 | ||
733 | return NULL; | |
734 | } | |
1362089d NB |
735 | |
736 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
737 | struct btrfs_super_block *disk_super) | |
738 | { | |
739 | struct btrfs_fs_devices *fs_devices; | |
740 | ||
741 | /* | |
742 | * Handle the case where the scanned device is part of an fs whose last | |
743 | * metadata UUID change reverted it to the original FSID. At the same | |
744 | * time * fs_devices was first created by another constitutent device | |
745 | * which didn't fully observe the operation. This results in an | |
746 | * btrfs_fs_devices created with metadata/fsid different AND | |
747 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
748 | * fs_devices equal to the FSID of the disk. | |
749 | */ | |
750 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
751 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
752 | BTRFS_FSID_SIZE) != 0 && | |
753 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
754 | BTRFS_FSID_SIZE) == 0 && | |
755 | fs_devices->fsid_change) | |
756 | return fs_devices; | |
757 | } | |
758 | ||
759 | return NULL; | |
760 | } | |
60999ca4 DS |
761 | /* |
762 | * Add new device to list of registered devices | |
763 | * | |
764 | * Returns: | |
e124ece5 AJ |
765 | * device pointer which was just added or updated when successful |
766 | * error pointer when failed | |
60999ca4 | 767 | */ |
e124ece5 | 768 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
769 | struct btrfs_super_block *disk_super, |
770 | bool *new_device_added) | |
8a4b83cc CM |
771 | { |
772 | struct btrfs_device *device; | |
7a62d0f0 | 773 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 774 | struct rcu_string *name; |
8a4b83cc | 775 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 776 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
7239ff4b NB |
777 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
778 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
779 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
780 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 781 | |
cc5de4e7 | 782 | if (fsid_change_in_progress) { |
c0d81c7c | 783 | if (!has_metadata_uuid) |
cc5de4e7 | 784 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 785 | else |
cc5de4e7 | 786 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 787 | } else if (has_metadata_uuid) { |
c6730a0e | 788 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 789 | } else { |
1362089d NB |
790 | fs_devices = find_fsid_reverted_metadata(disk_super); |
791 | if (!fs_devices) | |
792 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
793 | } |
794 | ||
8a4b83cc | 795 | |
8a4b83cc | 796 | if (!fs_devices) { |
7239ff4b NB |
797 | if (has_metadata_uuid) |
798 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
799 | disk_super->metadata_uuid); | |
800 | else | |
801 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
802 | ||
2208a378 | 803 | if (IS_ERR(fs_devices)) |
e124ece5 | 804 | return ERR_CAST(fs_devices); |
2208a378 | 805 | |
92900e51 AV |
806 | fs_devices->fsid_change = fsid_change_in_progress; |
807 | ||
9c6d173e | 808 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 809 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 810 | |
8a4b83cc CM |
811 | device = NULL; |
812 | } else { | |
9c6d173e | 813 | mutex_lock(&fs_devices->device_list_mutex); |
09ba3bc9 | 814 | device = btrfs_find_device(fs_devices, devid, |
b2598edf | 815 | disk_super->dev_item.uuid, NULL); |
7a62d0f0 NB |
816 | |
817 | /* | |
818 | * If this disk has been pulled into an fs devices created by | |
819 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
820 | * metadata_uuid/fsid values of the fs_devices. | |
821 | */ | |
1362089d | 822 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
823 | found_transid > fs_devices->latest_generation) { |
824 | memcpy(fs_devices->fsid, disk_super->fsid, | |
825 | BTRFS_FSID_SIZE); | |
1362089d NB |
826 | |
827 | if (has_metadata_uuid) | |
828 | memcpy(fs_devices->metadata_uuid, | |
829 | disk_super->metadata_uuid, | |
830 | BTRFS_FSID_SIZE); | |
831 | else | |
832 | memcpy(fs_devices->metadata_uuid, | |
833 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
834 | |
835 | fs_devices->fsid_change = false; | |
836 | } | |
8a4b83cc | 837 | } |
443f24fe | 838 | |
8a4b83cc | 839 | if (!device) { |
9c6d173e AJ |
840 | if (fs_devices->opened) { |
841 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 842 | return ERR_PTR(-EBUSY); |
9c6d173e | 843 | } |
2b82032c | 844 | |
12bd2fc0 ID |
845 | device = btrfs_alloc_device(NULL, &devid, |
846 | disk_super->dev_item.uuid); | |
847 | if (IS_ERR(device)) { | |
9c6d173e | 848 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 849 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 850 | return device; |
8a4b83cc | 851 | } |
606686ee JB |
852 | |
853 | name = rcu_string_strdup(path, GFP_NOFS); | |
854 | if (!name) { | |
a425f9d4 | 855 | btrfs_free_device(device); |
9c6d173e | 856 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 857 | return ERR_PTR(-ENOMEM); |
8a4b83cc | 858 | } |
606686ee | 859 | rcu_assign_pointer(device->name, name); |
90519d66 | 860 | |
1f78160c | 861 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 862 | fs_devices->num_devices++; |
e5e9a520 | 863 | |
2b82032c | 864 | device->fs_devices = fs_devices; |
4306a974 | 865 | *new_device_added = true; |
327f18cc AJ |
866 | |
867 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
868 | pr_info( |
869 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
870 | disk_super->label, devid, found_transid, path, | |
871 | current->comm, task_pid_nr(current)); | |
327f18cc | 872 | else |
aa6c0df7 AJ |
873 | pr_info( |
874 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
875 | disk_super->fsid, devid, found_transid, path, | |
876 | current->comm, task_pid_nr(current)); | |
327f18cc | 877 | |
606686ee | 878 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
879 | /* |
880 | * When FS is already mounted. | |
881 | * 1. If you are here and if the device->name is NULL that | |
882 | * means this device was missing at time of FS mount. | |
883 | * 2. If you are here and if the device->name is different | |
884 | * from 'path' that means either | |
885 | * a. The same device disappeared and reappeared with | |
886 | * different name. or | |
887 | * b. The missing-disk-which-was-replaced, has | |
888 | * reappeared now. | |
889 | * | |
890 | * We must allow 1 and 2a above. But 2b would be a spurious | |
891 | * and unintentional. | |
892 | * | |
893 | * Further in case of 1 and 2a above, the disk at 'path' | |
894 | * would have missed some transaction when it was away and | |
895 | * in case of 2a the stale bdev has to be updated as well. | |
896 | * 2b must not be allowed at all time. | |
897 | */ | |
898 | ||
899 | /* | |
0f23ae74 CM |
900 | * For now, we do allow update to btrfs_fs_device through the |
901 | * btrfs dev scan cli after FS has been mounted. We're still | |
902 | * tracking a problem where systems fail mount by subvolume id | |
903 | * when we reject replacement on a mounted FS. | |
b96de000 | 904 | */ |
0f23ae74 | 905 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
906 | /* |
907 | * That is if the FS is _not_ mounted and if you | |
908 | * are here, that means there is more than one | |
909 | * disk with same uuid and devid.We keep the one | |
910 | * with larger generation number or the last-in if | |
911 | * generation are equal. | |
912 | */ | |
9c6d173e | 913 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 914 | return ERR_PTR(-EEXIST); |
77bdae4d | 915 | } |
b96de000 | 916 | |
a9261d41 AJ |
917 | /* |
918 | * We are going to replace the device path for a given devid, | |
919 | * make sure it's the same device if the device is mounted | |
920 | */ | |
921 | if (device->bdev) { | |
4e7b5671 CH |
922 | int error; |
923 | dev_t path_dev; | |
a9261d41 | 924 | |
4e7b5671 CH |
925 | error = lookup_bdev(path, &path_dev); |
926 | if (error) { | |
a9261d41 | 927 | mutex_unlock(&fs_devices->device_list_mutex); |
4e7b5671 | 928 | return ERR_PTR(error); |
a9261d41 AJ |
929 | } |
930 | ||
4e7b5671 | 931 | if (device->bdev->bd_dev != path_dev) { |
a9261d41 | 932 | mutex_unlock(&fs_devices->device_list_mutex); |
0697d9a6 JT |
933 | /* |
934 | * device->fs_info may not be reliable here, so | |
935 | * pass in a NULL instead. This avoids a | |
936 | * possible use-after-free when the fs_info and | |
937 | * fs_info->sb are already torn down. | |
938 | */ | |
939 | btrfs_warn_in_rcu(NULL, | |
79dae17d AJ |
940 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
941 | path, devid, found_transid, | |
942 | current->comm, | |
943 | task_pid_nr(current)); | |
a9261d41 AJ |
944 | return ERR_PTR(-EEXIST); |
945 | } | |
a9261d41 | 946 | btrfs_info_in_rcu(device->fs_info, |
79dae17d AJ |
947 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
948 | devid, rcu_str_deref(device->name), | |
949 | path, current->comm, | |
950 | task_pid_nr(current)); | |
a9261d41 AJ |
951 | } |
952 | ||
606686ee | 953 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
954 | if (!name) { |
955 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 956 | return ERR_PTR(-ENOMEM); |
9c6d173e | 957 | } |
606686ee JB |
958 | rcu_string_free(device->name); |
959 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 960 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 961 | fs_devices->missing_devices--; |
e6e674bd | 962 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 963 | } |
8a4b83cc CM |
964 | } |
965 | ||
77bdae4d AJ |
966 | /* |
967 | * Unmount does not free the btrfs_device struct but would zero | |
968 | * generation along with most of the other members. So just update | |
969 | * it back. We need it to pick the disk with largest generation | |
970 | * (as above). | |
971 | */ | |
d1a63002 | 972 | if (!fs_devices->opened) { |
77bdae4d | 973 | device->generation = found_transid; |
d1a63002 NB |
974 | fs_devices->latest_generation = max_t(u64, found_transid, |
975 | fs_devices->latest_generation); | |
976 | } | |
77bdae4d | 977 | |
f2788d2f AJ |
978 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
979 | ||
9c6d173e | 980 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 981 | return device; |
8a4b83cc CM |
982 | } |
983 | ||
e4404d6e YZ |
984 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
985 | { | |
986 | struct btrfs_fs_devices *fs_devices; | |
987 | struct btrfs_device *device; | |
988 | struct btrfs_device *orig_dev; | |
d2979aa2 | 989 | int ret = 0; |
e4404d6e | 990 | |
7239ff4b | 991 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
992 | if (IS_ERR(fs_devices)) |
993 | return fs_devices; | |
e4404d6e | 994 | |
adbbb863 | 995 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 996 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
997 | |
998 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
606686ee JB |
999 | struct rcu_string *name; |
1000 | ||
12bd2fc0 ID |
1001 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
1002 | orig_dev->uuid); | |
d2979aa2 AJ |
1003 | if (IS_ERR(device)) { |
1004 | ret = PTR_ERR(device); | |
e4404d6e | 1005 | goto error; |
d2979aa2 | 1006 | } |
e4404d6e | 1007 | |
606686ee JB |
1008 | /* |
1009 | * This is ok to do without rcu read locked because we hold the | |
1010 | * uuid mutex so nothing we touch in here is going to disappear. | |
1011 | */ | |
e755f780 | 1012 | if (orig_dev->name) { |
78f2c9e6 DS |
1013 | name = rcu_string_strdup(orig_dev->name->str, |
1014 | GFP_KERNEL); | |
e755f780 | 1015 | if (!name) { |
a425f9d4 | 1016 | btrfs_free_device(device); |
d2979aa2 | 1017 | ret = -ENOMEM; |
e755f780 AJ |
1018 | goto error; |
1019 | } | |
1020 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 1021 | } |
e4404d6e | 1022 | |
e4404d6e YZ |
1023 | list_add(&device->dev_list, &fs_devices->devices); |
1024 | device->fs_devices = fs_devices; | |
1025 | fs_devices->num_devices++; | |
1026 | } | |
adbbb863 | 1027 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
1028 | return fs_devices; |
1029 | error: | |
adbbb863 | 1030 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e | 1031 | free_fs_devices(fs_devices); |
d2979aa2 | 1032 | return ERR_PTR(ret); |
e4404d6e YZ |
1033 | } |
1034 | ||
3712ccb7 | 1035 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
bacce86a | 1036 | struct btrfs_device **latest_dev) |
dfe25020 | 1037 | { |
c6e30871 | 1038 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1039 | |
46224705 | 1040 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1041 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1042 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1043 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1044 | &device->dev_state) && |
998a0671 AJ |
1045 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1046 | &device->dev_state) && | |
3712ccb7 NB |
1047 | (!*latest_dev || |
1048 | device->generation > (*latest_dev)->generation)) { | |
1049 | *latest_dev = device; | |
a6b0d5c8 | 1050 | } |
2b82032c | 1051 | continue; |
a6b0d5c8 | 1052 | } |
2b82032c | 1053 | |
cf89af14 AJ |
1054 | /* |
1055 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, | |
1056 | * in btrfs_init_dev_replace() so just continue. | |
1057 | */ | |
1058 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) | |
1059 | continue; | |
1060 | ||
2b82032c | 1061 | if (device->bdev) { |
d4d77629 | 1062 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1063 | device->bdev = NULL; |
1064 | fs_devices->open_devices--; | |
1065 | } | |
ebbede42 | 1066 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1067 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1068 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
b2a61667 | 1069 | fs_devices->rw_devices--; |
2b82032c | 1070 | } |
e4404d6e YZ |
1071 | list_del_init(&device->dev_list); |
1072 | fs_devices->num_devices--; | |
a425f9d4 | 1073 | btrfs_free_device(device); |
dfe25020 | 1074 | } |
2b82032c | 1075 | |
3712ccb7 NB |
1076 | } |
1077 | ||
1078 | /* | |
1079 | * After we have read the system tree and know devids belonging to this | |
1080 | * filesystem, remove the device which does not belong there. | |
1081 | */ | |
bacce86a | 1082 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
3712ccb7 NB |
1083 | { |
1084 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1085 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1086 | |
1087 | mutex_lock(&uuid_mutex); | |
bacce86a | 1088 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
944d3f9f NB |
1089 | |
1090 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
bacce86a | 1091 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
2b82032c | 1092 | |
443f24fe | 1093 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 1094 | |
dfe25020 | 1095 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1096 | } |
a0af469b | 1097 | |
14238819 AJ |
1098 | static void btrfs_close_bdev(struct btrfs_device *device) |
1099 | { | |
08ffcae8 DS |
1100 | if (!device->bdev) |
1101 | return; | |
1102 | ||
ebbede42 | 1103 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1104 | sync_blockdev(device->bdev); |
1105 | invalidate_bdev(device->bdev); | |
1106 | } | |
1107 | ||
08ffcae8 | 1108 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1109 | } |
1110 | ||
959b1c04 | 1111 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1112 | { |
1113 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1114 | |
ebbede42 | 1115 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1116 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1117 | list_del_init(&device->dev_alloc_list); | |
1118 | fs_devices->rw_devices--; | |
1119 | } | |
1120 | ||
0d977e0e DCZX |
1121 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
1122 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); | |
1123 | ||
e6e674bd | 1124 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
f448341a AJ |
1125 | fs_devices->missing_devices--; |
1126 | ||
959b1c04 | 1127 | btrfs_close_bdev(device); |
321f69f8 | 1128 | if (device->bdev) { |
3fff3975 | 1129 | fs_devices->open_devices--; |
321f69f8 | 1130 | device->bdev = NULL; |
f448341a | 1131 | } |
321f69f8 | 1132 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
5b316468 | 1133 | btrfs_destroy_dev_zone_info(device); |
f448341a | 1134 | |
321f69f8 JT |
1135 | device->fs_info = NULL; |
1136 | atomic_set(&device->dev_stats_ccnt, 0); | |
1137 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1138 | |
321f69f8 JT |
1139 | /* Verify the device is back in a pristine state */ |
1140 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1141 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1142 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1143 | ASSERT(list_empty(&device->post_commit_list)); | |
1144 | ASSERT(atomic_read(&device->reada_in_flight) == 0); | |
f448341a AJ |
1145 | } |
1146 | ||
54eed6ae | 1147 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1148 | { |
2037a093 | 1149 | struct btrfs_device *device, *tmp; |
e4404d6e | 1150 | |
425c6ed6 JB |
1151 | lockdep_assert_held(&uuid_mutex); |
1152 | ||
2b82032c | 1153 | if (--fs_devices->opened > 0) |
54eed6ae | 1154 | return; |
8a4b83cc | 1155 | |
425c6ed6 | 1156 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1157 | btrfs_close_one_device(device); |
c9513edb | 1158 | |
e4404d6e YZ |
1159 | WARN_ON(fs_devices->open_devices); |
1160 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1161 | fs_devices->opened = 0; |
0395d84f | 1162 | fs_devices->seeding = false; |
c4989c2f | 1163 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1164 | } |
1165 | ||
54eed6ae | 1166 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1167 | { |
944d3f9f NB |
1168 | LIST_HEAD(list); |
1169 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1170 | |
1171 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1172 | close_fs_devices(fs_devices); |
944d3f9f NB |
1173 | if (!fs_devices->opened) |
1174 | list_splice_init(&fs_devices->seed_list, &list); | |
e4404d6e | 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, |
e4404d6e | 1185 | fmode_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; |
8a4b83cc | 1190 | |
d4d77629 TH |
1191 | flags |= FMODE_EXCL; |
1192 | ||
96c2e067 AJ |
1193 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
1194 | dev_list) { | |
1195 | int ret; | |
a0af469b | 1196 | |
96c2e067 AJ |
1197 | ret = btrfs_open_one_device(fs_devices, device, flags, holder); |
1198 | if (ret == 0 && | |
1199 | (!latest_dev || device->generation > latest_dev->generation)) { | |
9f050db4 | 1200 | latest_dev = device; |
96c2e067 AJ |
1201 | } else if (ret == -ENODATA) { |
1202 | fs_devices->num_devices--; | |
1203 | list_del(&device->dev_list); | |
1204 | btrfs_free_device(device); | |
1205 | } | |
8a4b83cc | 1206 | } |
1ed802c9 AJ |
1207 | if (fs_devices->open_devices == 0) |
1208 | return -EINVAL; | |
1209 | ||
2b82032c | 1210 | fs_devices->opened = 1; |
443f24fe | 1211 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1212 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1213 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
33fd2f71 | 1214 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
1ed802c9 AJ |
1215 | |
1216 | return 0; | |
2b82032c YZ |
1217 | } |
1218 | ||
4f0f586b ST |
1219 | static int devid_cmp(void *priv, const struct list_head *a, |
1220 | const struct list_head *b) | |
f8e10cd3 | 1221 | { |
214cc184 | 1222 | const struct btrfs_device *dev1, *dev2; |
f8e10cd3 AJ |
1223 | |
1224 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1225 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1226 | ||
1227 | if (dev1->devid < dev2->devid) | |
1228 | return -1; | |
1229 | else if (dev1->devid > dev2->devid) | |
1230 | return 1; | |
1231 | return 0; | |
1232 | } | |
1233 | ||
2b82032c | 1234 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1235 | fmode_t flags, void *holder) |
2b82032c YZ |
1236 | { |
1237 | int ret; | |
1238 | ||
f5194e34 | 1239 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1240 | /* |
1241 | * The device_list_mutex cannot be taken here in case opening the | |
a8698707 | 1242 | * underlying device takes further locks like open_mutex. |
18c850fd JB |
1243 | * |
1244 | * We also don't need the lock here as this is called during mount and | |
1245 | * exclusion is provided by uuid_mutex | |
1246 | */ | |
f5194e34 | 1247 | |
2b82032c | 1248 | if (fs_devices->opened) { |
e4404d6e YZ |
1249 | fs_devices->opened++; |
1250 | ret = 0; | |
2b82032c | 1251 | } else { |
f8e10cd3 | 1252 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1253 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1254 | } |
542c5908 | 1255 | |
8a4b83cc CM |
1256 | return ret; |
1257 | } | |
1258 | ||
8f32380d | 1259 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1260 | { |
8f32380d JT |
1261 | struct page *page = virt_to_page(super); |
1262 | ||
6cf86a00 AJ |
1263 | put_page(page); |
1264 | } | |
1265 | ||
b335eab8 | 1266 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
12659251 | 1267 | u64 bytenr, u64 bytenr_orig) |
6cf86a00 | 1268 | { |
b335eab8 NB |
1269 | struct btrfs_super_block *disk_super; |
1270 | struct page *page; | |
6cf86a00 AJ |
1271 | void *p; |
1272 | pgoff_t index; | |
1273 | ||
1274 | /* make sure our super fits in the device */ | |
1275 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
b335eab8 | 1276 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1277 | |
1278 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1279 | if (sizeof(*disk_super) > PAGE_SIZE) |
1280 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1281 | |
1282 | /* make sure our super doesn't straddle pages on disk */ | |
1283 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1284 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1285 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1286 | |
1287 | /* pull in the page with our super */ | |
b335eab8 | 1288 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1289 | |
b335eab8 NB |
1290 | if (IS_ERR(page)) |
1291 | return ERR_CAST(page); | |
6cf86a00 | 1292 | |
b335eab8 | 1293 | p = page_address(page); |
6cf86a00 AJ |
1294 | |
1295 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1296 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1297 | |
12659251 | 1298 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
b335eab8 | 1299 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
8f32380d | 1300 | btrfs_release_disk_super(p); |
b335eab8 | 1301 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1302 | } |
1303 | ||
b335eab8 NB |
1304 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1305 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1306 | |
b335eab8 | 1307 | return disk_super; |
6cf86a00 AJ |
1308 | } |
1309 | ||
228a73ab AJ |
1310 | int btrfs_forget_devices(const char *path) |
1311 | { | |
1312 | int ret; | |
1313 | ||
1314 | mutex_lock(&uuid_mutex); | |
1315 | ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL); | |
1316 | mutex_unlock(&uuid_mutex); | |
1317 | ||
1318 | return ret; | |
1319 | } | |
1320 | ||
6f60cbd3 DS |
1321 | /* |
1322 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1323 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1324 | * is read via pagecache | |
1325 | */ | |
36350e95 GJ |
1326 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1327 | void *holder) | |
8a4b83cc CM |
1328 | { |
1329 | struct btrfs_super_block *disk_super; | |
4306a974 | 1330 | bool new_device_added = false; |
36350e95 | 1331 | struct btrfs_device *device = NULL; |
8a4b83cc | 1332 | struct block_device *bdev; |
12659251 NA |
1333 | u64 bytenr, bytenr_orig; |
1334 | int ret; | |
8a4b83cc | 1335 | |
899f9307 DS |
1336 | lockdep_assert_held(&uuid_mutex); |
1337 | ||
6f60cbd3 DS |
1338 | /* |
1339 | * we would like to check all the supers, but that would make | |
1340 | * a btrfs mount succeed after a mkfs from a different FS. | |
1341 | * So, we need to add a special mount option to scan for | |
1342 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1343 | */ | |
d4d77629 | 1344 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1345 | |
1346 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1347 | if (IS_ERR(bdev)) |
36350e95 | 1348 | return ERR_CAST(bdev); |
6f60cbd3 | 1349 | |
12659251 NA |
1350 | bytenr_orig = btrfs_sb_offset(0); |
1351 | ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); | |
1352 | if (ret) | |
1353 | return ERR_PTR(ret); | |
1354 | ||
1355 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); | |
b335eab8 NB |
1356 | if (IS_ERR(disk_super)) { |
1357 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1358 | goto error_bdev_put; |
05a5c55d | 1359 | } |
6f60cbd3 | 1360 | |
4306a974 | 1361 | device = device_list_add(path, disk_super, &new_device_added); |
36350e95 | 1362 | if (!IS_ERR(device)) { |
4306a974 AJ |
1363 | if (new_device_added) |
1364 | btrfs_free_stale_devices(path, device); | |
1365 | } | |
6f60cbd3 | 1366 | |
8f32380d | 1367 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1368 | |
1369 | error_bdev_put: | |
d4d77629 | 1370 | blkdev_put(bdev, flags); |
b6ed73bc | 1371 | |
36350e95 | 1372 | return device; |
8a4b83cc | 1373 | } |
0b86a832 | 1374 | |
1c11b63e JM |
1375 | /* |
1376 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1377 | * such is found, record the end of it in *start | |
1378 | */ | |
1c11b63e JM |
1379 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1380 | u64 len) | |
6df9a95e | 1381 | { |
1c11b63e | 1382 | u64 physical_start, physical_end; |
6df9a95e | 1383 | |
1c11b63e | 1384 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1385 | |
1c11b63e JM |
1386 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1387 | &physical_start, &physical_end, | |
1388 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1389 | |
1c11b63e JM |
1390 | if (in_range(physical_start, *start, len) || |
1391 | in_range(*start, physical_start, | |
1392 | physical_end - physical_start)) { | |
1393 | *start = physical_end + 1; | |
1394 | return true; | |
6df9a95e JB |
1395 | } |
1396 | } | |
1c11b63e | 1397 | return false; |
6df9a95e JB |
1398 | } |
1399 | ||
3b4ffa40 NA |
1400 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1401 | { | |
1402 | switch (device->fs_devices->chunk_alloc_policy) { | |
1403 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1404 | /* | |
1405 | * We don't want to overwrite the superblock on the drive nor | |
1406 | * any area used by the boot loader (grub for example), so we | |
1407 | * make sure to start at an offset of at least 1MB. | |
1408 | */ | |
1409 | return max_t(u64, start, SZ_1M); | |
1cd6121f NA |
1410 | case BTRFS_CHUNK_ALLOC_ZONED: |
1411 | /* | |
1412 | * We don't care about the starting region like regular | |
1413 | * allocator, because we anyway use/reserve the first two zones | |
1414 | * for superblock logging. | |
1415 | */ | |
1416 | return ALIGN(start, device->zone_info->zone_size); | |
3b4ffa40 NA |
1417 | default: |
1418 | BUG(); | |
1419 | } | |
1420 | } | |
1421 | ||
1cd6121f NA |
1422 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
1423 | u64 *hole_start, u64 *hole_size, | |
1424 | u64 num_bytes) | |
1425 | { | |
1426 | u64 zone_size = device->zone_info->zone_size; | |
1427 | u64 pos; | |
1428 | int ret; | |
1429 | bool changed = false; | |
1430 | ||
1431 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); | |
1432 | ||
1433 | while (*hole_size > 0) { | |
1434 | pos = btrfs_find_allocatable_zones(device, *hole_start, | |
1435 | *hole_start + *hole_size, | |
1436 | num_bytes); | |
1437 | if (pos != *hole_start) { | |
1438 | *hole_size = *hole_start + *hole_size - pos; | |
1439 | *hole_start = pos; | |
1440 | changed = true; | |
1441 | if (*hole_size < num_bytes) | |
1442 | break; | |
1443 | } | |
1444 | ||
1445 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); | |
1446 | ||
1447 | /* Range is ensured to be empty */ | |
1448 | if (!ret) | |
1449 | return changed; | |
1450 | ||
1451 | /* Given hole range was invalid (outside of device) */ | |
1452 | if (ret == -ERANGE) { | |
1453 | *hole_start += *hole_size; | |
d6f67afb | 1454 | *hole_size = 0; |
7000babd | 1455 | return true; |
1cd6121f NA |
1456 | } |
1457 | ||
1458 | *hole_start += zone_size; | |
1459 | *hole_size -= zone_size; | |
1460 | changed = true; | |
1461 | } | |
1462 | ||
1463 | return changed; | |
1464 | } | |
1465 | ||
3b4ffa40 NA |
1466 | /** |
1467 | * dev_extent_hole_check - check if specified hole is suitable for allocation | |
1468 | * @device: the device which we have the hole | |
1469 | * @hole_start: starting position of the hole | |
1470 | * @hole_size: the size of the hole | |
1471 | * @num_bytes: the size of the free space that we need | |
1472 | * | |
1cd6121f | 1473 | * This function may modify @hole_start and @hole_size to reflect the suitable |
3b4ffa40 NA |
1474 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
1475 | */ | |
1476 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1477 | u64 *hole_size, u64 num_bytes) | |
1478 | { | |
1479 | bool changed = false; | |
1480 | u64 hole_end = *hole_start + *hole_size; | |
1481 | ||
1cd6121f NA |
1482 | for (;;) { |
1483 | /* | |
1484 | * Check before we set max_hole_start, otherwise we could end up | |
1485 | * sending back this offset anyway. | |
1486 | */ | |
1487 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1488 | if (hole_end >= *hole_start) | |
1489 | *hole_size = hole_end - *hole_start; | |
1490 | else | |
1491 | *hole_size = 0; | |
1492 | changed = true; | |
1493 | } | |
1494 | ||
1495 | switch (device->fs_devices->chunk_alloc_policy) { | |
1496 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1497 | /* No extra check */ | |
1498 | break; | |
1499 | case BTRFS_CHUNK_ALLOC_ZONED: | |
1500 | if (dev_extent_hole_check_zoned(device, hole_start, | |
1501 | hole_size, num_bytes)) { | |
1502 | changed = true; | |
1503 | /* | |
1504 | * The changed hole can contain pending extent. | |
1505 | * Loop again to check that. | |
1506 | */ | |
1507 | continue; | |
1508 | } | |
1509 | break; | |
1510 | default: | |
1511 | BUG(); | |
1512 | } | |
3b4ffa40 | 1513 | |
3b4ffa40 | 1514 | break; |
3b4ffa40 NA |
1515 | } |
1516 | ||
1517 | return changed; | |
1518 | } | |
6df9a95e | 1519 | |
0b86a832 | 1520 | /* |
499f377f JM |
1521 | * find_free_dev_extent_start - find free space in the specified device |
1522 | * @device: the device which we search the free space in | |
1523 | * @num_bytes: the size of the free space that we need | |
1524 | * @search_start: the position from which to begin the search | |
1525 | * @start: store the start of the free space. | |
1526 | * @len: the size of the free space. that we find, or the size | |
1527 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1528 | * |
0b86a832 CM |
1529 | * this uses a pretty simple search, the expectation is that it is |
1530 | * called very infrequently and that a given device has a small number | |
1531 | * of extents | |
7bfc837d MX |
1532 | * |
1533 | * @start is used to store the start of the free space if we find. But if we | |
1534 | * don't find suitable free space, it will be used to store the start position | |
1535 | * of the max free space. | |
1536 | * | |
1537 | * @len is used to store the size of the free space that we find. | |
1538 | * But if we don't find suitable free space, it is used to store the size of | |
1539 | * the max free space. | |
135da976 QW |
1540 | * |
1541 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1542 | * check to ensure dev extents are not double allocated. | |
1543 | * This makes the function safe to allocate dev extents but may not report | |
1544 | * correct usable device space, as device extent freed in current transaction | |
1a9fd417 | 1545 | * is not reported as available. |
0b86a832 | 1546 | */ |
9e3246a5 QW |
1547 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1548 | u64 num_bytes, u64 search_start, u64 *start, | |
1549 | u64 *len) | |
0b86a832 | 1550 | { |
0b246afa JM |
1551 | struct btrfs_fs_info *fs_info = device->fs_info; |
1552 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1553 | struct btrfs_key key; |
7bfc837d | 1554 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1555 | struct btrfs_path *path; |
7bfc837d MX |
1556 | u64 hole_size; |
1557 | u64 max_hole_start; | |
1558 | u64 max_hole_size; | |
1559 | u64 extent_end; | |
0b86a832 CM |
1560 | u64 search_end = device->total_bytes; |
1561 | int ret; | |
7bfc837d | 1562 | int slot; |
0b86a832 | 1563 | struct extent_buffer *l; |
8cdc7c5b | 1564 | |
3b4ffa40 | 1565 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1566 | |
1cd6121f NA |
1567 | WARN_ON(device->zone_info && |
1568 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); | |
1569 | ||
6df9a95e JB |
1570 | path = btrfs_alloc_path(); |
1571 | if (!path) | |
1572 | return -ENOMEM; | |
f2ab7618 | 1573 | |
7bfc837d MX |
1574 | max_hole_start = search_start; |
1575 | max_hole_size = 0; | |
1576 | ||
f2ab7618 | 1577 | again: |
401e29c1 AJ |
1578 | if (search_start >= search_end || |
1579 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1580 | ret = -ENOSPC; |
6df9a95e | 1581 | goto out; |
7bfc837d MX |
1582 | } |
1583 | ||
e4058b54 | 1584 | path->reada = READA_FORWARD; |
6df9a95e JB |
1585 | path->search_commit_root = 1; |
1586 | path->skip_locking = 1; | |
7bfc837d | 1587 | |
0b86a832 CM |
1588 | key.objectid = device->devid; |
1589 | key.offset = search_start; | |
1590 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1591 | |
0ff40a91 | 1592 | ret = btrfs_search_backwards(root, &key, path); |
0b86a832 | 1593 | if (ret < 0) |
7bfc837d | 1594 | goto out; |
7bfc837d | 1595 | |
0b86a832 CM |
1596 | while (1) { |
1597 | l = path->nodes[0]; | |
1598 | slot = path->slots[0]; | |
1599 | if (slot >= btrfs_header_nritems(l)) { | |
1600 | ret = btrfs_next_leaf(root, path); | |
1601 | if (ret == 0) | |
1602 | continue; | |
1603 | if (ret < 0) | |
7bfc837d MX |
1604 | goto out; |
1605 | ||
1606 | break; | |
0b86a832 CM |
1607 | } |
1608 | btrfs_item_key_to_cpu(l, &key, slot); | |
1609 | ||
1610 | if (key.objectid < device->devid) | |
1611 | goto next; | |
1612 | ||
1613 | if (key.objectid > device->devid) | |
7bfc837d | 1614 | break; |
0b86a832 | 1615 | |
962a298f | 1616 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1617 | goto next; |
9779b72f | 1618 | |
7bfc837d MX |
1619 | if (key.offset > search_start) { |
1620 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1621 | dev_extent_hole_check(device, &search_start, &hole_size, |
1622 | num_bytes); | |
6df9a95e | 1623 | |
7bfc837d MX |
1624 | if (hole_size > max_hole_size) { |
1625 | max_hole_start = search_start; | |
1626 | max_hole_size = hole_size; | |
1627 | } | |
9779b72f | 1628 | |
7bfc837d MX |
1629 | /* |
1630 | * If this free space is greater than which we need, | |
1631 | * it must be the max free space that we have found | |
1632 | * until now, so max_hole_start must point to the start | |
1633 | * of this free space and the length of this free space | |
1634 | * is stored in max_hole_size. Thus, we return | |
1635 | * max_hole_start and max_hole_size and go back to the | |
1636 | * caller. | |
1637 | */ | |
1638 | if (hole_size >= num_bytes) { | |
1639 | ret = 0; | |
1640 | goto out; | |
0b86a832 CM |
1641 | } |
1642 | } | |
0b86a832 | 1643 | |
0b86a832 | 1644 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1645 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1646 | dev_extent); | |
1647 | if (extent_end > search_start) | |
1648 | search_start = extent_end; | |
0b86a832 CM |
1649 | next: |
1650 | path->slots[0]++; | |
1651 | cond_resched(); | |
1652 | } | |
0b86a832 | 1653 | |
38c01b96 | 1654 | /* |
1655 | * At this point, search_start should be the end of | |
1656 | * allocated dev extents, and when shrinking the device, | |
1657 | * search_end may be smaller than search_start. | |
1658 | */ | |
f2ab7618 | 1659 | if (search_end > search_start) { |
38c01b96 | 1660 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1661 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1662 | num_bytes)) { | |
f2ab7618 ZL |
1663 | btrfs_release_path(path); |
1664 | goto again; | |
1665 | } | |
0b86a832 | 1666 | |
f2ab7618 ZL |
1667 | if (hole_size > max_hole_size) { |
1668 | max_hole_start = search_start; | |
1669 | max_hole_size = hole_size; | |
1670 | } | |
6df9a95e JB |
1671 | } |
1672 | ||
7bfc837d | 1673 | /* See above. */ |
f2ab7618 | 1674 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1675 | ret = -ENOSPC; |
1676 | else | |
1677 | ret = 0; | |
1678 | ||
1679 | out: | |
2b82032c | 1680 | btrfs_free_path(path); |
7bfc837d | 1681 | *start = max_hole_start; |
b2117a39 | 1682 | if (len) |
7bfc837d | 1683 | *len = max_hole_size; |
0b86a832 CM |
1684 | return ret; |
1685 | } | |
1686 | ||
60dfdf25 | 1687 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1688 | u64 *start, u64 *len) |
1689 | { | |
499f377f | 1690 | /* FIXME use last free of some kind */ |
60dfdf25 | 1691 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1692 | } |
1693 | ||
b2950863 | 1694 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1695 | struct btrfs_device *device, |
2196d6e8 | 1696 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1697 | { |
0b246afa JM |
1698 | struct btrfs_fs_info *fs_info = device->fs_info; |
1699 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1700 | int ret; |
1701 | struct btrfs_path *path; | |
8f18cf13 | 1702 | struct btrfs_key key; |
a061fc8d CM |
1703 | struct btrfs_key found_key; |
1704 | struct extent_buffer *leaf = NULL; | |
1705 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1706 | |
1707 | path = btrfs_alloc_path(); | |
1708 | if (!path) | |
1709 | return -ENOMEM; | |
1710 | ||
1711 | key.objectid = device->devid; | |
1712 | key.offset = start; | |
1713 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1714 | again: |
8f18cf13 | 1715 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1716 | if (ret > 0) { |
1717 | ret = btrfs_previous_item(root, path, key.objectid, | |
1718 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1719 | if (ret) |
1720 | goto out; | |
a061fc8d CM |
1721 | leaf = path->nodes[0]; |
1722 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1723 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1724 | struct btrfs_dev_extent); | |
1725 | BUG_ON(found_key.offset > start || found_key.offset + | |
1726 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1727 | key = found_key; |
1728 | btrfs_release_path(path); | |
1729 | goto again; | |
a061fc8d CM |
1730 | } else if (ret == 0) { |
1731 | leaf = path->nodes[0]; | |
1732 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1733 | struct btrfs_dev_extent); | |
79787eaa | 1734 | } else { |
79787eaa | 1735 | goto out; |
a061fc8d | 1736 | } |
8f18cf13 | 1737 | |
2196d6e8 MX |
1738 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1739 | ||
8f18cf13 | 1740 | ret = btrfs_del_item(trans, root, path); |
79bd3712 | 1741 | if (ret == 0) |
3204d33c | 1742 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
b0b802d7 | 1743 | out: |
8f18cf13 CM |
1744 | btrfs_free_path(path); |
1745 | return ret; | |
1746 | } | |
1747 | ||
6df9a95e | 1748 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1749 | { |
6df9a95e JB |
1750 | struct extent_map_tree *em_tree; |
1751 | struct extent_map *em; | |
1752 | struct rb_node *n; | |
1753 | u64 ret = 0; | |
0b86a832 | 1754 | |
c8bf1b67 | 1755 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1756 | read_lock(&em_tree->lock); |
07e1ce09 | 1757 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1758 | if (n) { |
1759 | em = rb_entry(n, struct extent_map, rb_node); | |
1760 | ret = em->start + em->len; | |
0b86a832 | 1761 | } |
6df9a95e JB |
1762 | read_unlock(&em_tree->lock); |
1763 | ||
0b86a832 CM |
1764 | return ret; |
1765 | } | |
1766 | ||
53f10659 ID |
1767 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1768 | u64 *devid_ret) | |
0b86a832 CM |
1769 | { |
1770 | int ret; | |
1771 | struct btrfs_key key; | |
1772 | struct btrfs_key found_key; | |
2b82032c YZ |
1773 | struct btrfs_path *path; |
1774 | ||
2b82032c YZ |
1775 | path = btrfs_alloc_path(); |
1776 | if (!path) | |
1777 | return -ENOMEM; | |
0b86a832 CM |
1778 | |
1779 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1780 | key.type = BTRFS_DEV_ITEM_KEY; | |
1781 | key.offset = (u64)-1; | |
1782 | ||
53f10659 | 1783 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1784 | if (ret < 0) |
1785 | goto error; | |
1786 | ||
a06dee4d AJ |
1787 | if (ret == 0) { |
1788 | /* Corruption */ | |
1789 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1790 | ret = -EUCLEAN; | |
1791 | goto error; | |
1792 | } | |
0b86a832 | 1793 | |
53f10659 ID |
1794 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1795 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1796 | BTRFS_DEV_ITEM_KEY); |
1797 | if (ret) { | |
53f10659 | 1798 | *devid_ret = 1; |
0b86a832 CM |
1799 | } else { |
1800 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1801 | path->slots[0]); | |
53f10659 | 1802 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1803 | } |
1804 | ret = 0; | |
1805 | error: | |
2b82032c | 1806 | btrfs_free_path(path); |
0b86a832 CM |
1807 | return ret; |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * the device information is stored in the chunk root | |
1812 | * the btrfs_device struct should be fully filled in | |
1813 | */ | |
c74a0b02 | 1814 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1815 | struct btrfs_device *device) |
0b86a832 CM |
1816 | { |
1817 | int ret; | |
1818 | struct btrfs_path *path; | |
1819 | struct btrfs_dev_item *dev_item; | |
1820 | struct extent_buffer *leaf; | |
1821 | struct btrfs_key key; | |
1822 | unsigned long ptr; | |
0b86a832 | 1823 | |
0b86a832 CM |
1824 | path = btrfs_alloc_path(); |
1825 | if (!path) | |
1826 | return -ENOMEM; | |
1827 | ||
0b86a832 CM |
1828 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1829 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1830 | key.offset = device->devid; |
0b86a832 | 1831 | |
8e87e856 NB |
1832 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1833 | &key, sizeof(*dev_item)); | |
0b86a832 CM |
1834 | if (ret) |
1835 | goto out; | |
1836 | ||
1837 | leaf = path->nodes[0]; | |
1838 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1839 | ||
1840 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1841 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1842 | btrfs_set_device_type(leaf, dev_item, device->type); |
1843 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1844 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1845 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1846 | btrfs_set_device_total_bytes(leaf, dev_item, |
1847 | btrfs_device_get_disk_total_bytes(device)); | |
1848 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1849 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1850 | btrfs_set_device_group(leaf, dev_item, 0); |
1851 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1852 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1853 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1854 | |
410ba3a2 | 1855 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1856 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1857 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1858 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1859 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1860 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1861 | |
2b82032c | 1862 | ret = 0; |
0b86a832 CM |
1863 | out: |
1864 | btrfs_free_path(path); | |
1865 | return ret; | |
1866 | } | |
8f18cf13 | 1867 | |
5a1972bd QW |
1868 | /* |
1869 | * Function to update ctime/mtime for a given device path. | |
1870 | * Mainly used for ctime/mtime based probe like libblkid. | |
1871 | */ | |
da353f6b | 1872 | static void update_dev_time(const char *path_name) |
5a1972bd QW |
1873 | { |
1874 | struct file *filp; | |
1875 | ||
1876 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1877 | if (IS_ERR(filp)) |
5a1972bd QW |
1878 | return; |
1879 | file_update_time(filp); | |
1880 | filp_close(filp, NULL); | |
5a1972bd QW |
1881 | } |
1882 | ||
f331a952 | 1883 | static int btrfs_rm_dev_item(struct btrfs_device *device) |
a061fc8d | 1884 | { |
f331a952 | 1885 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1886 | int ret; |
1887 | struct btrfs_path *path; | |
a061fc8d | 1888 | struct btrfs_key key; |
a061fc8d CM |
1889 | struct btrfs_trans_handle *trans; |
1890 | ||
a061fc8d CM |
1891 | path = btrfs_alloc_path(); |
1892 | if (!path) | |
1893 | return -ENOMEM; | |
1894 | ||
a22285a6 | 1895 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1896 | if (IS_ERR(trans)) { |
1897 | btrfs_free_path(path); | |
1898 | return PTR_ERR(trans); | |
1899 | } | |
a061fc8d CM |
1900 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1901 | key.type = BTRFS_DEV_ITEM_KEY; | |
1902 | key.offset = device->devid; | |
1903 | ||
1904 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
5e9f2ad5 NB |
1905 | if (ret) { |
1906 | if (ret > 0) | |
1907 | ret = -ENOENT; | |
1908 | btrfs_abort_transaction(trans, ret); | |
1909 | btrfs_end_transaction(trans); | |
a061fc8d CM |
1910 | goto out; |
1911 | } | |
1912 | ||
1913 | ret = btrfs_del_item(trans, root, path); | |
5e9f2ad5 NB |
1914 | if (ret) { |
1915 | btrfs_abort_transaction(trans, ret); | |
1916 | btrfs_end_transaction(trans); | |
1917 | } | |
1918 | ||
a061fc8d CM |
1919 | out: |
1920 | btrfs_free_path(path); | |
5e9f2ad5 NB |
1921 | if (!ret) |
1922 | ret = btrfs_commit_transaction(trans); | |
a061fc8d CM |
1923 | return ret; |
1924 | } | |
1925 | ||
3cc31a0d DS |
1926 | /* |
1927 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1928 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1929 | * replace. | |
1930 | */ | |
1931 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1932 | u64 num_devices) | |
a061fc8d | 1933 | { |
a061fc8d | 1934 | u64 all_avail; |
de98ced9 | 1935 | unsigned seq; |
418775a2 | 1936 | int i; |
a061fc8d | 1937 | |
de98ced9 | 1938 | do { |
bd45ffbc | 1939 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1940 | |
bd45ffbc AJ |
1941 | all_avail = fs_info->avail_data_alloc_bits | |
1942 | fs_info->avail_system_alloc_bits | | |
1943 | fs_info->avail_metadata_alloc_bits; | |
1944 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1945 | |
418775a2 | 1946 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1947 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1948 | continue; |
a061fc8d | 1949 | |
efc222f8 AJ |
1950 | if (num_devices < btrfs_raid_array[i].devs_min) |
1951 | return btrfs_raid_array[i].mindev_error; | |
53b381b3 DW |
1952 | } |
1953 | ||
bd45ffbc | 1954 | return 0; |
f1fa7f26 AJ |
1955 | } |
1956 | ||
c9162bdf OS |
1957 | static struct btrfs_device * btrfs_find_next_active_device( |
1958 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1959 | { |
2b82032c | 1960 | struct btrfs_device *next_device; |
88acff64 AJ |
1961 | |
1962 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1963 | if (next_device != device && | |
e6e674bd AJ |
1964 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1965 | && next_device->bdev) | |
88acff64 AJ |
1966 | return next_device; |
1967 | } | |
1968 | ||
1969 | return NULL; | |
1970 | } | |
1971 | ||
1972 | /* | |
1973 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1974 | * and replace it with the provided or the next active device, in the context | |
1975 | * where this function called, there should be always be another device (or | |
1976 | * this_dev) which is active. | |
1977 | */ | |
b105e927 | 1978 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
e493e8f9 | 1979 | struct btrfs_device *next_device) |
88acff64 | 1980 | { |
d6507cf1 | 1981 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 | 1982 | |
e493e8f9 | 1983 | if (!next_device) |
88acff64 | 1984 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
e493e8f9 | 1985 | device); |
88acff64 AJ |
1986 | ASSERT(next_device); |
1987 | ||
1988 | if (fs_info->sb->s_bdev && | |
1989 | (fs_info->sb->s_bdev == device->bdev)) | |
1990 | fs_info->sb->s_bdev = next_device->bdev; | |
1991 | ||
1992 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1993 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1994 | } | |
1995 | ||
1da73967 AJ |
1996 | /* |
1997 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
1998 | * currently replaced. | |
1999 | */ | |
2000 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2001 | { | |
2002 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2003 | ||
cb5583dd | 2004 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2005 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2006 | ASSERT(num_devices > 1); | |
2007 | num_devices--; | |
2008 | } | |
cb5583dd | 2009 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2010 | |
2011 | return num_devices; | |
2012 | } | |
2013 | ||
313b0858 JB |
2014 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2015 | struct block_device *bdev, | |
2016 | const char *device_path) | |
6fbceb9f | 2017 | { |
6fbceb9f JT |
2018 | struct btrfs_super_block *disk_super; |
2019 | int copy_num; | |
2020 | ||
2021 | if (!bdev) | |
2022 | return; | |
2023 | ||
2024 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
8f32380d JT |
2025 | struct page *page; |
2026 | int ret; | |
6fbceb9f | 2027 | |
8f32380d JT |
2028 | disk_super = btrfs_read_dev_one_super(bdev, copy_num); |
2029 | if (IS_ERR(disk_super)) | |
2030 | continue; | |
6fbceb9f | 2031 | |
12659251 NA |
2032 | if (bdev_is_zoned(bdev)) { |
2033 | btrfs_reset_sb_log_zones(bdev, copy_num); | |
2034 | continue; | |
2035 | } | |
2036 | ||
6fbceb9f | 2037 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); |
8f32380d JT |
2038 | |
2039 | page = virt_to_page(disk_super); | |
2040 | set_page_dirty(page); | |
2041 | lock_page(page); | |
2042 | /* write_on_page() unlocks the page */ | |
2043 | ret = write_one_page(page); | |
2044 | if (ret) | |
2045 | btrfs_warn(fs_info, | |
2046 | "error clearing superblock number %d (%d)", | |
2047 | copy_num, ret); | |
2048 | btrfs_release_disk_super(disk_super); | |
2049 | ||
6fbceb9f JT |
2050 | } |
2051 | ||
2052 | /* Notify udev that device has changed */ | |
2053 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2054 | ||
2055 | /* Update ctime/mtime for device path for libblkid */ | |
2056 | update_dev_time(device_path); | |
2057 | } | |
2058 | ||
da353f6b | 2059 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path, |
53f8a74c | 2060 | u64 devid) |
f1fa7f26 AJ |
2061 | { |
2062 | struct btrfs_device *device; | |
1f78160c | 2063 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2064 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2065 | u64 num_devices; |
a061fc8d CM |
2066 | int ret = 0; |
2067 | ||
a061fc8d CM |
2068 | mutex_lock(&uuid_mutex); |
2069 | ||
1da73967 | 2070 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2071 | |
0b246afa | 2072 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2073 | if (ret) |
a061fc8d | 2074 | goto out; |
a061fc8d | 2075 | |
a27a94c2 NB |
2076 | device = btrfs_find_device_by_devspec(fs_info, devid, device_path); |
2077 | ||
2078 | if (IS_ERR(device)) { | |
2079 | if (PTR_ERR(device) == -ENOENT && | |
e4571b8c | 2080 | device_path && strcmp(device_path, "missing") == 0) |
a27a94c2 NB |
2081 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
2082 | else | |
2083 | ret = PTR_ERR(device); | |
53b381b3 | 2084 | goto out; |
a27a94c2 | 2085 | } |
dfe25020 | 2086 | |
eede2bf3 OS |
2087 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2088 | btrfs_warn_in_rcu(fs_info, | |
2089 | "cannot remove device %s (devid %llu) due to active swapfile", | |
2090 | rcu_str_deref(device->name), device->devid); | |
2091 | ret = -ETXTBSY; | |
2092 | goto out; | |
2093 | } | |
2094 | ||
401e29c1 | 2095 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
183860f6 | 2096 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 2097 | goto out; |
63a212ab SB |
2098 | } |
2099 | ||
ebbede42 AJ |
2100 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
2101 | fs_info->fs_devices->rw_devices == 1) { | |
183860f6 | 2102 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 2103 | goto out; |
2b82032c YZ |
2104 | } |
2105 | ||
ebbede42 | 2106 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2107 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2108 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2109 | device->fs_devices->rw_devices--; |
34441361 | 2110 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2111 | } |
a061fc8d | 2112 | |
d7901554 | 2113 | mutex_unlock(&uuid_mutex); |
a061fc8d | 2114 | ret = btrfs_shrink_device(device, 0); |
66d204a1 FM |
2115 | if (!ret) |
2116 | btrfs_reada_remove_dev(device); | |
d7901554 | 2117 | mutex_lock(&uuid_mutex); |
a061fc8d | 2118 | if (ret) |
9b3517e9 | 2119 | goto error_undo; |
a061fc8d | 2120 | |
63a212ab SB |
2121 | /* |
2122 | * TODO: the superblock still includes this device in its num_devices | |
2123 | * counter although write_all_supers() is not locked out. This | |
2124 | * could give a filesystem state which requires a degraded mount. | |
2125 | */ | |
f331a952 | 2126 | ret = btrfs_rm_dev_item(device); |
a061fc8d | 2127 | if (ret) |
9b3517e9 | 2128 | goto error_undo; |
a061fc8d | 2129 | |
e12c9621 | 2130 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2131 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2132 | |
2133 | /* | |
2134 | * the device list mutex makes sure that we don't change | |
2135 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2136 | * the device supers. Whoever is writing all supers, should |
2137 | * lock the device list mutex before getting the number of | |
2138 | * devices in the super block (super_copy). Conversely, | |
2139 | * whoever updates the number of devices in the super block | |
2140 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2141 | */ |
1f78160c | 2142 | |
41a52a0f AJ |
2143 | /* |
2144 | * In normal cases the cur_devices == fs_devices. But in case | |
2145 | * of deleting a seed device, the cur_devices should point to | |
2146 | * its own fs_devices listed under the fs_devices->seed. | |
2147 | */ | |
1f78160c | 2148 | cur_devices = device->fs_devices; |
b5185197 | 2149 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2150 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2151 | |
41a52a0f AJ |
2152 | cur_devices->num_devices--; |
2153 | cur_devices->total_devices--; | |
b4993e64 AJ |
2154 | /* Update total_devices of the parent fs_devices if it's seed */ |
2155 | if (cur_devices != fs_devices) | |
2156 | fs_devices->total_devices--; | |
2b82032c | 2157 | |
e6e674bd | 2158 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2159 | cur_devices->missing_devices--; |
cd02dca5 | 2160 | |
d6507cf1 | 2161 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2162 | |
0bfaa9c5 | 2163 | if (device->bdev) { |
41a52a0f | 2164 | cur_devices->open_devices--; |
0bfaa9c5 | 2165 | /* remove sysfs entry */ |
53f8a74c | 2166 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2167 | } |
99994cde | 2168 | |
0b246afa JM |
2169 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2170 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2171 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2172 | |
cea67ab9 JM |
2173 | /* |
2174 | * at this point, the device is zero sized and detached from | |
2175 | * the devices list. All that's left is to zero out the old | |
2176 | * supers and free the device. | |
2177 | */ | |
ebbede42 | 2178 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
8f32380d JT |
2179 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2180 | device->name->str); | |
cea67ab9 JM |
2181 | |
2182 | btrfs_close_bdev(device); | |
8e75fd89 NB |
2183 | synchronize_rcu(); |
2184 | btrfs_free_device(device); | |
cea67ab9 | 2185 | |
1f78160c | 2186 | if (cur_devices->open_devices == 0) { |
944d3f9f | 2187 | list_del_init(&cur_devices->seed_list); |
0226e0eb | 2188 | close_fs_devices(cur_devices); |
1f78160c | 2189 | free_fs_devices(cur_devices); |
2b82032c YZ |
2190 | } |
2191 | ||
a061fc8d CM |
2192 | out: |
2193 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 2194 | return ret; |
24fc572f | 2195 | |
9b3517e9 | 2196 | error_undo: |
66d204a1 | 2197 | btrfs_reada_undo_remove_dev(device); |
ebbede42 | 2198 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2199 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2200 | list_add(&device->dev_alloc_list, |
b5185197 | 2201 | &fs_devices->alloc_list); |
c3929c36 | 2202 | device->fs_devices->rw_devices++; |
34441361 | 2203 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2204 | } |
24fc572f | 2205 | goto out; |
a061fc8d CM |
2206 | } |
2207 | ||
68a9db5f | 2208 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2209 | { |
d51908ce AJ |
2210 | struct btrfs_fs_devices *fs_devices; |
2211 | ||
68a9db5f | 2212 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2213 | |
25e8e911 AJ |
2214 | /* |
2215 | * in case of fs with no seed, srcdev->fs_devices will point | |
2216 | * to fs_devices of fs_info. However when the dev being replaced is | |
2217 | * a seed dev it will point to the seed's local fs_devices. In short | |
2218 | * srcdev will have its correct fs_devices in both the cases. | |
2219 | */ | |
2220 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2221 | |
e93c89c1 | 2222 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2223 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2224 | fs_devices->num_devices--; |
e6e674bd | 2225 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2226 | fs_devices->missing_devices--; |
e93c89c1 | 2227 | |
ebbede42 | 2228 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2229 | fs_devices->rw_devices--; |
1357272f | 2230 | |
82372bc8 | 2231 | if (srcdev->bdev) |
d51908ce | 2232 | fs_devices->open_devices--; |
084b6e7c QW |
2233 | } |
2234 | ||
65237ee3 | 2235 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2236 | { |
2237 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2238 | |
a466c85e JB |
2239 | mutex_lock(&uuid_mutex); |
2240 | ||
14238819 | 2241 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2242 | synchronize_rcu(); |
2243 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2244 | |
94d5f0c2 AJ |
2245 | /* if this is no devs we rather delete the fs_devices */ |
2246 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2247 | /* |
2248 | * On a mounted FS, num_devices can't be zero unless it's a | |
2249 | * seed. In case of a seed device being replaced, the replace | |
2250 | * target added to the sprout FS, so there will be no more | |
2251 | * device left under the seed FS. | |
2252 | */ | |
2253 | ASSERT(fs_devices->seeding); | |
2254 | ||
944d3f9f | 2255 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2256 | close_fs_devices(fs_devices); |
8bef8401 | 2257 | free_fs_devices(fs_devices); |
94d5f0c2 | 2258 | } |
a466c85e | 2259 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2260 | } |
2261 | ||
4f5ad7bd | 2262 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2263 | { |
4f5ad7bd | 2264 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2265 | |
d9a071f0 | 2266 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2267 | |
53f8a74c | 2268 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2269 | |
779bf3fe | 2270 | if (tgtdev->bdev) |
d9a071f0 | 2271 | fs_devices->open_devices--; |
779bf3fe | 2272 | |
d9a071f0 | 2273 | fs_devices->num_devices--; |
e93c89c1 | 2274 | |
d6507cf1 | 2275 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2276 | |
e93c89c1 | 2277 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2278 | |
d9a071f0 | 2279 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe AJ |
2280 | |
2281 | /* | |
2282 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2283 | * may lead to a call to btrfs_show_devname() which will try | |
2284 | * to hold device_list_mutex. And here this device | |
2285 | * is already out of device list, so we don't have to hold | |
2286 | * the device_list_mutex lock. | |
2287 | */ | |
8f32380d JT |
2288 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2289 | tgtdev->name->str); | |
14238819 AJ |
2290 | |
2291 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2292 | synchronize_rcu(); |
2293 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2294 | } |
2295 | ||
b444ad46 NB |
2296 | static struct btrfs_device *btrfs_find_device_by_path( |
2297 | struct btrfs_fs_info *fs_info, const char *device_path) | |
7ba15b7d SB |
2298 | { |
2299 | int ret = 0; | |
2300 | struct btrfs_super_block *disk_super; | |
2301 | u64 devid; | |
2302 | u8 *dev_uuid; | |
2303 | struct block_device *bdev; | |
b444ad46 | 2304 | struct btrfs_device *device; |
7ba15b7d | 2305 | |
7ba15b7d | 2306 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, |
8f32380d | 2307 | fs_info->bdev_holder, 0, &bdev, &disk_super); |
7ba15b7d | 2308 | if (ret) |
b444ad46 | 2309 | return ERR_PTR(ret); |
8f32380d | 2310 | |
7ba15b7d SB |
2311 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2312 | dev_uuid = disk_super->dev_item.uuid; | |
7239ff4b | 2313 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
e4319cd9 | 2314 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2315 | disk_super->metadata_uuid); |
7239ff4b | 2316 | else |
e4319cd9 | 2317 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2318 | disk_super->fsid); |
7239ff4b | 2319 | |
8f32380d | 2320 | btrfs_release_disk_super(disk_super); |
b444ad46 NB |
2321 | if (!device) |
2322 | device = ERR_PTR(-ENOENT); | |
7ba15b7d | 2323 | blkdev_put(bdev, FMODE_READ); |
b444ad46 | 2324 | return device; |
7ba15b7d SB |
2325 | } |
2326 | ||
5c5c0df0 DS |
2327 | /* |
2328 | * Lookup a device given by device id, or the path if the id is 0. | |
2329 | */ | |
a27a94c2 | 2330 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2331 | struct btrfs_fs_info *fs_info, u64 devid, |
2332 | const char *device_path) | |
24e0474b | 2333 | { |
a27a94c2 | 2334 | struct btrfs_device *device; |
24e0474b | 2335 | |
5c5c0df0 | 2336 | if (devid) { |
e4319cd9 | 2337 | device = btrfs_find_device(fs_info->fs_devices, devid, NULL, |
b2598edf | 2338 | NULL); |
a27a94c2 NB |
2339 | if (!device) |
2340 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2341 | return device; |
2342 | } | |
2343 | ||
2344 | if (!device_path || !device_path[0]) | |
2345 | return ERR_PTR(-EINVAL); | |
2346 | ||
2347 | if (strcmp(device_path, "missing") == 0) { | |
2348 | /* Find first missing device */ | |
2349 | list_for_each_entry(device, &fs_info->fs_devices->devices, | |
2350 | dev_list) { | |
2351 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, | |
2352 | &device->dev_state) && !device->bdev) | |
2353 | return device; | |
d95a830c | 2354 | } |
6e927ceb | 2355 | return ERR_PTR(-ENOENT); |
24e0474b | 2356 | } |
6e927ceb AJ |
2357 | |
2358 | return btrfs_find_device_by_path(fs_info, device_path); | |
24e0474b AJ |
2359 | } |
2360 | ||
2b82032c YZ |
2361 | /* |
2362 | * does all the dirty work required for changing file system's UUID. | |
2363 | */ | |
2ff7e61e | 2364 | static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2365 | { |
0b246afa | 2366 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2367 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2368 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2369 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2370 | struct btrfs_device *device; |
2371 | u64 super_flags; | |
2372 | ||
a32bf9a3 | 2373 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2374 | if (!fs_devices->seeding) |
2b82032c YZ |
2375 | return -EINVAL; |
2376 | ||
427c8fdd NB |
2377 | /* |
2378 | * Private copy of the seed devices, anchored at | |
2379 | * fs_info->fs_devices->seed_list | |
2380 | */ | |
7239ff4b | 2381 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 ID |
2382 | if (IS_ERR(seed_devices)) |
2383 | return PTR_ERR(seed_devices); | |
2b82032c | 2384 | |
427c8fdd NB |
2385 | /* |
2386 | * It's necessary to retain a copy of the original seed fs_devices in | |
2387 | * fs_uuids so that filesystems which have been seeded can successfully | |
2388 | * reference the seed device from open_seed_devices. This also supports | |
2389 | * multiple fs seed. | |
2390 | */ | |
e4404d6e YZ |
2391 | old_devices = clone_fs_devices(fs_devices); |
2392 | if (IS_ERR(old_devices)) { | |
2393 | kfree(seed_devices); | |
2394 | return PTR_ERR(old_devices); | |
2b82032c | 2395 | } |
e4404d6e | 2396 | |
c4babc5e | 2397 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2398 | |
e4404d6e YZ |
2399 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2400 | seed_devices->opened = 1; | |
2401 | INIT_LIST_HEAD(&seed_devices->devices); | |
2402 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2403 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2404 | |
321a4bf7 | 2405 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c XG |
2406 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2407 | synchronize_rcu); | |
2196d6e8 MX |
2408 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2409 | device->fs_devices = seed_devices; | |
c9513edb | 2410 | |
0395d84f | 2411 | fs_devices->seeding = false; |
2b82032c YZ |
2412 | fs_devices->num_devices = 0; |
2413 | fs_devices->open_devices = 0; | |
69611ac8 | 2414 | fs_devices->missing_devices = 0; |
7f0432d0 | 2415 | fs_devices->rotating = false; |
944d3f9f | 2416 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2417 | |
2418 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2419 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2420 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
321a4bf7 | 2421 | mutex_unlock(&fs_devices->device_list_mutex); |
f7171750 | 2422 | |
2b82032c YZ |
2423 | super_flags = btrfs_super_flags(disk_super) & |
2424 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2425 | btrfs_set_super_flags(disk_super, super_flags); | |
2426 | ||
2427 | return 0; | |
2428 | } | |
2429 | ||
2430 | /* | |
01327610 | 2431 | * Store the expected generation for seed devices in device items. |
2b82032c | 2432 | */ |
5c466629 | 2433 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2434 | { |
5c466629 | 2435 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2436 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2437 | struct btrfs_path *path; |
2438 | struct extent_buffer *leaf; | |
2439 | struct btrfs_dev_item *dev_item; | |
2440 | struct btrfs_device *device; | |
2441 | struct btrfs_key key; | |
44880fdc | 2442 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c YZ |
2443 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2444 | u64 devid; | |
2445 | int ret; | |
2446 | ||
2447 | path = btrfs_alloc_path(); | |
2448 | if (!path) | |
2449 | return -ENOMEM; | |
2450 | ||
2b82032c YZ |
2451 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2452 | key.offset = 0; | |
2453 | key.type = BTRFS_DEV_ITEM_KEY; | |
2454 | ||
2455 | while (1) { | |
2456 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2457 | if (ret < 0) | |
2458 | goto error; | |
2459 | ||
2460 | leaf = path->nodes[0]; | |
2461 | next_slot: | |
2462 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2463 | ret = btrfs_next_leaf(root, path); | |
2464 | if (ret > 0) | |
2465 | break; | |
2466 | if (ret < 0) | |
2467 | goto error; | |
2468 | leaf = path->nodes[0]; | |
2469 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2470 | btrfs_release_path(path); |
2b82032c YZ |
2471 | continue; |
2472 | } | |
2473 | ||
2474 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2475 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2476 | key.type != BTRFS_DEV_ITEM_KEY) | |
2477 | break; | |
2478 | ||
2479 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2480 | struct btrfs_dev_item); | |
2481 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2482 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2483 | BTRFS_UUID_SIZE); |
1473b24e | 2484 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2485 | BTRFS_FSID_SIZE); |
e4319cd9 | 2486 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 2487 | fs_uuid); |
79787eaa | 2488 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2489 | |
2490 | if (device->fs_devices->seeding) { | |
2491 | btrfs_set_device_generation(leaf, dev_item, | |
2492 | device->generation); | |
2493 | btrfs_mark_buffer_dirty(leaf); | |
2494 | } | |
2495 | ||
2496 | path->slots[0]++; | |
2497 | goto next_slot; | |
2498 | } | |
2499 | ret = 0; | |
2500 | error: | |
2501 | btrfs_free_path(path); | |
2502 | return ret; | |
2503 | } | |
2504 | ||
da353f6b | 2505 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2506 | { |
5112febb | 2507 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2508 | struct request_queue *q; |
788f20eb CM |
2509 | struct btrfs_trans_handle *trans; |
2510 | struct btrfs_device *device; | |
2511 | struct block_device *bdev; | |
0b246afa | 2512 | struct super_block *sb = fs_info->sb; |
606686ee | 2513 | struct rcu_string *name; |
5da54bc1 | 2514 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
39379faa NA |
2515 | u64 orig_super_total_bytes; |
2516 | u64 orig_super_num_devices; | |
2b82032c | 2517 | int seeding_dev = 0; |
788f20eb | 2518 | int ret = 0; |
44cab9ba | 2519 | bool locked = false; |
788f20eb | 2520 | |
5da54bc1 | 2521 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2522 | return -EROFS; |
788f20eb | 2523 | |
a5d16333 | 2524 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2525 | fs_info->bdev_holder); |
7f59203a JB |
2526 | if (IS_ERR(bdev)) |
2527 | return PTR_ERR(bdev); | |
a2135011 | 2528 | |
b70f5097 NA |
2529 | if (!btrfs_check_device_zone_type(fs_info, bdev)) { |
2530 | ret = -EINVAL; | |
2531 | goto error; | |
2532 | } | |
2533 | ||
5da54bc1 | 2534 | if (fs_devices->seeding) { |
2b82032c YZ |
2535 | seeding_dev = 1; |
2536 | down_write(&sb->s_umount); | |
2537 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2538 | locked = true; |
2b82032c YZ |
2539 | } |
2540 | ||
b9ba017f | 2541 | sync_blockdev(bdev); |
a2135011 | 2542 | |
f4cfa9bd NB |
2543 | rcu_read_lock(); |
2544 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
788f20eb CM |
2545 | if (device->bdev == bdev) { |
2546 | ret = -EEXIST; | |
f4cfa9bd | 2547 | rcu_read_unlock(); |
2b82032c | 2548 | goto error; |
788f20eb CM |
2549 | } |
2550 | } | |
f4cfa9bd | 2551 | rcu_read_unlock(); |
788f20eb | 2552 | |
0b246afa | 2553 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2554 | if (IS_ERR(device)) { |
788f20eb | 2555 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2556 | ret = PTR_ERR(device); |
2b82032c | 2557 | goto error; |
788f20eb CM |
2558 | } |
2559 | ||
78f2c9e6 | 2560 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2561 | if (!name) { |
2b82032c | 2562 | ret = -ENOMEM; |
5c4cf6c9 | 2563 | goto error_free_device; |
788f20eb | 2564 | } |
606686ee | 2565 | rcu_assign_pointer(device->name, name); |
2b82032c | 2566 | |
5b316468 NA |
2567 | device->fs_info = fs_info; |
2568 | device->bdev = bdev; | |
2569 | ||
2570 | ret = btrfs_get_dev_zone_info(device); | |
2571 | if (ret) | |
2572 | goto error_free_device; | |
2573 | ||
a22285a6 | 2574 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2575 | if (IS_ERR(trans)) { |
98d5dc13 | 2576 | ret = PTR_ERR(trans); |
5b316468 | 2577 | goto error_free_zone; |
98d5dc13 TI |
2578 | } |
2579 | ||
d5e2003c | 2580 | q = bdev_get_queue(bdev); |
ebbede42 | 2581 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2582 | device->generation = trans->transid; |
0b246afa JM |
2583 | device->io_width = fs_info->sectorsize; |
2584 | device->io_align = fs_info->sectorsize; | |
2585 | device->sector_size = fs_info->sectorsize; | |
7dfb8be1 NB |
2586 | device->total_bytes = round_down(i_size_read(bdev->bd_inode), |
2587 | fs_info->sectorsize); | |
2cc3c559 | 2588 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2589 | device->commit_total_bytes = device->total_bytes; |
e12c9621 | 2590 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2591 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2592 | device->mode = FMODE_EXCL; |
27087f37 | 2593 | device->dev_stats_valid = 1; |
9f6d2510 | 2594 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2595 | |
2b82032c | 2596 | if (seeding_dev) { |
a0a1db70 | 2597 | btrfs_clear_sb_rdonly(sb); |
2ff7e61e | 2598 | ret = btrfs_prepare_sprout(fs_info); |
d31c32f6 AJ |
2599 | if (ret) { |
2600 | btrfs_abort_transaction(trans, ret); | |
2601 | goto error_trans; | |
2602 | } | |
2b82032c | 2603 | } |
788f20eb | 2604 | |
5da54bc1 | 2605 | device->fs_devices = fs_devices; |
e5e9a520 | 2606 | |
5da54bc1 | 2607 | mutex_lock(&fs_devices->device_list_mutex); |
34441361 | 2608 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2609 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2610 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2611 | fs_devices->num_devices++; | |
2612 | fs_devices->open_devices++; | |
2613 | fs_devices->rw_devices++; | |
2614 | fs_devices->total_devices++; | |
2615 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2616 | |
a5ed45f8 | 2617 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2618 | |
e884f4f0 | 2619 | if (!blk_queue_nonrot(q)) |
7f0432d0 | 2620 | fs_devices->rotating = true; |
c289811c | 2621 | |
39379faa | 2622 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2623 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2624 | round_down(orig_super_total_bytes + device->total_bytes, |
2625 | fs_info->sectorsize)); | |
788f20eb | 2626 | |
39379faa NA |
2627 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2628 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2629 | orig_super_num_devices + 1); | |
0d39376a | 2630 | |
2196d6e8 MX |
2631 | /* |
2632 | * we've got more storage, clear any full flags on the space | |
2633 | * infos | |
2634 | */ | |
0b246afa | 2635 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2636 | |
34441361 | 2637 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2638 | |
2639 | /* Add sysfs device entry */ | |
cd36da2e | 2640 | btrfs_sysfs_add_device(device); |
ca10845a | 2641 | |
5da54bc1 | 2642 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2643 | |
2b82032c | 2644 | if (seeding_dev) { |
34441361 | 2645 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2646 | ret = init_first_rw_device(trans); |
34441361 | 2647 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2648 | if (ret) { |
66642832 | 2649 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2650 | goto error_sysfs; |
005d6427 | 2651 | } |
2196d6e8 MX |
2652 | } |
2653 | ||
8e87e856 | 2654 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2655 | if (ret) { |
66642832 | 2656 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2657 | goto error_sysfs; |
2196d6e8 MX |
2658 | } |
2659 | ||
2660 | if (seeding_dev) { | |
5c466629 | 2661 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2662 | if (ret) { |
66642832 | 2663 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2664 | goto error_sysfs; |
005d6427 | 2665 | } |
b2373f25 | 2666 | |
8e560081 NB |
2667 | /* |
2668 | * fs_devices now represents the newly sprouted filesystem and | |
2669 | * its fsid has been changed by btrfs_prepare_sprout | |
2670 | */ | |
2671 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2672 | } |
2673 | ||
3a45bb20 | 2674 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2675 | |
2b82032c YZ |
2676 | if (seeding_dev) { |
2677 | mutex_unlock(&uuid_mutex); | |
2678 | up_write(&sb->s_umount); | |
44cab9ba | 2679 | locked = false; |
788f20eb | 2680 | |
79787eaa JM |
2681 | if (ret) /* transaction commit */ |
2682 | return ret; | |
2683 | ||
2ff7e61e | 2684 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2685 | if (ret < 0) |
0b246afa | 2686 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2687 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2688 | trans = btrfs_attach_transaction(root); |
2689 | if (IS_ERR(trans)) { | |
2690 | if (PTR_ERR(trans) == -ENOENT) | |
2691 | return 0; | |
7132a262 AJ |
2692 | ret = PTR_ERR(trans); |
2693 | trans = NULL; | |
2694 | goto error_sysfs; | |
671415b7 | 2695 | } |
3a45bb20 | 2696 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2697 | } |
c9e9f97b | 2698 | |
7f551d96 AJ |
2699 | /* |
2700 | * Now that we have written a new super block to this device, check all | |
2701 | * other fs_devices list if device_path alienates any other scanned | |
2702 | * device. | |
2703 | * We can ignore the return value as it typically returns -EINVAL and | |
2704 | * only succeeds if the device was an alien. | |
2705 | */ | |
2706 | btrfs_forget_devices(device_path); | |
2707 | ||
2708 | /* Update ctime/mtime for blkid or udev */ | |
5a1972bd | 2709 | update_dev_time(device_path); |
7f551d96 | 2710 | |
2b82032c | 2711 | return ret; |
79787eaa | 2712 | |
d31c32f6 | 2713 | error_sysfs: |
53f8a74c | 2714 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2715 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2716 | mutex_lock(&fs_info->chunk_mutex); | |
2717 | list_del_rcu(&device->dev_list); | |
2718 | list_del(&device->dev_alloc_list); | |
2719 | fs_info->fs_devices->num_devices--; | |
2720 | fs_info->fs_devices->open_devices--; | |
2721 | fs_info->fs_devices->rw_devices--; | |
2722 | fs_info->fs_devices->total_devices--; | |
2723 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2724 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2725 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2726 | orig_super_total_bytes); | |
2727 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2728 | orig_super_num_devices); | |
2729 | mutex_unlock(&fs_info->chunk_mutex); | |
2730 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2731 | error_trans: |
0af2c4bf | 2732 | if (seeding_dev) |
a0a1db70 | 2733 | btrfs_set_sb_rdonly(sb); |
7132a262 AJ |
2734 | if (trans) |
2735 | btrfs_end_transaction(trans); | |
5b316468 NA |
2736 | error_free_zone: |
2737 | btrfs_destroy_dev_zone_info(device); | |
5c4cf6c9 | 2738 | error_free_device: |
a425f9d4 | 2739 | btrfs_free_device(device); |
2b82032c | 2740 | error: |
e525fd89 | 2741 | blkdev_put(bdev, FMODE_EXCL); |
44cab9ba | 2742 | if (locked) { |
2b82032c YZ |
2743 | mutex_unlock(&uuid_mutex); |
2744 | up_write(&sb->s_umount); | |
2745 | } | |
c9e9f97b | 2746 | return ret; |
788f20eb CM |
2747 | } |
2748 | ||
d397712b CM |
2749 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2750 | struct btrfs_device *device) | |
0b86a832 CM |
2751 | { |
2752 | int ret; | |
2753 | struct btrfs_path *path; | |
0b246afa | 2754 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2755 | struct btrfs_dev_item *dev_item; |
2756 | struct extent_buffer *leaf; | |
2757 | struct btrfs_key key; | |
2758 | ||
0b86a832 CM |
2759 | path = btrfs_alloc_path(); |
2760 | if (!path) | |
2761 | return -ENOMEM; | |
2762 | ||
2763 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2764 | key.type = BTRFS_DEV_ITEM_KEY; | |
2765 | key.offset = device->devid; | |
2766 | ||
2767 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2768 | if (ret < 0) | |
2769 | goto out; | |
2770 | ||
2771 | if (ret > 0) { | |
2772 | ret = -ENOENT; | |
2773 | goto out; | |
2774 | } | |
2775 | ||
2776 | leaf = path->nodes[0]; | |
2777 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2778 | ||
2779 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2780 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2781 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2782 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2783 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2784 | btrfs_set_device_total_bytes(leaf, dev_item, |
2785 | btrfs_device_get_disk_total_bytes(device)); | |
2786 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2787 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2788 | btrfs_mark_buffer_dirty(leaf); |
2789 | ||
2790 | out: | |
2791 | btrfs_free_path(path); | |
2792 | return ret; | |
2793 | } | |
2794 | ||
2196d6e8 | 2795 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2796 | struct btrfs_device *device, u64 new_size) |
2797 | { | |
0b246afa JM |
2798 | struct btrfs_fs_info *fs_info = device->fs_info; |
2799 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2800 | u64 old_total; |
2801 | u64 diff; | |
8f18cf13 | 2802 | |
ebbede42 | 2803 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2804 | return -EACCES; |
2196d6e8 | 2805 | |
7dfb8be1 NB |
2806 | new_size = round_down(new_size, fs_info->sectorsize); |
2807 | ||
34441361 | 2808 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2809 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2810 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2811 | |
63a212ab | 2812 | if (new_size <= device->total_bytes || |
401e29c1 | 2813 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2814 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2815 | return -EINVAL; |
2196d6e8 | 2816 | } |
2b82032c | 2817 | |
7dfb8be1 NB |
2818 | btrfs_set_super_total_bytes(super_copy, |
2819 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2820 | device->fs_devices->total_rw_bytes += diff; |
2821 | ||
7cc8e58d MX |
2822 | btrfs_device_set_total_bytes(device, new_size); |
2823 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2824 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2825 | if (list_empty(&device->post_commit_list)) |
2826 | list_add_tail(&device->post_commit_list, | |
2827 | &trans->transaction->dev_update_list); | |
34441361 | 2828 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2829 | |
8f18cf13 CM |
2830 | return btrfs_update_device(trans, device); |
2831 | } | |
2832 | ||
f4208794 | 2833 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2834 | { |
f4208794 | 2835 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2836 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2837 | int ret; |
2838 | struct btrfs_path *path; | |
2839 | struct btrfs_key key; | |
2840 | ||
8f18cf13 CM |
2841 | path = btrfs_alloc_path(); |
2842 | if (!path) | |
2843 | return -ENOMEM; | |
2844 | ||
408fbf19 | 2845 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2846 | key.offset = chunk_offset; |
2847 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2848 | ||
2849 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2850 | if (ret < 0) |
2851 | goto out; | |
2852 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2853 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2854 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2855 | ret = -ENOENT; |
2856 | goto out; | |
2857 | } | |
8f18cf13 CM |
2858 | |
2859 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2860 | if (ret < 0) |
0b246afa JM |
2861 | btrfs_handle_fs_error(fs_info, ret, |
2862 | "Failed to delete chunk item."); | |
79787eaa | 2863 | out: |
8f18cf13 | 2864 | btrfs_free_path(path); |
65a246c5 | 2865 | return ret; |
8f18cf13 CM |
2866 | } |
2867 | ||
408fbf19 | 2868 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2869 | { |
0b246afa | 2870 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2871 | struct btrfs_disk_key *disk_key; |
2872 | struct btrfs_chunk *chunk; | |
2873 | u8 *ptr; | |
2874 | int ret = 0; | |
2875 | u32 num_stripes; | |
2876 | u32 array_size; | |
2877 | u32 len = 0; | |
2878 | u32 cur; | |
2879 | struct btrfs_key key; | |
2880 | ||
79bd3712 | 2881 | lockdep_assert_held(&fs_info->chunk_mutex); |
8f18cf13 CM |
2882 | array_size = btrfs_super_sys_array_size(super_copy); |
2883 | ||
2884 | ptr = super_copy->sys_chunk_array; | |
2885 | cur = 0; | |
2886 | ||
2887 | while (cur < array_size) { | |
2888 | disk_key = (struct btrfs_disk_key *)ptr; | |
2889 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2890 | ||
2891 | len = sizeof(*disk_key); | |
2892 | ||
2893 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2894 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2895 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2896 | len += btrfs_chunk_item_size(num_stripes); | |
2897 | } else { | |
2898 | ret = -EIO; | |
2899 | break; | |
2900 | } | |
408fbf19 | 2901 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2902 | key.offset == chunk_offset) { |
2903 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2904 | array_size -= len; | |
2905 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2906 | } else { | |
2907 | ptr += len; | |
2908 | cur += len; | |
2909 | } | |
2910 | } | |
2911 | return ret; | |
2912 | } | |
2913 | ||
60ca842e OS |
2914 | /* |
2915 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
2916 | * @logical: Logical block offset in bytes. | |
2917 | * @length: Length of extent in bytes. | |
2918 | * | |
2919 | * Return: Chunk mapping or ERR_PTR. | |
2920 | */ | |
2921 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
2922 | u64 logical, u64 length) | |
592d92ee LB |
2923 | { |
2924 | struct extent_map_tree *em_tree; | |
2925 | struct extent_map *em; | |
2926 | ||
c8bf1b67 | 2927 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
2928 | read_lock(&em_tree->lock); |
2929 | em = lookup_extent_mapping(em_tree, logical, length); | |
2930 | read_unlock(&em_tree->lock); | |
2931 | ||
2932 | if (!em) { | |
2933 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
2934 | logical, length); | |
2935 | return ERR_PTR(-EINVAL); | |
2936 | } | |
2937 | ||
2938 | if (em->start > logical || em->start + em->len < logical) { | |
2939 | btrfs_crit(fs_info, | |
2940 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
2941 | logical, length, em->start, em->start + em->len); | |
2942 | free_extent_map(em); | |
2943 | return ERR_PTR(-EINVAL); | |
2944 | } | |
2945 | ||
2946 | /* callers are responsible for dropping em's ref. */ | |
2947 | return em; | |
2948 | } | |
2949 | ||
79bd3712 FM |
2950 | static int remove_chunk_item(struct btrfs_trans_handle *trans, |
2951 | struct map_lookup *map, u64 chunk_offset) | |
2952 | { | |
2953 | int i; | |
2954 | ||
2955 | /* | |
2956 | * Removing chunk items and updating the device items in the chunks btree | |
2957 | * requires holding the chunk_mutex. | |
2958 | * See the comment at btrfs_chunk_alloc() for the details. | |
2959 | */ | |
2960 | lockdep_assert_held(&trans->fs_info->chunk_mutex); | |
2961 | ||
2962 | for (i = 0; i < map->num_stripes; i++) { | |
2963 | int ret; | |
2964 | ||
2965 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2966 | if (ret) | |
2967 | return ret; | |
2968 | } | |
2969 | ||
2970 | return btrfs_free_chunk(trans, chunk_offset); | |
2971 | } | |
2972 | ||
97aff912 | 2973 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2974 | { |
97aff912 | 2975 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
2976 | struct extent_map *em; |
2977 | struct map_lookup *map; | |
2196d6e8 | 2978 | u64 dev_extent_len = 0; |
47ab2a6c | 2979 | int i, ret = 0; |
0b246afa | 2980 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2981 | |
60ca842e | 2982 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 2983 | if (IS_ERR(em)) { |
47ab2a6c JB |
2984 | /* |
2985 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2986 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2987 | * do anything we still error out. |
2988 | */ | |
2989 | ASSERT(0); | |
592d92ee | 2990 | return PTR_ERR(em); |
47ab2a6c | 2991 | } |
95617d69 | 2992 | map = em->map_lookup; |
8f18cf13 | 2993 | |
57ba4cb8 | 2994 | /* |
79bd3712 FM |
2995 | * First delete the device extent items from the devices btree. |
2996 | * We take the device_list_mutex to avoid racing with the finishing phase | |
2997 | * of a device replace operation. See the comment below before acquiring | |
2998 | * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex | |
2999 | * because that can result in a deadlock when deleting the device extent | |
3000 | * items from the devices btree - COWing an extent buffer from the btree | |
3001 | * may result in allocating a new metadata chunk, which would attempt to | |
3002 | * lock again fs_info->chunk_mutex. | |
57ba4cb8 FM |
3003 | */ |
3004 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3005 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3006 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3007 | ret = btrfs_free_dev_extent(trans, device, |
3008 | map->stripes[i].physical, | |
3009 | &dev_extent_len); | |
47ab2a6c | 3010 | if (ret) { |
57ba4cb8 | 3011 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3012 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3013 | goto out; |
3014 | } | |
a061fc8d | 3015 | |
2196d6e8 | 3016 | if (device->bytes_used > 0) { |
34441361 | 3017 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3018 | btrfs_device_set_bytes_used(device, |
3019 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3020 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3021 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3022 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3023 | } |
79bd3712 FM |
3024 | } |
3025 | mutex_unlock(&fs_devices->device_list_mutex); | |
a061fc8d | 3026 | |
79bd3712 FM |
3027 | /* |
3028 | * We acquire fs_info->chunk_mutex for 2 reasons: | |
3029 | * | |
3030 | * 1) Just like with the first phase of the chunk allocation, we must | |
3031 | * reserve system space, do all chunk btree updates and deletions, and | |
3032 | * update the system chunk array in the superblock while holding this | |
3033 | * mutex. This is for similar reasons as explained on the comment at | |
3034 | * the top of btrfs_chunk_alloc(); | |
3035 | * | |
3036 | * 2) Prevent races with the final phase of a device replace operation | |
3037 | * that replaces the device object associated with the map's stripes, | |
3038 | * because the device object's id can change at any time during that | |
3039 | * final phase of the device replace operation | |
3040 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
3041 | * replaced device and then see it with an ID of | |
3042 | * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating | |
3043 | * the device item, which does not exists on the chunk btree. | |
3044 | * The finishing phase of device replace acquires both the | |
3045 | * device_list_mutex and the chunk_mutex, in that order, so we are | |
3046 | * safe by just acquiring the chunk_mutex. | |
3047 | */ | |
3048 | trans->removing_chunk = true; | |
3049 | mutex_lock(&fs_info->chunk_mutex); | |
3050 | ||
3051 | check_system_chunk(trans, map->type); | |
3052 | ||
3053 | ret = remove_chunk_item(trans, map, chunk_offset); | |
3054 | /* | |
3055 | * Normally we should not get -ENOSPC since we reserved space before | |
3056 | * through the call to check_system_chunk(). | |
3057 | * | |
3058 | * Despite our system space_info having enough free space, we may not | |
3059 | * be able to allocate extents from its block groups, because all have | |
3060 | * an incompatible profile, which will force us to allocate a new system | |
3061 | * block group with the right profile, or right after we called | |
3062 | * check_system_space() above, a scrub turned the only system block group | |
3063 | * with enough free space into RO mode. | |
3064 | * This is explained with more detail at do_chunk_alloc(). | |
3065 | * | |
3066 | * So if we get -ENOSPC, allocate a new system chunk and retry once. | |
3067 | */ | |
3068 | if (ret == -ENOSPC) { | |
3069 | const u64 sys_flags = btrfs_system_alloc_profile(fs_info); | |
3070 | struct btrfs_block_group *sys_bg; | |
3071 | ||
3072 | sys_bg = btrfs_alloc_chunk(trans, sys_flags); | |
3073 | if (IS_ERR(sys_bg)) { | |
3074 | ret = PTR_ERR(sys_bg); | |
3075 | btrfs_abort_transaction(trans, ret); | |
3076 | goto out; | |
3077 | } | |
3078 | ||
3079 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
64bc6c2a | 3080 | if (ret) { |
64bc6c2a NB |
3081 | btrfs_abort_transaction(trans, ret); |
3082 | goto out; | |
dfe25020 | 3083 | } |
57ba4cb8 | 3084 | |
79bd3712 FM |
3085 | ret = remove_chunk_item(trans, map, chunk_offset); |
3086 | if (ret) { | |
3087 | btrfs_abort_transaction(trans, ret); | |
3088 | goto out; | |
3089 | } | |
3090 | } else if (ret) { | |
66642832 | 3091 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3092 | goto out; |
3093 | } | |
8f18cf13 | 3094 | |
6bccf3ab | 3095 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3096 | |
8f18cf13 | 3097 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3098 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3099 | if (ret) { |
66642832 | 3100 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3101 | goto out; |
3102 | } | |
8f18cf13 CM |
3103 | } |
3104 | ||
79bd3712 FM |
3105 | mutex_unlock(&fs_info->chunk_mutex); |
3106 | trans->removing_chunk = false; | |
3107 | ||
3108 | /* | |
3109 | * We are done with chunk btree updates and deletions, so release the | |
3110 | * system space we previously reserved (with check_system_chunk()). | |
3111 | */ | |
3112 | btrfs_trans_release_chunk_metadata(trans); | |
3113 | ||
5a98ec01 | 3114 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3115 | if (ret) { |
66642832 | 3116 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3117 | goto out; |
3118 | } | |
2b82032c | 3119 | |
47ab2a6c | 3120 | out: |
79bd3712 FM |
3121 | if (trans->removing_chunk) { |
3122 | mutex_unlock(&fs_info->chunk_mutex); | |
3123 | trans->removing_chunk = false; | |
3124 | } | |
2b82032c YZ |
3125 | /* once for us */ |
3126 | free_extent_map(em); | |
47ab2a6c JB |
3127 | return ret; |
3128 | } | |
2b82032c | 3129 | |
18bb8bbf | 3130 | int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3131 | { |
5b4aacef | 3132 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3133 | struct btrfs_trans_handle *trans; |
b0643e59 | 3134 | struct btrfs_block_group *block_group; |
01e86008 | 3135 | u64 length; |
47ab2a6c | 3136 | int ret; |
2b82032c | 3137 | |
67c5e7d4 FM |
3138 | /* |
3139 | * Prevent races with automatic removal of unused block groups. | |
3140 | * After we relocate and before we remove the chunk with offset | |
3141 | * chunk_offset, automatic removal of the block group can kick in, | |
3142 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3143 | * | |
3144 | * Make sure to acquire this mutex before doing a tree search (dev | |
3145 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3146 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3147 | * we release the path used to search the chunk/dev tree and before | |
3148 | * the current task acquires this mutex and calls us. | |
3149 | */ | |
f3372065 | 3150 | lockdep_assert_held(&fs_info->reclaim_bgs_lock); |
67c5e7d4 | 3151 | |
47ab2a6c | 3152 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3153 | btrfs_scrub_pause(fs_info); |
0b246afa | 3154 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3155 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3156 | if (ret) |
3157 | return ret; | |
3158 | ||
b0643e59 DZ |
3159 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3160 | if (!block_group) | |
3161 | return -ENOENT; | |
3162 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
01e86008 | 3163 | length = block_group->length; |
b0643e59 DZ |
3164 | btrfs_put_block_group(block_group); |
3165 | ||
01e86008 JT |
3166 | /* |
3167 | * On a zoned file system, discard the whole block group, this will | |
3168 | * trigger a REQ_OP_ZONE_RESET operation on the device zone. If | |
3169 | * resetting the zone fails, don't treat it as a fatal problem from the | |
3170 | * filesystem's point of view. | |
3171 | */ | |
3172 | if (btrfs_is_zoned(fs_info)) { | |
3173 | ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); | |
3174 | if (ret) | |
3175 | btrfs_info(fs_info, | |
3176 | "failed to reset zone %llu after relocation", | |
3177 | chunk_offset); | |
3178 | } | |
3179 | ||
19c4d2f9 CM |
3180 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3181 | chunk_offset); | |
3182 | if (IS_ERR(trans)) { | |
3183 | ret = PTR_ERR(trans); | |
3184 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3185 | return ret; | |
3186 | } | |
3187 | ||
47ab2a6c | 3188 | /* |
19c4d2f9 CM |
3189 | * step two, delete the device extents and the |
3190 | * chunk tree entries | |
47ab2a6c | 3191 | */ |
97aff912 | 3192 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3193 | btrfs_end_transaction(trans); |
19c4d2f9 | 3194 | return ret; |
2b82032c YZ |
3195 | } |
3196 | ||
2ff7e61e | 3197 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3198 | { |
0b246afa | 3199 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3200 | struct btrfs_path *path; |
3201 | struct extent_buffer *leaf; | |
3202 | struct btrfs_chunk *chunk; | |
3203 | struct btrfs_key key; | |
3204 | struct btrfs_key found_key; | |
2b82032c | 3205 | u64 chunk_type; |
ba1bf481 JB |
3206 | bool retried = false; |
3207 | int failed = 0; | |
2b82032c YZ |
3208 | int ret; |
3209 | ||
3210 | path = btrfs_alloc_path(); | |
3211 | if (!path) | |
3212 | return -ENOMEM; | |
3213 | ||
ba1bf481 | 3214 | again: |
2b82032c YZ |
3215 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3216 | key.offset = (u64)-1; | |
3217 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3218 | ||
3219 | while (1) { | |
f3372065 | 3220 | mutex_lock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3221 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3222 | if (ret < 0) { |
f3372065 | 3223 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3224 | goto error; |
67c5e7d4 | 3225 | } |
79787eaa | 3226 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3227 | |
3228 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3229 | key.type); | |
67c5e7d4 | 3230 | if (ret) |
f3372065 | 3231 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c YZ |
3232 | if (ret < 0) |
3233 | goto error; | |
3234 | if (ret > 0) | |
3235 | break; | |
1a40e23b | 3236 | |
2b82032c YZ |
3237 | leaf = path->nodes[0]; |
3238 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3239 | |
2b82032c YZ |
3240 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3241 | struct btrfs_chunk); | |
3242 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3243 | btrfs_release_path(path); |
8f18cf13 | 3244 | |
2b82032c | 3245 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3246 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3247 | if (ret == -ENOSPC) |
3248 | failed++; | |
14586651 HS |
3249 | else |
3250 | BUG_ON(ret); | |
2b82032c | 3251 | } |
f3372065 | 3252 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 3253 | |
2b82032c YZ |
3254 | if (found_key.offset == 0) |
3255 | break; | |
3256 | key.offset = found_key.offset - 1; | |
3257 | } | |
3258 | ret = 0; | |
ba1bf481 JB |
3259 | if (failed && !retried) { |
3260 | failed = 0; | |
3261 | retried = true; | |
3262 | goto again; | |
fae7f21c | 3263 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3264 | ret = -ENOSPC; |
3265 | } | |
2b82032c YZ |
3266 | error: |
3267 | btrfs_free_path(path); | |
3268 | return ret; | |
8f18cf13 CM |
3269 | } |
3270 | ||
a6f93c71 LB |
3271 | /* |
3272 | * return 1 : allocate a data chunk successfully, | |
3273 | * return <0: errors during allocating a data chunk, | |
3274 | * return 0 : no need to allocate a data chunk. | |
3275 | */ | |
3276 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3277 | u64 chunk_offset) | |
3278 | { | |
32da5386 | 3279 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3280 | u64 bytes_used; |
3281 | u64 chunk_type; | |
3282 | ||
3283 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3284 | ASSERT(cache); | |
3285 | chunk_type = cache->flags; | |
3286 | btrfs_put_block_group(cache); | |
3287 | ||
5ae21692 JT |
3288 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3289 | return 0; | |
3290 | ||
3291 | spin_lock(&fs_info->data_sinfo->lock); | |
3292 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3293 | spin_unlock(&fs_info->data_sinfo->lock); | |
3294 | ||
3295 | if (!bytes_used) { | |
3296 | struct btrfs_trans_handle *trans; | |
3297 | int ret; | |
3298 | ||
3299 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3300 | if (IS_ERR(trans)) | |
3301 | return PTR_ERR(trans); | |
3302 | ||
3303 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3304 | btrfs_end_transaction(trans); | |
3305 | if (ret < 0) | |
3306 | return ret; | |
3307 | return 1; | |
a6f93c71 | 3308 | } |
5ae21692 | 3309 | |
a6f93c71 LB |
3310 | return 0; |
3311 | } | |
3312 | ||
6bccf3ab | 3313 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3314 | struct btrfs_balance_control *bctl) |
3315 | { | |
6bccf3ab | 3316 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3317 | struct btrfs_trans_handle *trans; |
3318 | struct btrfs_balance_item *item; | |
3319 | struct btrfs_disk_balance_args disk_bargs; | |
3320 | struct btrfs_path *path; | |
3321 | struct extent_buffer *leaf; | |
3322 | struct btrfs_key key; | |
3323 | int ret, err; | |
3324 | ||
3325 | path = btrfs_alloc_path(); | |
3326 | if (!path) | |
3327 | return -ENOMEM; | |
3328 | ||
3329 | trans = btrfs_start_transaction(root, 0); | |
3330 | if (IS_ERR(trans)) { | |
3331 | btrfs_free_path(path); | |
3332 | return PTR_ERR(trans); | |
3333 | } | |
3334 | ||
3335 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3336 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3337 | key.offset = 0; |
3338 | ||
3339 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3340 | sizeof(*item)); | |
3341 | if (ret) | |
3342 | goto out; | |
3343 | ||
3344 | leaf = path->nodes[0]; | |
3345 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3346 | ||
b159fa28 | 3347 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3348 | |
3349 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3350 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3351 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3352 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3353 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3354 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3355 | ||
3356 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3357 | ||
3358 | btrfs_mark_buffer_dirty(leaf); | |
3359 | out: | |
3360 | btrfs_free_path(path); | |
3a45bb20 | 3361 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3362 | if (err && !ret) |
3363 | ret = err; | |
3364 | return ret; | |
3365 | } | |
3366 | ||
6bccf3ab | 3367 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3368 | { |
6bccf3ab | 3369 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3370 | struct btrfs_trans_handle *trans; |
3371 | struct btrfs_path *path; | |
3372 | struct btrfs_key key; | |
3373 | int ret, err; | |
3374 | ||
3375 | path = btrfs_alloc_path(); | |
3376 | if (!path) | |
3377 | return -ENOMEM; | |
3378 | ||
3502a8c0 | 3379 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3380 | if (IS_ERR(trans)) { |
3381 | btrfs_free_path(path); | |
3382 | return PTR_ERR(trans); | |
3383 | } | |
3384 | ||
3385 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3386 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3387 | key.offset = 0; |
3388 | ||
3389 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3390 | if (ret < 0) | |
3391 | goto out; | |
3392 | if (ret > 0) { | |
3393 | ret = -ENOENT; | |
3394 | goto out; | |
3395 | } | |
3396 | ||
3397 | ret = btrfs_del_item(trans, root, path); | |
3398 | out: | |
3399 | btrfs_free_path(path); | |
3a45bb20 | 3400 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3401 | if (err && !ret) |
3402 | ret = err; | |
3403 | return ret; | |
3404 | } | |
3405 | ||
59641015 ID |
3406 | /* |
3407 | * This is a heuristic used to reduce the number of chunks balanced on | |
3408 | * resume after balance was interrupted. | |
3409 | */ | |
3410 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3411 | { | |
3412 | /* | |
3413 | * Turn on soft mode for chunk types that were being converted. | |
3414 | */ | |
3415 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3416 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3417 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3418 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3419 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3420 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3421 | ||
3422 | /* | |
3423 | * Turn on usage filter if is not already used. The idea is | |
3424 | * that chunks that we have already balanced should be | |
3425 | * reasonably full. Don't do it for chunks that are being | |
3426 | * converted - that will keep us from relocating unconverted | |
3427 | * (albeit full) chunks. | |
3428 | */ | |
3429 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3430 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3431 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3432 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3433 | bctl->data.usage = 90; | |
3434 | } | |
3435 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3436 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3437 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3438 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3439 | bctl->sys.usage = 90; | |
3440 | } | |
3441 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3442 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3443 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3444 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3445 | bctl->meta.usage = 90; | |
3446 | } | |
3447 | } | |
3448 | ||
149196a2 DS |
3449 | /* |
3450 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3451 | */ | |
3452 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3453 | { |
3454 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3455 | int ret; |
c9e9f97b ID |
3456 | |
3457 | BUG_ON(!fs_info->balance_ctl); | |
3458 | ||
3459 | spin_lock(&fs_info->balance_lock); | |
3460 | fs_info->balance_ctl = NULL; | |
3461 | spin_unlock(&fs_info->balance_lock); | |
3462 | ||
3463 | kfree(bctl); | |
149196a2 DS |
3464 | ret = del_balance_item(fs_info); |
3465 | if (ret) | |
3466 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3467 | } |
3468 | ||
ed25e9b2 ID |
3469 | /* |
3470 | * Balance filters. Return 1 if chunk should be filtered out | |
3471 | * (should not be balanced). | |
3472 | */ | |
899c81ea | 3473 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3474 | struct btrfs_balance_args *bargs) |
3475 | { | |
899c81ea ID |
3476 | chunk_type = chunk_to_extended(chunk_type) & |
3477 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3478 | |
899c81ea | 3479 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3480 | return 0; |
3481 | ||
3482 | return 1; | |
3483 | } | |
3484 | ||
dba72cb3 | 3485 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3486 | struct btrfs_balance_args *bargs) |
bc309467 | 3487 | { |
32da5386 | 3488 | struct btrfs_block_group *cache; |
bc309467 DS |
3489 | u64 chunk_used; |
3490 | u64 user_thresh_min; | |
3491 | u64 user_thresh_max; | |
3492 | int ret = 1; | |
3493 | ||
3494 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3495 | chunk_used = cache->used; |
bc309467 DS |
3496 | |
3497 | if (bargs->usage_min == 0) | |
3498 | user_thresh_min = 0; | |
3499 | else | |
b3470b5d DS |
3500 | user_thresh_min = div_factor_fine(cache->length, |
3501 | bargs->usage_min); | |
bc309467 DS |
3502 | |
3503 | if (bargs->usage_max == 0) | |
3504 | user_thresh_max = 1; | |
3505 | else if (bargs->usage_max > 100) | |
b3470b5d | 3506 | user_thresh_max = cache->length; |
bc309467 | 3507 | else |
b3470b5d DS |
3508 | user_thresh_max = div_factor_fine(cache->length, |
3509 | bargs->usage_max); | |
bc309467 DS |
3510 | |
3511 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3512 | ret = 0; | |
3513 | ||
3514 | btrfs_put_block_group(cache); | |
3515 | return ret; | |
3516 | } | |
3517 | ||
dba72cb3 | 3518 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3519 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3520 | { |
32da5386 | 3521 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3522 | u64 chunk_used, user_thresh; |
3523 | int ret = 1; | |
3524 | ||
3525 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3526 | chunk_used = cache->used; |
5ce5b3c0 | 3527 | |
bc309467 | 3528 | if (bargs->usage_min == 0) |
3e39cea6 | 3529 | user_thresh = 1; |
a105bb88 | 3530 | else if (bargs->usage > 100) |
b3470b5d | 3531 | user_thresh = cache->length; |
a105bb88 | 3532 | else |
b3470b5d | 3533 | user_thresh = div_factor_fine(cache->length, bargs->usage); |
a105bb88 | 3534 | |
5ce5b3c0 ID |
3535 | if (chunk_used < user_thresh) |
3536 | ret = 0; | |
3537 | ||
3538 | btrfs_put_block_group(cache); | |
3539 | return ret; | |
3540 | } | |
3541 | ||
409d404b ID |
3542 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3543 | struct btrfs_chunk *chunk, | |
3544 | struct btrfs_balance_args *bargs) | |
3545 | { | |
3546 | struct btrfs_stripe *stripe; | |
3547 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3548 | int i; | |
3549 | ||
3550 | for (i = 0; i < num_stripes; i++) { | |
3551 | stripe = btrfs_stripe_nr(chunk, i); | |
3552 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3553 | return 0; | |
3554 | } | |
3555 | ||
3556 | return 1; | |
3557 | } | |
3558 | ||
946c9256 DS |
3559 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3560 | { | |
3561 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3562 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3563 | const int nparity = btrfs_raid_array[index].nparity; | |
3564 | ||
d58ede8d | 3565 | return (num_stripes - nparity) / ncopies; |
946c9256 DS |
3566 | } |
3567 | ||
94e60d5a ID |
3568 | /* [pstart, pend) */ |
3569 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3570 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3571 | struct btrfs_balance_args *bargs) |
3572 | { | |
3573 | struct btrfs_stripe *stripe; | |
3574 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3575 | u64 stripe_offset; | |
3576 | u64 stripe_length; | |
946c9256 | 3577 | u64 type; |
94e60d5a ID |
3578 | int factor; |
3579 | int i; | |
3580 | ||
3581 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3582 | return 0; | |
3583 | ||
946c9256 DS |
3584 | type = btrfs_chunk_type(leaf, chunk); |
3585 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3586 | |
3587 | for (i = 0; i < num_stripes; i++) { | |
3588 | stripe = btrfs_stripe_nr(chunk, i); | |
3589 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3590 | continue; | |
3591 | ||
3592 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3593 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3594 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3595 | |
3596 | if (stripe_offset < bargs->pend && | |
3597 | stripe_offset + stripe_length > bargs->pstart) | |
3598 | return 0; | |
3599 | } | |
3600 | ||
3601 | return 1; | |
3602 | } | |
3603 | ||
ea67176a ID |
3604 | /* [vstart, vend) */ |
3605 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3606 | struct btrfs_chunk *chunk, | |
3607 | u64 chunk_offset, | |
3608 | struct btrfs_balance_args *bargs) | |
3609 | { | |
3610 | if (chunk_offset < bargs->vend && | |
3611 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3612 | /* at least part of the chunk is inside this vrange */ | |
3613 | return 0; | |
3614 | ||
3615 | return 1; | |
3616 | } | |
3617 | ||
dee32d0a GAP |
3618 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3619 | struct btrfs_chunk *chunk, | |
3620 | struct btrfs_balance_args *bargs) | |
3621 | { | |
3622 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3623 | ||
3624 | if (bargs->stripes_min <= num_stripes | |
3625 | && num_stripes <= bargs->stripes_max) | |
3626 | return 0; | |
3627 | ||
3628 | return 1; | |
3629 | } | |
3630 | ||
899c81ea | 3631 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3632 | struct btrfs_balance_args *bargs) |
3633 | { | |
3634 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3635 | return 0; | |
3636 | ||
899c81ea ID |
3637 | chunk_type = chunk_to_extended(chunk_type) & |
3638 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3639 | |
899c81ea | 3640 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3641 | return 1; |
3642 | ||
3643 | return 0; | |
3644 | } | |
3645 | ||
6ec0896c | 3646 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3647 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3648 | { | |
6ec0896c | 3649 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3650 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3651 | struct btrfs_balance_args *bargs = NULL; |
3652 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3653 | ||
3654 | /* type filter */ | |
3655 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3656 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3657 | return 0; | |
3658 | } | |
3659 | ||
3660 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3661 | bargs = &bctl->data; | |
3662 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3663 | bargs = &bctl->sys; | |
3664 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3665 | bargs = &bctl->meta; | |
3666 | ||
ed25e9b2 ID |
3667 | /* profiles filter */ |
3668 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3669 | chunk_profiles_filter(chunk_type, bargs)) { | |
3670 | return 0; | |
5ce5b3c0 ID |
3671 | } |
3672 | ||
3673 | /* usage filter */ | |
3674 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3675 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3676 | return 0; |
bc309467 | 3677 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3678 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3679 | return 0; |
409d404b ID |
3680 | } |
3681 | ||
3682 | /* devid filter */ | |
3683 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3684 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3685 | return 0; | |
94e60d5a ID |
3686 | } |
3687 | ||
3688 | /* drange filter, makes sense only with devid filter */ | |
3689 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3690 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3691 | return 0; |
ea67176a ID |
3692 | } |
3693 | ||
3694 | /* vrange filter */ | |
3695 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3696 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3697 | return 0; | |
ed25e9b2 ID |
3698 | } |
3699 | ||
dee32d0a GAP |
3700 | /* stripes filter */ |
3701 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3702 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3703 | return 0; | |
3704 | } | |
3705 | ||
cfa4c961 ID |
3706 | /* soft profile changing mode */ |
3707 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3708 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3709 | return 0; | |
3710 | } | |
3711 | ||
7d824b6f DS |
3712 | /* |
3713 | * limited by count, must be the last filter | |
3714 | */ | |
3715 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3716 | if (bargs->limit == 0) | |
3717 | return 0; | |
3718 | else | |
3719 | bargs->limit--; | |
12907fc7 DS |
3720 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3721 | /* | |
3722 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3723 | * determined here because we do not have the global information |
12907fc7 DS |
3724 | * about the count of all chunks that satisfy the filters. |
3725 | */ | |
3726 | if (bargs->limit_max == 0) | |
3727 | return 0; | |
3728 | else | |
3729 | bargs->limit_max--; | |
7d824b6f DS |
3730 | } |
3731 | ||
f43ffb60 ID |
3732 | return 1; |
3733 | } | |
3734 | ||
c9e9f97b | 3735 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3736 | { |
19a39dce | 3737 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3738 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3739 | u64 chunk_type; |
f43ffb60 | 3740 | struct btrfs_chunk *chunk; |
5a488b9d | 3741 | struct btrfs_path *path = NULL; |
ec44a35c | 3742 | struct btrfs_key key; |
ec44a35c | 3743 | struct btrfs_key found_key; |
f43ffb60 ID |
3744 | struct extent_buffer *leaf; |
3745 | int slot; | |
c9e9f97b ID |
3746 | int ret; |
3747 | int enospc_errors = 0; | |
19a39dce | 3748 | bool counting = true; |
12907fc7 | 3749 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3750 | u64 limit_data = bctl->data.limit; |
3751 | u64 limit_meta = bctl->meta.limit; | |
3752 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3753 | u32 count_data = 0; |
3754 | u32 count_meta = 0; | |
3755 | u32 count_sys = 0; | |
2c9fe835 | 3756 | int chunk_reserved = 0; |
ec44a35c | 3757 | |
ec44a35c | 3758 | path = btrfs_alloc_path(); |
17e9f796 MF |
3759 | if (!path) { |
3760 | ret = -ENOMEM; | |
3761 | goto error; | |
3762 | } | |
19a39dce ID |
3763 | |
3764 | /* zero out stat counters */ | |
3765 | spin_lock(&fs_info->balance_lock); | |
3766 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3767 | spin_unlock(&fs_info->balance_lock); | |
3768 | again: | |
7d824b6f | 3769 | if (!counting) { |
12907fc7 DS |
3770 | /* |
3771 | * The single value limit and min/max limits use the same bytes | |
3772 | * in the | |
3773 | */ | |
7d824b6f DS |
3774 | bctl->data.limit = limit_data; |
3775 | bctl->meta.limit = limit_meta; | |
3776 | bctl->sys.limit = limit_sys; | |
3777 | } | |
ec44a35c CM |
3778 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3779 | key.offset = (u64)-1; | |
3780 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3781 | ||
d397712b | 3782 | while (1) { |
19a39dce | 3783 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3784 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3785 | ret = -ECANCELED; |
3786 | goto error; | |
3787 | } | |
3788 | ||
f3372065 | 3789 | mutex_lock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3790 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3791 | if (ret < 0) { |
f3372065 | 3792 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3793 | goto error; |
67c5e7d4 | 3794 | } |
ec44a35c CM |
3795 | |
3796 | /* | |
3797 | * this shouldn't happen, it means the last relocate | |
3798 | * failed | |
3799 | */ | |
3800 | if (ret == 0) | |
c9e9f97b | 3801 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3802 | |
3803 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3804 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3805 | if (ret) { |
f3372065 | 3806 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
c9e9f97b | 3807 | ret = 0; |
ec44a35c | 3808 | break; |
c9e9f97b | 3809 | } |
7d9eb12c | 3810 | |
f43ffb60 ID |
3811 | leaf = path->nodes[0]; |
3812 | slot = path->slots[0]; | |
3813 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3814 | |
67c5e7d4 | 3815 | if (found_key.objectid != key.objectid) { |
f3372065 | 3816 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3817 | break; |
67c5e7d4 | 3818 | } |
7d9eb12c | 3819 | |
f43ffb60 | 3820 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3821 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3822 | |
19a39dce ID |
3823 | if (!counting) { |
3824 | spin_lock(&fs_info->balance_lock); | |
3825 | bctl->stat.considered++; | |
3826 | spin_unlock(&fs_info->balance_lock); | |
3827 | } | |
3828 | ||
6ec0896c | 3829 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3830 | |
b3b4aa74 | 3831 | btrfs_release_path(path); |
67c5e7d4 | 3832 | if (!ret) { |
f3372065 | 3833 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f43ffb60 | 3834 | goto loop; |
67c5e7d4 | 3835 | } |
f43ffb60 | 3836 | |
19a39dce | 3837 | if (counting) { |
f3372065 | 3838 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3839 | spin_lock(&fs_info->balance_lock); |
3840 | bctl->stat.expected++; | |
3841 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3842 | |
3843 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3844 | count_data++; | |
3845 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3846 | count_sys++; | |
3847 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3848 | count_meta++; | |
3849 | ||
3850 | goto loop; | |
3851 | } | |
3852 | ||
3853 | /* | |
3854 | * Apply limit_min filter, no need to check if the LIMITS | |
3855 | * filter is used, limit_min is 0 by default | |
3856 | */ | |
3857 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3858 | count_data < bctl->data.limit_min) | |
3859 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3860 | count_meta < bctl->meta.limit_min) | |
3861 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3862 | count_sys < bctl->sys.limit_min)) { | |
f3372065 | 3863 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3864 | goto loop; |
3865 | } | |
3866 | ||
a6f93c71 LB |
3867 | if (!chunk_reserved) { |
3868 | /* | |
3869 | * We may be relocating the only data chunk we have, | |
3870 | * which could potentially end up with losing data's | |
3871 | * raid profile, so lets allocate an empty one in | |
3872 | * advance. | |
3873 | */ | |
3874 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3875 | found_key.offset); | |
2c9fe835 | 3876 | if (ret < 0) { |
f3372065 | 3877 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2c9fe835 | 3878 | goto error; |
a6f93c71 LB |
3879 | } else if (ret == 1) { |
3880 | chunk_reserved = 1; | |
2c9fe835 | 3881 | } |
2c9fe835 ZL |
3882 | } |
3883 | ||
5b4aacef | 3884 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
f3372065 | 3885 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce | 3886 | if (ret == -ENOSPC) { |
c9e9f97b | 3887 | enospc_errors++; |
eede2bf3 OS |
3888 | } else if (ret == -ETXTBSY) { |
3889 | btrfs_info(fs_info, | |
3890 | "skipping relocation of block group %llu due to active swapfile", | |
3891 | found_key.offset); | |
3892 | ret = 0; | |
3893 | } else if (ret) { | |
3894 | goto error; | |
19a39dce ID |
3895 | } else { |
3896 | spin_lock(&fs_info->balance_lock); | |
3897 | bctl->stat.completed++; | |
3898 | spin_unlock(&fs_info->balance_lock); | |
3899 | } | |
f43ffb60 | 3900 | loop: |
795a3321 ID |
3901 | if (found_key.offset == 0) |
3902 | break; | |
ba1bf481 | 3903 | key.offset = found_key.offset - 1; |
ec44a35c | 3904 | } |
c9e9f97b | 3905 | |
19a39dce ID |
3906 | if (counting) { |
3907 | btrfs_release_path(path); | |
3908 | counting = false; | |
3909 | goto again; | |
3910 | } | |
ec44a35c CM |
3911 | error: |
3912 | btrfs_free_path(path); | |
c9e9f97b | 3913 | if (enospc_errors) { |
efe120a0 | 3914 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3915 | enospc_errors); |
c9e9f97b ID |
3916 | if (!ret) |
3917 | ret = -ENOSPC; | |
3918 | } | |
3919 | ||
ec44a35c CM |
3920 | return ret; |
3921 | } | |
3922 | ||
0c460c0d ID |
3923 | /** |
3924 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3925 | * @flags: profile to validate | |
3926 | * @extended: if true @flags is treated as an extended profile | |
3927 | */ | |
3928 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3929 | { | |
3930 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3931 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3932 | ||
3933 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3934 | ||
3935 | /* 1) check that all other bits are zeroed */ | |
3936 | if (flags & ~mask) | |
3937 | return 0; | |
3938 | ||
3939 | /* 2) see if profile is reduced */ | |
3940 | if (flags == 0) | |
3941 | return !extended; /* "0" is valid for usual profiles */ | |
3942 | ||
c1499166 | 3943 | return has_single_bit_set(flags); |
0c460c0d ID |
3944 | } |
3945 | ||
837d5b6e ID |
3946 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3947 | { | |
a7e99c69 ID |
3948 | /* cancel requested || normal exit path */ |
3949 | return atomic_read(&fs_info->balance_cancel_req) || | |
3950 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3951 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3952 | } |
3953 | ||
5ba366c3 DS |
3954 | /* |
3955 | * Validate target profile against allowed profiles and return true if it's OK. | |
3956 | * Otherwise print the error message and return false. | |
3957 | */ | |
3958 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
3959 | const struct btrfs_balance_args *bargs, | |
3960 | u64 allowed, const char *type) | |
bdcd3c97 | 3961 | { |
5ba366c3 DS |
3962 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
3963 | return true; | |
3964 | ||
c8050b3b QW |
3965 | if (fs_info->sectorsize < PAGE_SIZE && |
3966 | bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
3967 | btrfs_err(fs_info, | |
3968 | "RAID56 is not yet supported for sectorsize %u with page size %lu", | |
3969 | fs_info->sectorsize, PAGE_SIZE); | |
3970 | return false; | |
3971 | } | |
5ba366c3 DS |
3972 | /* Profile is valid and does not have bits outside of the allowed set */ |
3973 | if (alloc_profile_is_valid(bargs->target, 1) && | |
3974 | (bargs->target & ~allowed) == 0) | |
3975 | return true; | |
3976 | ||
3977 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
3978 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
3979 | return false; | |
bdcd3c97 AM |
3980 | } |
3981 | ||
56fc37d9 AJ |
3982 | /* |
3983 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
3984 | * @size_buf including the terminating null. The output may be trimmed if it | |
3985 | * does not fit into the provided buffer. | |
3986 | */ | |
3987 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
3988 | u32 size_buf) | |
3989 | { | |
3990 | int ret; | |
3991 | u32 size_bp = size_buf; | |
3992 | char *bp = buf; | |
3993 | u64 flags = bargs->flags; | |
3994 | char tmp_buf[128] = {'\0'}; | |
3995 | ||
3996 | if (!flags) | |
3997 | return; | |
3998 | ||
3999 | #define CHECK_APPEND_NOARG(a) \ | |
4000 | do { \ | |
4001 | ret = snprintf(bp, size_bp, (a)); \ | |
4002 | if (ret < 0 || ret >= size_bp) \ | |
4003 | goto out_overflow; \ | |
4004 | size_bp -= ret; \ | |
4005 | bp += ret; \ | |
4006 | } while (0) | |
4007 | ||
4008 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4009 | do { \ | |
4010 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4011 | if (ret < 0 || ret >= size_bp) \ | |
4012 | goto out_overflow; \ | |
4013 | size_bp -= ret; \ | |
4014 | bp += ret; \ | |
4015 | } while (0) | |
4016 | ||
4017 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
4018 | do { \ | |
4019 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
4020 | if (ret < 0 || ret >= size_bp) \ | |
4021 | goto out_overflow; \ | |
4022 | size_bp -= ret; \ | |
4023 | bp += ret; \ | |
4024 | } while (0) | |
4025 | ||
158da513 DS |
4026 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
4027 | CHECK_APPEND_1ARG("convert=%s,", | |
4028 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
4029 | |
4030 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
4031 | CHECK_APPEND_NOARG("soft,"); | |
4032 | ||
4033 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
4034 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
4035 | sizeof(tmp_buf)); | |
4036 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
4037 | } | |
4038 | ||
4039 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
4040 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
4041 | ||
4042 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
4043 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
4044 | bargs->usage_min, bargs->usage_max); | |
4045 | ||
4046 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
4047 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
4048 | ||
4049 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4050 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4051 | bargs->pstart, bargs->pend); | |
4052 | ||
4053 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4054 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4055 | bargs->vstart, bargs->vend); | |
4056 | ||
4057 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4058 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4059 | ||
4060 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4061 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4062 | bargs->limit_min, bargs->limit_max); | |
4063 | ||
4064 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4065 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4066 | bargs->stripes_min, bargs->stripes_max); | |
4067 | ||
4068 | #undef CHECK_APPEND_2ARG | |
4069 | #undef CHECK_APPEND_1ARG | |
4070 | #undef CHECK_APPEND_NOARG | |
4071 | ||
4072 | out_overflow: | |
4073 | ||
4074 | if (size_bp < size_buf) | |
4075 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4076 | else | |
4077 | buf[0] = '\0'; | |
4078 | } | |
4079 | ||
4080 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4081 | { | |
4082 | u32 size_buf = 1024; | |
4083 | char tmp_buf[192] = {'\0'}; | |
4084 | char *buf; | |
4085 | char *bp; | |
4086 | u32 size_bp = size_buf; | |
4087 | int ret; | |
4088 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4089 | ||
4090 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4091 | if (!buf) | |
4092 | return; | |
4093 | ||
4094 | bp = buf; | |
4095 | ||
4096 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4097 | do { \ | |
4098 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4099 | if (ret < 0 || ret >= size_bp) \ | |
4100 | goto out_overflow; \ | |
4101 | size_bp -= ret; \ | |
4102 | bp += ret; \ | |
4103 | } while (0) | |
4104 | ||
4105 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4106 | CHECK_APPEND_1ARG("%s", "-f "); | |
4107 | ||
4108 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4109 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4110 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4111 | } | |
4112 | ||
4113 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4114 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4115 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4116 | } | |
4117 | ||
4118 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4119 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4120 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4121 | } | |
4122 | ||
4123 | #undef CHECK_APPEND_1ARG | |
4124 | ||
4125 | out_overflow: | |
4126 | ||
4127 | if (size_bp < size_buf) | |
4128 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4129 | btrfs_info(fs_info, "balance: %s %s", | |
4130 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4131 | "resume" : "start", buf); | |
4132 | ||
4133 | kfree(buf); | |
4134 | } | |
4135 | ||
c9e9f97b | 4136 | /* |
dccdb07b | 4137 | * Should be called with balance mutexe held |
c9e9f97b | 4138 | */ |
6fcf6e2b DS |
4139 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4140 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4141 | struct btrfs_ioctl_balance_args *bargs) |
4142 | { | |
14506127 | 4143 | u64 meta_target, data_target; |
f43ffb60 | 4144 | u64 allowed; |
e4837f8f | 4145 | int mixed = 0; |
c9e9f97b | 4146 | int ret; |
8dabb742 | 4147 | u64 num_devices; |
de98ced9 | 4148 | unsigned seq; |
e62869be | 4149 | bool reducing_redundancy; |
081db89b | 4150 | int i; |
c9e9f97b | 4151 | |
837d5b6e | 4152 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4153 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4154 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4155 | ret = -EINVAL; |
4156 | goto out; | |
4157 | } | |
4158 | ||
e4837f8f ID |
4159 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4160 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4161 | mixed = 1; | |
4162 | ||
f43ffb60 ID |
4163 | /* |
4164 | * In case of mixed groups both data and meta should be picked, | |
4165 | * and identical options should be given for both of them. | |
4166 | */ | |
e4837f8f ID |
4167 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4168 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4169 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4170 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4171 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4172 | btrfs_err(fs_info, |
6dac13f8 | 4173 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4174 | ret = -EINVAL; |
4175 | goto out; | |
4176 | } | |
4177 | } | |
4178 | ||
b35cf1f0 JB |
4179 | /* |
4180 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4181 | * are exclusive |
b35cf1f0 JB |
4182 | */ |
4183 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4184 | |
4185 | /* | |
4186 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4187 | * special bit for it, to make it easier to distinguish. Thus we need | |
4188 | * to set it manually, or balance would refuse the profile. | |
4189 | */ | |
4190 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4191 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4192 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4193 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4194 | |
5ba366c3 DS |
4195 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4196 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4197 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4198 | ret = -EINVAL; |
4199 | goto out; | |
4200 | } | |
4201 | ||
6079e12c DS |
4202 | /* |
4203 | * Allow to reduce metadata or system integrity only if force set for | |
4204 | * profiles with redundancy (copies, parity) | |
4205 | */ | |
4206 | allowed = 0; | |
4207 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4208 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4209 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4210 | allowed |= btrfs_raid_array[i].bg_flag; | |
4211 | } | |
de98ced9 MX |
4212 | do { |
4213 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4214 | ||
4215 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4216 | (fs_info->avail_system_alloc_bits & allowed) && | |
4217 | !(bctl->sys.target & allowed)) || | |
4218 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4219 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4220 | !(bctl->meta.target & allowed))) |
e62869be | 4221 | reducing_redundancy = true; |
5a8067c0 | 4222 | else |
e62869be | 4223 | reducing_redundancy = false; |
5a8067c0 FM |
4224 | |
4225 | /* if we're not converting, the target field is uninitialized */ | |
4226 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4227 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4228 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4229 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4230 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4231 | |
e62869be | 4232 | if (reducing_redundancy) { |
5a8067c0 FM |
4233 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4234 | btrfs_info(fs_info, | |
e62869be | 4235 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4236 | } else { |
4237 | btrfs_err(fs_info, | |
e62869be | 4238 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4239 | ret = -EINVAL; |
4240 | goto out; | |
4241 | } | |
4242 | } | |
4243 | ||
14506127 AB |
4244 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4245 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4246 | btrfs_warn(fs_info, |
6dac13f8 | 4247 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4248 | btrfs_bg_type_to_raid_name(meta_target), |
4249 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4250 | } |
4251 | ||
6bccf3ab | 4252 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4253 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4254 | goto out; |
4255 | ||
59641015 ID |
4256 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4257 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4258 | BUG_ON(fs_info->balance_ctl); |
4259 | spin_lock(&fs_info->balance_lock); | |
4260 | fs_info->balance_ctl = bctl; | |
4261 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4262 | } else { |
4263 | BUG_ON(ret != -EEXIST); | |
4264 | spin_lock(&fs_info->balance_lock); | |
4265 | update_balance_args(bctl); | |
4266 | spin_unlock(&fs_info->balance_lock); | |
4267 | } | |
c9e9f97b | 4268 | |
3009a62f DS |
4269 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4270 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4271 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4272 | mutex_unlock(&fs_info->balance_mutex); |
4273 | ||
4274 | ret = __btrfs_balance(fs_info); | |
4275 | ||
4276 | mutex_lock(&fs_info->balance_mutex); | |
7333bd02 AJ |
4277 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) |
4278 | btrfs_info(fs_info, "balance: paused"); | |
44d354ab QW |
4279 | /* |
4280 | * Balance can be canceled by: | |
4281 | * | |
4282 | * - Regular cancel request | |
4283 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4284 | * | |
4285 | * - Fatal signal to "btrfs" process | |
4286 | * Either the signal caught by wait_reserve_ticket() and callers | |
4287 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4288 | * got -ECANCELED. | |
4289 | * Either way, in this case balance_cancel_req = 0, and | |
4290 | * ret == -EINTR or ret == -ECANCELED. | |
4291 | * | |
4292 | * So here we only check the return value to catch canceled balance. | |
4293 | */ | |
4294 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4295 | btrfs_info(fs_info, "balance: canceled"); |
4296 | else | |
4297 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4298 | ||
3009a62f | 4299 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4300 | |
4301 | if (bargs) { | |
4302 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4303 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4304 | } |
4305 | ||
3a01aa7a ID |
4306 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4307 | balance_need_close(fs_info)) { | |
149196a2 | 4308 | reset_balance_state(fs_info); |
c3e1f96c | 4309 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4310 | } |
4311 | ||
837d5b6e | 4312 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4313 | |
4314 | return ret; | |
4315 | out: | |
59641015 | 4316 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4317 | reset_balance_state(fs_info); |
a17c95df | 4318 | else |
59641015 | 4319 | kfree(bctl); |
c3e1f96c | 4320 | btrfs_exclop_finish(fs_info); |
a17c95df | 4321 | |
59641015 ID |
4322 | return ret; |
4323 | } | |
4324 | ||
4325 | static int balance_kthread(void *data) | |
4326 | { | |
2b6ba629 | 4327 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4328 | int ret = 0; |
59641015 | 4329 | |
59641015 | 4330 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4331 | if (fs_info->balance_ctl) |
6fcf6e2b | 4332 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4333 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4334 | |
59641015 ID |
4335 | return ret; |
4336 | } | |
4337 | ||
2b6ba629 ID |
4338 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4339 | { | |
4340 | struct task_struct *tsk; | |
4341 | ||
1354e1a1 | 4342 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4343 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4344 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4345 | return 0; |
4346 | } | |
1354e1a1 | 4347 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4348 | |
3cdde224 | 4349 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4350 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4351 | return 0; |
4352 | } | |
4353 | ||
02ee654d AJ |
4354 | /* |
4355 | * A ro->rw remount sequence should continue with the paused balance | |
4356 | * regardless of who pauses it, system or the user as of now, so set | |
4357 | * the resume flag. | |
4358 | */ | |
4359 | spin_lock(&fs_info->balance_lock); | |
4360 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4361 | spin_unlock(&fs_info->balance_lock); | |
4362 | ||
2b6ba629 | 4363 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4364 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4365 | } |
4366 | ||
68310a5e | 4367 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4368 | { |
59641015 ID |
4369 | struct btrfs_balance_control *bctl; |
4370 | struct btrfs_balance_item *item; | |
4371 | struct btrfs_disk_balance_args disk_bargs; | |
4372 | struct btrfs_path *path; | |
4373 | struct extent_buffer *leaf; | |
4374 | struct btrfs_key key; | |
4375 | int ret; | |
4376 | ||
4377 | path = btrfs_alloc_path(); | |
4378 | if (!path) | |
4379 | return -ENOMEM; | |
4380 | ||
59641015 | 4381 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4382 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4383 | key.offset = 0; |
4384 | ||
68310a5e | 4385 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4386 | if (ret < 0) |
68310a5e | 4387 | goto out; |
59641015 ID |
4388 | if (ret > 0) { /* ret = -ENOENT; */ |
4389 | ret = 0; | |
68310a5e ID |
4390 | goto out; |
4391 | } | |
4392 | ||
4393 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4394 | if (!bctl) { | |
4395 | ret = -ENOMEM; | |
4396 | goto out; | |
59641015 ID |
4397 | } |
4398 | ||
4399 | leaf = path->nodes[0]; | |
4400 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4401 | ||
68310a5e ID |
4402 | bctl->flags = btrfs_balance_flags(leaf, item); |
4403 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4404 | |
4405 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4406 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4407 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4408 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4409 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4410 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4411 | ||
eee95e3f DS |
4412 | /* |
4413 | * This should never happen, as the paused balance state is recovered | |
4414 | * during mount without any chance of other exclusive ops to collide. | |
4415 | * | |
4416 | * This gives the exclusive op status to balance and keeps in paused | |
4417 | * state until user intervention (cancel or umount). If the ownership | |
4418 | * cannot be assigned, show a message but do not fail. The balance | |
4419 | * is in a paused state and must have fs_info::balance_ctl properly | |
4420 | * set up. | |
4421 | */ | |
c3e1f96c | 4422 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) |
eee95e3f | 4423 | btrfs_warn(fs_info, |
6dac13f8 | 4424 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4425 | |
fb286100 JB |
4426 | btrfs_release_path(path); |
4427 | ||
68310a5e | 4428 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4429 | BUG_ON(fs_info->balance_ctl); |
4430 | spin_lock(&fs_info->balance_lock); | |
4431 | fs_info->balance_ctl = bctl; | |
4432 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4433 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4434 | out: |
4435 | btrfs_free_path(path); | |
ec44a35c CM |
4436 | return ret; |
4437 | } | |
4438 | ||
837d5b6e ID |
4439 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4440 | { | |
4441 | int ret = 0; | |
4442 | ||
4443 | mutex_lock(&fs_info->balance_mutex); | |
4444 | if (!fs_info->balance_ctl) { | |
4445 | mutex_unlock(&fs_info->balance_mutex); | |
4446 | return -ENOTCONN; | |
4447 | } | |
4448 | ||
3009a62f | 4449 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4450 | atomic_inc(&fs_info->balance_pause_req); |
4451 | mutex_unlock(&fs_info->balance_mutex); | |
4452 | ||
4453 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4454 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4455 | |
4456 | mutex_lock(&fs_info->balance_mutex); | |
4457 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4458 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4459 | atomic_dec(&fs_info->balance_pause_req); |
4460 | } else { | |
4461 | ret = -ENOTCONN; | |
4462 | } | |
4463 | ||
4464 | mutex_unlock(&fs_info->balance_mutex); | |
4465 | return ret; | |
4466 | } | |
4467 | ||
a7e99c69 ID |
4468 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4469 | { | |
4470 | mutex_lock(&fs_info->balance_mutex); | |
4471 | if (!fs_info->balance_ctl) { | |
4472 | mutex_unlock(&fs_info->balance_mutex); | |
4473 | return -ENOTCONN; | |
4474 | } | |
4475 | ||
cf7d20f4 DS |
4476 | /* |
4477 | * A paused balance with the item stored on disk can be resumed at | |
4478 | * mount time if the mount is read-write. Otherwise it's still paused | |
4479 | * and we must not allow cancelling as it deletes the item. | |
4480 | */ | |
4481 | if (sb_rdonly(fs_info->sb)) { | |
4482 | mutex_unlock(&fs_info->balance_mutex); | |
4483 | return -EROFS; | |
4484 | } | |
4485 | ||
a7e99c69 ID |
4486 | atomic_inc(&fs_info->balance_cancel_req); |
4487 | /* | |
4488 | * if we are running just wait and return, balance item is | |
4489 | * deleted in btrfs_balance in this case | |
4490 | */ | |
3009a62f | 4491 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4492 | mutex_unlock(&fs_info->balance_mutex); |
4493 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4494 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4495 | mutex_lock(&fs_info->balance_mutex); |
4496 | } else { | |
a7e99c69 | 4497 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4498 | /* |
4499 | * Lock released to allow other waiters to continue, we'll | |
4500 | * reexamine the status again. | |
4501 | */ | |
a7e99c69 ID |
4502 | mutex_lock(&fs_info->balance_mutex); |
4503 | ||
a17c95df | 4504 | if (fs_info->balance_ctl) { |
149196a2 | 4505 | reset_balance_state(fs_info); |
c3e1f96c | 4506 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4507 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4508 | } |
a7e99c69 ID |
4509 | } |
4510 | ||
3009a62f DS |
4511 | BUG_ON(fs_info->balance_ctl || |
4512 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4513 | atomic_dec(&fs_info->balance_cancel_req); |
4514 | mutex_unlock(&fs_info->balance_mutex); | |
4515 | return 0; | |
4516 | } | |
4517 | ||
97f4dd09 | 4518 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4519 | { |
4520 | struct btrfs_fs_info *fs_info = data; | |
4521 | struct btrfs_root *root = fs_info->tree_root; | |
4522 | struct btrfs_key key; | |
803b2f54 SB |
4523 | struct btrfs_path *path = NULL; |
4524 | int ret = 0; | |
4525 | struct extent_buffer *eb; | |
4526 | int slot; | |
4527 | struct btrfs_root_item root_item; | |
4528 | u32 item_size; | |
f45388f3 | 4529 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4530 | bool closing = false; |
803b2f54 SB |
4531 | |
4532 | path = btrfs_alloc_path(); | |
4533 | if (!path) { | |
4534 | ret = -ENOMEM; | |
4535 | goto out; | |
4536 | } | |
4537 | ||
4538 | key.objectid = 0; | |
4539 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4540 | key.offset = 0; | |
4541 | ||
803b2f54 | 4542 | while (1) { |
c94bec2c JB |
4543 | if (btrfs_fs_closing(fs_info)) { |
4544 | closing = true; | |
4545 | break; | |
4546 | } | |
7c829b72 AJ |
4547 | ret = btrfs_search_forward(root, &key, path, |
4548 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4549 | if (ret) { |
4550 | if (ret > 0) | |
4551 | ret = 0; | |
4552 | break; | |
4553 | } | |
4554 | ||
4555 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4556 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4557 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4558 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4559 | goto skip; | |
4560 | ||
4561 | eb = path->nodes[0]; | |
4562 | slot = path->slots[0]; | |
4563 | item_size = btrfs_item_size_nr(eb, slot); | |
4564 | if (item_size < sizeof(root_item)) | |
4565 | goto skip; | |
4566 | ||
803b2f54 SB |
4567 | read_extent_buffer(eb, &root_item, |
4568 | btrfs_item_ptr_offset(eb, slot), | |
4569 | (int)sizeof(root_item)); | |
4570 | if (btrfs_root_refs(&root_item) == 0) | |
4571 | goto skip; | |
f45388f3 FDBM |
4572 | |
4573 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4574 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4575 | if (trans) | |
4576 | goto update_tree; | |
4577 | ||
4578 | btrfs_release_path(path); | |
803b2f54 SB |
4579 | /* |
4580 | * 1 - subvol uuid item | |
4581 | * 1 - received_subvol uuid item | |
4582 | */ | |
4583 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4584 | if (IS_ERR(trans)) { | |
4585 | ret = PTR_ERR(trans); | |
4586 | break; | |
4587 | } | |
f45388f3 FDBM |
4588 | continue; |
4589 | } else { | |
4590 | goto skip; | |
4591 | } | |
4592 | update_tree: | |
9771a5cf | 4593 | btrfs_release_path(path); |
f45388f3 | 4594 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4595 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4596 | BTRFS_UUID_KEY_SUBVOL, |
4597 | key.objectid); | |
4598 | if (ret < 0) { | |
efe120a0 | 4599 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4600 | ret); |
803b2f54 SB |
4601 | break; |
4602 | } | |
4603 | } | |
4604 | ||
4605 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4606 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4607 | root_item.received_uuid, |
4608 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4609 | key.objectid); | |
4610 | if (ret < 0) { | |
efe120a0 | 4611 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4612 | ret); |
803b2f54 SB |
4613 | break; |
4614 | } | |
4615 | } | |
4616 | ||
f45388f3 | 4617 | skip: |
9771a5cf | 4618 | btrfs_release_path(path); |
803b2f54 | 4619 | if (trans) { |
3a45bb20 | 4620 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4621 | trans = NULL; |
803b2f54 SB |
4622 | if (ret) |
4623 | break; | |
4624 | } | |
4625 | ||
803b2f54 SB |
4626 | if (key.offset < (u64)-1) { |
4627 | key.offset++; | |
4628 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4629 | key.offset = 0; | |
4630 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4631 | } else if (key.objectid < (u64)-1) { | |
4632 | key.offset = 0; | |
4633 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4634 | key.objectid++; | |
4635 | } else { | |
4636 | break; | |
4637 | } | |
4638 | cond_resched(); | |
4639 | } | |
4640 | ||
4641 | out: | |
4642 | btrfs_free_path(path); | |
f45388f3 | 4643 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4644 | btrfs_end_transaction(trans); |
803b2f54 | 4645 | if (ret) |
efe120a0 | 4646 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4647 | else if (!closing) |
afcdd129 | 4648 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4649 | up(&fs_info->uuid_tree_rescan_sem); |
4650 | return 0; | |
4651 | } | |
4652 | ||
f7a81ea4 SB |
4653 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4654 | { | |
4655 | struct btrfs_trans_handle *trans; | |
4656 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4657 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4658 | struct task_struct *task; |
4659 | int ret; | |
f7a81ea4 SB |
4660 | |
4661 | /* | |
4662 | * 1 - root node | |
4663 | * 1 - root item | |
4664 | */ | |
4665 | trans = btrfs_start_transaction(tree_root, 2); | |
4666 | if (IS_ERR(trans)) | |
4667 | return PTR_ERR(trans); | |
4668 | ||
9b7a2440 | 4669 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4670 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4671 | ret = PTR_ERR(uuid_root); |
66642832 | 4672 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4673 | btrfs_end_transaction(trans); |
6d13f549 | 4674 | return ret; |
f7a81ea4 SB |
4675 | } |
4676 | ||
4677 | fs_info->uuid_root = uuid_root; | |
4678 | ||
3a45bb20 | 4679 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4680 | if (ret) |
4681 | return ret; | |
4682 | ||
4683 | down(&fs_info->uuid_tree_rescan_sem); | |
4684 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4685 | if (IS_ERR(task)) { | |
70f80175 | 4686 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4687 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4688 | up(&fs_info->uuid_tree_rescan_sem); |
4689 | return PTR_ERR(task); | |
4690 | } | |
4691 | ||
4692 | return 0; | |
f7a81ea4 | 4693 | } |
803b2f54 | 4694 | |
8f18cf13 CM |
4695 | /* |
4696 | * shrinking a device means finding all of the device extents past | |
4697 | * the new size, and then following the back refs to the chunks. | |
4698 | * The chunk relocation code actually frees the device extent | |
4699 | */ | |
4700 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4701 | { | |
0b246afa JM |
4702 | struct btrfs_fs_info *fs_info = device->fs_info; |
4703 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4704 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4705 | struct btrfs_dev_extent *dev_extent = NULL; |
4706 | struct btrfs_path *path; | |
4707 | u64 length; | |
8f18cf13 CM |
4708 | u64 chunk_offset; |
4709 | int ret; | |
4710 | int slot; | |
ba1bf481 JB |
4711 | int failed = 0; |
4712 | bool retried = false; | |
8f18cf13 CM |
4713 | struct extent_buffer *l; |
4714 | struct btrfs_key key; | |
0b246afa | 4715 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4716 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4717 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4718 | u64 diff; |
61d0d0d2 | 4719 | u64 start; |
7dfb8be1 NB |
4720 | |
4721 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4722 | start = new_size; |
0e4324a4 | 4723 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4724 | |
401e29c1 | 4725 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4726 | return -EINVAL; |
4727 | ||
8f18cf13 CM |
4728 | path = btrfs_alloc_path(); |
4729 | if (!path) | |
4730 | return -ENOMEM; | |
4731 | ||
0338dff6 | 4732 | path->reada = READA_BACK; |
8f18cf13 | 4733 | |
61d0d0d2 NB |
4734 | trans = btrfs_start_transaction(root, 0); |
4735 | if (IS_ERR(trans)) { | |
4736 | btrfs_free_path(path); | |
4737 | return PTR_ERR(trans); | |
4738 | } | |
4739 | ||
34441361 | 4740 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4741 | |
7cc8e58d | 4742 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4743 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4744 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4745 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4746 | } |
61d0d0d2 NB |
4747 | |
4748 | /* | |
4749 | * Once the device's size has been set to the new size, ensure all | |
4750 | * in-memory chunks are synced to disk so that the loop below sees them | |
4751 | * and relocates them accordingly. | |
4752 | */ | |
1c11b63e | 4753 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4754 | mutex_unlock(&fs_info->chunk_mutex); |
4755 | ret = btrfs_commit_transaction(trans); | |
4756 | if (ret) | |
4757 | goto done; | |
4758 | } else { | |
4759 | mutex_unlock(&fs_info->chunk_mutex); | |
4760 | btrfs_end_transaction(trans); | |
4761 | } | |
8f18cf13 | 4762 | |
ba1bf481 | 4763 | again: |
8f18cf13 CM |
4764 | key.objectid = device->devid; |
4765 | key.offset = (u64)-1; | |
4766 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4767 | ||
213e64da | 4768 | do { |
f3372065 | 4769 | mutex_lock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4770 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4771 | if (ret < 0) { |
f3372065 | 4772 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4773 | goto done; |
67c5e7d4 | 4774 | } |
8f18cf13 CM |
4775 | |
4776 | ret = btrfs_previous_item(root, path, 0, key.type); | |
8f18cf13 | 4777 | if (ret) { |
f3372065 | 4778 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
7056bf69 NB |
4779 | if (ret < 0) |
4780 | goto done; | |
8f18cf13 | 4781 | ret = 0; |
b3b4aa74 | 4782 | btrfs_release_path(path); |
bf1fb512 | 4783 | break; |
8f18cf13 CM |
4784 | } |
4785 | ||
4786 | l = path->nodes[0]; | |
4787 | slot = path->slots[0]; | |
4788 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4789 | ||
ba1bf481 | 4790 | if (key.objectid != device->devid) { |
f3372065 | 4791 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4792 | btrfs_release_path(path); |
bf1fb512 | 4793 | break; |
ba1bf481 | 4794 | } |
8f18cf13 CM |
4795 | |
4796 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4797 | length = btrfs_dev_extent_length(l, dev_extent); | |
4798 | ||
ba1bf481 | 4799 | if (key.offset + length <= new_size) { |
f3372065 | 4800 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4801 | btrfs_release_path(path); |
d6397bae | 4802 | break; |
ba1bf481 | 4803 | } |
8f18cf13 | 4804 | |
8f18cf13 | 4805 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4806 | btrfs_release_path(path); |
8f18cf13 | 4807 | |
a6f93c71 LB |
4808 | /* |
4809 | * We may be relocating the only data chunk we have, | |
4810 | * which could potentially end up with losing data's | |
4811 | * raid profile, so lets allocate an empty one in | |
4812 | * advance. | |
4813 | */ | |
4814 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4815 | if (ret < 0) { | |
f3372065 | 4816 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
a6f93c71 LB |
4817 | goto done; |
4818 | } | |
4819 | ||
0b246afa | 4820 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
f3372065 | 4821 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
eede2bf3 | 4822 | if (ret == -ENOSPC) { |
ba1bf481 | 4823 | failed++; |
eede2bf3 OS |
4824 | } else if (ret) { |
4825 | if (ret == -ETXTBSY) { | |
4826 | btrfs_warn(fs_info, | |
4827 | "could not shrink block group %llu due to active swapfile", | |
4828 | chunk_offset); | |
4829 | } | |
4830 | goto done; | |
4831 | } | |
213e64da | 4832 | } while (key.offset-- > 0); |
ba1bf481 JB |
4833 | |
4834 | if (failed && !retried) { | |
4835 | failed = 0; | |
4836 | retried = true; | |
4837 | goto again; | |
4838 | } else if (failed && retried) { | |
4839 | ret = -ENOSPC; | |
ba1bf481 | 4840 | goto done; |
8f18cf13 CM |
4841 | } |
4842 | ||
d6397bae | 4843 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4844 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4845 | if (IS_ERR(trans)) { |
4846 | ret = PTR_ERR(trans); | |
4847 | goto done; | |
4848 | } | |
4849 | ||
34441361 | 4850 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4851 | /* Clear all state bits beyond the shrunk device size */ |
4852 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4853 | CHUNK_STATE_MASK); | |
4854 | ||
7cc8e58d | 4855 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4856 | if (list_empty(&device->post_commit_list)) |
4857 | list_add_tail(&device->post_commit_list, | |
4858 | &trans->transaction->dev_update_list); | |
d6397bae | 4859 | |
d6397bae | 4860 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4861 | btrfs_set_super_total_bytes(super_copy, |
4862 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4863 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 MX |
4864 | |
4865 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4866 | ret = btrfs_update_device(trans, device); | |
801660b0 AJ |
4867 | if (ret < 0) { |
4868 | btrfs_abort_transaction(trans, ret); | |
4869 | btrfs_end_transaction(trans); | |
4870 | } else { | |
4871 | ret = btrfs_commit_transaction(trans); | |
4872 | } | |
8f18cf13 CM |
4873 | done: |
4874 | btrfs_free_path(path); | |
53e489bc | 4875 | if (ret) { |
34441361 | 4876 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4877 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4878 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4879 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4880 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4881 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4882 | } |
8f18cf13 CM |
4883 | return ret; |
4884 | } | |
4885 | ||
2ff7e61e | 4886 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4887 | struct btrfs_key *key, |
4888 | struct btrfs_chunk *chunk, int item_size) | |
4889 | { | |
0b246afa | 4890 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4891 | struct btrfs_disk_key disk_key; |
4892 | u32 array_size; | |
4893 | u8 *ptr; | |
4894 | ||
79bd3712 FM |
4895 | lockdep_assert_held(&fs_info->chunk_mutex); |
4896 | ||
0b86a832 | 4897 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4898 | if (array_size + item_size + sizeof(disk_key) |
79bd3712 | 4899 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) |
0b86a832 CM |
4900 | return -EFBIG; |
4901 | ||
4902 | ptr = super_copy->sys_chunk_array + array_size; | |
4903 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4904 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4905 | ptr += sizeof(disk_key); | |
4906 | memcpy(ptr, chunk, item_size); | |
4907 | item_size += sizeof(disk_key); | |
4908 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 | 4909 | |
0b86a832 CM |
4910 | return 0; |
4911 | } | |
4912 | ||
73c5de00 AJ |
4913 | /* |
4914 | * sort the devices in descending order by max_avail, total_avail | |
4915 | */ | |
4916 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4917 | { |
73c5de00 AJ |
4918 | const struct btrfs_device_info *di_a = a; |
4919 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4920 | |
73c5de00 | 4921 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4922 | return -1; |
73c5de00 | 4923 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4924 | return 1; |
73c5de00 AJ |
4925 | if (di_a->total_avail > di_b->total_avail) |
4926 | return -1; | |
4927 | if (di_a->total_avail < di_b->total_avail) | |
4928 | return 1; | |
4929 | return 0; | |
b2117a39 | 4930 | } |
0b86a832 | 4931 | |
53b381b3 DW |
4932 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4933 | { | |
ffe2d203 | 4934 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4935 | return; |
4936 | ||
ceda0864 | 4937 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4938 | } |
4939 | ||
cfbb825c DS |
4940 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
4941 | { | |
4942 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
4943 | return; | |
4944 | ||
4945 | btrfs_set_fs_incompat(info, RAID1C34); | |
4946 | } | |
4947 | ||
4f2bafe8 NA |
4948 | /* |
4949 | * Structure used internally for __btrfs_alloc_chunk() function. | |
4950 | * Wraps needed parameters. | |
4951 | */ | |
4952 | struct alloc_chunk_ctl { | |
4953 | u64 start; | |
4954 | u64 type; | |
4955 | /* Total number of stripes to allocate */ | |
4956 | int num_stripes; | |
4957 | /* sub_stripes info for map */ | |
4958 | int sub_stripes; | |
4959 | /* Stripes per device */ | |
4960 | int dev_stripes; | |
4961 | /* Maximum number of devices to use */ | |
4962 | int devs_max; | |
4963 | /* Minimum number of devices to use */ | |
4964 | int devs_min; | |
4965 | /* ndevs has to be a multiple of this */ | |
4966 | int devs_increment; | |
4967 | /* Number of copies */ | |
4968 | int ncopies; | |
4969 | /* Number of stripes worth of bytes to store parity information */ | |
4970 | int nparity; | |
4971 | u64 max_stripe_size; | |
4972 | u64 max_chunk_size; | |
6aafb303 | 4973 | u64 dev_extent_min; |
4f2bafe8 NA |
4974 | u64 stripe_size; |
4975 | u64 chunk_size; | |
4976 | int ndevs; | |
4977 | }; | |
4978 | ||
27c314d5 NA |
4979 | static void init_alloc_chunk_ctl_policy_regular( |
4980 | struct btrfs_fs_devices *fs_devices, | |
4981 | struct alloc_chunk_ctl *ctl) | |
4982 | { | |
4983 | u64 type = ctl->type; | |
4984 | ||
4985 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4986 | ctl->max_stripe_size = SZ_1G; | |
4987 | ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE; | |
4988 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4989 | /* For larger filesystems, use larger metadata chunks */ | |
4990 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) | |
4991 | ctl->max_stripe_size = SZ_1G; | |
4992 | else | |
4993 | ctl->max_stripe_size = SZ_256M; | |
4994 | ctl->max_chunk_size = ctl->max_stripe_size; | |
4995 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4996 | ctl->max_stripe_size = SZ_32M; | |
4997 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
4998 | ctl->devs_max = min_t(int, ctl->devs_max, | |
4999 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
5000 | } else { | |
5001 | BUG(); | |
5002 | } | |
5003 | ||
5004 | /* We don't want a chunk larger than 10% of writable space */ | |
5005 | ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
5006 | ctl->max_chunk_size); | |
6aafb303 | 5007 | ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; |
27c314d5 NA |
5008 | } |
5009 | ||
1cd6121f NA |
5010 | static void init_alloc_chunk_ctl_policy_zoned( |
5011 | struct btrfs_fs_devices *fs_devices, | |
5012 | struct alloc_chunk_ctl *ctl) | |
5013 | { | |
5014 | u64 zone_size = fs_devices->fs_info->zone_size; | |
5015 | u64 limit; | |
5016 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; | |
5017 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; | |
5018 | u64 min_chunk_size = min_data_stripes * zone_size; | |
5019 | u64 type = ctl->type; | |
5020 | ||
5021 | ctl->max_stripe_size = zone_size; | |
5022 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
5023 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, | |
5024 | zone_size); | |
5025 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
5026 | ctl->max_chunk_size = ctl->max_stripe_size; | |
5027 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
5028 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
5029 | ctl->devs_max = min_t(int, ctl->devs_max, | |
5030 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
bb05b298 AB |
5031 | } else { |
5032 | BUG(); | |
1cd6121f NA |
5033 | } |
5034 | ||
5035 | /* We don't want a chunk larger than 10% of writable space */ | |
5036 | limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1), | |
5037 | zone_size), | |
5038 | min_chunk_size); | |
5039 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); | |
5040 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; | |
5041 | } | |
5042 | ||
27c314d5 NA |
5043 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
5044 | struct alloc_chunk_ctl *ctl) | |
5045 | { | |
5046 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
5047 | ||
5048 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
5049 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5050 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
5051 | if (!ctl->devs_max) | |
5052 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
5053 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
5054 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
5055 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
5056 | ctl->nparity = btrfs_raid_array[index].nparity; | |
5057 | ctl->ndevs = 0; | |
5058 | ||
5059 | switch (fs_devices->chunk_alloc_policy) { | |
5060 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5061 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
5062 | break; | |
1cd6121f NA |
5063 | case BTRFS_CHUNK_ALLOC_ZONED: |
5064 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); | |
5065 | break; | |
27c314d5 NA |
5066 | default: |
5067 | BUG(); | |
5068 | } | |
5069 | } | |
5070 | ||
560156cb NA |
5071 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
5072 | struct alloc_chunk_ctl *ctl, | |
5073 | struct btrfs_device_info *devices_info) | |
b2117a39 | 5074 | { |
560156cb | 5075 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 5076 | struct btrfs_device *device; |
73c5de00 | 5077 | u64 total_avail; |
560156cb | 5078 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 5079 | int ret; |
560156cb NA |
5080 | int ndevs = 0; |
5081 | u64 max_avail; | |
5082 | u64 dev_offset; | |
0cad8a11 | 5083 | |
9f680ce0 | 5084 | /* |
73c5de00 AJ |
5085 | * in the first pass through the devices list, we gather information |
5086 | * about the available holes on each device. | |
9f680ce0 | 5087 | */ |
ebcc9301 | 5088 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 5089 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5090 | WARN(1, KERN_ERR |
efe120a0 | 5091 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5092 | continue; |
5093 | } | |
b2117a39 | 5094 | |
e12c9621 AJ |
5095 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5096 | &device->dev_state) || | |
401e29c1 | 5097 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5098 | continue; |
b2117a39 | 5099 | |
73c5de00 AJ |
5100 | if (device->total_bytes > device->bytes_used) |
5101 | total_avail = device->total_bytes - device->bytes_used; | |
5102 | else | |
5103 | total_avail = 0; | |
38c01b96 | 5104 | |
5105 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 5106 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 5107 | continue; |
b2117a39 | 5108 | |
560156cb NA |
5109 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
5110 | &max_avail); | |
73c5de00 | 5111 | if (ret && ret != -ENOSPC) |
560156cb | 5112 | return ret; |
b2117a39 | 5113 | |
73c5de00 | 5114 | if (ret == 0) |
560156cb | 5115 | max_avail = dev_extent_want; |
b2117a39 | 5116 | |
6aafb303 | 5117 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
5118 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
5119 | btrfs_debug(info, | |
560156cb | 5120 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 5121 | __func__, device->devid, max_avail, |
6aafb303 | 5122 | ctl->dev_extent_min); |
73c5de00 | 5123 | continue; |
4117f207 | 5124 | } |
b2117a39 | 5125 | |
063d006f ES |
5126 | if (ndevs == fs_devices->rw_devices) { |
5127 | WARN(1, "%s: found more than %llu devices\n", | |
5128 | __func__, fs_devices->rw_devices); | |
5129 | break; | |
5130 | } | |
73c5de00 AJ |
5131 | devices_info[ndevs].dev_offset = dev_offset; |
5132 | devices_info[ndevs].max_avail = max_avail; | |
5133 | devices_info[ndevs].total_avail = total_avail; | |
5134 | devices_info[ndevs].dev = device; | |
5135 | ++ndevs; | |
5136 | } | |
560156cb | 5137 | ctl->ndevs = ndevs; |
b2117a39 | 5138 | |
73c5de00 AJ |
5139 | /* |
5140 | * now sort the devices by hole size / available space | |
5141 | */ | |
560156cb | 5142 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 5143 | btrfs_cmp_device_info, NULL); |
b2117a39 | 5144 | |
560156cb NA |
5145 | return 0; |
5146 | } | |
5147 | ||
5badf512 NA |
5148 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
5149 | struct btrfs_device_info *devices_info) | |
5150 | { | |
5151 | /* Number of stripes that count for block group size */ | |
5152 | int data_stripes; | |
5153 | ||
5154 | /* | |
5155 | * The primary goal is to maximize the number of stripes, so use as | |
5156 | * many devices as possible, even if the stripes are not maximum sized. | |
5157 | * | |
5158 | * The DUP profile stores more than one stripe per device, the | |
5159 | * max_avail is the total size so we have to adjust. | |
5160 | */ | |
5161 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5162 | ctl->dev_stripes); | |
5163 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5164 | ||
5165 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5166 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5167 | ||
5168 | /* | |
5169 | * Use the number of data stripes to figure out how big this chunk is | |
5170 | * really going to be in terms of logical address space, and compare | |
5171 | * that answer with the max chunk size. If it's higher, we try to | |
5172 | * reduce stripe_size. | |
5173 | */ | |
5174 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5175 | /* | |
5176 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5177 | * then use it, unless it ends up being even bigger than the | |
5178 | * previous value we had already. | |
5179 | */ | |
5180 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5181 | data_stripes), SZ_16M), | |
5182 | ctl->stripe_size); | |
5183 | } | |
5184 | ||
5185 | /* Align to BTRFS_STRIPE_LEN */ | |
5186 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5187 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5188 | ||
5189 | return 0; | |
5190 | } | |
5191 | ||
1cd6121f NA |
5192 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
5193 | struct btrfs_device_info *devices_info) | |
5194 | { | |
5195 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; | |
5196 | /* Number of stripes that count for block group size */ | |
5197 | int data_stripes; | |
5198 | ||
5199 | /* | |
5200 | * It should hold because: | |
5201 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes | |
5202 | */ | |
5203 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); | |
5204 | ||
5205 | ctl->stripe_size = zone_size; | |
5206 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5207 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5208 | ||
5209 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ | |
5210 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5211 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, | |
5212 | ctl->stripe_size) + ctl->nparity, | |
5213 | ctl->dev_stripes); | |
5214 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5215 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5216 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); | |
5217 | } | |
5218 | ||
5219 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5220 | ||
5221 | return 0; | |
5222 | } | |
5223 | ||
5badf512 NA |
5224 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
5225 | struct alloc_chunk_ctl *ctl, | |
5226 | struct btrfs_device_info *devices_info) | |
5227 | { | |
5228 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5229 | ||
5230 | /* | |
5231 | * Round down to number of usable stripes, devs_increment can be any | |
5232 | * number so we can't use round_down() that requires power of 2, while | |
5233 | * rounddown is safe. | |
5234 | */ | |
5235 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5236 | ||
5237 | if (ctl->ndevs < ctl->devs_min) { | |
5238 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5239 | btrfs_debug(info, | |
5240 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5241 | __func__, ctl->ndevs, ctl->devs_min); | |
5242 | } | |
5243 | return -ENOSPC; | |
5244 | } | |
5245 | ||
5246 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5247 | ||
5248 | switch (fs_devices->chunk_alloc_policy) { | |
5249 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5250 | return decide_stripe_size_regular(ctl, devices_info); | |
1cd6121f NA |
5251 | case BTRFS_CHUNK_ALLOC_ZONED: |
5252 | return decide_stripe_size_zoned(ctl, devices_info); | |
5badf512 NA |
5253 | default: |
5254 | BUG(); | |
5255 | } | |
5256 | } | |
5257 | ||
79bd3712 | 5258 | static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, |
dce580ca NA |
5259 | struct alloc_chunk_ctl *ctl, |
5260 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5261 | { |
5262 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5263 | struct map_lookup *map = NULL; |
5264 | struct extent_map_tree *em_tree; | |
79bd3712 | 5265 | struct btrfs_block_group *block_group; |
560156cb | 5266 | struct extent_map *em; |
dce580ca NA |
5267 | u64 start = ctl->start; |
5268 | u64 type = ctl->type; | |
560156cb NA |
5269 | int ret; |
5270 | int i; | |
5271 | int j; | |
5272 | ||
dce580ca NA |
5273 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5274 | if (!map) | |
79bd3712 | 5275 | return ERR_PTR(-ENOMEM); |
dce580ca | 5276 | map->num_stripes = ctl->num_stripes; |
560156cb | 5277 | |
dce580ca NA |
5278 | for (i = 0; i < ctl->ndevs; ++i) { |
5279 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5280 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5281 | map->stripes[s].dev = devices_info[i].dev; |
5282 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5283 | j * ctl->stripe_size; |
6324fbf3 | 5284 | } |
6324fbf3 | 5285 | } |
500ceed8 NB |
5286 | map->stripe_len = BTRFS_STRIPE_LEN; |
5287 | map->io_align = BTRFS_STRIPE_LEN; | |
5288 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5289 | map->type = type; |
dce580ca | 5290 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5291 | |
dce580ca | 5292 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5293 | |
172ddd60 | 5294 | em = alloc_extent_map(); |
2b82032c | 5295 | if (!em) { |
298a8f9c | 5296 | kfree(map); |
79bd3712 | 5297 | return ERR_PTR(-ENOMEM); |
593060d7 | 5298 | } |
298a8f9c | 5299 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5300 | em->map_lookup = map; |
2b82032c | 5301 | em->start = start; |
dce580ca | 5302 | em->len = ctl->chunk_size; |
2b82032c YZ |
5303 | em->block_start = 0; |
5304 | em->block_len = em->len; | |
dce580ca | 5305 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5306 | |
c8bf1b67 | 5307 | em_tree = &info->mapping_tree; |
890871be | 5308 | write_lock(&em_tree->lock); |
09a2a8f9 | 5309 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5310 | if (ret) { |
1efb72a3 | 5311 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5312 | free_extent_map(em); |
79bd3712 | 5313 | return ERR_PTR(ret); |
0f5d42b2 | 5314 | } |
1efb72a3 NB |
5315 | write_unlock(&em_tree->lock); |
5316 | ||
79bd3712 FM |
5317 | block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); |
5318 | if (IS_ERR(block_group)) | |
6df9a95e | 5319 | goto error_del_extent; |
2b82032c | 5320 | |
bbbf7243 NB |
5321 | for (i = 0; i < map->num_stripes; i++) { |
5322 | struct btrfs_device *dev = map->stripes[i].dev; | |
5323 | ||
4f2bafe8 | 5324 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5325 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5326 | if (list_empty(&dev->post_commit_list)) |
5327 | list_add_tail(&dev->post_commit_list, | |
5328 | &trans->transaction->dev_update_list); | |
5329 | } | |
43530c46 | 5330 | |
dce580ca | 5331 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5332 | &info->free_chunk_space); |
1c116187 | 5333 | |
0f5d42b2 | 5334 | free_extent_map(em); |
0b246afa | 5335 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5336 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5337 | |
79bd3712 | 5338 | return block_group; |
b2117a39 | 5339 | |
6df9a95e | 5340 | error_del_extent: |
0f5d42b2 JB |
5341 | write_lock(&em_tree->lock); |
5342 | remove_extent_mapping(em_tree, em); | |
5343 | write_unlock(&em_tree->lock); | |
5344 | ||
5345 | /* One for our allocation */ | |
5346 | free_extent_map(em); | |
5347 | /* One for the tree reference */ | |
5348 | free_extent_map(em); | |
dce580ca | 5349 | |
79bd3712 | 5350 | return block_group; |
dce580ca NA |
5351 | } |
5352 | ||
79bd3712 FM |
5353 | struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
5354 | u64 type) | |
dce580ca NA |
5355 | { |
5356 | struct btrfs_fs_info *info = trans->fs_info; | |
5357 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5358 | struct btrfs_device_info *devices_info = NULL; | |
5359 | struct alloc_chunk_ctl ctl; | |
79bd3712 | 5360 | struct btrfs_block_group *block_group; |
dce580ca NA |
5361 | int ret; |
5362 | ||
11c67b1a NB |
5363 | lockdep_assert_held(&info->chunk_mutex); |
5364 | ||
dce580ca NA |
5365 | if (!alloc_profile_is_valid(type, 0)) { |
5366 | ASSERT(0); | |
79bd3712 | 5367 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5368 | } |
5369 | ||
5370 | if (list_empty(&fs_devices->alloc_list)) { | |
5371 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5372 | btrfs_debug(info, "%s: no writable device", __func__); | |
79bd3712 | 5373 | return ERR_PTR(-ENOSPC); |
dce580ca NA |
5374 | } |
5375 | ||
5376 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5377 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5378 | ASSERT(0); | |
79bd3712 | 5379 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5380 | } |
5381 | ||
11c67b1a | 5382 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5383 | ctl.type = type; |
5384 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5385 | ||
5386 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5387 | GFP_NOFS); | |
5388 | if (!devices_info) | |
79bd3712 | 5389 | return ERR_PTR(-ENOMEM); |
dce580ca NA |
5390 | |
5391 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5392 | if (ret < 0) { |
5393 | block_group = ERR_PTR(ret); | |
dce580ca | 5394 | goto out; |
79bd3712 | 5395 | } |
dce580ca NA |
5396 | |
5397 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5398 | if (ret < 0) { |
5399 | block_group = ERR_PTR(ret); | |
dce580ca | 5400 | goto out; |
79bd3712 | 5401 | } |
dce580ca | 5402 | |
79bd3712 | 5403 | block_group = create_chunk(trans, &ctl, devices_info); |
dce580ca NA |
5404 | |
5405 | out: | |
b2117a39 | 5406 | kfree(devices_info); |
79bd3712 | 5407 | return block_group; |
2b82032c YZ |
5408 | } |
5409 | ||
79bd3712 FM |
5410 | /* |
5411 | * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the | |
5412 | * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system | |
5413 | * chunks. | |
5414 | * | |
5415 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
5416 | * phases. | |
5417 | */ | |
5418 | int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, | |
5419 | struct btrfs_block_group *bg) | |
5420 | { | |
5421 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
5422 | struct btrfs_root *extent_root = fs_info->extent_root; | |
5423 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
5424 | struct btrfs_key key; | |
5425 | struct btrfs_chunk *chunk; | |
5426 | struct btrfs_stripe *stripe; | |
5427 | struct extent_map *em; | |
5428 | struct map_lookup *map; | |
5429 | size_t item_size; | |
5430 | int i; | |
5431 | int ret; | |
5432 | ||
5433 | /* | |
5434 | * We take the chunk_mutex for 2 reasons: | |
5435 | * | |
5436 | * 1) Updates and insertions in the chunk btree must be done while holding | |
5437 | * the chunk_mutex, as well as updating the system chunk array in the | |
5438 | * superblock. See the comment on top of btrfs_chunk_alloc() for the | |
5439 | * details; | |
5440 | * | |
5441 | * 2) To prevent races with the final phase of a device replace operation | |
5442 | * that replaces the device object associated with the map's stripes, | |
5443 | * because the device object's id can change at any time during that | |
5444 | * final phase of the device replace operation | |
5445 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
5446 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
5447 | * which would cause a failure when updating the device item, which does | |
5448 | * not exists, or persisting a stripe of the chunk item with such ID. | |
5449 | * Here we can't use the device_list_mutex because our caller already | |
5450 | * has locked the chunk_mutex, and the final phase of device replace | |
5451 | * acquires both mutexes - first the device_list_mutex and then the | |
5452 | * chunk_mutex. Using any of those two mutexes protects us from a | |
5453 | * concurrent device replace. | |
5454 | */ | |
5455 | lockdep_assert_held(&fs_info->chunk_mutex); | |
5456 | ||
5457 | em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); | |
5458 | if (IS_ERR(em)) { | |
5459 | ret = PTR_ERR(em); | |
5460 | btrfs_abort_transaction(trans, ret); | |
5461 | return ret; | |
5462 | } | |
5463 | ||
5464 | map = em->map_lookup; | |
5465 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
5466 | ||
5467 | chunk = kzalloc(item_size, GFP_NOFS); | |
5468 | if (!chunk) { | |
5469 | ret = -ENOMEM; | |
5470 | btrfs_abort_transaction(trans, ret); | |
50460e37 | 5471 | goto out; |
2b82032c YZ |
5472 | } |
5473 | ||
79bd3712 FM |
5474 | for (i = 0; i < map->num_stripes; i++) { |
5475 | struct btrfs_device *device = map->stripes[i].dev; | |
5476 | ||
5477 | ret = btrfs_update_device(trans, device); | |
5478 | if (ret) | |
5479 | goto out; | |
5480 | } | |
5481 | ||
2b82032c | 5482 | stripe = &chunk->stripe; |
6df9a95e | 5483 | for (i = 0; i < map->num_stripes; i++) { |
79bd3712 FM |
5484 | struct btrfs_device *device = map->stripes[i].dev; |
5485 | const u64 dev_offset = map->stripes[i].physical; | |
0b86a832 | 5486 | |
e17cade2 CM |
5487 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5488 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5489 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5490 | stripe++; |
0b86a832 CM |
5491 | } |
5492 | ||
79bd3712 | 5493 | btrfs_set_stack_chunk_length(chunk, bg->length); |
0b86a832 | 5494 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
5495 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5496 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5497 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5498 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5499 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5500 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5501 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5502 | |
2b82032c YZ |
5503 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5504 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
79bd3712 | 5505 | key.offset = bg->start; |
0b86a832 | 5506 | |
2b82032c | 5507 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
79bd3712 FM |
5508 | if (ret) |
5509 | goto out; | |
5510 | ||
5511 | bg->chunk_item_inserted = 1; | |
5512 | ||
5513 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2ff7e61e | 5514 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
79bd3712 FM |
5515 | if (ret) |
5516 | goto out; | |
8f18cf13 | 5517 | } |
1abe9b8a | 5518 | |
6df9a95e | 5519 | out: |
0b86a832 | 5520 | kfree(chunk); |
6df9a95e | 5521 | free_extent_map(em); |
4ed1d16e | 5522 | return ret; |
2b82032c | 5523 | } |
0b86a832 | 5524 | |
6f8e0fc7 | 5525 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5526 | { |
6f8e0fc7 | 5527 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5528 | u64 alloc_profile; |
79bd3712 FM |
5529 | struct btrfs_block_group *meta_bg; |
5530 | struct btrfs_block_group *sys_bg; | |
5531 | ||
5532 | /* | |
5533 | * When adding a new device for sprouting, the seed device is read-only | |
5534 | * so we must first allocate a metadata and a system chunk. But before | |
5535 | * adding the block group items to the extent, device and chunk btrees, | |
5536 | * we must first: | |
5537 | * | |
5538 | * 1) Create both chunks without doing any changes to the btrees, as | |
5539 | * otherwise we would get -ENOSPC since the block groups from the | |
5540 | * seed device are read-only; | |
5541 | * | |
5542 | * 2) Add the device item for the new sprout device - finishing the setup | |
5543 | * of a new block group requires updating the device item in the chunk | |
5544 | * btree, so it must exist when we attempt to do it. The previous step | |
5545 | * ensures this does not fail with -ENOSPC. | |
5546 | * | |
5547 | * After that we can add the block group items to their btrees: | |
5548 | * update existing device item in the chunk btree, add a new block group | |
5549 | * item to the extent btree, add a new chunk item to the chunk btree and | |
5550 | * finally add the new device extent items to the devices btree. | |
5551 | */ | |
2b82032c | 5552 | |
1b86826d | 5553 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
79bd3712 FM |
5554 | meta_bg = btrfs_alloc_chunk(trans, alloc_profile); |
5555 | if (IS_ERR(meta_bg)) | |
5556 | return PTR_ERR(meta_bg); | |
2b82032c | 5557 | |
1b86826d | 5558 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
79bd3712 FM |
5559 | sys_bg = btrfs_alloc_chunk(trans, alloc_profile); |
5560 | if (IS_ERR(sys_bg)) | |
5561 | return PTR_ERR(sys_bg); | |
5562 | ||
5563 | return 0; | |
2b82032c YZ |
5564 | } |
5565 | ||
d20983b4 MX |
5566 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5567 | { | |
fc9a2ac7 | 5568 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5569 | |
fc9a2ac7 | 5570 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5571 | } |
5572 | ||
2ff7e61e | 5573 | int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5574 | { |
5575 | struct extent_map *em; | |
5576 | struct map_lookup *map; | |
2b82032c | 5577 | int readonly = 0; |
d20983b4 | 5578 | int miss_ndevs = 0; |
2b82032c YZ |
5579 | int i; |
5580 | ||
60ca842e | 5581 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5582 | if (IS_ERR(em)) |
2b82032c YZ |
5583 | return 1; |
5584 | ||
95617d69 | 5585 | map = em->map_lookup; |
2b82032c | 5586 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5587 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5588 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5589 | miss_ndevs++; |
5590 | continue; | |
5591 | } | |
ebbede42 AJ |
5592 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5593 | &map->stripes[i].dev->dev_state)) { | |
2b82032c | 5594 | readonly = 1; |
d20983b4 | 5595 | goto end; |
2b82032c YZ |
5596 | } |
5597 | } | |
d20983b4 MX |
5598 | |
5599 | /* | |
5600 | * If the number of missing devices is larger than max errors, | |
5601 | * we can not write the data into that chunk successfully, so | |
5602 | * set it readonly. | |
5603 | */ | |
5604 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5605 | readonly = 1; | |
5606 | end: | |
0b86a832 | 5607 | free_extent_map(em); |
2b82032c | 5608 | return readonly; |
0b86a832 CM |
5609 | } |
5610 | ||
c8bf1b67 | 5611 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5612 | { |
5613 | struct extent_map *em; | |
5614 | ||
d397712b | 5615 | while (1) { |
c8bf1b67 DS |
5616 | write_lock(&tree->lock); |
5617 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5618 | if (em) |
c8bf1b67 DS |
5619 | remove_extent_mapping(tree, em); |
5620 | write_unlock(&tree->lock); | |
0b86a832 CM |
5621 | if (!em) |
5622 | break; | |
0b86a832 CM |
5623 | /* once for us */ |
5624 | free_extent_map(em); | |
5625 | /* once for the tree */ | |
5626 | free_extent_map(em); | |
5627 | } | |
5628 | } | |
5629 | ||
5d964051 | 5630 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5631 | { |
5632 | struct extent_map *em; | |
5633 | struct map_lookup *map; | |
f188591e CM |
5634 | int ret; |
5635 | ||
60ca842e | 5636 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5637 | if (IS_ERR(em)) |
5638 | /* | |
5639 | * We could return errors for these cases, but that could get | |
5640 | * ugly and we'd probably do the same thing which is just not do | |
5641 | * anything else and exit, so return 1 so the callers don't try | |
5642 | * to use other copies. | |
5643 | */ | |
fb7669b5 | 5644 | return 1; |
fb7669b5 | 5645 | |
95617d69 | 5646 | map = em->map_lookup; |
c7369b3f | 5647 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK)) |
f188591e | 5648 | ret = map->num_stripes; |
321aecc6 CM |
5649 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5650 | ret = map->sub_stripes; | |
53b381b3 DW |
5651 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5652 | ret = 2; | |
5653 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5654 | /* |
5655 | * There could be two corrupted data stripes, we need | |
5656 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5657 | * |
8810f751 LB |
5658 | * Fail a stripe at a time on every retry except the |
5659 | * stripe under reconstruction. | |
5660 | */ | |
5661 | ret = map->num_stripes; | |
f188591e CM |
5662 | else |
5663 | ret = 1; | |
5664 | free_extent_map(em); | |
ad6d620e | 5665 | |
cb5583dd | 5666 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5667 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5668 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5669 | ret++; |
cb5583dd | 5670 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5671 | |
f188591e CM |
5672 | return ret; |
5673 | } | |
5674 | ||
2ff7e61e | 5675 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5676 | u64 logical) |
5677 | { | |
5678 | struct extent_map *em; | |
5679 | struct map_lookup *map; | |
0b246afa | 5680 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5681 | |
60ca842e | 5682 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5683 | |
69f03f13 NB |
5684 | if (!WARN_ON(IS_ERR(em))) { |
5685 | map = em->map_lookup; | |
5686 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5687 | len = map->stripe_len * nr_data_stripes(map); | |
5688 | free_extent_map(em); | |
5689 | } | |
53b381b3 DW |
5690 | return len; |
5691 | } | |
5692 | ||
e4ff5fb5 | 5693 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5694 | { |
5695 | struct extent_map *em; | |
5696 | struct map_lookup *map; | |
53b381b3 DW |
5697 | int ret = 0; |
5698 | ||
60ca842e | 5699 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5700 | |
69f03f13 NB |
5701 | if(!WARN_ON(IS_ERR(em))) { |
5702 | map = em->map_lookup; | |
5703 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5704 | ret = 1; | |
5705 | free_extent_map(em); | |
5706 | } | |
53b381b3 DW |
5707 | return ret; |
5708 | } | |
5709 | ||
30d9861f | 5710 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5711 | struct map_lookup *map, int first, |
8ba0ae78 | 5712 | int dev_replace_is_ongoing) |
dfe25020 CM |
5713 | { |
5714 | int i; | |
99f92a7c | 5715 | int num_stripes; |
8ba0ae78 | 5716 | int preferred_mirror; |
30d9861f SB |
5717 | int tolerance; |
5718 | struct btrfs_device *srcdev; | |
5719 | ||
99f92a7c | 5720 | ASSERT((map->type & |
c7369b3f | 5721 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5722 | |
5723 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5724 | num_stripes = map->sub_stripes; | |
5725 | else | |
5726 | num_stripes = map->num_stripes; | |
5727 | ||
33fd2f71 AJ |
5728 | switch (fs_info->fs_devices->read_policy) { |
5729 | default: | |
5730 | /* Shouldn't happen, just warn and use pid instead of failing */ | |
5731 | btrfs_warn_rl(fs_info, | |
5732 | "unknown read_policy type %u, reset to pid", | |
5733 | fs_info->fs_devices->read_policy); | |
5734 | fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; | |
5735 | fallthrough; | |
5736 | case BTRFS_READ_POLICY_PID: | |
5737 | preferred_mirror = first + (current->pid % num_stripes); | |
5738 | break; | |
5739 | } | |
8ba0ae78 | 5740 | |
30d9861f SB |
5741 | if (dev_replace_is_ongoing && |
5742 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5743 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5744 | srcdev = fs_info->dev_replace.srcdev; | |
5745 | else | |
5746 | srcdev = NULL; | |
5747 | ||
5748 | /* | |
5749 | * try to avoid the drive that is the source drive for a | |
5750 | * dev-replace procedure, only choose it if no other non-missing | |
5751 | * mirror is available | |
5752 | */ | |
5753 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5754 | if (map->stripes[preferred_mirror].dev->bdev && |
5755 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5756 | return preferred_mirror; | |
99f92a7c | 5757 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5758 | if (map->stripes[i].dev->bdev && |
5759 | (tolerance || map->stripes[i].dev != srcdev)) | |
5760 | return i; | |
5761 | } | |
dfe25020 | 5762 | } |
30d9861f | 5763 | |
dfe25020 CM |
5764 | /* we couldn't find one that doesn't fail. Just return something |
5765 | * and the io error handling code will clean up eventually | |
5766 | */ | |
8ba0ae78 | 5767 | return preferred_mirror; |
dfe25020 CM |
5768 | } |
5769 | ||
53b381b3 | 5770 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ |
8e5cfb55 | 5771 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 | 5772 | { |
53b381b3 | 5773 | int i; |
53b381b3 DW |
5774 | int again = 1; |
5775 | ||
5776 | while (again) { | |
5777 | again = 0; | |
cc7539ed | 5778 | for (i = 0; i < num_stripes - 1; i++) { |
eeb6f172 DS |
5779 | /* Swap if parity is on a smaller index */ |
5780 | if (bbio->raid_map[i] > bbio->raid_map[i + 1]) { | |
5781 | swap(bbio->stripes[i], bbio->stripes[i + 1]); | |
5782 | swap(bbio->raid_map[i], bbio->raid_map[i + 1]); | |
53b381b3 DW |
5783 | again = 1; |
5784 | } | |
5785 | } | |
5786 | } | |
5787 | } | |
5788 | ||
6e9606d2 ZL |
5789 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5790 | { | |
5791 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5792 | /* the size of the btrfs_bio */ |
6e9606d2 | 5793 | sizeof(struct btrfs_bio) + |
e57cf21e | 5794 | /* plus the variable array for the stripes */ |
6e9606d2 | 5795 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5796 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5797 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5798 | /* |
5799 | * plus the raid_map, which includes both the tgt dev | |
5800 | * and the stripes | |
5801 | */ | |
5802 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5803 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5804 | |
5805 | atomic_set(&bbio->error, 0); | |
140475ae | 5806 | refcount_set(&bbio->refs, 1); |
6e9606d2 | 5807 | |
608769a4 NB |
5808 | bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes); |
5809 | bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes); | |
5810 | ||
6e9606d2 ZL |
5811 | return bbio; |
5812 | } | |
5813 | ||
5814 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5815 | { | |
140475ae ER |
5816 | WARN_ON(!refcount_read(&bbio->refs)); |
5817 | refcount_inc(&bbio->refs); | |
6e9606d2 ZL |
5818 | } |
5819 | ||
5820 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5821 | { | |
5822 | if (!bbio) | |
5823 | return; | |
140475ae | 5824 | if (refcount_dec_and_test(&bbio->refs)) |
6e9606d2 ZL |
5825 | kfree(bbio); |
5826 | } | |
5827 | ||
0b3d4cd3 LB |
5828 | /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */ |
5829 | /* | |
5830 | * Please note that, discard won't be sent to target device of device | |
5831 | * replace. | |
5832 | */ | |
5833 | static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info, | |
6b7faadd | 5834 | u64 logical, u64 *length_ret, |
0b3d4cd3 LB |
5835 | struct btrfs_bio **bbio_ret) |
5836 | { | |
5837 | struct extent_map *em; | |
5838 | struct map_lookup *map; | |
5839 | struct btrfs_bio *bbio; | |
6b7faadd | 5840 | u64 length = *length_ret; |
0b3d4cd3 LB |
5841 | u64 offset; |
5842 | u64 stripe_nr; | |
5843 | u64 stripe_nr_end; | |
5844 | u64 stripe_end_offset; | |
5845 | u64 stripe_cnt; | |
5846 | u64 stripe_len; | |
5847 | u64 stripe_offset; | |
5848 | u64 num_stripes; | |
5849 | u32 stripe_index; | |
5850 | u32 factor = 0; | |
5851 | u32 sub_stripes = 0; | |
5852 | u64 stripes_per_dev = 0; | |
5853 | u32 remaining_stripes = 0; | |
5854 | u32 last_stripe = 0; | |
5855 | int ret = 0; | |
5856 | int i; | |
5857 | ||
5858 | /* discard always return a bbio */ | |
5859 | ASSERT(bbio_ret); | |
5860 | ||
60ca842e | 5861 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 LB |
5862 | if (IS_ERR(em)) |
5863 | return PTR_ERR(em); | |
5864 | ||
5865 | map = em->map_lookup; | |
5866 | /* we don't discard raid56 yet */ | |
5867 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5868 | ret = -EOPNOTSUPP; | |
5869 | goto out; | |
5870 | } | |
5871 | ||
5872 | offset = logical - em->start; | |
2d974619 | 5873 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5874 | *length_ret = length; |
0b3d4cd3 LB |
5875 | |
5876 | stripe_len = map->stripe_len; | |
5877 | /* | |
5878 | * stripe_nr counts the total number of stripes we have to stride | |
5879 | * to get to this block | |
5880 | */ | |
5881 | stripe_nr = div64_u64(offset, stripe_len); | |
5882 | ||
5883 | /* stripe_offset is the offset of this block in its stripe */ | |
5884 | stripe_offset = offset - stripe_nr * stripe_len; | |
5885 | ||
5886 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5887 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5888 | stripe_cnt = stripe_nr_end - stripe_nr; |
5889 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5890 | (offset + length); | |
5891 | /* | |
5892 | * after this, stripe_nr is the number of stripes on this | |
5893 | * device we have to walk to find the data, and stripe_index is | |
5894 | * the number of our device in the stripe array | |
5895 | */ | |
5896 | num_stripes = 1; | |
5897 | stripe_index = 0; | |
5898 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5899 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5900 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5901 | sub_stripes = 1; | |
5902 | else | |
5903 | sub_stripes = map->sub_stripes; | |
5904 | ||
5905 | factor = map->num_stripes / sub_stripes; | |
5906 | num_stripes = min_t(u64, map->num_stripes, | |
5907 | sub_stripes * stripe_cnt); | |
5908 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5909 | stripe_index *= sub_stripes; | |
5910 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
5911 | &remaining_stripes); | |
5912 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5913 | last_stripe *= sub_stripes; | |
c7369b3f | 5914 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 LB |
5915 | BTRFS_BLOCK_GROUP_DUP)) { |
5916 | num_stripes = map->num_stripes; | |
5917 | } else { | |
5918 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5919 | &stripe_index); | |
5920 | } | |
5921 | ||
5922 | bbio = alloc_btrfs_bio(num_stripes, 0); | |
5923 | if (!bbio) { | |
5924 | ret = -ENOMEM; | |
5925 | goto out; | |
5926 | } | |
5927 | ||
5928 | for (i = 0; i < num_stripes; i++) { | |
5929 | bbio->stripes[i].physical = | |
5930 | map->stripes[stripe_index].physical + | |
5931 | stripe_offset + stripe_nr * map->stripe_len; | |
5932 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5933 | ||
5934 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5935 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5936 | bbio->stripes[i].length = stripes_per_dev * | |
5937 | map->stripe_len; | |
5938 | ||
5939 | if (i / sub_stripes < remaining_stripes) | |
5940 | bbio->stripes[i].length += | |
5941 | map->stripe_len; | |
5942 | ||
5943 | /* | |
5944 | * Special for the first stripe and | |
5945 | * the last stripe: | |
5946 | * | |
5947 | * |-------|...|-------| | |
5948 | * |----------| | |
5949 | * off end_off | |
5950 | */ | |
5951 | if (i < sub_stripes) | |
5952 | bbio->stripes[i].length -= | |
5953 | stripe_offset; | |
5954 | ||
5955 | if (stripe_index >= last_stripe && | |
5956 | stripe_index <= (last_stripe + | |
5957 | sub_stripes - 1)) | |
5958 | bbio->stripes[i].length -= | |
5959 | stripe_end_offset; | |
5960 | ||
5961 | if (i == sub_stripes - 1) | |
5962 | stripe_offset = 0; | |
5963 | } else { | |
5964 | bbio->stripes[i].length = length; | |
5965 | } | |
5966 | ||
5967 | stripe_index++; | |
5968 | if (stripe_index == map->num_stripes) { | |
5969 | stripe_index = 0; | |
5970 | stripe_nr++; | |
5971 | } | |
5972 | } | |
5973 | ||
5974 | *bbio_ret = bbio; | |
5975 | bbio->map_type = map->type; | |
5976 | bbio->num_stripes = num_stripes; | |
5977 | out: | |
5978 | free_extent_map(em); | |
5979 | return ret; | |
5980 | } | |
5981 | ||
5ab56090 LB |
5982 | /* |
5983 | * In dev-replace case, for repair case (that's the only case where the mirror | |
5984 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
5985 | * left cursor can also be read from the target drive. | |
5986 | * | |
5987 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
5988 | * array of stripes. | |
5989 | * For READ, it also needs to be supported using the same mirror number. | |
5990 | * | |
5991 | * If the requested block is not left of the left cursor, EIO is returned. This | |
5992 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
5993 | * case. | |
5994 | */ | |
5995 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
5996 | u64 logical, u64 length, | |
5997 | u64 srcdev_devid, int *mirror_num, | |
5998 | u64 *physical) | |
5999 | { | |
6000 | struct btrfs_bio *bbio = NULL; | |
6001 | int num_stripes; | |
6002 | int index_srcdev = 0; | |
6003 | int found = 0; | |
6004 | u64 physical_of_found = 0; | |
6005 | int i; | |
6006 | int ret = 0; | |
6007 | ||
6008 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
6009 | logical, &length, &bbio, 0, 0); | |
6010 | if (ret) { | |
6011 | ASSERT(bbio == NULL); | |
6012 | return ret; | |
6013 | } | |
6014 | ||
6015 | num_stripes = bbio->num_stripes; | |
6016 | if (*mirror_num > num_stripes) { | |
6017 | /* | |
6018 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
6019 | * that means that the requested area is not left of the left | |
6020 | * cursor | |
6021 | */ | |
6022 | btrfs_put_bbio(bbio); | |
6023 | return -EIO; | |
6024 | } | |
6025 | ||
6026 | /* | |
6027 | * process the rest of the function using the mirror_num of the source | |
6028 | * drive. Therefore look it up first. At the end, patch the device | |
6029 | * pointer to the one of the target drive. | |
6030 | */ | |
6031 | for (i = 0; i < num_stripes; i++) { | |
6032 | if (bbio->stripes[i].dev->devid != srcdev_devid) | |
6033 | continue; | |
6034 | ||
6035 | /* | |
6036 | * In case of DUP, in order to keep it simple, only add the | |
6037 | * mirror with the lowest physical address | |
6038 | */ | |
6039 | if (found && | |
6040 | physical_of_found <= bbio->stripes[i].physical) | |
6041 | continue; | |
6042 | ||
6043 | index_srcdev = i; | |
6044 | found = 1; | |
6045 | physical_of_found = bbio->stripes[i].physical; | |
6046 | } | |
6047 | ||
6048 | btrfs_put_bbio(bbio); | |
6049 | ||
6050 | ASSERT(found); | |
6051 | if (!found) | |
6052 | return -EIO; | |
6053 | ||
6054 | *mirror_num = index_srcdev + 1; | |
6055 | *physical = physical_of_found; | |
6056 | return ret; | |
6057 | } | |
6058 | ||
6143c23c NA |
6059 | static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) |
6060 | { | |
6061 | struct btrfs_block_group *cache; | |
6062 | bool ret; | |
6063 | ||
de17addc | 6064 | /* Non zoned filesystem does not use "to_copy" flag */ |
6143c23c NA |
6065 | if (!btrfs_is_zoned(fs_info)) |
6066 | return false; | |
6067 | ||
6068 | cache = btrfs_lookup_block_group(fs_info, logical); | |
6069 | ||
6070 | spin_lock(&cache->lock); | |
6071 | ret = cache->to_copy; | |
6072 | spin_unlock(&cache->lock); | |
6073 | ||
6074 | btrfs_put_block_group(cache); | |
6075 | return ret; | |
6076 | } | |
6077 | ||
73c0f228 LB |
6078 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
6079 | struct btrfs_bio **bbio_ret, | |
6080 | struct btrfs_dev_replace *dev_replace, | |
6143c23c | 6081 | u64 logical, |
73c0f228 LB |
6082 | int *num_stripes_ret, int *max_errors_ret) |
6083 | { | |
6084 | struct btrfs_bio *bbio = *bbio_ret; | |
6085 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
6086 | int tgtdev_indexes = 0; | |
6087 | int num_stripes = *num_stripes_ret; | |
6088 | int max_errors = *max_errors_ret; | |
6089 | int i; | |
6090 | ||
6091 | if (op == BTRFS_MAP_WRITE) { | |
6092 | int index_where_to_add; | |
6093 | ||
6143c23c NA |
6094 | /* |
6095 | * A block group which have "to_copy" set will eventually | |
6096 | * copied by dev-replace process. We can avoid cloning IO here. | |
6097 | */ | |
6098 | if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) | |
6099 | return; | |
6100 | ||
73c0f228 LB |
6101 | /* |
6102 | * duplicate the write operations while the dev replace | |
6103 | * procedure is running. Since the copying of the old disk to | |
6104 | * the new disk takes place at run time while the filesystem is | |
6105 | * mounted writable, the regular write operations to the old | |
6106 | * disk have to be duplicated to go to the new disk as well. | |
6107 | * | |
6108 | * Note that device->missing is handled by the caller, and that | |
6109 | * the write to the old disk is already set up in the stripes | |
6110 | * array. | |
6111 | */ | |
6112 | index_where_to_add = num_stripes; | |
6113 | for (i = 0; i < num_stripes; i++) { | |
6114 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
6115 | /* write to new disk, too */ | |
6116 | struct btrfs_bio_stripe *new = | |
6117 | bbio->stripes + index_where_to_add; | |
6118 | struct btrfs_bio_stripe *old = | |
6119 | bbio->stripes + i; | |
6120 | ||
6121 | new->physical = old->physical; | |
6122 | new->length = old->length; | |
6123 | new->dev = dev_replace->tgtdev; | |
6124 | bbio->tgtdev_map[i] = index_where_to_add; | |
6125 | index_where_to_add++; | |
6126 | max_errors++; | |
6127 | tgtdev_indexes++; | |
6128 | } | |
6129 | } | |
6130 | num_stripes = index_where_to_add; | |
6131 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
6132 | int index_srcdev = 0; | |
6133 | int found = 0; | |
6134 | u64 physical_of_found = 0; | |
6135 | ||
6136 | /* | |
6137 | * During the dev-replace procedure, the target drive can also | |
6138 | * be used to read data in case it is needed to repair a corrupt | |
6139 | * block elsewhere. This is possible if the requested area is | |
6140 | * left of the left cursor. In this area, the target drive is a | |
6141 | * full copy of the source drive. | |
6142 | */ | |
6143 | for (i = 0; i < num_stripes; i++) { | |
6144 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
6145 | /* | |
6146 | * In case of DUP, in order to keep it simple, | |
6147 | * only add the mirror with the lowest physical | |
6148 | * address | |
6149 | */ | |
6150 | if (found && | |
6151 | physical_of_found <= | |
6152 | bbio->stripes[i].physical) | |
6153 | continue; | |
6154 | index_srcdev = i; | |
6155 | found = 1; | |
6156 | physical_of_found = bbio->stripes[i].physical; | |
6157 | } | |
6158 | } | |
6159 | if (found) { | |
6160 | struct btrfs_bio_stripe *tgtdev_stripe = | |
6161 | bbio->stripes + num_stripes; | |
6162 | ||
6163 | tgtdev_stripe->physical = physical_of_found; | |
6164 | tgtdev_stripe->length = | |
6165 | bbio->stripes[index_srcdev].length; | |
6166 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
6167 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
6168 | ||
6169 | tgtdev_indexes++; | |
6170 | num_stripes++; | |
6171 | } | |
6172 | } | |
6173 | ||
6174 | *num_stripes_ret = num_stripes; | |
6175 | *max_errors_ret = max_errors; | |
6176 | bbio->num_tgtdevs = tgtdev_indexes; | |
6177 | *bbio_ret = bbio; | |
6178 | } | |
6179 | ||
2b19a1fe LB |
6180 | static bool need_full_stripe(enum btrfs_map_op op) |
6181 | { | |
6182 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
6183 | } | |
6184 | ||
5f141126 | 6185 | /* |
42034313 MR |
6186 | * Calculate the geometry of a particular (address, len) tuple. This |
6187 | * information is used to calculate how big a particular bio can get before it | |
6188 | * straddles a stripe. | |
5f141126 | 6189 | * |
42034313 MR |
6190 | * @fs_info: the filesystem |
6191 | * @em: mapping containing the logical extent | |
6192 | * @op: type of operation - write or read | |
6193 | * @logical: address that we want to figure out the geometry of | |
42034313 | 6194 | * @io_geom: pointer used to return values |
5f141126 NB |
6195 | * |
6196 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
6197 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
6198 | */ | |
42034313 | 6199 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, |
43c0d1a5 | 6200 | enum btrfs_map_op op, u64 logical, |
42034313 | 6201 | struct btrfs_io_geometry *io_geom) |
5f141126 | 6202 | { |
5f141126 | 6203 | struct map_lookup *map; |
43c0d1a5 | 6204 | u64 len; |
5f141126 NB |
6205 | u64 offset; |
6206 | u64 stripe_offset; | |
6207 | u64 stripe_nr; | |
6208 | u64 stripe_len; | |
6209 | u64 raid56_full_stripe_start = (u64)-1; | |
6210 | int data_stripes; | |
6211 | ||
6212 | ASSERT(op != BTRFS_MAP_DISCARD); | |
6213 | ||
5f141126 NB |
6214 | map = em->map_lookup; |
6215 | /* Offset of this logical address in the chunk */ | |
6216 | offset = logical - em->start; | |
6217 | /* Len of a stripe in a chunk */ | |
6218 | stripe_len = map->stripe_len; | |
1a9fd417 | 6219 | /* Stripe where this block falls in */ |
5f141126 NB |
6220 | stripe_nr = div64_u64(offset, stripe_len); |
6221 | /* Offset of stripe in the chunk */ | |
6222 | stripe_offset = stripe_nr * stripe_len; | |
6223 | if (offset < stripe_offset) { | |
6224 | btrfs_crit(fs_info, | |
6225 | "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu", | |
6226 | stripe_offset, offset, em->start, logical, stripe_len); | |
42034313 | 6227 | return -EINVAL; |
5f141126 NB |
6228 | } |
6229 | ||
6230 | /* stripe_offset is the offset of this block in its stripe */ | |
6231 | stripe_offset = offset - stripe_offset; | |
6232 | data_stripes = nr_data_stripes(map); | |
6233 | ||
6234 | if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
6235 | u64 max_len = stripe_len - stripe_offset; | |
6236 | ||
6237 | /* | |
6238 | * In case of raid56, we need to know the stripe aligned start | |
6239 | */ | |
6240 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
6241 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
6242 | raid56_full_stripe_start = offset; | |
6243 | ||
6244 | /* | |
6245 | * Allow a write of a full stripe, but make sure we | |
6246 | * don't allow straddling of stripes | |
6247 | */ | |
6248 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
6249 | full_stripe_len); | |
6250 | raid56_full_stripe_start *= full_stripe_len; | |
6251 | ||
6252 | /* | |
6253 | * For writes to RAID[56], allow a full stripeset across | |
6254 | * all disks. For other RAID types and for RAID[56] | |
6255 | * reads, just allow a single stripe (on a single disk). | |
6256 | */ | |
6257 | if (op == BTRFS_MAP_WRITE) { | |
6258 | max_len = stripe_len * data_stripes - | |
6259 | (offset - raid56_full_stripe_start); | |
6260 | } | |
6261 | } | |
6262 | len = min_t(u64, em->len - offset, max_len); | |
6263 | } else { | |
6264 | len = em->len - offset; | |
6265 | } | |
6266 | ||
6267 | io_geom->len = len; | |
6268 | io_geom->offset = offset; | |
6269 | io_geom->stripe_len = stripe_len; | |
6270 | io_geom->stripe_nr = stripe_nr; | |
6271 | io_geom->stripe_offset = stripe_offset; | |
6272 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
6273 | ||
42034313 | 6274 | return 0; |
5f141126 NB |
6275 | } |
6276 | ||
cf8cddd3 CH |
6277 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
6278 | enum btrfs_map_op op, | |
f2d8d74d | 6279 | u64 logical, u64 *length, |
a1d3c478 | 6280 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 6281 | int mirror_num, int need_raid_map) |
0b86a832 CM |
6282 | { |
6283 | struct extent_map *em; | |
6284 | struct map_lookup *map; | |
593060d7 CM |
6285 | u64 stripe_offset; |
6286 | u64 stripe_nr; | |
53b381b3 | 6287 | u64 stripe_len; |
9d644a62 | 6288 | u32 stripe_index; |
cff82672 | 6289 | int data_stripes; |
cea9e445 | 6290 | int i; |
de11cc12 | 6291 | int ret = 0; |
f2d8d74d | 6292 | int num_stripes; |
a236aed1 | 6293 | int max_errors = 0; |
2c8cdd6e | 6294 | int tgtdev_indexes = 0; |
a1d3c478 | 6295 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
6296 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6297 | int dev_replace_is_ongoing = 0; | |
6298 | int num_alloc_stripes; | |
ad6d620e SB |
6299 | int patch_the_first_stripe_for_dev_replace = 0; |
6300 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6301 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6302 | struct btrfs_io_geometry geom; |
6303 | ||
6304 | ASSERT(bbio_ret); | |
75fb2e9e | 6305 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6306 | |
42034313 MR |
6307 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
6308 | ASSERT(!IS_ERR(em)); | |
6309 | ||
43c0d1a5 | 6310 | ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); |
89b798ad NB |
6311 | if (ret < 0) |
6312 | return ret; | |
0b86a832 | 6313 | |
95617d69 | 6314 | map = em->map_lookup; |
593060d7 | 6315 | |
89b798ad | 6316 | *length = geom.len; |
89b798ad NB |
6317 | stripe_len = geom.stripe_len; |
6318 | stripe_nr = geom.stripe_nr; | |
6319 | stripe_offset = geom.stripe_offset; | |
6320 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6321 | data_stripes = nr_data_stripes(map); |
593060d7 | 6322 | |
cb5583dd | 6323 | down_read(&dev_replace->rwsem); |
472262f3 | 6324 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6325 | /* |
6326 | * Hold the semaphore for read during the whole operation, write is | |
6327 | * requested at commit time but must wait. | |
6328 | */ | |
472262f3 | 6329 | if (!dev_replace_is_ongoing) |
cb5583dd | 6330 | up_read(&dev_replace->rwsem); |
472262f3 | 6331 | |
ad6d620e | 6332 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6333 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6334 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6335 | dev_replace->srcdev->devid, | |
6336 | &mirror_num, | |
6337 | &physical_to_patch_in_first_stripe); | |
6338 | if (ret) | |
ad6d620e | 6339 | goto out; |
5ab56090 LB |
6340 | else |
6341 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6342 | } else if (mirror_num > map->num_stripes) { |
6343 | mirror_num = 0; | |
6344 | } | |
6345 | ||
f2d8d74d | 6346 | num_stripes = 1; |
cea9e445 | 6347 | stripe_index = 0; |
fce3bb9a | 6348 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6349 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6350 | &stripe_index); | |
de483734 | 6351 | if (!need_full_stripe(op)) |
28e1cc7d | 6352 | mirror_num = 1; |
c7369b3f | 6353 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6354 | if (need_full_stripe(op)) |
f2d8d74d | 6355 | num_stripes = map->num_stripes; |
2fff734f | 6356 | else if (mirror_num) |
f188591e | 6357 | stripe_index = mirror_num - 1; |
dfe25020 | 6358 | else { |
30d9861f | 6359 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6360 | dev_replace_is_ongoing); |
a1d3c478 | 6361 | mirror_num = stripe_index + 1; |
dfe25020 | 6362 | } |
2fff734f | 6363 | |
611f0e00 | 6364 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6365 | if (need_full_stripe(op)) { |
f2d8d74d | 6366 | num_stripes = map->num_stripes; |
a1d3c478 | 6367 | } else if (mirror_num) { |
f188591e | 6368 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6369 | } else { |
6370 | mirror_num = 1; | |
6371 | } | |
2fff734f | 6372 | |
321aecc6 | 6373 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6374 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6375 | |
47c5713f | 6376 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6377 | stripe_index *= map->sub_stripes; |
6378 | ||
de483734 | 6379 | if (need_full_stripe(op)) |
f2d8d74d | 6380 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6381 | else if (mirror_num) |
6382 | stripe_index += mirror_num - 1; | |
dfe25020 | 6383 | else { |
3e74317a | 6384 | int old_stripe_index = stripe_index; |
30d9861f SB |
6385 | stripe_index = find_live_mirror(fs_info, map, |
6386 | stripe_index, | |
30d9861f | 6387 | dev_replace_is_ongoing); |
3e74317a | 6388 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6389 | } |
53b381b3 | 6390 | |
ffe2d203 | 6391 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6392 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6393 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6394 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6395 | stripe_len * data_stripes); |
53b381b3 DW |
6396 | |
6397 | /* RAID[56] write or recovery. Return all stripes */ | |
6398 | num_stripes = map->num_stripes; | |
6399 | max_errors = nr_parity_stripes(map); | |
6400 | ||
53b381b3 DW |
6401 | *length = map->stripe_len; |
6402 | stripe_index = 0; | |
6403 | stripe_offset = 0; | |
6404 | } else { | |
6405 | /* | |
6406 | * Mirror #0 or #1 means the original data block. | |
6407 | * Mirror #2 is RAID5 parity block. | |
6408 | * Mirror #3 is RAID6 Q block. | |
6409 | */ | |
47c5713f | 6410 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6411 | data_stripes, &stripe_index); |
53b381b3 | 6412 | if (mirror_num > 1) |
cff82672 | 6413 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6414 | |
6415 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6416 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6417 | &stripe_index); | |
de483734 | 6418 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6419 | mirror_num = 1; |
53b381b3 | 6420 | } |
8790d502 CM |
6421 | } else { |
6422 | /* | |
47c5713f DS |
6423 | * after this, stripe_nr is the number of stripes on this |
6424 | * device we have to walk to find the data, and stripe_index is | |
6425 | * the number of our device in the stripe array | |
8790d502 | 6426 | */ |
47c5713f DS |
6427 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6428 | &stripe_index); | |
a1d3c478 | 6429 | mirror_num = stripe_index + 1; |
8790d502 | 6430 | } |
e042d1ec | 6431 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6432 | btrfs_crit(fs_info, |
6433 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6434 | stripe_index, map->num_stripes); |
6435 | ret = -EINVAL; | |
6436 | goto out; | |
6437 | } | |
cea9e445 | 6438 | |
472262f3 | 6439 | num_alloc_stripes = num_stripes; |
6fad823f | 6440 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6441 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6442 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6443 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6444 | num_alloc_stripes++; |
2c8cdd6e | 6445 | tgtdev_indexes = num_stripes; |
ad6d620e | 6446 | } |
2c8cdd6e | 6447 | |
6e9606d2 | 6448 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
6449 | if (!bbio) { |
6450 | ret = -ENOMEM; | |
6451 | goto out; | |
6452 | } | |
608769a4 NB |
6453 | |
6454 | for (i = 0; i < num_stripes; i++) { | |
6455 | bbio->stripes[i].physical = map->stripes[stripe_index].physical + | |
6456 | stripe_offset + stripe_nr * map->stripe_len; | |
6457 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
6458 | stripe_index++; | |
6459 | } | |
de11cc12 | 6460 | |
8e5cfb55 | 6461 | /* build raid_map */ |
2b19a1fe LB |
6462 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6463 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6464 | u64 tmp; |
9d644a62 | 6465 | unsigned rot; |
8e5cfb55 | 6466 | |
8e5cfb55 | 6467 | /* Work out the disk rotation on this stripe-set */ |
47c5713f | 6468 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6469 | |
6470 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6471 | tmp = stripe_nr * data_stripes; |
6472 | for (i = 0; i < data_stripes; i++) | |
8e5cfb55 ZL |
6473 | bbio->raid_map[(i+rot) % num_stripes] = |
6474 | em->start + (tmp + i) * map->stripe_len; | |
6475 | ||
6476 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
6477 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
6478 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
6479 | RAID6_Q_STRIPE; | |
8e5cfb55 | 6480 | |
608769a4 | 6481 | sort_parity_stripes(bbio, num_stripes); |
593060d7 | 6482 | } |
de11cc12 | 6483 | |
2b19a1fe | 6484 | if (need_full_stripe(op)) |
d20983b4 | 6485 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6486 | |
73c0f228 | 6487 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6488 | need_full_stripe(op)) { |
6143c23c NA |
6489 | handle_ops_on_dev_replace(op, &bbio, dev_replace, logical, |
6490 | &num_stripes, &max_errors); | |
472262f3 SB |
6491 | } |
6492 | ||
de11cc12 | 6493 | *bbio_ret = bbio; |
10f11900 | 6494 | bbio->map_type = map->type; |
de11cc12 LZ |
6495 | bbio->num_stripes = num_stripes; |
6496 | bbio->max_errors = max_errors; | |
6497 | bbio->mirror_num = mirror_num; | |
ad6d620e SB |
6498 | |
6499 | /* | |
6500 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6501 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6502 | * available as a mirror | |
6503 | */ | |
6504 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6505 | WARN_ON(num_stripes > 1); | |
6506 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
6507 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6508 | bbio->mirror_num = map->num_stripes + 1; | |
6509 | } | |
cea9e445 | 6510 | out: |
73beece9 | 6511 | if (dev_replace_is_ongoing) { |
53176dde DS |
6512 | lockdep_assert_held(&dev_replace->rwsem); |
6513 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6514 | up_read(&dev_replace->rwsem); |
73beece9 | 6515 | } |
0b86a832 | 6516 | free_extent_map(em); |
de11cc12 | 6517 | return ret; |
0b86a832 CM |
6518 | } |
6519 | ||
cf8cddd3 | 6520 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6521 | u64 logical, u64 *length, |
a1d3c478 | 6522 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 6523 | { |
75fb2e9e DS |
6524 | if (op == BTRFS_MAP_DISCARD) |
6525 | return __btrfs_map_block_for_discard(fs_info, logical, | |
6526 | length, bbio_ret); | |
6527 | ||
b3d3fa51 | 6528 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 6529 | mirror_num, 0); |
f2d8d74d CM |
6530 | } |
6531 | ||
af8e2d1d | 6532 | /* For Scrub/replace */ |
cf8cddd3 | 6533 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6534 | u64 logical, u64 *length, |
825ad4c9 | 6535 | struct btrfs_bio **bbio_ret) |
af8e2d1d | 6536 | { |
825ad4c9 | 6537 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1); |
af8e2d1d MX |
6538 | } |
6539 | ||
4246a0b6 | 6540 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6541 | { |
326e1dbb MS |
6542 | bio->bi_private = bbio->private; |
6543 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6544 | bio_endio(bio); |
326e1dbb | 6545 | |
6e9606d2 | 6546 | btrfs_put_bbio(bbio); |
8408c716 MX |
6547 | } |
6548 | ||
4246a0b6 | 6549 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6550 | { |
9be3395b | 6551 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6552 | int is_orig_bio = 0; |
8790d502 | 6553 | |
4e4cbee9 | 6554 | if (bio->bi_status) { |
a1d3c478 | 6555 | atomic_inc(&bbio->error); |
4e4cbee9 CH |
6556 | if (bio->bi_status == BLK_STS_IOERR || |
6557 | bio->bi_status == BLK_STS_TARGET) { | |
c31efbdf | 6558 | struct btrfs_device *dev = btrfs_io_bio(bio)->device; |
442a4f63 | 6559 | |
3eee86c8 | 6560 | ASSERT(dev->bdev); |
cfe94440 | 6561 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) |
3eee86c8 | 6562 | btrfs_dev_stat_inc_and_print(dev, |
597a60fa | 6563 | BTRFS_DEV_STAT_WRITE_ERRS); |
3eee86c8 NB |
6564 | else if (!(bio->bi_opf & REQ_RAHEAD)) |
6565 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6566 | BTRFS_DEV_STAT_READ_ERRS); |
3eee86c8 NB |
6567 | if (bio->bi_opf & REQ_PREFLUSH) |
6568 | btrfs_dev_stat_inc_and_print(dev, | |
597a60fa | 6569 | BTRFS_DEV_STAT_FLUSH_ERRS); |
442a4f63 SB |
6570 | } |
6571 | } | |
8790d502 | 6572 | |
a1d3c478 | 6573 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6574 | is_orig_bio = 1; |
6575 | ||
c404e0dc MX |
6576 | btrfs_bio_counter_dec(bbio->fs_info); |
6577 | ||
a1d3c478 | 6578 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6579 | if (!is_orig_bio) { |
6580 | bio_put(bio); | |
a1d3c478 | 6581 | bio = bbio->orig_bio; |
7d2b4daa | 6582 | } |
c7b22bb1 | 6583 | |
9be3395b | 6584 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6585 | /* only send an error to the higher layers if it is |
53b381b3 | 6586 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6587 | */ |
a1d3c478 | 6588 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4e4cbee9 | 6589 | bio->bi_status = BLK_STS_IOERR; |
5dbc8fca | 6590 | } else { |
1259ab75 CM |
6591 | /* |
6592 | * this bio is actually up to date, we didn't | |
6593 | * go over the max number of errors | |
6594 | */ | |
2dbe0c77 | 6595 | bio->bi_status = BLK_STS_OK; |
1259ab75 | 6596 | } |
c55f1396 | 6597 | |
4246a0b6 | 6598 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6599 | } else if (!is_orig_bio) { |
8790d502 CM |
6600 | bio_put(bio); |
6601 | } | |
8790d502 CM |
6602 | } |
6603 | ||
2ff7e61e | 6604 | static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio, |
c31efbdf | 6605 | u64 physical, struct btrfs_device *dev) |
de1ee92a | 6606 | { |
2ff7e61e | 6607 | struct btrfs_fs_info *fs_info = bbio->fs_info; |
de1ee92a JB |
6608 | |
6609 | bio->bi_private = bbio; | |
c31efbdf | 6610 | btrfs_io_bio(bio)->device = dev; |
de1ee92a | 6611 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6612 | bio->bi_iter.bi_sector = physical >> 9; |
d8e3fb10 NA |
6613 | /* |
6614 | * For zone append writing, bi_sector must point the beginning of the | |
6615 | * zone | |
6616 | */ | |
6617 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { | |
6618 | if (btrfs_dev_is_sequential(dev, physical)) { | |
6619 | u64 zone_start = round_down(physical, fs_info->zone_size); | |
6620 | ||
6621 | bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT; | |
6622 | } else { | |
6623 | bio->bi_opf &= ~REQ_OP_ZONE_APPEND; | |
6624 | bio->bi_opf |= REQ_OP_WRITE; | |
6625 | } | |
6626 | } | |
672d5990 MT |
6627 | btrfs_debug_in_rcu(fs_info, |
6628 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
1201b58b | 6629 | bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector, |
1db45a35 DS |
6630 | (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name), |
6631 | dev->devid, bio->bi_iter.bi_size); | |
74d46992 | 6632 | bio_set_dev(bio, dev->bdev); |
c404e0dc | 6633 | |
2ff7e61e | 6634 | btrfs_bio_counter_inc_noblocked(fs_info); |
c404e0dc | 6635 | |
08635bae | 6636 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6637 | } |
6638 | ||
de1ee92a JB |
6639 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6640 | { | |
6641 | atomic_inc(&bbio->error); | |
6642 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6643 | /* Should be the original bio. */ |
8408c716 MX |
6644 | WARN_ON(bio != bbio->orig_bio); |
6645 | ||
9be3395b | 6646 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6647 | bio->bi_iter.bi_sector = logical >> 9; |
102ed2c5 AJ |
6648 | if (atomic_read(&bbio->error) > bbio->max_errors) |
6649 | bio->bi_status = BLK_STS_IOERR; | |
6650 | else | |
6651 | bio->bi_status = BLK_STS_OK; | |
4246a0b6 | 6652 | btrfs_end_bbio(bbio, bio); |
de1ee92a JB |
6653 | } |
6654 | } | |
6655 | ||
58efbc9f | 6656 | blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio, |
08635bae | 6657 | int mirror_num) |
0b86a832 | 6658 | { |
0b86a832 | 6659 | struct btrfs_device *dev; |
8790d502 | 6660 | struct bio *first_bio = bio; |
1201b58b | 6661 | u64 logical = bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6662 | u64 length = 0; |
6663 | u64 map_length; | |
0b86a832 | 6664 | int ret; |
08da757d ZL |
6665 | int dev_nr; |
6666 | int total_devs; | |
a1d3c478 | 6667 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6668 | |
4f024f37 | 6669 | length = bio->bi_iter.bi_size; |
0b86a832 | 6670 | map_length = length; |
cea9e445 | 6671 | |
0b246afa | 6672 | btrfs_bio_counter_inc_blocked(fs_info); |
bd7d63c2 | 6673 | ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical, |
37226b21 | 6674 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6675 | if (ret) { |
0b246afa | 6676 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6677 | return errno_to_blk_status(ret); |
c404e0dc | 6678 | } |
cea9e445 | 6679 | |
a1d3c478 | 6680 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6681 | bbio->orig_bio = first_bio; |
6682 | bbio->private = first_bio->bi_private; | |
6683 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6684 | bbio->fs_info = fs_info; |
53b381b3 DW |
6685 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6686 | ||
ad1ba2a0 | 6687 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cfe94440 | 6688 | ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6689 | /* In this case, map_length has been set to the length of |
6690 | a single stripe; not the whole write */ | |
cfe94440 | 6691 | if (btrfs_op(bio) == BTRFS_MAP_WRITE) { |
2ff7e61e JM |
6692 | ret = raid56_parity_write(fs_info, bio, bbio, |
6693 | map_length); | |
53b381b3 | 6694 | } else { |
2ff7e61e JM |
6695 | ret = raid56_parity_recover(fs_info, bio, bbio, |
6696 | map_length, mirror_num, 1); | |
53b381b3 | 6697 | } |
4245215d | 6698 | |
0b246afa | 6699 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6700 | return errno_to_blk_status(ret); |
53b381b3 DW |
6701 | } |
6702 | ||
cea9e445 | 6703 | if (map_length < length) { |
0b246afa | 6704 | btrfs_crit(fs_info, |
5d163e0e JM |
6705 | "mapping failed logical %llu bio len %llu len %llu", |
6706 | logical, length, map_length); | |
cea9e445 CM |
6707 | BUG(); |
6708 | } | |
a1d3c478 | 6709 | |
08da757d | 6710 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6711 | dev = bbio->stripes[dev_nr].dev; |
fc8a168a NB |
6712 | if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING, |
6713 | &dev->dev_state) || | |
cfe94440 | 6714 | (btrfs_op(first_bio) == BTRFS_MAP_WRITE && |
ebbede42 | 6715 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) { |
de1ee92a | 6716 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6717 | continue; |
6718 | } | |
6719 | ||
3aa8e074 | 6720 | if (dev_nr < total_devs - 1) |
8b6c1d56 | 6721 | bio = btrfs_bio_clone(first_bio); |
3aa8e074 | 6722 | else |
a1d3c478 | 6723 | bio = first_bio; |
de1ee92a | 6724 | |
c31efbdf | 6725 | submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev); |
8790d502 | 6726 | } |
0b246afa | 6727 | btrfs_bio_counter_dec(fs_info); |
58efbc9f | 6728 | return BLK_STS_OK; |
0b86a832 CM |
6729 | } |
6730 | ||
09ba3bc9 AJ |
6731 | /* |
6732 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6733 | * return NULL. | |
6734 | * | |
6735 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6736 | * only devid is used. | |
09ba3bc9 | 6737 | */ |
e4319cd9 | 6738 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices, |
b2598edf | 6739 | u64 devid, u8 *uuid, u8 *fsid) |
0b86a832 | 6740 | { |
2b82032c | 6741 | struct btrfs_device *device; |
944d3f9f NB |
6742 | struct btrfs_fs_devices *seed_devs; |
6743 | ||
6744 | if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { | |
6745 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6746 | if (device->devid == devid && | |
6747 | (!uuid || memcmp(device->uuid, uuid, | |
6748 | BTRFS_UUID_SIZE) == 0)) | |
6749 | return device; | |
6750 | } | |
6751 | } | |
2b82032c | 6752 | |
944d3f9f | 6753 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
2b82032c | 6754 | if (!fsid || |
944d3f9f NB |
6755 | !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) { |
6756 | list_for_each_entry(device, &seed_devs->devices, | |
09ba3bc9 AJ |
6757 | dev_list) { |
6758 | if (device->devid == devid && | |
6759 | (!uuid || memcmp(device->uuid, uuid, | |
6760 | BTRFS_UUID_SIZE) == 0)) | |
6761 | return device; | |
6762 | } | |
2b82032c | 6763 | } |
2b82032c | 6764 | } |
944d3f9f | 6765 | |
2b82032c | 6766 | return NULL; |
0b86a832 CM |
6767 | } |
6768 | ||
2ff7e61e | 6769 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6770 | u64 devid, u8 *dev_uuid) |
6771 | { | |
6772 | struct btrfs_device *device; | |
fccc0007 | 6773 | unsigned int nofs_flag; |
dfe25020 | 6774 | |
fccc0007 JB |
6775 | /* |
6776 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6777 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6778 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6779 | * places. | |
6780 | */ | |
6781 | nofs_flag = memalloc_nofs_save(); | |
12bd2fc0 | 6782 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
fccc0007 | 6783 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6784 | if (IS_ERR(device)) |
adfb69af | 6785 | return device; |
12bd2fc0 ID |
6786 | |
6787 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6788 | device->fs_devices = fs_devices; |
dfe25020 | 6789 | fs_devices->num_devices++; |
12bd2fc0 | 6790 | |
e6e674bd | 6791 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6792 | fs_devices->missing_devices++; |
12bd2fc0 | 6793 | |
dfe25020 CM |
6794 | return device; |
6795 | } | |
6796 | ||
12bd2fc0 ID |
6797 | /** |
6798 | * btrfs_alloc_device - allocate struct btrfs_device | |
6799 | * @fs_info: used only for generating a new devid, can be NULL if | |
6800 | * devid is provided (i.e. @devid != NULL). | |
6801 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6802 | * is generated. | |
6803 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6804 | * is generated. | |
6805 | * | |
6806 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6807 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6808 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6809 | */ |
6810 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6811 | const u64 *devid, | |
6812 | const u8 *uuid) | |
6813 | { | |
6814 | struct btrfs_device *dev; | |
6815 | u64 tmp; | |
6816 | ||
fae7f21c | 6817 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6818 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6819 | |
fe4f46d4 DS |
6820 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
6821 | if (!dev) | |
6822 | return ERR_PTR(-ENOMEM); | |
6823 | ||
6824 | /* | |
6825 | * Preallocate a bio that's always going to be used for flushing device | |
6826 | * barriers and matches the device lifespan | |
6827 | */ | |
6828 | dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0); | |
6829 | if (!dev->flush_bio) { | |
6830 | kfree(dev); | |
6831 | return ERR_PTR(-ENOMEM); | |
6832 | } | |
6833 | ||
6834 | INIT_LIST_HEAD(&dev->dev_list); | |
6835 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
6836 | INIT_LIST_HEAD(&dev->post_commit_list); | |
6837 | ||
6838 | atomic_set(&dev->reada_in_flight, 0); | |
6839 | atomic_set(&dev->dev_stats_ccnt, 0); | |
6840 | btrfs_device_data_ordered_init(dev); | |
6841 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
6842 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
6843 | extent_io_tree_init(fs_info, &dev->alloc_state, | |
6844 | IO_TREE_DEVICE_ALLOC_STATE, NULL); | |
12bd2fc0 ID |
6845 | |
6846 | if (devid) | |
6847 | tmp = *devid; | |
6848 | else { | |
6849 | int ret; | |
6850 | ||
6851 | ret = find_next_devid(fs_info, &tmp); | |
6852 | if (ret) { | |
a425f9d4 | 6853 | btrfs_free_device(dev); |
12bd2fc0 ID |
6854 | return ERR_PTR(ret); |
6855 | } | |
6856 | } | |
6857 | dev->devid = tmp; | |
6858 | ||
6859 | if (uuid) | |
6860 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6861 | else | |
6862 | generate_random_uuid(dev->uuid); | |
6863 | ||
12bd2fc0 ID |
6864 | return dev; |
6865 | } | |
6866 | ||
5a2b8e60 | 6867 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6868 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6869 | { |
2b902dfc AJ |
6870 | if (error) |
6871 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6872 | devid, uuid); | |
6873 | else | |
6874 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6875 | devid, uuid); | |
5a2b8e60 AJ |
6876 | } |
6877 | ||
39e264a4 NB |
6878 | static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes) |
6879 | { | |
d58ede8d | 6880 | const int data_stripes = calc_data_stripes(type, num_stripes); |
e4f6c6be | 6881 | |
39e264a4 NB |
6882 | return div_u64(chunk_len, data_stripes); |
6883 | } | |
6884 | ||
e9306ad4 QW |
6885 | #if BITS_PER_LONG == 32 |
6886 | /* | |
6887 | * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE | |
6888 | * can't be accessed on 32bit systems. | |
6889 | * | |
6890 | * This function do mount time check to reject the fs if it already has | |
6891 | * metadata chunk beyond that limit. | |
6892 | */ | |
6893 | static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6894 | u64 logical, u64 length, u64 type) | |
6895 | { | |
6896 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6897 | return 0; | |
6898 | ||
6899 | if (logical + length < MAX_LFS_FILESIZE) | |
6900 | return 0; | |
6901 | ||
6902 | btrfs_err_32bit_limit(fs_info); | |
6903 | return -EOVERFLOW; | |
6904 | } | |
6905 | ||
6906 | /* | |
6907 | * This is to give early warning for any metadata chunk reaching | |
6908 | * BTRFS_32BIT_EARLY_WARN_THRESHOLD. | |
6909 | * Although we can still access the metadata, it's not going to be possible | |
6910 | * once the limit is reached. | |
6911 | */ | |
6912 | static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6913 | u64 logical, u64 length, u64 type) | |
6914 | { | |
6915 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6916 | return; | |
6917 | ||
6918 | if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6919 | return; | |
6920 | ||
6921 | btrfs_warn_32bit_limit(fs_info); | |
6922 | } | |
6923 | #endif | |
6924 | ||
9690ac09 | 6925 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6926 | struct btrfs_chunk *chunk) |
6927 | { | |
9690ac09 | 6928 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6929 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6930 | struct map_lookup *map; |
6931 | struct extent_map *em; | |
6932 | u64 logical; | |
6933 | u64 length; | |
e06cd3dd | 6934 | u64 devid; |
e9306ad4 | 6935 | u64 type; |
e06cd3dd LB |
6936 | u8 uuid[BTRFS_UUID_SIZE]; |
6937 | int num_stripes; | |
6938 | int ret; | |
6939 | int i; | |
6940 | ||
6941 | logical = key->offset; | |
6942 | length = btrfs_chunk_length(leaf, chunk); | |
e9306ad4 | 6943 | type = btrfs_chunk_type(leaf, chunk); |
e06cd3dd LB |
6944 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6945 | ||
e9306ad4 QW |
6946 | #if BITS_PER_LONG == 32 |
6947 | ret = check_32bit_meta_chunk(fs_info, logical, length, type); | |
6948 | if (ret < 0) | |
6949 | return ret; | |
6950 | warn_32bit_meta_chunk(fs_info, logical, length, type); | |
6951 | #endif | |
6952 | ||
075cb3c7 QW |
6953 | /* |
6954 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6955 | * as chunk item in tree block is already verified by tree-checker. | |
6956 | */ | |
6957 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6958 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6959 | if (ret) |
6960 | return ret; | |
6961 | } | |
a061fc8d | 6962 | |
c8bf1b67 DS |
6963 | read_lock(&map_tree->lock); |
6964 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6965 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6966 | |
6967 | /* already mapped? */ | |
6968 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6969 | free_extent_map(em); | |
0b86a832 CM |
6970 | return 0; |
6971 | } else if (em) { | |
6972 | free_extent_map(em); | |
6973 | } | |
0b86a832 | 6974 | |
172ddd60 | 6975 | em = alloc_extent_map(); |
0b86a832 CM |
6976 | if (!em) |
6977 | return -ENOMEM; | |
593060d7 | 6978 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6979 | if (!map) { |
6980 | free_extent_map(em); | |
6981 | return -ENOMEM; | |
6982 | } | |
6983 | ||
298a8f9c | 6984 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6985 | em->map_lookup = map; |
0b86a832 CM |
6986 | em->start = logical; |
6987 | em->len = length; | |
70c8a91c | 6988 | em->orig_start = 0; |
0b86a832 | 6989 | em->block_start = 0; |
c8b97818 | 6990 | em->block_len = em->len; |
0b86a832 | 6991 | |
593060d7 CM |
6992 | map->num_stripes = num_stripes; |
6993 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6994 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 | 6995 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
e9306ad4 | 6996 | map->type = type; |
321aecc6 | 6997 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
cf90d884 | 6998 | map->verified_stripes = 0; |
e9306ad4 | 6999 | em->orig_block_len = calc_stripe_length(type, em->len, |
39e264a4 | 7000 | map->num_stripes); |
593060d7 CM |
7001 | for (i = 0; i < num_stripes; i++) { |
7002 | map->stripes[i].physical = | |
7003 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
7004 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
7005 | read_extent_buffer(leaf, uuid, (unsigned long) |
7006 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
7007 | BTRFS_UUID_SIZE); | |
e4319cd9 | 7008 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, |
b2598edf | 7009 | devid, uuid, NULL); |
3cdde224 | 7010 | if (!map->stripes[i].dev && |
0b246afa | 7011 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 | 7012 | free_extent_map(em); |
2b902dfc | 7013 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
45dbdbc9 | 7014 | return -ENOENT; |
593060d7 | 7015 | } |
dfe25020 CM |
7016 | if (!map->stripes[i].dev) { |
7017 | map->stripes[i].dev = | |
2ff7e61e JM |
7018 | add_missing_dev(fs_info->fs_devices, devid, |
7019 | uuid); | |
adfb69af | 7020 | if (IS_ERR(map->stripes[i].dev)) { |
dfe25020 | 7021 | free_extent_map(em); |
adfb69af AJ |
7022 | btrfs_err(fs_info, |
7023 | "failed to init missing dev %llu: %ld", | |
7024 | devid, PTR_ERR(map->stripes[i].dev)); | |
7025 | return PTR_ERR(map->stripes[i].dev); | |
dfe25020 | 7026 | } |
2b902dfc | 7027 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
dfe25020 | 7028 | } |
e12c9621 AJ |
7029 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
7030 | &(map->stripes[i].dev->dev_state)); | |
7031 | ||
0b86a832 CM |
7032 | } |
7033 | ||
c8bf1b67 DS |
7034 | write_lock(&map_tree->lock); |
7035 | ret = add_extent_mapping(map_tree, em, 0); | |
7036 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
7037 | if (ret < 0) { |
7038 | btrfs_err(fs_info, | |
7039 | "failed to add chunk map, start=%llu len=%llu: %d", | |
7040 | em->start, em->len, ret); | |
7041 | } | |
0b86a832 CM |
7042 | free_extent_map(em); |
7043 | ||
64f64f43 | 7044 | return ret; |
0b86a832 CM |
7045 | } |
7046 | ||
143bede5 | 7047 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
7048 | struct btrfs_dev_item *dev_item, |
7049 | struct btrfs_device *device) | |
7050 | { | |
7051 | unsigned long ptr; | |
0b86a832 CM |
7052 | |
7053 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
7054 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
7055 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 7056 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 7057 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 7058 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
7059 | device->type = btrfs_device_type(leaf, dev_item); |
7060 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
7061 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
7062 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 7063 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 7064 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 7065 | |
410ba3a2 | 7066 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 7067 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
7068 | } |
7069 | ||
2ff7e61e | 7070 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 7071 | u8 *fsid) |
2b82032c YZ |
7072 | { |
7073 | struct btrfs_fs_devices *fs_devices; | |
7074 | int ret; | |
7075 | ||
a32bf9a3 | 7076 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 7077 | ASSERT(fsid); |
2b82032c | 7078 | |
427c8fdd | 7079 | /* This will match only for multi-device seed fs */ |
944d3f9f | 7080 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 7081 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
7082 | return fs_devices; |
7083 | ||
2b82032c | 7084 | |
7239ff4b | 7085 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 7086 | if (!fs_devices) { |
0b246afa | 7087 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
7088 | return ERR_PTR(-ENOENT); |
7089 | ||
7239ff4b | 7090 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
7091 | if (IS_ERR(fs_devices)) |
7092 | return fs_devices; | |
7093 | ||
0395d84f | 7094 | fs_devices->seeding = true; |
5f375835 MX |
7095 | fs_devices->opened = 1; |
7096 | return fs_devices; | |
2b82032c | 7097 | } |
e4404d6e | 7098 | |
427c8fdd NB |
7099 | /* |
7100 | * Upon first call for a seed fs fsid, just create a private copy of the | |
7101 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
7102 | */ | |
e4404d6e | 7103 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
7104 | if (IS_ERR(fs_devices)) |
7105 | return fs_devices; | |
2b82032c | 7106 | |
897fb573 | 7107 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
7108 | if (ret) { |
7109 | free_fs_devices(fs_devices); | |
c83b60c0 | 7110 | return ERR_PTR(ret); |
48d28232 | 7111 | } |
2b82032c YZ |
7112 | |
7113 | if (!fs_devices->seeding) { | |
0226e0eb | 7114 | close_fs_devices(fs_devices); |
e4404d6e | 7115 | free_fs_devices(fs_devices); |
c83b60c0 | 7116 | return ERR_PTR(-EINVAL); |
2b82032c YZ |
7117 | } |
7118 | ||
944d3f9f | 7119 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
c83b60c0 | 7120 | |
5f375835 | 7121 | return fs_devices; |
2b82032c YZ |
7122 | } |
7123 | ||
17850759 | 7124 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
7125 | struct btrfs_dev_item *dev_item) |
7126 | { | |
17850759 | 7127 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 7128 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
7129 | struct btrfs_device *device; |
7130 | u64 devid; | |
7131 | int ret; | |
44880fdc | 7132 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
7133 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
7134 | ||
0b86a832 | 7135 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 7136 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 7137 | BTRFS_UUID_SIZE); |
1473b24e | 7138 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 7139 | BTRFS_FSID_SIZE); |
2b82032c | 7140 | |
de37aa51 | 7141 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 7142 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
7143 | if (IS_ERR(fs_devices)) |
7144 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
7145 | } |
7146 | ||
e4319cd9 | 7147 | device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid, |
b2598edf | 7148 | fs_uuid); |
5f375835 | 7149 | if (!device) { |
c5502451 | 7150 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
7151 | btrfs_report_missing_device(fs_info, devid, |
7152 | dev_uuid, true); | |
45dbdbc9 | 7153 | return -ENOENT; |
c5502451 | 7154 | } |
2b82032c | 7155 | |
2ff7e61e | 7156 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
7157 | if (IS_ERR(device)) { |
7158 | btrfs_err(fs_info, | |
7159 | "failed to add missing dev %llu: %ld", | |
7160 | devid, PTR_ERR(device)); | |
7161 | return PTR_ERR(device); | |
7162 | } | |
2b902dfc | 7163 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 7164 | } else { |
c5502451 | 7165 | if (!device->bdev) { |
2b902dfc AJ |
7166 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
7167 | btrfs_report_missing_device(fs_info, | |
7168 | devid, dev_uuid, true); | |
45dbdbc9 | 7169 | return -ENOENT; |
2b902dfc AJ |
7170 | } |
7171 | btrfs_report_missing_device(fs_info, devid, | |
7172 | dev_uuid, false); | |
c5502451 | 7173 | } |
5f375835 | 7174 | |
e6e674bd AJ |
7175 | if (!device->bdev && |
7176 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7177 | /* |
7178 | * this happens when a device that was properly setup | |
7179 | * in the device info lists suddenly goes bad. | |
7180 | * device->bdev is NULL, and so we have to set | |
7181 | * device->missing to one here | |
7182 | */ | |
5f375835 | 7183 | device->fs_devices->missing_devices++; |
e6e674bd | 7184 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7185 | } |
5f375835 MX |
7186 | |
7187 | /* Move the device to its own fs_devices */ | |
7188 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7189 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7190 | &device->dev_state)); | |
5f375835 MX |
7191 | |
7192 | list_move(&device->dev_list, &fs_devices->devices); | |
7193 | device->fs_devices->num_devices--; | |
7194 | fs_devices->num_devices++; | |
7195 | ||
7196 | device->fs_devices->missing_devices--; | |
7197 | fs_devices->missing_devices++; | |
7198 | ||
7199 | device->fs_devices = fs_devices; | |
7200 | } | |
2b82032c YZ |
7201 | } |
7202 | ||
0b246afa | 7203 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7204 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7205 | if (device->generation != |
7206 | btrfs_device_generation(leaf, dev_item)) | |
7207 | return -EINVAL; | |
6324fbf3 | 7208 | } |
0b86a832 CM |
7209 | |
7210 | fill_device_from_item(leaf, dev_item, device); | |
3a160a93 AJ |
7211 | if (device->bdev) { |
7212 | u64 max_total_bytes = i_size_read(device->bdev->bd_inode); | |
7213 | ||
7214 | if (device->total_bytes > max_total_bytes) { | |
7215 | btrfs_err(fs_info, | |
7216 | "device total_bytes should be at most %llu but found %llu", | |
7217 | max_total_bytes, device->total_bytes); | |
7218 | return -EINVAL; | |
7219 | } | |
7220 | } | |
e12c9621 | 7221 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7222 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7223 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7224 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7225 | atomic64_add(device->total_bytes - device->bytes_used, |
7226 | &fs_info->free_chunk_space); | |
2bf64758 | 7227 | } |
0b86a832 | 7228 | ret = 0; |
0b86a832 CM |
7229 | return ret; |
7230 | } | |
7231 | ||
6bccf3ab | 7232 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7233 | { |
6bccf3ab | 7234 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 7235 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7236 | struct extent_buffer *sb; |
0b86a832 | 7237 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7238 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7239 | u8 *array_ptr; |
7240 | unsigned long sb_array_offset; | |
84eed90f | 7241 | int ret = 0; |
0b86a832 CM |
7242 | u32 num_stripes; |
7243 | u32 array_size; | |
7244 | u32 len = 0; | |
1ffb22cf | 7245 | u32 cur_offset; |
e06cd3dd | 7246 | u64 type; |
84eed90f | 7247 | struct btrfs_key key; |
0b86a832 | 7248 | |
0b246afa | 7249 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
7250 | /* |
7251 | * This will create extent buffer of nodesize, superblock size is | |
7252 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
7253 | * overallocate but we can keep it as-is, only the first page is used. | |
7254 | */ | |
3fbaf258 JB |
7255 | sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET, |
7256 | root->root_key.objectid, 0); | |
c871b0f2 LB |
7257 | if (IS_ERR(sb)) |
7258 | return PTR_ERR(sb); | |
4db8c528 | 7259 | set_extent_buffer_uptodate(sb); |
8a334426 | 7260 | /* |
01327610 | 7261 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 7262 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
7263 | * pages up-to-date when the page is larger: extent does not cover the |
7264 | * whole page and consequently check_page_uptodate does not find all | |
7265 | * the page's extents up-to-date (the hole beyond sb), | |
7266 | * write_extent_buffer then triggers a WARN_ON. | |
7267 | * | |
7268 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
7269 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
7270 | * to silence the warning eg. on PowerPC 64. | |
7271 | */ | |
09cbfeaf | 7272 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 7273 | SetPageUptodate(sb->pages[0]); |
4008c04a | 7274 | |
a061fc8d | 7275 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7276 | array_size = btrfs_super_sys_array_size(super_copy); |
7277 | ||
1ffb22cf DS |
7278 | array_ptr = super_copy->sys_chunk_array; |
7279 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7280 | cur_offset = 0; | |
0b86a832 | 7281 | |
1ffb22cf DS |
7282 | while (cur_offset < array_size) { |
7283 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7284 | len = sizeof(*disk_key); |
7285 | if (cur_offset + len > array_size) | |
7286 | goto out_short_read; | |
7287 | ||
0b86a832 CM |
7288 | btrfs_disk_key_to_cpu(&key, disk_key); |
7289 | ||
1ffb22cf DS |
7290 | array_ptr += len; |
7291 | sb_array_offset += len; | |
7292 | cur_offset += len; | |
0b86a832 | 7293 | |
32ab3d1b JT |
7294 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
7295 | btrfs_err(fs_info, | |
7296 | "unexpected item type %u in sys_array at offset %u", | |
7297 | (u32)key.type, cur_offset); | |
7298 | ret = -EIO; | |
7299 | break; | |
7300 | } | |
f5cdedd7 | 7301 | |
32ab3d1b JT |
7302 | chunk = (struct btrfs_chunk *)sb_array_offset; |
7303 | /* | |
7304 | * At least one btrfs_chunk with one stripe must be present, | |
7305 | * exact stripe count check comes afterwards | |
7306 | */ | |
7307 | len = btrfs_chunk_item_size(1); | |
7308 | if (cur_offset + len > array_size) | |
7309 | goto out_short_read; | |
e06cd3dd | 7310 | |
32ab3d1b JT |
7311 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
7312 | if (!num_stripes) { | |
7313 | btrfs_err(fs_info, | |
7314 | "invalid number of stripes %u in sys_array at offset %u", | |
7315 | num_stripes, cur_offset); | |
7316 | ret = -EIO; | |
7317 | break; | |
7318 | } | |
e3540eab | 7319 | |
32ab3d1b JT |
7320 | type = btrfs_chunk_type(sb, chunk); |
7321 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7322 | btrfs_err(fs_info, |
32ab3d1b JT |
7323 | "invalid chunk type %llu in sys_array at offset %u", |
7324 | type, cur_offset); | |
84eed90f CM |
7325 | ret = -EIO; |
7326 | break; | |
0b86a832 | 7327 | } |
32ab3d1b JT |
7328 | |
7329 | len = btrfs_chunk_item_size(num_stripes); | |
7330 | if (cur_offset + len > array_size) | |
7331 | goto out_short_read; | |
7332 | ||
7333 | ret = read_one_chunk(&key, sb, chunk); | |
7334 | if (ret) | |
7335 | break; | |
7336 | ||
1ffb22cf DS |
7337 | array_ptr += len; |
7338 | sb_array_offset += len; | |
7339 | cur_offset += len; | |
0b86a832 | 7340 | } |
d865177a | 7341 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7342 | free_extent_buffer_stale(sb); |
84eed90f | 7343 | return ret; |
e3540eab DS |
7344 | |
7345 | out_short_read: | |
ab8d0fc4 | 7346 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7347 | len, cur_offset); |
d865177a | 7348 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7349 | free_extent_buffer_stale(sb); |
e3540eab | 7350 | return -EIO; |
0b86a832 CM |
7351 | } |
7352 | ||
21634a19 QW |
7353 | /* |
7354 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7355 | * | |
6528b99d AJ |
7356 | * If the @failing_dev is specified, it's accounted as missing. |
7357 | * | |
21634a19 QW |
7358 | * Return true if all chunks meet the minimal RW mount requirements. |
7359 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7360 | */ | |
6528b99d AJ |
7361 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7362 | struct btrfs_device *failing_dev) | |
21634a19 | 7363 | { |
c8bf1b67 | 7364 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7365 | struct extent_map *em; |
7366 | u64 next_start = 0; | |
7367 | bool ret = true; | |
7368 | ||
c8bf1b67 DS |
7369 | read_lock(&map_tree->lock); |
7370 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7371 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7372 | /* No chunk at all? Return false anyway */ |
7373 | if (!em) { | |
7374 | ret = false; | |
7375 | goto out; | |
7376 | } | |
7377 | while (em) { | |
7378 | struct map_lookup *map; | |
7379 | int missing = 0; | |
7380 | int max_tolerated; | |
7381 | int i; | |
7382 | ||
7383 | map = em->map_lookup; | |
7384 | max_tolerated = | |
7385 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7386 | map->type); | |
7387 | for (i = 0; i < map->num_stripes; i++) { | |
7388 | struct btrfs_device *dev = map->stripes[i].dev; | |
7389 | ||
e6e674bd AJ |
7390 | if (!dev || !dev->bdev || |
7391 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7392 | dev->last_flush_error) |
7393 | missing++; | |
6528b99d AJ |
7394 | else if (failing_dev && failing_dev == dev) |
7395 | missing++; | |
21634a19 QW |
7396 | } |
7397 | if (missing > max_tolerated) { | |
6528b99d AJ |
7398 | if (!failing_dev) |
7399 | btrfs_warn(fs_info, | |
52042d8e | 7400 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7401 | em->start, missing, max_tolerated); |
7402 | free_extent_map(em); | |
7403 | ret = false; | |
7404 | goto out; | |
7405 | } | |
7406 | next_start = extent_map_end(em); | |
7407 | free_extent_map(em); | |
7408 | ||
c8bf1b67 DS |
7409 | read_lock(&map_tree->lock); |
7410 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7411 | (u64)(-1) - next_start); |
c8bf1b67 | 7412 | read_unlock(&map_tree->lock); |
21634a19 QW |
7413 | } |
7414 | out: | |
7415 | return ret; | |
7416 | } | |
7417 | ||
d85327b1 DS |
7418 | static void readahead_tree_node_children(struct extent_buffer *node) |
7419 | { | |
7420 | int i; | |
7421 | const int nr_items = btrfs_header_nritems(node); | |
7422 | ||
bfb484d9 JB |
7423 | for (i = 0; i < nr_items; i++) |
7424 | btrfs_readahead_node_child(node, i); | |
d85327b1 DS |
7425 | } |
7426 | ||
5b4aacef | 7427 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7428 | { |
5b4aacef | 7429 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7430 | struct btrfs_path *path; |
7431 | struct extent_buffer *leaf; | |
7432 | struct btrfs_key key; | |
7433 | struct btrfs_key found_key; | |
7434 | int ret; | |
7435 | int slot; | |
99e3ecfc | 7436 | u64 total_dev = 0; |
d85327b1 | 7437 | u64 last_ra_node = 0; |
0b86a832 | 7438 | |
0b86a832 CM |
7439 | path = btrfs_alloc_path(); |
7440 | if (!path) | |
7441 | return -ENOMEM; | |
7442 | ||
3dd0f7a3 AJ |
7443 | /* |
7444 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7445 | * otherwise we don't need it. | |
7446 | */ | |
b367e47f | 7447 | mutex_lock(&uuid_mutex); |
b367e47f | 7448 | |
48cfa61b BB |
7449 | /* |
7450 | * It is possible for mount and umount to race in such a way that | |
7451 | * we execute this code path, but open_fs_devices failed to clear | |
7452 | * total_rw_bytes. We certainly want it cleared before reading the | |
7453 | * device items, so clear it here. | |
7454 | */ | |
7455 | fs_info->fs_devices->total_rw_bytes = 0; | |
7456 | ||
395927a9 FDBM |
7457 | /* |
7458 | * Read all device items, and then all the chunk items. All | |
7459 | * device items are found before any chunk item (their object id | |
7460 | * is smaller than the lowest possible object id for a chunk | |
7461 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7462 | */ |
7463 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7464 | key.offset = 0; | |
7465 | key.type = 0; | |
0b86a832 | 7466 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
7467 | if (ret < 0) |
7468 | goto error; | |
d397712b | 7469 | while (1) { |
d85327b1 DS |
7470 | struct extent_buffer *node; |
7471 | ||
0b86a832 CM |
7472 | leaf = path->nodes[0]; |
7473 | slot = path->slots[0]; | |
7474 | if (slot >= btrfs_header_nritems(leaf)) { | |
7475 | ret = btrfs_next_leaf(root, path); | |
7476 | if (ret == 0) | |
7477 | continue; | |
7478 | if (ret < 0) | |
7479 | goto error; | |
7480 | break; | |
7481 | } | |
d85327b1 DS |
7482 | /* |
7483 | * The nodes on level 1 are not locked but we don't need to do | |
7484 | * that during mount time as nothing else can access the tree | |
7485 | */ | |
7486 | node = path->nodes[1]; | |
7487 | if (node) { | |
7488 | if (last_ra_node != node->start) { | |
7489 | readahead_tree_node_children(node); | |
7490 | last_ra_node = node->start; | |
7491 | } | |
7492 | } | |
0b86a832 | 7493 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
395927a9 FDBM |
7494 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7495 | struct btrfs_dev_item *dev_item; | |
7496 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7497 | struct btrfs_dev_item); |
17850759 | 7498 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7499 | if (ret) |
7500 | goto error; | |
99e3ecfc | 7501 | total_dev++; |
0b86a832 CM |
7502 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7503 | struct btrfs_chunk *chunk; | |
79bd3712 FM |
7504 | |
7505 | /* | |
7506 | * We are only called at mount time, so no need to take | |
7507 | * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, | |
7508 | * we always lock first fs_info->chunk_mutex before | |
7509 | * acquiring any locks on the chunk tree. This is a | |
7510 | * requirement for chunk allocation, see the comment on | |
7511 | * top of btrfs_chunk_alloc() for details. | |
7512 | */ | |
7513 | ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); | |
0b86a832 | 7514 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
9690ac09 | 7515 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7516 | if (ret) |
7517 | goto error; | |
0b86a832 CM |
7518 | } |
7519 | path->slots[0]++; | |
7520 | } | |
99e3ecfc LB |
7521 | |
7522 | /* | |
7523 | * After loading chunk tree, we've got all device information, | |
7524 | * do another round of validation checks. | |
7525 | */ | |
0b246afa JM |
7526 | if (total_dev != fs_info->fs_devices->total_devices) { |
7527 | btrfs_err(fs_info, | |
99e3ecfc | 7528 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 7529 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
7530 | total_dev); |
7531 | ret = -EINVAL; | |
7532 | goto error; | |
7533 | } | |
0b246afa JM |
7534 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7535 | fs_info->fs_devices->total_rw_bytes) { | |
7536 | btrfs_err(fs_info, | |
99e3ecfc | 7537 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7538 | btrfs_super_total_bytes(fs_info->super_copy), |
7539 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7540 | ret = -EINVAL; |
7541 | goto error; | |
7542 | } | |
0b86a832 CM |
7543 | ret = 0; |
7544 | error: | |
b367e47f LZ |
7545 | mutex_unlock(&uuid_mutex); |
7546 | ||
2b82032c | 7547 | btrfs_free_path(path); |
0b86a832 CM |
7548 | return ret; |
7549 | } | |
442a4f63 | 7550 | |
cb517eab MX |
7551 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
7552 | { | |
944d3f9f | 7553 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab MX |
7554 | struct btrfs_device *device; |
7555 | ||
944d3f9f NB |
7556 | fs_devices->fs_info = fs_info; |
7557 | ||
7558 | mutex_lock(&fs_devices->device_list_mutex); | |
7559 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7560 | device->fs_info = fs_info; | |
944d3f9f NB |
7561 | |
7562 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
944d3f9f | 7563 | list_for_each_entry(device, &seed_devs->devices, dev_list) |
fb456252 | 7564 | device->fs_info = fs_info; |
29cc83f6 | 7565 | |
944d3f9f | 7566 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7567 | } |
e17125b5 | 7568 | mutex_unlock(&fs_devices->device_list_mutex); |
cb517eab MX |
7569 | } |
7570 | ||
1dc990df DS |
7571 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7572 | const struct btrfs_dev_stats_item *ptr, | |
7573 | int index) | |
7574 | { | |
7575 | u64 val; | |
7576 | ||
7577 | read_extent_buffer(eb, &val, | |
7578 | offsetof(struct btrfs_dev_stats_item, values) + | |
7579 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7580 | sizeof(val)); | |
7581 | return val; | |
7582 | } | |
7583 | ||
7584 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7585 | struct btrfs_dev_stats_item *ptr, | |
7586 | int index, u64 val) | |
7587 | { | |
7588 | write_extent_buffer(eb, &val, | |
7589 | offsetof(struct btrfs_dev_stats_item, values) + | |
7590 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7591 | sizeof(val)); | |
7592 | } | |
7593 | ||
92e26df4 JB |
7594 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
7595 | struct btrfs_path *path) | |
733f4fbb | 7596 | { |
124604eb | 7597 | struct btrfs_dev_stats_item *ptr; |
733f4fbb | 7598 | struct extent_buffer *eb; |
124604eb JB |
7599 | struct btrfs_key key; |
7600 | int item_size; | |
7601 | int i, ret, slot; | |
7602 | ||
82d62d06 JB |
7603 | if (!device->fs_info->dev_root) |
7604 | return 0; | |
7605 | ||
124604eb JB |
7606 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7607 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
7608 | key.offset = device->devid; | |
7609 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); | |
7610 | if (ret) { | |
7611 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7612 | btrfs_dev_stat_set(device, i, 0); | |
7613 | device->dev_stats_valid = 1; | |
7614 | btrfs_release_path(path); | |
92e26df4 | 7615 | return ret < 0 ? ret : 0; |
124604eb JB |
7616 | } |
7617 | slot = path->slots[0]; | |
7618 | eb = path->nodes[0]; | |
7619 | item_size = btrfs_item_size_nr(eb, slot); | |
7620 | ||
7621 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); | |
7622 | ||
7623 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7624 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7625 | btrfs_dev_stat_set(device, i, | |
7626 | btrfs_dev_stats_value(eb, ptr, i)); | |
7627 | else | |
7628 | btrfs_dev_stat_set(device, i, 0); | |
7629 | } | |
7630 | ||
7631 | device->dev_stats_valid = 1; | |
7632 | btrfs_dev_stat_print_on_load(device); | |
7633 | btrfs_release_path(path); | |
92e26df4 JB |
7634 | |
7635 | return 0; | |
124604eb JB |
7636 | } |
7637 | ||
7638 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7639 | { | |
7640 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
733f4fbb SB |
7641 | struct btrfs_device *device; |
7642 | struct btrfs_path *path = NULL; | |
92e26df4 | 7643 | int ret = 0; |
733f4fbb SB |
7644 | |
7645 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7646 | if (!path) |
7647 | return -ENOMEM; | |
733f4fbb SB |
7648 | |
7649 | mutex_lock(&fs_devices->device_list_mutex); | |
92e26df4 JB |
7650 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
7651 | ret = btrfs_device_init_dev_stats(device, path); | |
7652 | if (ret) | |
7653 | goto out; | |
7654 | } | |
124604eb | 7655 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
92e26df4 JB |
7656 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
7657 | ret = btrfs_device_init_dev_stats(device, path); | |
7658 | if (ret) | |
7659 | goto out; | |
7660 | } | |
733f4fbb | 7661 | } |
92e26df4 | 7662 | out: |
733f4fbb SB |
7663 | mutex_unlock(&fs_devices->device_list_mutex); |
7664 | ||
733f4fbb | 7665 | btrfs_free_path(path); |
92e26df4 | 7666 | return ret; |
733f4fbb SB |
7667 | } |
7668 | ||
7669 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7670 | struct btrfs_device *device) |
7671 | { | |
5495f195 | 7672 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7673 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7674 | struct btrfs_path *path; |
7675 | struct btrfs_key key; | |
7676 | struct extent_buffer *eb; | |
7677 | struct btrfs_dev_stats_item *ptr; | |
7678 | int ret; | |
7679 | int i; | |
7680 | ||
242e2956 DS |
7681 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7682 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7683 | key.offset = device->devid; |
7684 | ||
7685 | path = btrfs_alloc_path(); | |
fa252992 DS |
7686 | if (!path) |
7687 | return -ENOMEM; | |
733f4fbb SB |
7688 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7689 | if (ret < 0) { | |
0b246afa | 7690 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7691 | "error %d while searching for dev_stats item for device %s", |
606686ee | 7692 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
7693 | goto out; |
7694 | } | |
7695 | ||
7696 | if (ret == 0 && | |
7697 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
7698 | /* need to delete old one and insert a new one */ | |
7699 | ret = btrfs_del_item(trans, dev_root, path); | |
7700 | if (ret != 0) { | |
0b246afa | 7701 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7702 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7703 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7704 | goto out; |
7705 | } | |
7706 | ret = 1; | |
7707 | } | |
7708 | ||
7709 | if (ret == 1) { | |
7710 | /* need to insert a new item */ | |
7711 | btrfs_release_path(path); | |
7712 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7713 | &key, sizeof(*ptr)); | |
7714 | if (ret < 0) { | |
0b246afa | 7715 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7716 | "insert dev_stats item for device %s failed %d", |
7717 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7718 | goto out; |
7719 | } | |
7720 | } | |
7721 | ||
7722 | eb = path->nodes[0]; | |
7723 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7724 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7725 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7726 | btrfs_dev_stat_read(device, i)); | |
7727 | btrfs_mark_buffer_dirty(eb); | |
7728 | ||
7729 | out: | |
7730 | btrfs_free_path(path); | |
7731 | return ret; | |
7732 | } | |
7733 | ||
7734 | /* | |
7735 | * called from commit_transaction. Writes all changed device stats to disk. | |
7736 | */ | |
196c9d8d | 7737 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7738 | { |
196c9d8d | 7739 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7740 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7741 | struct btrfs_device *device; | |
addc3fa7 | 7742 | int stats_cnt; |
733f4fbb SB |
7743 | int ret = 0; |
7744 | ||
7745 | mutex_lock(&fs_devices->device_list_mutex); | |
7746 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7747 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7748 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7749 | continue; |
7750 | ||
9deae968 NB |
7751 | |
7752 | /* | |
7753 | * There is a LOAD-LOAD control dependency between the value of | |
7754 | * dev_stats_ccnt and updating the on-disk values which requires | |
7755 | * reading the in-memory counters. Such control dependencies | |
7756 | * require explicit read memory barriers. | |
7757 | * | |
7758 | * This memory barriers pairs with smp_mb__before_atomic in | |
7759 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7760 | * barrier implied by atomic_xchg in | |
7761 | * btrfs_dev_stats_read_and_reset | |
7762 | */ | |
7763 | smp_rmb(); | |
7764 | ||
5495f195 | 7765 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7766 | if (!ret) |
addc3fa7 | 7767 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7768 | } |
7769 | mutex_unlock(&fs_devices->device_list_mutex); | |
7770 | ||
7771 | return ret; | |
7772 | } | |
7773 | ||
442a4f63 SB |
7774 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7775 | { | |
7776 | btrfs_dev_stat_inc(dev, index); | |
7777 | btrfs_dev_stat_print_on_error(dev); | |
7778 | } | |
7779 | ||
48a3b636 | 7780 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7781 | { |
733f4fbb SB |
7782 | if (!dev->dev_stats_valid) |
7783 | return; | |
fb456252 | 7784 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7785 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7786 | rcu_str_deref(dev->name), |
442a4f63 SB |
7787 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7788 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7789 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7790 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7791 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7792 | } |
c11d2c23 | 7793 | |
733f4fbb SB |
7794 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7795 | { | |
a98cdb85 SB |
7796 | int i; |
7797 | ||
7798 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7799 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7800 | break; | |
7801 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7802 | return; /* all values == 0, suppress message */ | |
7803 | ||
fb456252 | 7804 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7805 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7806 | rcu_str_deref(dev->name), |
733f4fbb SB |
7807 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7808 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7809 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7810 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7811 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7812 | } | |
7813 | ||
2ff7e61e | 7814 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7815 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
7816 | { |
7817 | struct btrfs_device *dev; | |
0b246afa | 7818 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7819 | int i; |
7820 | ||
7821 | mutex_lock(&fs_devices->device_list_mutex); | |
b2598edf | 7822 | dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL); |
c11d2c23 SB |
7823 | mutex_unlock(&fs_devices->device_list_mutex); |
7824 | ||
7825 | if (!dev) { | |
0b246afa | 7826 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7827 | return -ENODEV; |
733f4fbb | 7828 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7829 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7830 | return -ENODEV; |
b27f7c0c | 7831 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7832 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7833 | if (stats->nr_items > i) | |
7834 | stats->values[i] = | |
7835 | btrfs_dev_stat_read_and_reset(dev, i); | |
7836 | else | |
4e411a7d | 7837 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7838 | } |
a69976bc AJ |
7839 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7840 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7841 | } else { |
7842 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7843 | if (stats->nr_items > i) | |
7844 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7845 | } | |
7846 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7847 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7848 | return 0; | |
7849 | } | |
a8a6dab7 | 7850 | |
935e5cc9 | 7851 | /* |
bbbf7243 NB |
7852 | * Update the size and bytes used for each device where it changed. This is |
7853 | * delayed since we would otherwise get errors while writing out the | |
7854 | * superblocks. | |
7855 | * | |
7856 | * Must be invoked during transaction commit. | |
935e5cc9 | 7857 | */ |
bbbf7243 | 7858 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7859 | { |
935e5cc9 MX |
7860 | struct btrfs_device *curr, *next; |
7861 | ||
bbbf7243 | 7862 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7863 | |
bbbf7243 | 7864 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7865 | return; |
7866 | ||
bbbf7243 NB |
7867 | /* |
7868 | * We don't need the device_list_mutex here. This list is owned by the | |
7869 | * transaction and the transaction must complete before the device is | |
7870 | * released. | |
7871 | */ | |
7872 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7873 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7874 | post_commit_list) { | |
7875 | list_del_init(&curr->post_commit_list); | |
7876 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7877 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7878 | } |
bbbf7243 | 7879 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7880 | } |
5a13f430 | 7881 | |
46df06b8 DS |
7882 | /* |
7883 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7884 | */ | |
7885 | int btrfs_bg_type_to_factor(u64 flags) | |
7886 | { | |
44b28ada DS |
7887 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7888 | ||
7889 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7890 | } |
cf90d884 QW |
7891 | |
7892 | ||
cf90d884 QW |
7893 | |
7894 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7895 | u64 chunk_offset, u64 devid, | |
7896 | u64 physical_offset, u64 physical_len) | |
7897 | { | |
c8bf1b67 | 7898 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7899 | struct extent_map *em; |
7900 | struct map_lookup *map; | |
05a37c48 | 7901 | struct btrfs_device *dev; |
cf90d884 QW |
7902 | u64 stripe_len; |
7903 | bool found = false; | |
7904 | int ret = 0; | |
7905 | int i; | |
7906 | ||
7907 | read_lock(&em_tree->lock); | |
7908 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7909 | read_unlock(&em_tree->lock); | |
7910 | ||
7911 | if (!em) { | |
7912 | btrfs_err(fs_info, | |
7913 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7914 | physical_offset, devid); | |
7915 | ret = -EUCLEAN; | |
7916 | goto out; | |
7917 | } | |
7918 | ||
7919 | map = em->map_lookup; | |
7920 | stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes); | |
7921 | if (physical_len != stripe_len) { | |
7922 | btrfs_err(fs_info, | |
7923 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7924 | physical_offset, devid, em->start, physical_len, | |
7925 | stripe_len); | |
7926 | ret = -EUCLEAN; | |
7927 | goto out; | |
7928 | } | |
7929 | ||
7930 | for (i = 0; i < map->num_stripes; i++) { | |
7931 | if (map->stripes[i].dev->devid == devid && | |
7932 | map->stripes[i].physical == physical_offset) { | |
7933 | found = true; | |
7934 | if (map->verified_stripes >= map->num_stripes) { | |
7935 | btrfs_err(fs_info, | |
7936 | "too many dev extents for chunk %llu found", | |
7937 | em->start); | |
7938 | ret = -EUCLEAN; | |
7939 | goto out; | |
7940 | } | |
7941 | map->verified_stripes++; | |
7942 | break; | |
7943 | } | |
7944 | } | |
7945 | if (!found) { | |
7946 | btrfs_err(fs_info, | |
7947 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7948 | physical_offset, devid); | |
7949 | ret = -EUCLEAN; | |
7950 | } | |
05a37c48 | 7951 | |
1a9fd417 | 7952 | /* Make sure no dev extent is beyond device boundary */ |
b2598edf | 7953 | dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL); |
05a37c48 QW |
7954 | if (!dev) { |
7955 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7956 | ret = -EUCLEAN; | |
7957 | goto out; | |
7958 | } | |
1b3922a8 | 7959 | |
05a37c48 QW |
7960 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7961 | btrfs_err(fs_info, | |
7962 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7963 | devid, physical_offset, physical_len, | |
7964 | dev->disk_total_bytes); | |
7965 | ret = -EUCLEAN; | |
7966 | goto out; | |
7967 | } | |
381a696e NA |
7968 | |
7969 | if (dev->zone_info) { | |
7970 | u64 zone_size = dev->zone_info->zone_size; | |
7971 | ||
7972 | if (!IS_ALIGNED(physical_offset, zone_size) || | |
7973 | !IS_ALIGNED(physical_len, zone_size)) { | |
7974 | btrfs_err(fs_info, | |
7975 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", | |
7976 | devid, physical_offset, physical_len); | |
7977 | ret = -EUCLEAN; | |
7978 | goto out; | |
7979 | } | |
7980 | } | |
7981 | ||
cf90d884 QW |
7982 | out: |
7983 | free_extent_map(em); | |
7984 | return ret; | |
7985 | } | |
7986 | ||
7987 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7988 | { | |
c8bf1b67 | 7989 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7990 | struct extent_map *em; |
7991 | struct rb_node *node; | |
7992 | int ret = 0; | |
7993 | ||
7994 | read_lock(&em_tree->lock); | |
07e1ce09 | 7995 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7996 | em = rb_entry(node, struct extent_map, rb_node); |
7997 | if (em->map_lookup->num_stripes != | |
7998 | em->map_lookup->verified_stripes) { | |
7999 | btrfs_err(fs_info, | |
8000 | "chunk %llu has missing dev extent, have %d expect %d", | |
8001 | em->start, em->map_lookup->verified_stripes, | |
8002 | em->map_lookup->num_stripes); | |
8003 | ret = -EUCLEAN; | |
8004 | goto out; | |
8005 | } | |
8006 | } | |
8007 | out: | |
8008 | read_unlock(&em_tree->lock); | |
8009 | return ret; | |
8010 | } | |
8011 | ||
8012 | /* | |
8013 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
8014 | * later chunk allocation/free would cause unexpected behavior. | |
8015 | * | |
8016 | * NOTE: This will iterate through the whole device tree, which should be of | |
8017 | * the same size level as the chunk tree. This slightly increases mount time. | |
8018 | */ | |
8019 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
8020 | { | |
8021 | struct btrfs_path *path; | |
8022 | struct btrfs_root *root = fs_info->dev_root; | |
8023 | struct btrfs_key key; | |
5eb19381 QW |
8024 | u64 prev_devid = 0; |
8025 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
8026 | int ret = 0; |
8027 | ||
42437a63 JB |
8028 | /* |
8029 | * We don't have a dev_root because we mounted with ignorebadroots and | |
8030 | * failed to load the root, so we want to skip the verification in this | |
8031 | * case for sure. | |
8032 | * | |
8033 | * However if the dev root is fine, but the tree itself is corrupted | |
8034 | * we'd still fail to mount. This verification is only to make sure | |
8035 | * writes can happen safely, so instead just bypass this check | |
8036 | * completely in the case of IGNOREBADROOTS. | |
8037 | */ | |
8038 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) | |
8039 | return 0; | |
8040 | ||
cf90d884 QW |
8041 | key.objectid = 1; |
8042 | key.type = BTRFS_DEV_EXTENT_KEY; | |
8043 | key.offset = 0; | |
8044 | ||
8045 | path = btrfs_alloc_path(); | |
8046 | if (!path) | |
8047 | return -ENOMEM; | |
8048 | ||
8049 | path->reada = READA_FORWARD; | |
8050 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
8051 | if (ret < 0) | |
8052 | goto out; | |
8053 | ||
8054 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 8055 | ret = btrfs_next_leaf(root, path); |
cf90d884 QW |
8056 | if (ret < 0) |
8057 | goto out; | |
8058 | /* No dev extents at all? Not good */ | |
8059 | if (ret > 0) { | |
8060 | ret = -EUCLEAN; | |
8061 | goto out; | |
8062 | } | |
8063 | } | |
8064 | while (1) { | |
8065 | struct extent_buffer *leaf = path->nodes[0]; | |
8066 | struct btrfs_dev_extent *dext; | |
8067 | int slot = path->slots[0]; | |
8068 | u64 chunk_offset; | |
8069 | u64 physical_offset; | |
8070 | u64 physical_len; | |
8071 | u64 devid; | |
8072 | ||
8073 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
8074 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
8075 | break; | |
8076 | devid = key.objectid; | |
8077 | physical_offset = key.offset; | |
8078 | ||
8079 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
8080 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
8081 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
8082 | ||
5eb19381 QW |
8083 | /* Check if this dev extent overlaps with the previous one */ |
8084 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
8085 | btrfs_err(fs_info, | |
8086 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
8087 | devid, physical_offset, prev_dev_ext_end); | |
8088 | ret = -EUCLEAN; | |
8089 | goto out; | |
8090 | } | |
8091 | ||
cf90d884 QW |
8092 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
8093 | physical_offset, physical_len); | |
8094 | if (ret < 0) | |
8095 | goto out; | |
5eb19381 QW |
8096 | prev_devid = devid; |
8097 | prev_dev_ext_end = physical_offset + physical_len; | |
8098 | ||
cf90d884 QW |
8099 | ret = btrfs_next_item(root, path); |
8100 | if (ret < 0) | |
8101 | goto out; | |
8102 | if (ret > 0) { | |
8103 | ret = 0; | |
8104 | break; | |
8105 | } | |
8106 | } | |
8107 | ||
8108 | /* Ensure all chunks have corresponding dev extents */ | |
8109 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
8110 | out: | |
8111 | btrfs_free_path(path); | |
8112 | return ret; | |
8113 | } | |
eede2bf3 OS |
8114 | |
8115 | /* | |
8116 | * Check whether the given block group or device is pinned by any inode being | |
8117 | * used as a swapfile. | |
8118 | */ | |
8119 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
8120 | { | |
8121 | struct btrfs_swapfile_pin *sp; | |
8122 | struct rb_node *node; | |
8123 | ||
8124 | spin_lock(&fs_info->swapfile_pins_lock); | |
8125 | node = fs_info->swapfile_pins.rb_node; | |
8126 | while (node) { | |
8127 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
8128 | if (ptr < sp->ptr) | |
8129 | node = node->rb_left; | |
8130 | else if (ptr > sp->ptr) | |
8131 | node = node->rb_right; | |
8132 | else | |
8133 | break; | |
8134 | } | |
8135 | spin_unlock(&fs_info->swapfile_pins_lock); | |
8136 | return node != NULL; | |
8137 | } | |
f7ef5287 NA |
8138 | |
8139 | static int relocating_repair_kthread(void *data) | |
8140 | { | |
8141 | struct btrfs_block_group *cache = (struct btrfs_block_group *)data; | |
8142 | struct btrfs_fs_info *fs_info = cache->fs_info; | |
8143 | u64 target; | |
8144 | int ret = 0; | |
8145 | ||
8146 | target = cache->start; | |
8147 | btrfs_put_block_group(cache); | |
8148 | ||
8149 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { | |
8150 | btrfs_info(fs_info, | |
8151 | "zoned: skip relocating block group %llu to repair: EBUSY", | |
8152 | target); | |
8153 | return -EBUSY; | |
8154 | } | |
8155 | ||
f3372065 | 8156 | mutex_lock(&fs_info->reclaim_bgs_lock); |
f7ef5287 NA |
8157 | |
8158 | /* Ensure block group still exists */ | |
8159 | cache = btrfs_lookup_block_group(fs_info, target); | |
8160 | if (!cache) | |
8161 | goto out; | |
8162 | ||
8163 | if (!cache->relocating_repair) | |
8164 | goto out; | |
8165 | ||
8166 | ret = btrfs_may_alloc_data_chunk(fs_info, target); | |
8167 | if (ret < 0) | |
8168 | goto out; | |
8169 | ||
8170 | btrfs_info(fs_info, | |
8171 | "zoned: relocating block group %llu to repair IO failure", | |
8172 | target); | |
8173 | ret = btrfs_relocate_chunk(fs_info, target); | |
8174 | ||
8175 | out: | |
8176 | if (cache) | |
8177 | btrfs_put_block_group(cache); | |
f3372065 | 8178 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f7ef5287 NA |
8179 | btrfs_exclop_finish(fs_info); |
8180 | ||
8181 | return ret; | |
8182 | } | |
8183 | ||
8184 | int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) | |
8185 | { | |
8186 | struct btrfs_block_group *cache; | |
8187 | ||
8188 | /* Do not attempt to repair in degraded state */ | |
8189 | if (btrfs_test_opt(fs_info, DEGRADED)) | |
8190 | return 0; | |
8191 | ||
8192 | cache = btrfs_lookup_block_group(fs_info, logical); | |
8193 | if (!cache) | |
8194 | return 0; | |
8195 | ||
8196 | spin_lock(&cache->lock); | |
8197 | if (cache->relocating_repair) { | |
8198 | spin_unlock(&cache->lock); | |
8199 | btrfs_put_block_group(cache); | |
8200 | return 0; | |
8201 | } | |
8202 | cache->relocating_repair = 1; | |
8203 | spin_unlock(&cache->lock); | |
8204 | ||
8205 | kthread_run(relocating_repair_kthread, cache, | |
8206 | "btrfs-relocating-repair"); | |
8207 | ||
8208 | return 0; | |
8209 | } |