Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
0b86a832 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
0b86a832 | 4 | */ |
c1d7c514 | 5 | |
0b86a832 | 6 | #include <linux/sched.h> |
fccc0007 | 7 | #include <linux/sched/mm.h> |
5a0e3ad6 | 8 | #include <linux/slab.h> |
442a4f63 | 9 | #include <linux/ratelimit.h> |
59641015 | 10 | #include <linux/kthread.h> |
803b2f54 | 11 | #include <linux/semaphore.h> |
8da4b8c4 | 12 | #include <linux/uuid.h> |
f8e10cd3 | 13 | #include <linux/list_sort.h> |
54fde91f | 14 | #include <linux/namei.h> |
784352fe | 15 | #include "misc.h" |
0b86a832 CM |
16 | #include "ctree.h" |
17 | #include "extent_map.h" | |
18 | #include "disk-io.h" | |
19 | #include "transaction.h" | |
20 | #include "print-tree.h" | |
21 | #include "volumes.h" | |
53b381b3 | 22 | #include "raid56.h" |
606686ee | 23 | #include "rcu-string.h" |
8dabb742 | 24 | #include "dev-replace.h" |
99994cde | 25 | #include "sysfs.h" |
82fc28fb | 26 | #include "tree-checker.h" |
8719aaae | 27 | #include "space-info.h" |
aac0023c | 28 | #include "block-group.h" |
b0643e59 | 29 | #include "discard.h" |
5b316468 | 30 | #include "zoned.h" |
c7f13d42 | 31 | #include "fs.h" |
07e81dc9 | 32 | #include "accessors.h" |
c7a03b52 | 33 | #include "uuid-tree.h" |
7572dec8 | 34 | #include "ioctl.h" |
67707479 | 35 | #include "relocation.h" |
2fc6822c | 36 | #include "scrub.h" |
7f0add25 | 37 | #include "super.h" |
0b86a832 | 38 | |
bf08387f QW |
39 | #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ |
40 | BTRFS_BLOCK_GROUP_RAID10 | \ | |
41 | BTRFS_BLOCK_GROUP_RAID56_MASK) | |
42 | ||
af902047 ZL |
43 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
44 | [BTRFS_RAID_RAID10] = { | |
45 | .sub_stripes = 2, | |
46 | .dev_stripes = 1, | |
47 | .devs_max = 0, /* 0 == as many as possible */ | |
b2f78e88 | 48 | .devs_min = 2, |
8789f4fe | 49 | .tolerated_failures = 1, |
af902047 ZL |
50 | .devs_increment = 2, |
51 | .ncopies = 2, | |
b50836ed | 52 | .nparity = 0, |
ed23467b | 53 | .raid_name = "raid10", |
41a6e891 | 54 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
f9fbcaa2 | 55 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
af902047 ZL |
56 | }, |
57 | [BTRFS_RAID_RAID1] = { | |
58 | .sub_stripes = 1, | |
59 | .dev_stripes = 1, | |
60 | .devs_max = 2, | |
61 | .devs_min = 2, | |
8789f4fe | 62 | .tolerated_failures = 1, |
af902047 ZL |
63 | .devs_increment = 2, |
64 | .ncopies = 2, | |
b50836ed | 65 | .nparity = 0, |
ed23467b | 66 | .raid_name = "raid1", |
41a6e891 | 67 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
f9fbcaa2 | 68 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
af902047 | 69 | }, |
47e6f742 DS |
70 | [BTRFS_RAID_RAID1C3] = { |
71 | .sub_stripes = 1, | |
72 | .dev_stripes = 1, | |
cf93e15e | 73 | .devs_max = 3, |
47e6f742 DS |
74 | .devs_min = 3, |
75 | .tolerated_failures = 2, | |
76 | .devs_increment = 3, | |
77 | .ncopies = 3, | |
db26a024 | 78 | .nparity = 0, |
47e6f742 DS |
79 | .raid_name = "raid1c3", |
80 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, | |
81 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, | |
82 | }, | |
8d6fac00 DS |
83 | [BTRFS_RAID_RAID1C4] = { |
84 | .sub_stripes = 1, | |
85 | .dev_stripes = 1, | |
cf93e15e | 86 | .devs_max = 4, |
8d6fac00 DS |
87 | .devs_min = 4, |
88 | .tolerated_failures = 3, | |
89 | .devs_increment = 4, | |
90 | .ncopies = 4, | |
db26a024 | 91 | .nparity = 0, |
8d6fac00 DS |
92 | .raid_name = "raid1c4", |
93 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, | |
94 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, | |
95 | }, | |
af902047 ZL |
96 | [BTRFS_RAID_DUP] = { |
97 | .sub_stripes = 1, | |
98 | .dev_stripes = 2, | |
99 | .devs_max = 1, | |
100 | .devs_min = 1, | |
8789f4fe | 101 | .tolerated_failures = 0, |
af902047 ZL |
102 | .devs_increment = 1, |
103 | .ncopies = 2, | |
b50836ed | 104 | .nparity = 0, |
ed23467b | 105 | .raid_name = "dup", |
41a6e891 | 106 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
f9fbcaa2 | 107 | .mindev_error = 0, |
af902047 ZL |
108 | }, |
109 | [BTRFS_RAID_RAID0] = { | |
110 | .sub_stripes = 1, | |
111 | .dev_stripes = 1, | |
112 | .devs_max = 0, | |
b2f78e88 | 113 | .devs_min = 1, |
8789f4fe | 114 | .tolerated_failures = 0, |
af902047 ZL |
115 | .devs_increment = 1, |
116 | .ncopies = 1, | |
b50836ed | 117 | .nparity = 0, |
ed23467b | 118 | .raid_name = "raid0", |
41a6e891 | 119 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
f9fbcaa2 | 120 | .mindev_error = 0, |
af902047 ZL |
121 | }, |
122 | [BTRFS_RAID_SINGLE] = { | |
123 | .sub_stripes = 1, | |
124 | .dev_stripes = 1, | |
125 | .devs_max = 1, | |
126 | .devs_min = 1, | |
8789f4fe | 127 | .tolerated_failures = 0, |
af902047 ZL |
128 | .devs_increment = 1, |
129 | .ncopies = 1, | |
b50836ed | 130 | .nparity = 0, |
ed23467b | 131 | .raid_name = "single", |
41a6e891 | 132 | .bg_flag = 0, |
f9fbcaa2 | 133 | .mindev_error = 0, |
af902047 ZL |
134 | }, |
135 | [BTRFS_RAID_RAID5] = { | |
136 | .sub_stripes = 1, | |
137 | .dev_stripes = 1, | |
138 | .devs_max = 0, | |
139 | .devs_min = 2, | |
8789f4fe | 140 | .tolerated_failures = 1, |
af902047 | 141 | .devs_increment = 1, |
da612e31 | 142 | .ncopies = 1, |
b50836ed | 143 | .nparity = 1, |
ed23467b | 144 | .raid_name = "raid5", |
41a6e891 | 145 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
f9fbcaa2 | 146 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
af902047 ZL |
147 | }, |
148 | [BTRFS_RAID_RAID6] = { | |
149 | .sub_stripes = 1, | |
150 | .dev_stripes = 1, | |
151 | .devs_max = 0, | |
152 | .devs_min = 3, | |
8789f4fe | 153 | .tolerated_failures = 2, |
af902047 | 154 | .devs_increment = 1, |
da612e31 | 155 | .ncopies = 1, |
b50836ed | 156 | .nparity = 2, |
ed23467b | 157 | .raid_name = "raid6", |
41a6e891 | 158 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
f9fbcaa2 | 159 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
af902047 ZL |
160 | }, |
161 | }; | |
162 | ||
500a44c9 DS |
163 | /* |
164 | * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which | |
165 | * can be used as index to access btrfs_raid_array[]. | |
166 | */ | |
167 | enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) | |
168 | { | |
719fae89 | 169 | const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); |
500a44c9 | 170 | |
719fae89 QW |
171 | if (!profile) |
172 | return BTRFS_RAID_SINGLE; | |
173 | ||
174 | return BTRFS_BG_FLAG_TO_INDEX(profile); | |
500a44c9 DS |
175 | } |
176 | ||
158da513 | 177 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
ed23467b | 178 | { |
158da513 DS |
179 | const int index = btrfs_bg_flags_to_raid_index(flags); |
180 | ||
181 | if (index >= BTRFS_NR_RAID_TYPES) | |
ed23467b AJ |
182 | return NULL; |
183 | ||
158da513 | 184 | return btrfs_raid_array[index].raid_name; |
ed23467b AJ |
185 | } |
186 | ||
0b30f719 QW |
187 | int btrfs_nr_parity_stripes(u64 type) |
188 | { | |
189 | enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); | |
190 | ||
191 | return btrfs_raid_array[index].nparity; | |
192 | } | |
193 | ||
f89e09cf AJ |
194 | /* |
195 | * Fill @buf with textual description of @bg_flags, no more than @size_buf | |
196 | * bytes including terminating null byte. | |
197 | */ | |
198 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) | |
199 | { | |
200 | int i; | |
201 | int ret; | |
202 | char *bp = buf; | |
203 | u64 flags = bg_flags; | |
204 | u32 size_bp = size_buf; | |
205 | ||
206 | if (!flags) { | |
207 | strcpy(bp, "NONE"); | |
208 | return; | |
209 | } | |
210 | ||
211 | #define DESCRIBE_FLAG(flag, desc) \ | |
212 | do { \ | |
213 | if (flags & (flag)) { \ | |
214 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ | |
215 | if (ret < 0 || ret >= size_bp) \ | |
216 | goto out_overflow; \ | |
217 | size_bp -= ret; \ | |
218 | bp += ret; \ | |
219 | flags &= ~(flag); \ | |
220 | } \ | |
221 | } while (0) | |
222 | ||
223 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); | |
224 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); | |
225 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); | |
226 | ||
227 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); | |
228 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
229 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, | |
230 | btrfs_raid_array[i].raid_name); | |
231 | #undef DESCRIBE_FLAG | |
232 | ||
233 | if (flags) { | |
234 | ret = snprintf(bp, size_bp, "0x%llx|", flags); | |
235 | size_bp -= ret; | |
236 | } | |
237 | ||
238 | if (size_bp < size_buf) | |
239 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ | |
240 | ||
241 | /* | |
242 | * The text is trimmed, it's up to the caller to provide sufficiently | |
243 | * large buffer | |
244 | */ | |
245 | out_overflow:; | |
246 | } | |
247 | ||
6f8e0fc7 | 248 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
2ff7e61e | 249 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
733f4fbb | 250 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 251 | |
9c6b1c4d DS |
252 | /* |
253 | * Device locking | |
254 | * ============== | |
255 | * | |
256 | * There are several mutexes that protect manipulation of devices and low-level | |
257 | * structures like chunks but not block groups, extents or files | |
258 | * | |
259 | * uuid_mutex (global lock) | |
260 | * ------------------------ | |
261 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from | |
262 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first | |
263 | * device) or requested by the device= mount option | |
264 | * | |
265 | * the mutex can be very coarse and can cover long-running operations | |
266 | * | |
267 | * protects: updates to fs_devices counters like missing devices, rw devices, | |
52042d8e | 268 | * seeding, structure cloning, opening/closing devices at mount/umount time |
9c6b1c4d DS |
269 | * |
270 | * global::fs_devs - add, remove, updates to the global list | |
271 | * | |
18c850fd JB |
272 | * does not protect: manipulation of the fs_devices::devices list in general |
273 | * but in mount context it could be used to exclude list modifications by eg. | |
274 | * scan ioctl | |
9c6b1c4d DS |
275 | * |
276 | * btrfs_device::name - renames (write side), read is RCU | |
277 | * | |
278 | * fs_devices::device_list_mutex (per-fs, with RCU) | |
279 | * ------------------------------------------------ | |
280 | * protects updates to fs_devices::devices, ie. adding and deleting | |
281 | * | |
282 | * simple list traversal with read-only actions can be done with RCU protection | |
283 | * | |
284 | * may be used to exclude some operations from running concurrently without any | |
285 | * modifications to the list (see write_all_supers) | |
286 | * | |
18c850fd JB |
287 | * Is not required at mount and close times, because our device list is |
288 | * protected by the uuid_mutex at that point. | |
289 | * | |
9c6b1c4d DS |
290 | * balance_mutex |
291 | * ------------- | |
292 | * protects balance structures (status, state) and context accessed from | |
293 | * several places (internally, ioctl) | |
294 | * | |
295 | * chunk_mutex | |
296 | * ----------- | |
297 | * protects chunks, adding or removing during allocation, trim or when a new | |
0b6f5d40 NB |
298 | * device is added/removed. Additionally it also protects post_commit_list of |
299 | * individual devices, since they can be added to the transaction's | |
300 | * post_commit_list only with chunk_mutex held. | |
9c6b1c4d DS |
301 | * |
302 | * cleaner_mutex | |
303 | * ------------- | |
304 | * a big lock that is held by the cleaner thread and prevents running subvolume | |
305 | * cleaning together with relocation or delayed iputs | |
306 | * | |
307 | * | |
308 | * Lock nesting | |
309 | * ============ | |
310 | * | |
311 | * uuid_mutex | |
ae3e715f AJ |
312 | * device_list_mutex |
313 | * chunk_mutex | |
314 | * balance_mutex | |
89595e80 AJ |
315 | * |
316 | * | |
c3e1f96c GR |
317 | * Exclusive operations |
318 | * ==================== | |
89595e80 AJ |
319 | * |
320 | * Maintains the exclusivity of the following operations that apply to the | |
321 | * whole filesystem and cannot run in parallel. | |
322 | * | |
323 | * - Balance (*) | |
324 | * - Device add | |
325 | * - Device remove | |
326 | * - Device replace (*) | |
327 | * - Resize | |
328 | * | |
329 | * The device operations (as above) can be in one of the following states: | |
330 | * | |
331 | * - Running state | |
332 | * - Paused state | |
333 | * - Completed state | |
334 | * | |
335 | * Only device operations marked with (*) can go into the Paused state for the | |
336 | * following reasons: | |
337 | * | |
338 | * - ioctl (only Balance can be Paused through ioctl) | |
339 | * - filesystem remounted as read-only | |
340 | * - filesystem unmounted and mounted as read-only | |
341 | * - system power-cycle and filesystem mounted as read-only | |
342 | * - filesystem or device errors leading to forced read-only | |
343 | * | |
c3e1f96c GR |
344 | * The status of exclusive operation is set and cleared atomically. |
345 | * During the course of Paused state, fs_info::exclusive_operation remains set. | |
89595e80 AJ |
346 | * A device operation in Paused or Running state can be canceled or resumed |
347 | * either by ioctl (Balance only) or when remounted as read-write. | |
c3e1f96c | 348 | * The exclusive status is cleared when the device operation is canceled or |
89595e80 | 349 | * completed. |
9c6b1c4d DS |
350 | */ |
351 | ||
67a2c45e | 352 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 353 | static LIST_HEAD(fs_uuids); |
4143cb8b | 354 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
c73eccf7 AJ |
355 | { |
356 | return &fs_uuids; | |
357 | } | |
8a4b83cc | 358 | |
2dfeca9b DS |
359 | /* |
360 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
7239ff4b NB |
361 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid |
362 | * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid | |
2dfeca9b DS |
363 | * |
364 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). | |
365 | * The returned struct is not linked onto any lists and can be destroyed with | |
366 | * kfree() right away. | |
367 | */ | |
7239ff4b NB |
368 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid, |
369 | const u8 *metadata_fsid) | |
2208a378 ID |
370 | { |
371 | struct btrfs_fs_devices *fs_devs; | |
372 | ||
78f2c9e6 | 373 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
374 | if (!fs_devs) |
375 | return ERR_PTR(-ENOMEM); | |
376 | ||
377 | mutex_init(&fs_devs->device_list_mutex); | |
378 | ||
379 | INIT_LIST_HEAD(&fs_devs->devices); | |
380 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
c4babc5e | 381 | INIT_LIST_HEAD(&fs_devs->fs_list); |
944d3f9f | 382 | INIT_LIST_HEAD(&fs_devs->seed_list); |
2208a378 ID |
383 | if (fsid) |
384 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
2208a378 | 385 | |
7239ff4b NB |
386 | if (metadata_fsid) |
387 | memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE); | |
388 | else if (fsid) | |
389 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); | |
390 | ||
2208a378 ID |
391 | return fs_devs; |
392 | } | |
393 | ||
a425f9d4 | 394 | void btrfs_free_device(struct btrfs_device *device) |
48dae9cf | 395 | { |
bbbf7243 | 396 | WARN_ON(!list_empty(&device->post_commit_list)); |
48dae9cf | 397 | rcu_string_free(device->name); |
1c11b63e | 398 | extent_io_tree_release(&device->alloc_state); |
5b316468 | 399 | btrfs_destroy_dev_zone_info(device); |
48dae9cf DS |
400 | kfree(device); |
401 | } | |
402 | ||
e4404d6e YZ |
403 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
404 | { | |
405 | struct btrfs_device *device; | |
406 | WARN_ON(fs_devices->opened); | |
407 | while (!list_empty(&fs_devices->devices)) { | |
408 | device = list_entry(fs_devices->devices.next, | |
409 | struct btrfs_device, dev_list); | |
410 | list_del(&device->dev_list); | |
a425f9d4 | 411 | btrfs_free_device(device); |
e4404d6e YZ |
412 | } |
413 | kfree(fs_devices); | |
414 | } | |
415 | ||
ffc5a379 | 416 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
417 | { |
418 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 419 | |
2b82032c YZ |
420 | while (!list_empty(&fs_uuids)) { |
421 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
422 | struct btrfs_fs_devices, fs_list); |
423 | list_del(&fs_devices->fs_list); | |
e4404d6e | 424 | free_fs_devices(fs_devices); |
8a4b83cc | 425 | } |
8a4b83cc CM |
426 | } |
427 | ||
7239ff4b NB |
428 | static noinline struct btrfs_fs_devices *find_fsid( |
429 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 430 | { |
8a4b83cc CM |
431 | struct btrfs_fs_devices *fs_devices; |
432 | ||
7239ff4b NB |
433 | ASSERT(fsid); |
434 | ||
7a62d0f0 | 435 | /* Handle non-split brain cases */ |
c4babc5e | 436 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
437 | if (metadata_fsid) { |
438 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
439 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
440 | BTRFS_FSID_SIZE) == 0) | |
441 | return fs_devices; | |
442 | } else { | |
443 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
444 | return fs_devices; | |
445 | } | |
8a4b83cc CM |
446 | } |
447 | return NULL; | |
448 | } | |
449 | ||
c6730a0e SY |
450 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
451 | struct btrfs_super_block *disk_super) | |
452 | { | |
453 | ||
454 | struct btrfs_fs_devices *fs_devices; | |
455 | ||
456 | /* | |
457 | * Handle scanned device having completed its fsid change but | |
458 | * belonging to a fs_devices that was created by first scanning | |
459 | * a device which didn't have its fsid/metadata_uuid changed | |
460 | * at all and the CHANGING_FSID_V2 flag set. | |
461 | */ | |
462 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
463 | if (fs_devices->fsid_change && | |
464 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
465 | BTRFS_FSID_SIZE) == 0 && | |
466 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
467 | BTRFS_FSID_SIZE) == 0) { | |
468 | return fs_devices; | |
469 | } | |
470 | } | |
471 | /* | |
472 | * Handle scanned device having completed its fsid change but | |
473 | * belonging to a fs_devices that was created by a device that | |
474 | * has an outdated pair of fsid/metadata_uuid and | |
475 | * CHANGING_FSID_V2 flag set. | |
476 | */ | |
477 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
478 | if (fs_devices->fsid_change && | |
479 | memcmp(fs_devices->metadata_uuid, | |
480 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
481 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
482 | BTRFS_FSID_SIZE) == 0) { | |
483 | return fs_devices; | |
484 | } | |
485 | } | |
486 | ||
487 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
488 | } | |
489 | ||
490 | ||
beaf8ab3 SB |
491 | static int |
492 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
493 | int flush, struct block_device **bdev, | |
8f32380d | 494 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
495 | { |
496 | int ret; | |
497 | ||
498 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
499 | ||
500 | if (IS_ERR(*bdev)) { | |
501 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
502 | goto error; |
503 | } | |
504 | ||
505 | if (flush) | |
1226dfff | 506 | sync_blockdev(*bdev); |
9f6d2510 | 507 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 SB |
508 | if (ret) { |
509 | blkdev_put(*bdev, flags); | |
510 | goto error; | |
511 | } | |
512 | invalidate_bdev(*bdev); | |
8f32380d JT |
513 | *disk_super = btrfs_read_dev_super(*bdev); |
514 | if (IS_ERR(*disk_super)) { | |
515 | ret = PTR_ERR(*disk_super); | |
beaf8ab3 SB |
516 | blkdev_put(*bdev, flags); |
517 | goto error; | |
518 | } | |
519 | ||
520 | return 0; | |
521 | ||
522 | error: | |
523 | *bdev = NULL; | |
beaf8ab3 SB |
524 | return ret; |
525 | } | |
526 | ||
43dd529a DS |
527 | /* |
528 | * Search and remove all stale devices (which are not mounted). When both | |
529 | * inputs are NULL, it will search and release all stale devices. | |
16cab91a | 530 | * |
43dd529a DS |
531 | * @devt: Optional. When provided will it release all unmounted devices |
532 | * matching this devt only. | |
16cab91a | 533 | * @skip_device: Optional. Will skip this device when searching for the stale |
43dd529a | 534 | * devices. |
16cab91a AJ |
535 | * |
536 | * Return: 0 for success or if @devt is 0. | |
537 | * -EBUSY if @devt is a mounted device. | |
538 | * -ENOENT if @devt does not match any device in the list. | |
d8367db3 | 539 | */ |
16cab91a | 540 | static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) |
4fde46f0 | 541 | { |
fa6d2ae5 AJ |
542 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
543 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
544 | int ret = 0; |
545 | ||
c1247069 AJ |
546 | lockdep_assert_held(&uuid_mutex); |
547 | ||
16cab91a | 548 | if (devt) |
70bc7088 | 549 | ret = -ENOENT; |
4fde46f0 | 550 | |
fa6d2ae5 | 551 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 552 | |
70bc7088 | 553 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
554 | list_for_each_entry_safe(device, tmp_device, |
555 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 556 | if (skip_device && skip_device == device) |
d8367db3 | 557 | continue; |
330a5bf4 | 558 | if (devt && devt != device->devt) |
38cf665d | 559 | continue; |
70bc7088 AJ |
560 | if (fs_devices->opened) { |
561 | /* for an already deleted device return 0 */ | |
16cab91a | 562 | if (devt && ret != 0) |
70bc7088 AJ |
563 | ret = -EBUSY; |
564 | break; | |
565 | } | |
4fde46f0 | 566 | |
4fde46f0 | 567 | /* delete the stale device */ |
7bcb8164 AJ |
568 | fs_devices->num_devices--; |
569 | list_del(&device->dev_list); | |
570 | btrfs_free_device(device); | |
571 | ||
70bc7088 | 572 | ret = 0; |
7bcb8164 AJ |
573 | } |
574 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 575 | |
7bcb8164 AJ |
576 | if (fs_devices->num_devices == 0) { |
577 | btrfs_sysfs_remove_fsid(fs_devices); | |
578 | list_del(&fs_devices->fs_list); | |
579 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
580 | } |
581 | } | |
70bc7088 AJ |
582 | |
583 | return ret; | |
4fde46f0 AJ |
584 | } |
585 | ||
18c850fd JB |
586 | /* |
587 | * This is only used on mount, and we are protected from competing things | |
588 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
589 | * fs_devices->device_list_mutex here. | |
590 | */ | |
0fb08bcc AJ |
591 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
592 | struct btrfs_device *device, fmode_t flags, | |
593 | void *holder) | |
594 | { | |
0fb08bcc | 595 | struct block_device *bdev; |
0fb08bcc AJ |
596 | struct btrfs_super_block *disk_super; |
597 | u64 devid; | |
598 | int ret; | |
599 | ||
600 | if (device->bdev) | |
601 | return -EINVAL; | |
602 | if (!device->name) | |
603 | return -EINVAL; | |
604 | ||
605 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 606 | &bdev, &disk_super); |
0fb08bcc AJ |
607 | if (ret) |
608 | return ret; | |
609 | ||
0fb08bcc AJ |
610 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
611 | if (devid != device->devid) | |
8f32380d | 612 | goto error_free_page; |
0fb08bcc AJ |
613 | |
614 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 615 | goto error_free_page; |
0fb08bcc AJ |
616 | |
617 | device->generation = btrfs_super_generation(disk_super); | |
618 | ||
619 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
620 | if (btrfs_super_incompat_flags(disk_super) & |
621 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
622 | pr_err( | |
623 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 624 | goto error_free_page; |
7239ff4b NB |
625 | } |
626 | ||
ebbede42 | 627 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 628 | fs_devices->seeding = true; |
0fb08bcc | 629 | } else { |
ebbede42 AJ |
630 | if (bdev_read_only(bdev)) |
631 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
632 | else | |
633 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
634 | } |
635 | ||
10f0d2a5 | 636 | if (!bdev_nonrot(bdev)) |
7f0432d0 | 637 | fs_devices->rotating = true; |
0fb08bcc | 638 | |
63a7cb13 DS |
639 | if (bdev_max_discard_sectors(bdev)) |
640 | fs_devices->discardable = true; | |
641 | ||
0fb08bcc | 642 | device->bdev = bdev; |
e12c9621 | 643 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
0fb08bcc AJ |
644 | device->mode = flags; |
645 | ||
646 | fs_devices->open_devices++; | |
ebbede42 AJ |
647 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
648 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 649 | fs_devices->rw_devices++; |
b1b8e386 | 650 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 651 | } |
8f32380d | 652 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
653 | |
654 | return 0; | |
655 | ||
8f32380d JT |
656 | error_free_page: |
657 | btrfs_release_disk_super(disk_super); | |
0fb08bcc AJ |
658 | blkdev_put(bdev, flags); |
659 | ||
660 | return -EINVAL; | |
661 | } | |
662 | ||
7a62d0f0 NB |
663 | /* |
664 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
665 | * being created with a disk that has already completed its fsid change. Such |
666 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
667 | * Handle both cases here. | |
7a62d0f0 NB |
668 | */ |
669 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
670 | struct btrfs_super_block *disk_super) | |
671 | { | |
672 | struct btrfs_fs_devices *fs_devices; | |
673 | ||
674 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
675 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
676 | BTRFS_FSID_SIZE) != 0 && | |
677 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
678 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
679 | return fs_devices; | |
680 | } | |
681 | } | |
682 | ||
c0d81c7c | 683 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
684 | } |
685 | ||
cc5de4e7 NB |
686 | |
687 | static struct btrfs_fs_devices *find_fsid_changed( | |
688 | struct btrfs_super_block *disk_super) | |
689 | { | |
690 | struct btrfs_fs_devices *fs_devices; | |
691 | ||
692 | /* | |
693 | * Handles the case where scanned device is part of an fs that had | |
1a9fd417 | 694 | * multiple successful changes of FSID but currently device didn't |
05840710 NB |
695 | * observe it. Meaning our fsid will be different than theirs. We need |
696 | * to handle two subcases : | |
697 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
698 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
699 | * are equal). | |
cc5de4e7 NB |
700 | */ |
701 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 702 | /* Changed UUIDs */ |
cc5de4e7 NB |
703 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
704 | BTRFS_FSID_SIZE) != 0 && | |
705 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
706 | BTRFS_FSID_SIZE) == 0 && | |
707 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
708 | BTRFS_FSID_SIZE) != 0) |
709 | return fs_devices; | |
710 | ||
711 | /* Unchanged UUIDs */ | |
712 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
713 | BTRFS_FSID_SIZE) == 0 && | |
714 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
715 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 716 | return fs_devices; |
cc5de4e7 NB |
717 | } |
718 | ||
719 | return NULL; | |
720 | } | |
1362089d NB |
721 | |
722 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
723 | struct btrfs_super_block *disk_super) | |
724 | { | |
725 | struct btrfs_fs_devices *fs_devices; | |
726 | ||
727 | /* | |
728 | * Handle the case where the scanned device is part of an fs whose last | |
729 | * metadata UUID change reverted it to the original FSID. At the same | |
730 | * time * fs_devices was first created by another constitutent device | |
731 | * which didn't fully observe the operation. This results in an | |
732 | * btrfs_fs_devices created with metadata/fsid different AND | |
733 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
734 | * fs_devices equal to the FSID of the disk. | |
735 | */ | |
736 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
737 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
738 | BTRFS_FSID_SIZE) != 0 && | |
739 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
740 | BTRFS_FSID_SIZE) == 0 && | |
741 | fs_devices->fsid_change) | |
742 | return fs_devices; | |
743 | } | |
744 | ||
745 | return NULL; | |
746 | } | |
60999ca4 DS |
747 | /* |
748 | * Add new device to list of registered devices | |
749 | * | |
750 | * Returns: | |
e124ece5 AJ |
751 | * device pointer which was just added or updated when successful |
752 | * error pointer when failed | |
60999ca4 | 753 | */ |
e124ece5 | 754 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
755 | struct btrfs_super_block *disk_super, |
756 | bool *new_device_added) | |
8a4b83cc CM |
757 | { |
758 | struct btrfs_device *device; | |
7a62d0f0 | 759 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 760 | struct rcu_string *name; |
8a4b83cc | 761 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 762 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
4889bc05 AJ |
763 | dev_t path_devt; |
764 | int error; | |
7239ff4b NB |
765 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
766 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
767 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
768 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 769 | |
4889bc05 | 770 | error = lookup_bdev(path, &path_devt); |
ed02363f QW |
771 | if (error) { |
772 | btrfs_err(NULL, "failed to lookup block device for path %s: %d", | |
773 | path, error); | |
4889bc05 | 774 | return ERR_PTR(error); |
ed02363f | 775 | } |
4889bc05 | 776 | |
cc5de4e7 | 777 | if (fsid_change_in_progress) { |
c0d81c7c | 778 | if (!has_metadata_uuid) |
cc5de4e7 | 779 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 780 | else |
cc5de4e7 | 781 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 782 | } else if (has_metadata_uuid) { |
c6730a0e | 783 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 784 | } else { |
1362089d NB |
785 | fs_devices = find_fsid_reverted_metadata(disk_super); |
786 | if (!fs_devices) | |
787 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
788 | } |
789 | ||
8a4b83cc | 790 | |
8a4b83cc | 791 | if (!fs_devices) { |
7239ff4b NB |
792 | if (has_metadata_uuid) |
793 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
794 | disk_super->metadata_uuid); | |
795 | else | |
796 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
797 | ||
2208a378 | 798 | if (IS_ERR(fs_devices)) |
e124ece5 | 799 | return ERR_CAST(fs_devices); |
2208a378 | 800 | |
92900e51 AV |
801 | fs_devices->fsid_change = fsid_change_in_progress; |
802 | ||
9c6d173e | 803 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 804 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 805 | |
8a4b83cc CM |
806 | device = NULL; |
807 | } else { | |
562d7b15 JB |
808 | struct btrfs_dev_lookup_args args = { |
809 | .devid = devid, | |
810 | .uuid = disk_super->dev_item.uuid, | |
811 | }; | |
812 | ||
9c6d173e | 813 | mutex_lock(&fs_devices->device_list_mutex); |
562d7b15 | 814 | device = btrfs_find_device(fs_devices, &args); |
7a62d0f0 NB |
815 | |
816 | /* | |
817 | * If this disk has been pulled into an fs devices created by | |
818 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
819 | * metadata_uuid/fsid values of the fs_devices. | |
820 | */ | |
1362089d | 821 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
822 | found_transid > fs_devices->latest_generation) { |
823 | memcpy(fs_devices->fsid, disk_super->fsid, | |
824 | BTRFS_FSID_SIZE); | |
1362089d NB |
825 | |
826 | if (has_metadata_uuid) | |
827 | memcpy(fs_devices->metadata_uuid, | |
828 | disk_super->metadata_uuid, | |
829 | BTRFS_FSID_SIZE); | |
830 | else | |
831 | memcpy(fs_devices->metadata_uuid, | |
832 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
833 | |
834 | fs_devices->fsid_change = false; | |
835 | } | |
8a4b83cc | 836 | } |
443f24fe | 837 | |
8a4b83cc | 838 | if (!device) { |
bb21e302 AJ |
839 | unsigned int nofs_flag; |
840 | ||
9c6d173e | 841 | if (fs_devices->opened) { |
ed02363f QW |
842 | btrfs_err(NULL, |
843 | "device %s belongs to fsid %pU, and the fs is already mounted", | |
844 | path, fs_devices->fsid); | |
9c6d173e | 845 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 846 | return ERR_PTR(-EBUSY); |
9c6d173e | 847 | } |
2b82032c | 848 | |
bb21e302 | 849 | nofs_flag = memalloc_nofs_save(); |
12bd2fc0 | 850 | device = btrfs_alloc_device(NULL, &devid, |
bb21e302 AJ |
851 | disk_super->dev_item.uuid, path); |
852 | memalloc_nofs_restore(nofs_flag); | |
12bd2fc0 | 853 | if (IS_ERR(device)) { |
9c6d173e | 854 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 855 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 856 | return device; |
8a4b83cc | 857 | } |
606686ee | 858 | |
4889bc05 | 859 | device->devt = path_devt; |
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); |
ed02363f QW |
914 | btrfs_err(NULL, |
915 | "device %s already registered with a higher generation, found %llu expect %llu", | |
916 | path, found_transid, device->generation); | |
e124ece5 | 917 | return ERR_PTR(-EEXIST); |
77bdae4d | 918 | } |
b96de000 | 919 | |
a9261d41 AJ |
920 | /* |
921 | * We are going to replace the device path for a given devid, | |
922 | * make sure it's the same device if the device is mounted | |
79c9234b DM |
923 | * |
924 | * NOTE: the device->fs_info may not be reliable here so pass | |
925 | * in a NULL to message helpers instead. This avoids a possible | |
926 | * use-after-free when the fs_info and fs_info->sb are already | |
927 | * torn down. | |
a9261d41 AJ |
928 | */ |
929 | if (device->bdev) { | |
4889bc05 | 930 | if (device->devt != path_devt) { |
a9261d41 | 931 | mutex_unlock(&fs_devices->device_list_mutex); |
0697d9a6 | 932 | btrfs_warn_in_rcu(NULL, |
79dae17d AJ |
933 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
934 | path, devid, found_transid, | |
935 | current->comm, | |
936 | task_pid_nr(current)); | |
a9261d41 AJ |
937 | return ERR_PTR(-EEXIST); |
938 | } | |
79c9234b | 939 | btrfs_info_in_rcu(NULL, |
79dae17d | 940 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
cb3e217b | 941 | devid, btrfs_dev_name(device), |
79dae17d AJ |
942 | path, current->comm, |
943 | task_pid_nr(current)); | |
a9261d41 AJ |
944 | } |
945 | ||
606686ee | 946 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
947 | if (!name) { |
948 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 949 | return ERR_PTR(-ENOMEM); |
9c6d173e | 950 | } |
606686ee JB |
951 | rcu_string_free(device->name); |
952 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 953 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 954 | fs_devices->missing_devices--; |
e6e674bd | 955 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 956 | } |
4889bc05 | 957 | device->devt = path_devt; |
8a4b83cc CM |
958 | } |
959 | ||
77bdae4d AJ |
960 | /* |
961 | * Unmount does not free the btrfs_device struct but would zero | |
962 | * generation along with most of the other members. So just update | |
963 | * it back. We need it to pick the disk with largest generation | |
964 | * (as above). | |
965 | */ | |
d1a63002 | 966 | if (!fs_devices->opened) { |
77bdae4d | 967 | device->generation = found_transid; |
d1a63002 NB |
968 | fs_devices->latest_generation = max_t(u64, found_transid, |
969 | fs_devices->latest_generation); | |
970 | } | |
77bdae4d | 971 | |
f2788d2f AJ |
972 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
973 | ||
9c6d173e | 974 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 975 | return device; |
8a4b83cc CM |
976 | } |
977 | ||
e4404d6e YZ |
978 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
979 | { | |
980 | struct btrfs_fs_devices *fs_devices; | |
981 | struct btrfs_device *device; | |
982 | struct btrfs_device *orig_dev; | |
d2979aa2 | 983 | int ret = 0; |
e4404d6e | 984 | |
c1247069 AJ |
985 | lockdep_assert_held(&uuid_mutex); |
986 | ||
7239ff4b | 987 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
988 | if (IS_ERR(fs_devices)) |
989 | return fs_devices; | |
e4404d6e | 990 | |
02db0844 | 991 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
992 | |
993 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
bb21e302 AJ |
994 | const char *dev_path = NULL; |
995 | ||
996 | /* | |
997 | * This is ok to do without RCU read locked because we hold the | |
998 | * uuid mutex so nothing we touch in here is going to disappear. | |
999 | */ | |
1000 | if (orig_dev->name) | |
1001 | dev_path = orig_dev->name->str; | |
606686ee | 1002 | |
12bd2fc0 | 1003 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
bb21e302 | 1004 | orig_dev->uuid, dev_path); |
d2979aa2 AJ |
1005 | if (IS_ERR(device)) { |
1006 | ret = PTR_ERR(device); | |
e4404d6e | 1007 | goto error; |
d2979aa2 | 1008 | } |
e4404d6e | 1009 | |
21e61ec6 JT |
1010 | if (orig_dev->zone_info) { |
1011 | struct btrfs_zoned_device_info *zone_info; | |
1012 | ||
1013 | zone_info = btrfs_clone_dev_zone_info(orig_dev); | |
1014 | if (!zone_info) { | |
1015 | btrfs_free_device(device); | |
1016 | ret = -ENOMEM; | |
1017 | goto error; | |
1018 | } | |
1019 | device->zone_info = zone_info; | |
1020 | } | |
1021 | ||
e4404d6e YZ |
1022 | list_add(&device->dev_list, &fs_devices->devices); |
1023 | device->fs_devices = fs_devices; | |
1024 | fs_devices->num_devices++; | |
1025 | } | |
1026 | return fs_devices; | |
1027 | error: | |
1028 | free_fs_devices(fs_devices); | |
d2979aa2 | 1029 | return ERR_PTR(ret); |
e4404d6e YZ |
1030 | } |
1031 | ||
3712ccb7 | 1032 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
bacce86a | 1033 | struct btrfs_device **latest_dev) |
dfe25020 | 1034 | { |
c6e30871 | 1035 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1036 | |
46224705 | 1037 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1038 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1039 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1040 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1041 | &device->dev_state) && |
998a0671 AJ |
1042 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1043 | &device->dev_state) && | |
3712ccb7 NB |
1044 | (!*latest_dev || |
1045 | device->generation > (*latest_dev)->generation)) { | |
1046 | *latest_dev = device; | |
a6b0d5c8 | 1047 | } |
2b82032c | 1048 | continue; |
a6b0d5c8 | 1049 | } |
2b82032c | 1050 | |
cf89af14 AJ |
1051 | /* |
1052 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, | |
1053 | * in btrfs_init_dev_replace() so just continue. | |
1054 | */ | |
1055 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) | |
1056 | continue; | |
1057 | ||
2b82032c | 1058 | if (device->bdev) { |
d4d77629 | 1059 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
1060 | device->bdev = NULL; |
1061 | fs_devices->open_devices--; | |
1062 | } | |
ebbede42 | 1063 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1064 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1065 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
b2a61667 | 1066 | fs_devices->rw_devices--; |
2b82032c | 1067 | } |
e4404d6e YZ |
1068 | list_del_init(&device->dev_list); |
1069 | fs_devices->num_devices--; | |
a425f9d4 | 1070 | btrfs_free_device(device); |
dfe25020 | 1071 | } |
2b82032c | 1072 | |
3712ccb7 NB |
1073 | } |
1074 | ||
1075 | /* | |
1076 | * After we have read the system tree and know devids belonging to this | |
1077 | * filesystem, remove the device which does not belong there. | |
1078 | */ | |
bacce86a | 1079 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
3712ccb7 NB |
1080 | { |
1081 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1082 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1083 | |
1084 | mutex_lock(&uuid_mutex); | |
bacce86a | 1085 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
944d3f9f NB |
1086 | |
1087 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
bacce86a | 1088 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
2b82032c | 1089 | |
d24fa5c1 | 1090 | fs_devices->latest_dev = latest_dev; |
a6b0d5c8 | 1091 | |
dfe25020 | 1092 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1093 | } |
a0af469b | 1094 | |
14238819 AJ |
1095 | static void btrfs_close_bdev(struct btrfs_device *device) |
1096 | { | |
08ffcae8 DS |
1097 | if (!device->bdev) |
1098 | return; | |
1099 | ||
ebbede42 | 1100 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1101 | sync_blockdev(device->bdev); |
1102 | invalidate_bdev(device->bdev); | |
1103 | } | |
1104 | ||
08ffcae8 | 1105 | blkdev_put(device->bdev, device->mode); |
14238819 AJ |
1106 | } |
1107 | ||
959b1c04 | 1108 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1109 | { |
1110 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1111 | |
ebbede42 | 1112 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1113 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1114 | list_del_init(&device->dev_alloc_list); | |
1115 | fs_devices->rw_devices--; | |
1116 | } | |
1117 | ||
0d977e0e DCZX |
1118 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
1119 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); | |
1120 | ||
5d03dbeb LZ |
1121 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
1122 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); | |
f448341a | 1123 | fs_devices->missing_devices--; |
5d03dbeb | 1124 | } |
f448341a | 1125 | |
959b1c04 | 1126 | btrfs_close_bdev(device); |
321f69f8 | 1127 | if (device->bdev) { |
3fff3975 | 1128 | fs_devices->open_devices--; |
321f69f8 | 1129 | device->bdev = NULL; |
f448341a | 1130 | } |
321f69f8 | 1131 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
5b316468 | 1132 | btrfs_destroy_dev_zone_info(device); |
f448341a | 1133 | |
321f69f8 JT |
1134 | device->fs_info = NULL; |
1135 | atomic_set(&device->dev_stats_ccnt, 0); | |
1136 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1137 | |
6b225baa FM |
1138 | /* |
1139 | * Reset the flush error record. We might have a transient flush error | |
1140 | * in this mount, and if so we aborted the current transaction and set | |
1141 | * the fs to an error state, guaranteeing no super blocks can be further | |
1142 | * committed. However that error might be transient and if we unmount the | |
1143 | * filesystem and mount it again, we should allow the mount to succeed | |
1144 | * (btrfs_check_rw_degradable() should not fail) - if after mounting the | |
1145 | * filesystem again we still get flush errors, then we will again abort | |
1146 | * any transaction and set the error state, guaranteeing no commits of | |
1147 | * unsafe super blocks. | |
1148 | */ | |
1149 | device->last_flush_error = 0; | |
1150 | ||
321f69f8 JT |
1151 | /* Verify the device is back in a pristine state */ |
1152 | ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); | |
1153 | ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1154 | ASSERT(list_empty(&device->dev_alloc_list)); | |
1155 | ASSERT(list_empty(&device->post_commit_list)); | |
f448341a AJ |
1156 | } |
1157 | ||
54eed6ae | 1158 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1159 | { |
2037a093 | 1160 | struct btrfs_device *device, *tmp; |
e4404d6e | 1161 | |
425c6ed6 JB |
1162 | lockdep_assert_held(&uuid_mutex); |
1163 | ||
2b82032c | 1164 | if (--fs_devices->opened > 0) |
54eed6ae | 1165 | return; |
8a4b83cc | 1166 | |
425c6ed6 | 1167 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1168 | btrfs_close_one_device(device); |
c9513edb | 1169 | |
e4404d6e YZ |
1170 | WARN_ON(fs_devices->open_devices); |
1171 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1172 | fs_devices->opened = 0; |
0395d84f | 1173 | fs_devices->seeding = false; |
c4989c2f | 1174 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1175 | } |
1176 | ||
54eed6ae | 1177 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1178 | { |
944d3f9f NB |
1179 | LIST_HEAD(list); |
1180 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1181 | |
1182 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1183 | close_fs_devices(fs_devices); |
944d3f9f NB |
1184 | if (!fs_devices->opened) |
1185 | list_splice_init(&fs_devices->seed_list, &list); | |
e4404d6e | 1186 | |
944d3f9f | 1187 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
0226e0eb | 1188 | close_fs_devices(fs_devices); |
944d3f9f | 1189 | list_del(&fs_devices->seed_list); |
e4404d6e YZ |
1190 | free_fs_devices(fs_devices); |
1191 | } | |
425c6ed6 | 1192 | mutex_unlock(&uuid_mutex); |
2b82032c YZ |
1193 | } |
1194 | ||
897fb573 | 1195 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
e4404d6e | 1196 | fmode_t flags, void *holder) |
8a4b83cc | 1197 | { |
8a4b83cc | 1198 | struct btrfs_device *device; |
443f24fe | 1199 | struct btrfs_device *latest_dev = NULL; |
96c2e067 | 1200 | struct btrfs_device *tmp_device; |
8a4b83cc | 1201 | |
d4d77629 TH |
1202 | flags |= FMODE_EXCL; |
1203 | ||
96c2e067 AJ |
1204 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
1205 | dev_list) { | |
1206 | int ret; | |
a0af469b | 1207 | |
96c2e067 AJ |
1208 | ret = btrfs_open_one_device(fs_devices, device, flags, holder); |
1209 | if (ret == 0 && | |
1210 | (!latest_dev || device->generation > latest_dev->generation)) { | |
9f050db4 | 1211 | latest_dev = device; |
96c2e067 AJ |
1212 | } else if (ret == -ENODATA) { |
1213 | fs_devices->num_devices--; | |
1214 | list_del(&device->dev_list); | |
1215 | btrfs_free_device(device); | |
1216 | } | |
8a4b83cc | 1217 | } |
1ed802c9 AJ |
1218 | if (fs_devices->open_devices == 0) |
1219 | return -EINVAL; | |
1220 | ||
2b82032c | 1221 | fs_devices->opened = 1; |
d24fa5c1 | 1222 | fs_devices->latest_dev = latest_dev; |
2b82032c | 1223 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1224 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
33fd2f71 | 1225 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
1ed802c9 AJ |
1226 | |
1227 | return 0; | |
2b82032c YZ |
1228 | } |
1229 | ||
4f0f586b ST |
1230 | static int devid_cmp(void *priv, const struct list_head *a, |
1231 | const struct list_head *b) | |
f8e10cd3 | 1232 | { |
214cc184 | 1233 | const struct btrfs_device *dev1, *dev2; |
f8e10cd3 AJ |
1234 | |
1235 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1236 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1237 | ||
1238 | if (dev1->devid < dev2->devid) | |
1239 | return -1; | |
1240 | else if (dev1->devid > dev2->devid) | |
1241 | return 1; | |
1242 | return 0; | |
1243 | } | |
1244 | ||
2b82032c | 1245 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
97288f2c | 1246 | fmode_t flags, void *holder) |
2b82032c YZ |
1247 | { |
1248 | int ret; | |
1249 | ||
f5194e34 | 1250 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1251 | /* |
1252 | * The device_list_mutex cannot be taken here in case opening the | |
a8698707 | 1253 | * underlying device takes further locks like open_mutex. |
18c850fd JB |
1254 | * |
1255 | * We also don't need the lock here as this is called during mount and | |
1256 | * exclusion is provided by uuid_mutex | |
1257 | */ | |
f5194e34 | 1258 | |
2b82032c | 1259 | if (fs_devices->opened) { |
e4404d6e YZ |
1260 | fs_devices->opened++; |
1261 | ret = 0; | |
2b82032c | 1262 | } else { |
f8e10cd3 | 1263 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1264 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1265 | } |
542c5908 | 1266 | |
8a4b83cc CM |
1267 | return ret; |
1268 | } | |
1269 | ||
8f32380d | 1270 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1271 | { |
8f32380d JT |
1272 | struct page *page = virt_to_page(super); |
1273 | ||
6cf86a00 AJ |
1274 | put_page(page); |
1275 | } | |
1276 | ||
b335eab8 | 1277 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
12659251 | 1278 | u64 bytenr, u64 bytenr_orig) |
6cf86a00 | 1279 | { |
b335eab8 NB |
1280 | struct btrfs_super_block *disk_super; |
1281 | struct page *page; | |
6cf86a00 AJ |
1282 | void *p; |
1283 | pgoff_t index; | |
1284 | ||
1285 | /* make sure our super fits in the device */ | |
cda00eba | 1286 | if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) |
b335eab8 | 1287 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1288 | |
1289 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1290 | if (sizeof(*disk_super) > PAGE_SIZE) |
1291 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1292 | |
1293 | /* make sure our super doesn't straddle pages on disk */ | |
1294 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1295 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1296 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1297 | |
1298 | /* pull in the page with our super */ | |
b335eab8 | 1299 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1300 | |
b335eab8 NB |
1301 | if (IS_ERR(page)) |
1302 | return ERR_CAST(page); | |
6cf86a00 | 1303 | |
b335eab8 | 1304 | p = page_address(page); |
6cf86a00 AJ |
1305 | |
1306 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1307 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1308 | |
12659251 | 1309 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
b335eab8 | 1310 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
8f32380d | 1311 | btrfs_release_disk_super(p); |
b335eab8 | 1312 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1313 | } |
1314 | ||
b335eab8 NB |
1315 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1316 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1317 | |
b335eab8 | 1318 | return disk_super; |
6cf86a00 AJ |
1319 | } |
1320 | ||
16cab91a | 1321 | int btrfs_forget_devices(dev_t devt) |
228a73ab AJ |
1322 | { |
1323 | int ret; | |
1324 | ||
1325 | mutex_lock(&uuid_mutex); | |
16cab91a | 1326 | ret = btrfs_free_stale_devices(devt, NULL); |
228a73ab AJ |
1327 | mutex_unlock(&uuid_mutex); |
1328 | ||
1329 | return ret; | |
1330 | } | |
1331 | ||
6f60cbd3 DS |
1332 | /* |
1333 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1334 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1335 | * is read via pagecache | |
1336 | */ | |
36350e95 GJ |
1337 | struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags, |
1338 | void *holder) | |
8a4b83cc CM |
1339 | { |
1340 | struct btrfs_super_block *disk_super; | |
4306a974 | 1341 | bool new_device_added = false; |
36350e95 | 1342 | struct btrfs_device *device = NULL; |
8a4b83cc | 1343 | struct block_device *bdev; |
12659251 NA |
1344 | u64 bytenr, bytenr_orig; |
1345 | int ret; | |
8a4b83cc | 1346 | |
899f9307 DS |
1347 | lockdep_assert_held(&uuid_mutex); |
1348 | ||
6f60cbd3 DS |
1349 | /* |
1350 | * we would like to check all the supers, but that would make | |
1351 | * a btrfs mount succeed after a mkfs from a different FS. | |
1352 | * So, we need to add a special mount option to scan for | |
1353 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1354 | */ | |
d4d77629 | 1355 | flags |= FMODE_EXCL; |
6f60cbd3 DS |
1356 | |
1357 | bdev = blkdev_get_by_path(path, flags, holder); | |
b6ed73bc | 1358 | if (IS_ERR(bdev)) |
36350e95 | 1359 | return ERR_CAST(bdev); |
6f60cbd3 | 1360 | |
12659251 NA |
1361 | bytenr_orig = btrfs_sb_offset(0); |
1362 | ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); | |
4989d4a0 SK |
1363 | if (ret) { |
1364 | device = ERR_PTR(ret); | |
1365 | goto error_bdev_put; | |
1366 | } | |
12659251 NA |
1367 | |
1368 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); | |
b335eab8 NB |
1369 | if (IS_ERR(disk_super)) { |
1370 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1371 | goto error_bdev_put; |
05a5c55d | 1372 | } |
6f60cbd3 | 1373 | |
4306a974 | 1374 | device = device_list_add(path, disk_super, &new_device_added); |
4889bc05 AJ |
1375 | if (!IS_ERR(device) && new_device_added) |
1376 | btrfs_free_stale_devices(device->devt, device); | |
6f60cbd3 | 1377 | |
8f32380d | 1378 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1379 | |
1380 | error_bdev_put: | |
d4d77629 | 1381 | blkdev_put(bdev, flags); |
b6ed73bc | 1382 | |
36350e95 | 1383 | return device; |
8a4b83cc | 1384 | } |
0b86a832 | 1385 | |
1c11b63e JM |
1386 | /* |
1387 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1388 | * such is found, record the end of it in *start | |
1389 | */ | |
1c11b63e JM |
1390 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1391 | u64 len) | |
6df9a95e | 1392 | { |
1c11b63e | 1393 | u64 physical_start, physical_end; |
6df9a95e | 1394 | |
1c11b63e | 1395 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1396 | |
1c11b63e JM |
1397 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1398 | &physical_start, &physical_end, | |
1399 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1400 | |
1c11b63e JM |
1401 | if (in_range(physical_start, *start, len) || |
1402 | in_range(*start, physical_start, | |
1403 | physical_end - physical_start)) { | |
1404 | *start = physical_end + 1; | |
1405 | return true; | |
6df9a95e JB |
1406 | } |
1407 | } | |
1c11b63e | 1408 | return false; |
6df9a95e JB |
1409 | } |
1410 | ||
3b4ffa40 NA |
1411 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1412 | { | |
1413 | switch (device->fs_devices->chunk_alloc_policy) { | |
1414 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
37f85ec3 | 1415 | return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); |
1cd6121f NA |
1416 | case BTRFS_CHUNK_ALLOC_ZONED: |
1417 | /* | |
1418 | * We don't care about the starting region like regular | |
1419 | * allocator, because we anyway use/reserve the first two zones | |
1420 | * for superblock logging. | |
1421 | */ | |
1422 | return ALIGN(start, device->zone_info->zone_size); | |
3b4ffa40 NA |
1423 | default: |
1424 | BUG(); | |
1425 | } | |
1426 | } | |
1427 | ||
1cd6121f NA |
1428 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
1429 | u64 *hole_start, u64 *hole_size, | |
1430 | u64 num_bytes) | |
1431 | { | |
1432 | u64 zone_size = device->zone_info->zone_size; | |
1433 | u64 pos; | |
1434 | int ret; | |
1435 | bool changed = false; | |
1436 | ||
1437 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); | |
1438 | ||
1439 | while (*hole_size > 0) { | |
1440 | pos = btrfs_find_allocatable_zones(device, *hole_start, | |
1441 | *hole_start + *hole_size, | |
1442 | num_bytes); | |
1443 | if (pos != *hole_start) { | |
1444 | *hole_size = *hole_start + *hole_size - pos; | |
1445 | *hole_start = pos; | |
1446 | changed = true; | |
1447 | if (*hole_size < num_bytes) | |
1448 | break; | |
1449 | } | |
1450 | ||
1451 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); | |
1452 | ||
1453 | /* Range is ensured to be empty */ | |
1454 | if (!ret) | |
1455 | return changed; | |
1456 | ||
1457 | /* Given hole range was invalid (outside of device) */ | |
1458 | if (ret == -ERANGE) { | |
1459 | *hole_start += *hole_size; | |
d6f67afb | 1460 | *hole_size = 0; |
7000babd | 1461 | return true; |
1cd6121f NA |
1462 | } |
1463 | ||
1464 | *hole_start += zone_size; | |
1465 | *hole_size -= zone_size; | |
1466 | changed = true; | |
1467 | } | |
1468 | ||
1469 | return changed; | |
1470 | } | |
1471 | ||
43dd529a DS |
1472 | /* |
1473 | * Check if specified hole is suitable for allocation. | |
1474 | * | |
3b4ffa40 NA |
1475 | * @device: the device which we have the hole |
1476 | * @hole_start: starting position of the hole | |
1477 | * @hole_size: the size of the hole | |
1478 | * @num_bytes: the size of the free space that we need | |
1479 | * | |
1cd6121f | 1480 | * This function may modify @hole_start and @hole_size to reflect the suitable |
3b4ffa40 NA |
1481 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
1482 | */ | |
1483 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1484 | u64 *hole_size, u64 num_bytes) | |
1485 | { | |
1486 | bool changed = false; | |
1487 | u64 hole_end = *hole_start + *hole_size; | |
1488 | ||
1cd6121f NA |
1489 | for (;;) { |
1490 | /* | |
1491 | * Check before we set max_hole_start, otherwise we could end up | |
1492 | * sending back this offset anyway. | |
1493 | */ | |
1494 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1495 | if (hole_end >= *hole_start) | |
1496 | *hole_size = hole_end - *hole_start; | |
1497 | else | |
1498 | *hole_size = 0; | |
1499 | changed = true; | |
1500 | } | |
1501 | ||
1502 | switch (device->fs_devices->chunk_alloc_policy) { | |
1503 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1504 | /* No extra check */ | |
1505 | break; | |
1506 | case BTRFS_CHUNK_ALLOC_ZONED: | |
1507 | if (dev_extent_hole_check_zoned(device, hole_start, | |
1508 | hole_size, num_bytes)) { | |
1509 | changed = true; | |
1510 | /* | |
1511 | * The changed hole can contain pending extent. | |
1512 | * Loop again to check that. | |
1513 | */ | |
1514 | continue; | |
1515 | } | |
1516 | break; | |
1517 | default: | |
1518 | BUG(); | |
1519 | } | |
3b4ffa40 | 1520 | |
3b4ffa40 | 1521 | break; |
3b4ffa40 NA |
1522 | } |
1523 | ||
1524 | return changed; | |
1525 | } | |
6df9a95e | 1526 | |
0b86a832 | 1527 | /* |
43dd529a DS |
1528 | * Find free space in the specified device. |
1529 | * | |
499f377f JM |
1530 | * @device: the device which we search the free space in |
1531 | * @num_bytes: the size of the free space that we need | |
1532 | * @search_start: the position from which to begin the search | |
1533 | * @start: store the start of the free space. | |
1534 | * @len: the size of the free space. that we find, or the size | |
1535 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1536 | * |
43dd529a DS |
1537 | * This does a pretty simple search, the expectation is that it is called very |
1538 | * infrequently and that a given device has a small number of extents. | |
7bfc837d MX |
1539 | * |
1540 | * @start is used to store the start of the free space if we find. But if we | |
1541 | * don't find suitable free space, it will be used to store the start position | |
1542 | * of the max free space. | |
1543 | * | |
1544 | * @len is used to store the size of the free space that we find. | |
1545 | * But if we don't find suitable free space, it is used to store the size of | |
1546 | * the max free space. | |
135da976 QW |
1547 | * |
1548 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1549 | * check to ensure dev extents are not double allocated. | |
1550 | * This makes the function safe to allocate dev extents but may not report | |
1551 | * correct usable device space, as device extent freed in current transaction | |
1a9fd417 | 1552 | * is not reported as available. |
0b86a832 | 1553 | */ |
9e3246a5 QW |
1554 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1555 | u64 num_bytes, u64 search_start, u64 *start, | |
1556 | u64 *len) | |
0b86a832 | 1557 | { |
0b246afa JM |
1558 | struct btrfs_fs_info *fs_info = device->fs_info; |
1559 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1560 | struct btrfs_key key; |
7bfc837d | 1561 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1562 | struct btrfs_path *path; |
7bfc837d MX |
1563 | u64 hole_size; |
1564 | u64 max_hole_start; | |
1565 | u64 max_hole_size; | |
1566 | u64 extent_end; | |
0b86a832 CM |
1567 | u64 search_end = device->total_bytes; |
1568 | int ret; | |
7bfc837d | 1569 | int slot; |
0b86a832 | 1570 | struct extent_buffer *l; |
8cdc7c5b | 1571 | |
3b4ffa40 | 1572 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1573 | |
1cd6121f NA |
1574 | WARN_ON(device->zone_info && |
1575 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); | |
1576 | ||
6df9a95e JB |
1577 | path = btrfs_alloc_path(); |
1578 | if (!path) | |
1579 | return -ENOMEM; | |
f2ab7618 | 1580 | |
7bfc837d MX |
1581 | max_hole_start = search_start; |
1582 | max_hole_size = 0; | |
1583 | ||
f2ab7618 | 1584 | again: |
401e29c1 AJ |
1585 | if (search_start >= search_end || |
1586 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1587 | ret = -ENOSPC; |
6df9a95e | 1588 | goto out; |
7bfc837d MX |
1589 | } |
1590 | ||
e4058b54 | 1591 | path->reada = READA_FORWARD; |
6df9a95e JB |
1592 | path->search_commit_root = 1; |
1593 | path->skip_locking = 1; | |
7bfc837d | 1594 | |
0b86a832 CM |
1595 | key.objectid = device->devid; |
1596 | key.offset = search_start; | |
1597 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1598 | |
0ff40a91 | 1599 | ret = btrfs_search_backwards(root, &key, path); |
0b86a832 | 1600 | if (ret < 0) |
7bfc837d | 1601 | goto out; |
7bfc837d | 1602 | |
3c538de0 | 1603 | while (search_start < search_end) { |
0b86a832 CM |
1604 | l = path->nodes[0]; |
1605 | slot = path->slots[0]; | |
1606 | if (slot >= btrfs_header_nritems(l)) { | |
1607 | ret = btrfs_next_leaf(root, path); | |
1608 | if (ret == 0) | |
1609 | continue; | |
1610 | if (ret < 0) | |
7bfc837d MX |
1611 | goto out; |
1612 | ||
1613 | break; | |
0b86a832 CM |
1614 | } |
1615 | btrfs_item_key_to_cpu(l, &key, slot); | |
1616 | ||
1617 | if (key.objectid < device->devid) | |
1618 | goto next; | |
1619 | ||
1620 | if (key.objectid > device->devid) | |
7bfc837d | 1621 | break; |
0b86a832 | 1622 | |
962a298f | 1623 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1624 | goto next; |
9779b72f | 1625 | |
3c538de0 JB |
1626 | if (key.offset > search_end) |
1627 | break; | |
1628 | ||
7bfc837d MX |
1629 | if (key.offset > search_start) { |
1630 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1631 | dev_extent_hole_check(device, &search_start, &hole_size, |
1632 | num_bytes); | |
6df9a95e | 1633 | |
7bfc837d MX |
1634 | if (hole_size > max_hole_size) { |
1635 | max_hole_start = search_start; | |
1636 | max_hole_size = hole_size; | |
1637 | } | |
9779b72f | 1638 | |
7bfc837d MX |
1639 | /* |
1640 | * If this free space is greater than which we need, | |
1641 | * it must be the max free space that we have found | |
1642 | * until now, so max_hole_start must point to the start | |
1643 | * of this free space and the length of this free space | |
1644 | * is stored in max_hole_size. Thus, we return | |
1645 | * max_hole_start and max_hole_size and go back to the | |
1646 | * caller. | |
1647 | */ | |
1648 | if (hole_size >= num_bytes) { | |
1649 | ret = 0; | |
1650 | goto out; | |
0b86a832 CM |
1651 | } |
1652 | } | |
0b86a832 | 1653 | |
0b86a832 | 1654 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1655 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1656 | dev_extent); | |
1657 | if (extent_end > search_start) | |
1658 | search_start = extent_end; | |
0b86a832 CM |
1659 | next: |
1660 | path->slots[0]++; | |
1661 | cond_resched(); | |
1662 | } | |
0b86a832 | 1663 | |
38c01b96 | 1664 | /* |
1665 | * At this point, search_start should be the end of | |
1666 | * allocated dev extents, and when shrinking the device, | |
1667 | * search_end may be smaller than search_start. | |
1668 | */ | |
f2ab7618 | 1669 | if (search_end > search_start) { |
38c01b96 | 1670 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1671 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1672 | num_bytes)) { | |
f2ab7618 ZL |
1673 | btrfs_release_path(path); |
1674 | goto again; | |
1675 | } | |
0b86a832 | 1676 | |
f2ab7618 ZL |
1677 | if (hole_size > max_hole_size) { |
1678 | max_hole_start = search_start; | |
1679 | max_hole_size = hole_size; | |
1680 | } | |
6df9a95e JB |
1681 | } |
1682 | ||
7bfc837d | 1683 | /* See above. */ |
f2ab7618 | 1684 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1685 | ret = -ENOSPC; |
1686 | else | |
1687 | ret = 0; | |
1688 | ||
3c538de0 | 1689 | ASSERT(max_hole_start + max_hole_size <= search_end); |
7bfc837d | 1690 | out: |
2b82032c | 1691 | btrfs_free_path(path); |
7bfc837d | 1692 | *start = max_hole_start; |
b2117a39 | 1693 | if (len) |
7bfc837d | 1694 | *len = max_hole_size; |
0b86a832 CM |
1695 | return ret; |
1696 | } | |
1697 | ||
60dfdf25 | 1698 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1699 | u64 *start, u64 *len) |
1700 | { | |
499f377f | 1701 | /* FIXME use last free of some kind */ |
60dfdf25 | 1702 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1703 | } |
1704 | ||
b2950863 | 1705 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1706 | struct btrfs_device *device, |
2196d6e8 | 1707 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1708 | { |
0b246afa JM |
1709 | struct btrfs_fs_info *fs_info = device->fs_info; |
1710 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1711 | int ret; |
1712 | struct btrfs_path *path; | |
8f18cf13 | 1713 | struct btrfs_key key; |
a061fc8d CM |
1714 | struct btrfs_key found_key; |
1715 | struct extent_buffer *leaf = NULL; | |
1716 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1717 | |
1718 | path = btrfs_alloc_path(); | |
1719 | if (!path) | |
1720 | return -ENOMEM; | |
1721 | ||
1722 | key.objectid = device->devid; | |
1723 | key.offset = start; | |
1724 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1725 | again: |
8f18cf13 | 1726 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1727 | if (ret > 0) { |
1728 | ret = btrfs_previous_item(root, path, key.objectid, | |
1729 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1730 | if (ret) |
1731 | goto out; | |
a061fc8d CM |
1732 | leaf = path->nodes[0]; |
1733 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1734 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1735 | struct btrfs_dev_extent); | |
1736 | BUG_ON(found_key.offset > start || found_key.offset + | |
1737 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1738 | key = found_key; |
1739 | btrfs_release_path(path); | |
1740 | goto again; | |
a061fc8d CM |
1741 | } else if (ret == 0) { |
1742 | leaf = path->nodes[0]; | |
1743 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1744 | struct btrfs_dev_extent); | |
79787eaa | 1745 | } else { |
79787eaa | 1746 | goto out; |
a061fc8d | 1747 | } |
8f18cf13 | 1748 | |
2196d6e8 MX |
1749 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1750 | ||
8f18cf13 | 1751 | ret = btrfs_del_item(trans, root, path); |
79bd3712 | 1752 | if (ret == 0) |
3204d33c | 1753 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
b0b802d7 | 1754 | out: |
8f18cf13 CM |
1755 | btrfs_free_path(path); |
1756 | return ret; | |
1757 | } | |
1758 | ||
6df9a95e | 1759 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1760 | { |
6df9a95e JB |
1761 | struct extent_map_tree *em_tree; |
1762 | struct extent_map *em; | |
1763 | struct rb_node *n; | |
1764 | u64 ret = 0; | |
0b86a832 | 1765 | |
c8bf1b67 | 1766 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1767 | read_lock(&em_tree->lock); |
07e1ce09 | 1768 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1769 | if (n) { |
1770 | em = rb_entry(n, struct extent_map, rb_node); | |
1771 | ret = em->start + em->len; | |
0b86a832 | 1772 | } |
6df9a95e JB |
1773 | read_unlock(&em_tree->lock); |
1774 | ||
0b86a832 CM |
1775 | return ret; |
1776 | } | |
1777 | ||
53f10659 ID |
1778 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1779 | u64 *devid_ret) | |
0b86a832 CM |
1780 | { |
1781 | int ret; | |
1782 | struct btrfs_key key; | |
1783 | struct btrfs_key found_key; | |
2b82032c YZ |
1784 | struct btrfs_path *path; |
1785 | ||
2b82032c YZ |
1786 | path = btrfs_alloc_path(); |
1787 | if (!path) | |
1788 | return -ENOMEM; | |
0b86a832 CM |
1789 | |
1790 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1791 | key.type = BTRFS_DEV_ITEM_KEY; | |
1792 | key.offset = (u64)-1; | |
1793 | ||
53f10659 | 1794 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1795 | if (ret < 0) |
1796 | goto error; | |
1797 | ||
a06dee4d AJ |
1798 | if (ret == 0) { |
1799 | /* Corruption */ | |
1800 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1801 | ret = -EUCLEAN; | |
1802 | goto error; | |
1803 | } | |
0b86a832 | 1804 | |
53f10659 ID |
1805 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1806 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1807 | BTRFS_DEV_ITEM_KEY); |
1808 | if (ret) { | |
53f10659 | 1809 | *devid_ret = 1; |
0b86a832 CM |
1810 | } else { |
1811 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1812 | path->slots[0]); | |
53f10659 | 1813 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1814 | } |
1815 | ret = 0; | |
1816 | error: | |
2b82032c | 1817 | btrfs_free_path(path); |
0b86a832 CM |
1818 | return ret; |
1819 | } | |
1820 | ||
1821 | /* | |
1822 | * the device information is stored in the chunk root | |
1823 | * the btrfs_device struct should be fully filled in | |
1824 | */ | |
c74a0b02 | 1825 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1826 | struct btrfs_device *device) |
0b86a832 CM |
1827 | { |
1828 | int ret; | |
1829 | struct btrfs_path *path; | |
1830 | struct btrfs_dev_item *dev_item; | |
1831 | struct extent_buffer *leaf; | |
1832 | struct btrfs_key key; | |
1833 | unsigned long ptr; | |
0b86a832 | 1834 | |
0b86a832 CM |
1835 | path = btrfs_alloc_path(); |
1836 | if (!path) | |
1837 | return -ENOMEM; | |
1838 | ||
0b86a832 CM |
1839 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1840 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1841 | key.offset = device->devid; |
0b86a832 | 1842 | |
2bb2e00e | 1843 | btrfs_reserve_chunk_metadata(trans, true); |
8e87e856 NB |
1844 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1845 | &key, sizeof(*dev_item)); | |
2bb2e00e | 1846 | btrfs_trans_release_chunk_metadata(trans); |
0b86a832 CM |
1847 | if (ret) |
1848 | goto out; | |
1849 | ||
1850 | leaf = path->nodes[0]; | |
1851 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1852 | ||
1853 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1854 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1855 | btrfs_set_device_type(leaf, dev_item, device->type); |
1856 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1857 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1858 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1859 | btrfs_set_device_total_bytes(leaf, dev_item, |
1860 | btrfs_device_get_disk_total_bytes(device)); | |
1861 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1862 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1863 | btrfs_set_device_group(leaf, dev_item, 0); |
1864 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1865 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1866 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1867 | |
410ba3a2 | 1868 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1869 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1870 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1871 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1872 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1873 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1874 | |
2b82032c | 1875 | ret = 0; |
0b86a832 CM |
1876 | out: |
1877 | btrfs_free_path(path); | |
1878 | return ret; | |
1879 | } | |
8f18cf13 | 1880 | |
5a1972bd QW |
1881 | /* |
1882 | * Function to update ctime/mtime for a given device path. | |
1883 | * Mainly used for ctime/mtime based probe like libblkid. | |
54fde91f JB |
1884 | * |
1885 | * We don't care about errors here, this is just to be kind to userspace. | |
5a1972bd | 1886 | */ |
54fde91f | 1887 | static void update_dev_time(const char *device_path) |
5a1972bd | 1888 | { |
54fde91f | 1889 | struct path path; |
8f96a5bf | 1890 | struct timespec64 now; |
54fde91f | 1891 | int ret; |
5a1972bd | 1892 | |
54fde91f JB |
1893 | ret = kern_path(device_path, LOOKUP_FOLLOW, &path); |
1894 | if (ret) | |
5a1972bd | 1895 | return; |
8f96a5bf | 1896 | |
54fde91f JB |
1897 | now = current_time(d_inode(path.dentry)); |
1898 | inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); | |
1899 | path_put(&path); | |
5a1972bd QW |
1900 | } |
1901 | ||
bbac5869 QW |
1902 | static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, |
1903 | struct btrfs_device *device) | |
a061fc8d | 1904 | { |
f331a952 | 1905 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1906 | int ret; |
1907 | struct btrfs_path *path; | |
a061fc8d | 1908 | struct btrfs_key key; |
a061fc8d | 1909 | |
a061fc8d CM |
1910 | path = btrfs_alloc_path(); |
1911 | if (!path) | |
1912 | return -ENOMEM; | |
1913 | ||
a061fc8d CM |
1914 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1915 | key.type = BTRFS_DEV_ITEM_KEY; | |
1916 | key.offset = device->devid; | |
1917 | ||
2bb2e00e | 1918 | btrfs_reserve_chunk_metadata(trans, false); |
a061fc8d | 1919 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
2bb2e00e | 1920 | btrfs_trans_release_chunk_metadata(trans); |
5e9f2ad5 NB |
1921 | if (ret) { |
1922 | if (ret > 0) | |
1923 | ret = -ENOENT; | |
a061fc8d CM |
1924 | goto out; |
1925 | } | |
1926 | ||
1927 | ret = btrfs_del_item(trans, root, path); | |
a061fc8d CM |
1928 | out: |
1929 | btrfs_free_path(path); | |
a061fc8d CM |
1930 | return ret; |
1931 | } | |
1932 | ||
3cc31a0d DS |
1933 | /* |
1934 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1935 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1936 | * replace. | |
1937 | */ | |
1938 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1939 | u64 num_devices) | |
a061fc8d | 1940 | { |
a061fc8d | 1941 | u64 all_avail; |
de98ced9 | 1942 | unsigned seq; |
418775a2 | 1943 | int i; |
a061fc8d | 1944 | |
de98ced9 | 1945 | do { |
bd45ffbc | 1946 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1947 | |
bd45ffbc AJ |
1948 | all_avail = fs_info->avail_data_alloc_bits | |
1949 | fs_info->avail_system_alloc_bits | | |
1950 | fs_info->avail_metadata_alloc_bits; | |
1951 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1952 | |
418775a2 | 1953 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1954 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1955 | continue; |
a061fc8d | 1956 | |
efc222f8 AJ |
1957 | if (num_devices < btrfs_raid_array[i].devs_min) |
1958 | return btrfs_raid_array[i].mindev_error; | |
53b381b3 DW |
1959 | } |
1960 | ||
bd45ffbc | 1961 | return 0; |
f1fa7f26 AJ |
1962 | } |
1963 | ||
c9162bdf OS |
1964 | static struct btrfs_device * btrfs_find_next_active_device( |
1965 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1966 | { |
2b82032c | 1967 | struct btrfs_device *next_device; |
88acff64 AJ |
1968 | |
1969 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1970 | if (next_device != device && | |
e6e674bd AJ |
1971 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1972 | && next_device->bdev) | |
88acff64 AJ |
1973 | return next_device; |
1974 | } | |
1975 | ||
1976 | return NULL; | |
1977 | } | |
1978 | ||
1979 | /* | |
d24fa5c1 | 1980 | * Helper function to check if the given device is part of s_bdev / latest_dev |
88acff64 AJ |
1981 | * and replace it with the provided or the next active device, in the context |
1982 | * where this function called, there should be always be another device (or | |
1983 | * this_dev) which is active. | |
1984 | */ | |
b105e927 | 1985 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
e493e8f9 | 1986 | struct btrfs_device *next_device) |
88acff64 | 1987 | { |
d6507cf1 | 1988 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 | 1989 | |
e493e8f9 | 1990 | if (!next_device) |
88acff64 | 1991 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
e493e8f9 | 1992 | device); |
88acff64 AJ |
1993 | ASSERT(next_device); |
1994 | ||
1995 | if (fs_info->sb->s_bdev && | |
1996 | (fs_info->sb->s_bdev == device->bdev)) | |
1997 | fs_info->sb->s_bdev = next_device->bdev; | |
1998 | ||
d24fa5c1 AJ |
1999 | if (fs_info->fs_devices->latest_dev->bdev == device->bdev) |
2000 | fs_info->fs_devices->latest_dev = next_device; | |
88acff64 AJ |
2001 | } |
2002 | ||
1da73967 AJ |
2003 | /* |
2004 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2005 | * currently replaced. | |
2006 | */ | |
2007 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2008 | { | |
2009 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2010 | ||
cb5583dd | 2011 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2012 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2013 | ASSERT(num_devices > 1); | |
2014 | num_devices--; | |
2015 | } | |
cb5583dd | 2016 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2017 | |
2018 | return num_devices; | |
2019 | } | |
2020 | ||
0e0078f7 CH |
2021 | static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, |
2022 | struct block_device *bdev, int copy_num) | |
2023 | { | |
2024 | struct btrfs_super_block *disk_super; | |
26ecf243 CH |
2025 | const size_t len = sizeof(disk_super->magic); |
2026 | const u64 bytenr = btrfs_sb_offset(copy_num); | |
0e0078f7 CH |
2027 | int ret; |
2028 | ||
26ecf243 | 2029 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); |
0e0078f7 CH |
2030 | if (IS_ERR(disk_super)) |
2031 | return; | |
2032 | ||
26ecf243 CH |
2033 | memset(&disk_super->magic, 0, len); |
2034 | folio_mark_dirty(virt_to_folio(disk_super)); | |
2035 | btrfs_release_disk_super(disk_super); | |
2036 | ||
2037 | ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); | |
0e0078f7 CH |
2038 | if (ret) |
2039 | btrfs_warn(fs_info, "error clearing superblock number %d (%d)", | |
2040 | copy_num, ret); | |
0e0078f7 CH |
2041 | } |
2042 | ||
313b0858 JB |
2043 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2044 | struct block_device *bdev, | |
2045 | const char *device_path) | |
6fbceb9f | 2046 | { |
6fbceb9f JT |
2047 | int copy_num; |
2048 | ||
2049 | if (!bdev) | |
2050 | return; | |
2051 | ||
2052 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
0e0078f7 | 2053 | if (bdev_is_zoned(bdev)) |
12659251 | 2054 | btrfs_reset_sb_log_zones(bdev, copy_num); |
0e0078f7 CH |
2055 | else |
2056 | btrfs_scratch_superblock(fs_info, bdev, copy_num); | |
6fbceb9f JT |
2057 | } |
2058 | ||
2059 | /* Notify udev that device has changed */ | |
2060 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2061 | ||
2062 | /* Update ctime/mtime for device path for libblkid */ | |
54fde91f | 2063 | update_dev_time(device_path); |
6fbceb9f JT |
2064 | } |
2065 | ||
1a15eb72 JB |
2066 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, |
2067 | struct btrfs_dev_lookup_args *args, | |
2068 | struct block_device **bdev, fmode_t *mode) | |
f1fa7f26 | 2069 | { |
bbac5869 | 2070 | struct btrfs_trans_handle *trans; |
f1fa7f26 | 2071 | struct btrfs_device *device; |
1f78160c | 2072 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2073 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2074 | u64 num_devices; |
a061fc8d CM |
2075 | int ret = 0; |
2076 | ||
914a519b JB |
2077 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2078 | btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); | |
2079 | return -EINVAL; | |
2080 | } | |
2081 | ||
8ef9dc0f JB |
2082 | /* |
2083 | * The device list in fs_devices is accessed without locks (neither | |
2084 | * uuid_mutex nor device_list_mutex) as it won't change on a mounted | |
2085 | * filesystem and another device rm cannot run. | |
2086 | */ | |
1da73967 | 2087 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2088 | |
0b246afa | 2089 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2090 | if (ret) |
bbac5869 | 2091 | return ret; |
a061fc8d | 2092 | |
1a15eb72 JB |
2093 | device = btrfs_find_device(fs_info->fs_devices, args); |
2094 | if (!device) { | |
2095 | if (args->missing) | |
a27a94c2 NB |
2096 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
2097 | else | |
1a15eb72 | 2098 | ret = -ENOENT; |
bbac5869 | 2099 | return ret; |
a27a94c2 | 2100 | } |
dfe25020 | 2101 | |
eede2bf3 OS |
2102 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2103 | btrfs_warn_in_rcu(fs_info, | |
2104 | "cannot remove device %s (devid %llu) due to active swapfile", | |
cb3e217b | 2105 | btrfs_dev_name(device), device->devid); |
bbac5869 | 2106 | return -ETXTBSY; |
eede2bf3 OS |
2107 | } |
2108 | ||
bbac5869 QW |
2109 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
2110 | return BTRFS_ERROR_DEV_TGT_REPLACE; | |
63a212ab | 2111 | |
ebbede42 | 2112 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
bbac5869 QW |
2113 | fs_info->fs_devices->rw_devices == 1) |
2114 | return BTRFS_ERROR_DEV_ONLY_WRITABLE; | |
2b82032c | 2115 | |
ebbede42 | 2116 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2117 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2118 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2119 | device->fs_devices->rw_devices--; |
34441361 | 2120 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2121 | } |
a061fc8d CM |
2122 | |
2123 | ret = btrfs_shrink_device(device, 0); | |
2124 | if (ret) | |
9b3517e9 | 2125 | goto error_undo; |
a061fc8d | 2126 | |
bbac5869 QW |
2127 | trans = btrfs_start_transaction(fs_info->chunk_root, 0); |
2128 | if (IS_ERR(trans)) { | |
2129 | ret = PTR_ERR(trans); | |
9b3517e9 | 2130 | goto error_undo; |
bbac5869 QW |
2131 | } |
2132 | ||
2133 | ret = btrfs_rm_dev_item(trans, device); | |
2134 | if (ret) { | |
2135 | /* Any error in dev item removal is critical */ | |
2136 | btrfs_crit(fs_info, | |
2137 | "failed to remove device item for devid %llu: %d", | |
2138 | device->devid, ret); | |
2139 | btrfs_abort_transaction(trans, ret); | |
2140 | btrfs_end_transaction(trans); | |
2141 | return ret; | |
2142 | } | |
a061fc8d | 2143 | |
e12c9621 | 2144 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2145 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2146 | |
2147 | /* | |
2148 | * the device list mutex makes sure that we don't change | |
2149 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2150 | * the device supers. Whoever is writing all supers, should |
2151 | * lock the device list mutex before getting the number of | |
2152 | * devices in the super block (super_copy). Conversely, | |
2153 | * whoever updates the number of devices in the super block | |
2154 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2155 | */ |
1f78160c | 2156 | |
41a52a0f AJ |
2157 | /* |
2158 | * In normal cases the cur_devices == fs_devices. But in case | |
2159 | * of deleting a seed device, the cur_devices should point to | |
9675ea8c | 2160 | * its own fs_devices listed under the fs_devices->seed_list. |
41a52a0f | 2161 | */ |
1f78160c | 2162 | cur_devices = device->fs_devices; |
b5185197 | 2163 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2164 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2165 | |
41a52a0f AJ |
2166 | cur_devices->num_devices--; |
2167 | cur_devices->total_devices--; | |
b4993e64 AJ |
2168 | /* Update total_devices of the parent fs_devices if it's seed */ |
2169 | if (cur_devices != fs_devices) | |
2170 | fs_devices->total_devices--; | |
2b82032c | 2171 | |
e6e674bd | 2172 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2173 | cur_devices->missing_devices--; |
cd02dca5 | 2174 | |
d6507cf1 | 2175 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2176 | |
0bfaa9c5 | 2177 | if (device->bdev) { |
41a52a0f | 2178 | cur_devices->open_devices--; |
0bfaa9c5 | 2179 | /* remove sysfs entry */ |
53f8a74c | 2180 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2181 | } |
99994cde | 2182 | |
0b246afa JM |
2183 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2184 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2185 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2186 | |
cea67ab9 | 2187 | /* |
3fa421de JB |
2188 | * At this point, the device is zero sized and detached from the |
2189 | * devices list. All that's left is to zero out the old supers and | |
2190 | * free the device. | |
2191 | * | |
2192 | * We cannot call btrfs_close_bdev() here because we're holding the sb | |
2193 | * write lock, and blkdev_put() will pull in the ->open_mutex on the | |
2194 | * block device and it's dependencies. Instead just flush the device | |
2195 | * and let the caller do the final blkdev_put. | |
cea67ab9 | 2196 | */ |
3fa421de | 2197 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
8f32380d JT |
2198 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2199 | device->name->str); | |
3fa421de JB |
2200 | if (device->bdev) { |
2201 | sync_blockdev(device->bdev); | |
2202 | invalidate_bdev(device->bdev); | |
2203 | } | |
2204 | } | |
cea67ab9 | 2205 | |
3fa421de JB |
2206 | *bdev = device->bdev; |
2207 | *mode = device->mode; | |
8e75fd89 NB |
2208 | synchronize_rcu(); |
2209 | btrfs_free_device(device); | |
cea67ab9 | 2210 | |
8b41393f JB |
2211 | /* |
2212 | * This can happen if cur_devices is the private seed devices list. We | |
2213 | * cannot call close_fs_devices() here because it expects the uuid_mutex | |
2214 | * to be held, but in fact we don't need that for the private | |
2215 | * seed_devices, we can simply decrement cur_devices->opened and then | |
2216 | * remove it from our list and free the fs_devices. | |
2217 | */ | |
8e906945 | 2218 | if (cur_devices->num_devices == 0) { |
944d3f9f | 2219 | list_del_init(&cur_devices->seed_list); |
8b41393f JB |
2220 | ASSERT(cur_devices->opened == 1); |
2221 | cur_devices->opened--; | |
1f78160c | 2222 | free_fs_devices(cur_devices); |
2b82032c YZ |
2223 | } |
2224 | ||
bbac5869 QW |
2225 | ret = btrfs_commit_transaction(trans); |
2226 | ||
a061fc8d | 2227 | return ret; |
24fc572f | 2228 | |
9b3517e9 | 2229 | error_undo: |
ebbede42 | 2230 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2231 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2232 | list_add(&device->dev_alloc_list, |
b5185197 | 2233 | &fs_devices->alloc_list); |
c3929c36 | 2234 | device->fs_devices->rw_devices++; |
34441361 | 2235 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2236 | } |
bbac5869 | 2237 | return ret; |
a061fc8d CM |
2238 | } |
2239 | ||
68a9db5f | 2240 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2241 | { |
d51908ce AJ |
2242 | struct btrfs_fs_devices *fs_devices; |
2243 | ||
68a9db5f | 2244 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2245 | |
25e8e911 AJ |
2246 | /* |
2247 | * in case of fs with no seed, srcdev->fs_devices will point | |
2248 | * to fs_devices of fs_info. However when the dev being replaced is | |
2249 | * a seed dev it will point to the seed's local fs_devices. In short | |
2250 | * srcdev will have its correct fs_devices in both the cases. | |
2251 | */ | |
2252 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2253 | |
e93c89c1 | 2254 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2255 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2256 | fs_devices->num_devices--; |
e6e674bd | 2257 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2258 | fs_devices->missing_devices--; |
e93c89c1 | 2259 | |
ebbede42 | 2260 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2261 | fs_devices->rw_devices--; |
1357272f | 2262 | |
82372bc8 | 2263 | if (srcdev->bdev) |
d51908ce | 2264 | fs_devices->open_devices--; |
084b6e7c QW |
2265 | } |
2266 | ||
65237ee3 | 2267 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2268 | { |
2269 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2270 | |
a466c85e JB |
2271 | mutex_lock(&uuid_mutex); |
2272 | ||
14238819 | 2273 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2274 | synchronize_rcu(); |
2275 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2276 | |
94d5f0c2 AJ |
2277 | /* if this is no devs we rather delete the fs_devices */ |
2278 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2279 | /* |
2280 | * On a mounted FS, num_devices can't be zero unless it's a | |
2281 | * seed. In case of a seed device being replaced, the replace | |
2282 | * target added to the sprout FS, so there will be no more | |
2283 | * device left under the seed FS. | |
2284 | */ | |
2285 | ASSERT(fs_devices->seeding); | |
2286 | ||
944d3f9f | 2287 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2288 | close_fs_devices(fs_devices); |
8bef8401 | 2289 | free_fs_devices(fs_devices); |
94d5f0c2 | 2290 | } |
a466c85e | 2291 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2292 | } |
2293 | ||
4f5ad7bd | 2294 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2295 | { |
4f5ad7bd | 2296 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2297 | |
d9a071f0 | 2298 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2299 | |
53f8a74c | 2300 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2301 | |
779bf3fe | 2302 | if (tgtdev->bdev) |
d9a071f0 | 2303 | fs_devices->open_devices--; |
779bf3fe | 2304 | |
d9a071f0 | 2305 | fs_devices->num_devices--; |
e93c89c1 | 2306 | |
d6507cf1 | 2307 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2308 | |
e93c89c1 | 2309 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2310 | |
d9a071f0 | 2311 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe | 2312 | |
8f32380d JT |
2313 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2314 | tgtdev->name->str); | |
14238819 AJ |
2315 | |
2316 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2317 | synchronize_rcu(); |
2318 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2319 | } |
2320 | ||
43dd529a DS |
2321 | /* |
2322 | * Populate args from device at path. | |
faa775c4 JB |
2323 | * |
2324 | * @fs_info: the filesystem | |
2325 | * @args: the args to populate | |
2326 | * @path: the path to the device | |
2327 | * | |
2328 | * This will read the super block of the device at @path and populate @args with | |
2329 | * the devid, fsid, and uuid. This is meant to be used for ioctls that need to | |
2330 | * lookup a device to operate on, but need to do it before we take any locks. | |
2331 | * This properly handles the special case of "missing" that a user may pass in, | |
2332 | * and does some basic sanity checks. The caller must make sure that @path is | |
2333 | * properly NUL terminated before calling in, and must call | |
2334 | * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and | |
2335 | * uuid buffers. | |
2336 | * | |
2337 | * Return: 0 for success, -errno for failure | |
2338 | */ | |
2339 | int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, | |
2340 | struct btrfs_dev_lookup_args *args, | |
2341 | const char *path) | |
7ba15b7d | 2342 | { |
7ba15b7d | 2343 | struct btrfs_super_block *disk_super; |
7ba15b7d | 2344 | struct block_device *bdev; |
faa775c4 | 2345 | int ret; |
7ba15b7d | 2346 | |
faa775c4 JB |
2347 | if (!path || !path[0]) |
2348 | return -EINVAL; | |
2349 | if (!strcmp(path, "missing")) { | |
2350 | args->missing = true; | |
2351 | return 0; | |
2352 | } | |
8f32380d | 2353 | |
faa775c4 JB |
2354 | args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); |
2355 | args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); | |
2356 | if (!args->uuid || !args->fsid) { | |
2357 | btrfs_put_dev_args_from_path(args); | |
2358 | return -ENOMEM; | |
2359 | } | |
8f32380d | 2360 | |
faa775c4 JB |
2361 | ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0, |
2362 | &bdev, &disk_super); | |
9ea0106a ZF |
2363 | if (ret) { |
2364 | btrfs_put_dev_args_from_path(args); | |
faa775c4 | 2365 | return ret; |
9ea0106a ZF |
2366 | } |
2367 | ||
faa775c4 JB |
2368 | args->devid = btrfs_stack_device_id(&disk_super->dev_item); |
2369 | memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); | |
7239ff4b | 2370 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
faa775c4 | 2371 | memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); |
7239ff4b | 2372 | else |
faa775c4 | 2373 | memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); |
8f32380d | 2374 | btrfs_release_disk_super(disk_super); |
7ba15b7d | 2375 | blkdev_put(bdev, FMODE_READ); |
faa775c4 | 2376 | return 0; |
7ba15b7d SB |
2377 | } |
2378 | ||
5c5c0df0 | 2379 | /* |
faa775c4 JB |
2380 | * Only use this jointly with btrfs_get_dev_args_from_path() because we will |
2381 | * allocate our ->uuid and ->fsid pointers, everybody else uses local variables | |
2382 | * that don't need to be freed. | |
5c5c0df0 | 2383 | */ |
faa775c4 JB |
2384 | void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) |
2385 | { | |
2386 | kfree(args->uuid); | |
2387 | kfree(args->fsid); | |
2388 | args->uuid = NULL; | |
2389 | args->fsid = NULL; | |
2390 | } | |
2391 | ||
a27a94c2 | 2392 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2393 | struct btrfs_fs_info *fs_info, u64 devid, |
2394 | const char *device_path) | |
24e0474b | 2395 | { |
562d7b15 | 2396 | BTRFS_DEV_LOOKUP_ARGS(args); |
a27a94c2 | 2397 | struct btrfs_device *device; |
faa775c4 | 2398 | int ret; |
24e0474b | 2399 | |
5c5c0df0 | 2400 | if (devid) { |
562d7b15 JB |
2401 | args.devid = devid; |
2402 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
a27a94c2 NB |
2403 | if (!device) |
2404 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2405 | return device; |
2406 | } | |
2407 | ||
faa775c4 JB |
2408 | ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); |
2409 | if (ret) | |
2410 | return ERR_PTR(ret); | |
2411 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
2412 | btrfs_put_dev_args_from_path(&args); | |
2413 | if (!device) | |
6e927ceb | 2414 | return ERR_PTR(-ENOENT); |
faa775c4 | 2415 | return device; |
24e0474b AJ |
2416 | } |
2417 | ||
849eae5e | 2418 | static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2419 | { |
0b246afa | 2420 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2421 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2422 | struct btrfs_fs_devices *seed_devices; |
2b82032c | 2423 | |
a32bf9a3 | 2424 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2425 | if (!fs_devices->seeding) |
849eae5e | 2426 | return ERR_PTR(-EINVAL); |
2b82032c | 2427 | |
427c8fdd NB |
2428 | /* |
2429 | * Private copy of the seed devices, anchored at | |
2430 | * fs_info->fs_devices->seed_list | |
2431 | */ | |
7239ff4b | 2432 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 | 2433 | if (IS_ERR(seed_devices)) |
849eae5e | 2434 | return seed_devices; |
2b82032c | 2435 | |
427c8fdd NB |
2436 | /* |
2437 | * It's necessary to retain a copy of the original seed fs_devices in | |
2438 | * fs_uuids so that filesystems which have been seeded can successfully | |
2439 | * reference the seed device from open_seed_devices. This also supports | |
2440 | * multiple fs seed. | |
2441 | */ | |
e4404d6e YZ |
2442 | old_devices = clone_fs_devices(fs_devices); |
2443 | if (IS_ERR(old_devices)) { | |
2444 | kfree(seed_devices); | |
849eae5e | 2445 | return old_devices; |
2b82032c | 2446 | } |
e4404d6e | 2447 | |
c4babc5e | 2448 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2449 | |
e4404d6e YZ |
2450 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2451 | seed_devices->opened = 1; | |
2452 | INIT_LIST_HEAD(&seed_devices->devices); | |
2453 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2454 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2455 | |
849eae5e AJ |
2456 | return seed_devices; |
2457 | } | |
2458 | ||
2459 | /* | |
2460 | * Splice seed devices into the sprout fs_devices. | |
2461 | * Generate a new fsid for the sprouted read-write filesystem. | |
2462 | */ | |
2463 | static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, | |
2464 | struct btrfs_fs_devices *seed_devices) | |
2465 | { | |
2466 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2467 | struct btrfs_super_block *disk_super = fs_info->super_copy; | |
2468 | struct btrfs_device *device; | |
2469 | u64 super_flags; | |
2470 | ||
2471 | /* | |
2472 | * We are updating the fsid, the thread leading to device_list_add() | |
2473 | * could race, so uuid_mutex is needed. | |
2474 | */ | |
2475 | lockdep_assert_held(&uuid_mutex); | |
2476 | ||
2477 | /* | |
2478 | * The threads listed below may traverse dev_list but can do that without | |
2479 | * device_list_mutex: | |
2480 | * - All device ops and balance - as we are in btrfs_exclop_start. | |
2481 | * - Various dev_list readers - are using RCU. | |
2482 | * - btrfs_ioctl_fitrim() - is using RCU. | |
2483 | * | |
2484 | * For-read threads as below are using device_list_mutex: | |
2485 | * - Readonly scrub btrfs_scrub_dev() | |
2486 | * - Readonly scrub btrfs_scrub_progress() | |
2487 | * - btrfs_get_dev_stats() | |
2488 | */ | |
2489 | lockdep_assert_held(&fs_devices->device_list_mutex); | |
2490 | ||
1f78160c XG |
2491 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2492 | synchronize_rcu); | |
2196d6e8 MX |
2493 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2494 | device->fs_devices = seed_devices; | |
c9513edb | 2495 | |
0395d84f | 2496 | fs_devices->seeding = false; |
2b82032c YZ |
2497 | fs_devices->num_devices = 0; |
2498 | fs_devices->open_devices = 0; | |
69611ac8 | 2499 | fs_devices->missing_devices = 0; |
7f0432d0 | 2500 | fs_devices->rotating = false; |
944d3f9f | 2501 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2502 | |
2503 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2504 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2505 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
f7171750 | 2506 | |
2b82032c YZ |
2507 | super_flags = btrfs_super_flags(disk_super) & |
2508 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2509 | btrfs_set_super_flags(disk_super, super_flags); | |
2b82032c YZ |
2510 | } |
2511 | ||
2512 | /* | |
01327610 | 2513 | * Store the expected generation for seed devices in device items. |
2b82032c | 2514 | */ |
5c466629 | 2515 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2516 | { |
562d7b15 | 2517 | BTRFS_DEV_LOOKUP_ARGS(args); |
5c466629 | 2518 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2519 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2520 | struct btrfs_path *path; |
2521 | struct extent_buffer *leaf; | |
2522 | struct btrfs_dev_item *dev_item; | |
2523 | struct btrfs_device *device; | |
2524 | struct btrfs_key key; | |
44880fdc | 2525 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c | 2526 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2b82032c YZ |
2527 | int ret; |
2528 | ||
2529 | path = btrfs_alloc_path(); | |
2530 | if (!path) | |
2531 | return -ENOMEM; | |
2532 | ||
2b82032c YZ |
2533 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2534 | key.offset = 0; | |
2535 | key.type = BTRFS_DEV_ITEM_KEY; | |
2536 | ||
2537 | while (1) { | |
2bb2e00e | 2538 | btrfs_reserve_chunk_metadata(trans, false); |
2b82032c | 2539 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
2bb2e00e | 2540 | btrfs_trans_release_chunk_metadata(trans); |
2b82032c YZ |
2541 | if (ret < 0) |
2542 | goto error; | |
2543 | ||
2544 | leaf = path->nodes[0]; | |
2545 | next_slot: | |
2546 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2547 | ret = btrfs_next_leaf(root, path); | |
2548 | if (ret > 0) | |
2549 | break; | |
2550 | if (ret < 0) | |
2551 | goto error; | |
2552 | leaf = path->nodes[0]; | |
2553 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2554 | btrfs_release_path(path); |
2b82032c YZ |
2555 | continue; |
2556 | } | |
2557 | ||
2558 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2559 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2560 | key.type != BTRFS_DEV_ITEM_KEY) | |
2561 | break; | |
2562 | ||
2563 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2564 | struct btrfs_dev_item); | |
562d7b15 | 2565 | args.devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 2566 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2567 | BTRFS_UUID_SIZE); |
1473b24e | 2568 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2569 | BTRFS_FSID_SIZE); |
562d7b15 JB |
2570 | args.uuid = dev_uuid; |
2571 | args.fsid = fs_uuid; | |
2572 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
79787eaa | 2573 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2574 | |
2575 | if (device->fs_devices->seeding) { | |
2576 | btrfs_set_device_generation(leaf, dev_item, | |
2577 | device->generation); | |
2578 | btrfs_mark_buffer_dirty(leaf); | |
2579 | } | |
2580 | ||
2581 | path->slots[0]++; | |
2582 | goto next_slot; | |
2583 | } | |
2584 | ret = 0; | |
2585 | error: | |
2586 | btrfs_free_path(path); | |
2587 | return ret; | |
2588 | } | |
2589 | ||
da353f6b | 2590 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2591 | { |
5112febb | 2592 | struct btrfs_root *root = fs_info->dev_root; |
788f20eb CM |
2593 | struct btrfs_trans_handle *trans; |
2594 | struct btrfs_device *device; | |
2595 | struct block_device *bdev; | |
0b246afa | 2596 | struct super_block *sb = fs_info->sb; |
5da54bc1 | 2597 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
849eae5e | 2598 | struct btrfs_fs_devices *seed_devices; |
39379faa NA |
2599 | u64 orig_super_total_bytes; |
2600 | u64 orig_super_num_devices; | |
788f20eb | 2601 | int ret = 0; |
fd880809 | 2602 | bool seeding_dev = false; |
44cab9ba | 2603 | bool locked = false; |
788f20eb | 2604 | |
5da54bc1 | 2605 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2606 | return -EROFS; |
788f20eb | 2607 | |
a5d16333 | 2608 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2609 | fs_info->bdev_holder); |
7f59203a JB |
2610 | if (IS_ERR(bdev)) |
2611 | return PTR_ERR(bdev); | |
a2135011 | 2612 | |
b70f5097 NA |
2613 | if (!btrfs_check_device_zone_type(fs_info, bdev)) { |
2614 | ret = -EINVAL; | |
2615 | goto error; | |
2616 | } | |
2617 | ||
5da54bc1 | 2618 | if (fs_devices->seeding) { |
fd880809 | 2619 | seeding_dev = true; |
2b82032c YZ |
2620 | down_write(&sb->s_umount); |
2621 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2622 | locked = true; |
2b82032c YZ |
2623 | } |
2624 | ||
b9ba017f | 2625 | sync_blockdev(bdev); |
a2135011 | 2626 | |
f4cfa9bd NB |
2627 | rcu_read_lock(); |
2628 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
788f20eb CM |
2629 | if (device->bdev == bdev) { |
2630 | ret = -EEXIST; | |
f4cfa9bd | 2631 | rcu_read_unlock(); |
2b82032c | 2632 | goto error; |
788f20eb CM |
2633 | } |
2634 | } | |
f4cfa9bd | 2635 | rcu_read_unlock(); |
788f20eb | 2636 | |
bb21e302 | 2637 | device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); |
12bd2fc0 | 2638 | if (IS_ERR(device)) { |
788f20eb | 2639 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2640 | ret = PTR_ERR(device); |
2b82032c | 2641 | goto error; |
788f20eb CM |
2642 | } |
2643 | ||
5b316468 NA |
2644 | device->fs_info = fs_info; |
2645 | device->bdev = bdev; | |
4889bc05 AJ |
2646 | ret = lookup_bdev(device_path, &device->devt); |
2647 | if (ret) | |
2648 | goto error_free_device; | |
5b316468 | 2649 | |
16beac87 | 2650 | ret = btrfs_get_dev_zone_info(device, false); |
5b316468 NA |
2651 | if (ret) |
2652 | goto error_free_device; | |
2653 | ||
a22285a6 | 2654 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2655 | if (IS_ERR(trans)) { |
98d5dc13 | 2656 | ret = PTR_ERR(trans); |
5b316468 | 2657 | goto error_free_zone; |
98d5dc13 TI |
2658 | } |
2659 | ||
ebbede42 | 2660 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2661 | device->generation = trans->transid; |
0b246afa JM |
2662 | device->io_width = fs_info->sectorsize; |
2663 | device->io_align = fs_info->sectorsize; | |
2664 | device->sector_size = fs_info->sectorsize; | |
cda00eba CH |
2665 | device->total_bytes = |
2666 | round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); | |
2cc3c559 | 2667 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2668 | device->commit_total_bytes = device->total_bytes; |
e12c9621 | 2669 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2670 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
fb01aa85 | 2671 | device->mode = FMODE_EXCL; |
27087f37 | 2672 | device->dev_stats_valid = 1; |
9f6d2510 | 2673 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2674 | |
2b82032c | 2675 | if (seeding_dev) { |
a0a1db70 | 2676 | btrfs_clear_sb_rdonly(sb); |
849eae5e AJ |
2677 | |
2678 | /* GFP_KERNEL allocation must not be under device_list_mutex */ | |
2679 | seed_devices = btrfs_init_sprout(fs_info); | |
2680 | if (IS_ERR(seed_devices)) { | |
2681 | ret = PTR_ERR(seed_devices); | |
d31c32f6 AJ |
2682 | btrfs_abort_transaction(trans, ret); |
2683 | goto error_trans; | |
2684 | } | |
849eae5e AJ |
2685 | } |
2686 | ||
2687 | mutex_lock(&fs_devices->device_list_mutex); | |
2688 | if (seeding_dev) { | |
2689 | btrfs_setup_sprout(fs_info, seed_devices); | |
b7cb29e6 AJ |
2690 | btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, |
2691 | device); | |
2b82032c | 2692 | } |
788f20eb | 2693 | |
5da54bc1 | 2694 | device->fs_devices = fs_devices; |
e5e9a520 | 2695 | |
34441361 | 2696 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2697 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2698 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2699 | fs_devices->num_devices++; | |
2700 | fs_devices->open_devices++; | |
2701 | fs_devices->rw_devices++; | |
2702 | fs_devices->total_devices++; | |
2703 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2704 | |
a5ed45f8 | 2705 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2706 | |
10f0d2a5 | 2707 | if (!bdev_nonrot(bdev)) |
7f0432d0 | 2708 | fs_devices->rotating = true; |
c289811c | 2709 | |
39379faa | 2710 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2711 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2712 | round_down(orig_super_total_bytes + device->total_bytes, |
2713 | fs_info->sectorsize)); | |
788f20eb | 2714 | |
39379faa NA |
2715 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2716 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2717 | orig_super_num_devices + 1); | |
0d39376a | 2718 | |
2196d6e8 MX |
2719 | /* |
2720 | * we've got more storage, clear any full flags on the space | |
2721 | * infos | |
2722 | */ | |
0b246afa | 2723 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2724 | |
34441361 | 2725 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2726 | |
2727 | /* Add sysfs device entry */ | |
cd36da2e | 2728 | btrfs_sysfs_add_device(device); |
ca10845a | 2729 | |
5da54bc1 | 2730 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2731 | |
2b82032c | 2732 | if (seeding_dev) { |
34441361 | 2733 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2734 | ret = init_first_rw_device(trans); |
34441361 | 2735 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2736 | if (ret) { |
66642832 | 2737 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2738 | goto error_sysfs; |
005d6427 | 2739 | } |
2196d6e8 MX |
2740 | } |
2741 | ||
8e87e856 | 2742 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2743 | if (ret) { |
66642832 | 2744 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2745 | goto error_sysfs; |
2196d6e8 MX |
2746 | } |
2747 | ||
2748 | if (seeding_dev) { | |
5c466629 | 2749 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2750 | if (ret) { |
66642832 | 2751 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2752 | goto error_sysfs; |
005d6427 | 2753 | } |
b2373f25 | 2754 | |
8e560081 NB |
2755 | /* |
2756 | * fs_devices now represents the newly sprouted filesystem and | |
849eae5e | 2757 | * its fsid has been changed by btrfs_sprout_splice(). |
8e560081 NB |
2758 | */ |
2759 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2760 | } |
2761 | ||
3a45bb20 | 2762 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2763 | |
2b82032c YZ |
2764 | if (seeding_dev) { |
2765 | mutex_unlock(&uuid_mutex); | |
2766 | up_write(&sb->s_umount); | |
44cab9ba | 2767 | locked = false; |
788f20eb | 2768 | |
79787eaa JM |
2769 | if (ret) /* transaction commit */ |
2770 | return ret; | |
2771 | ||
2ff7e61e | 2772 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2773 | if (ret < 0) |
0b246afa | 2774 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2775 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2776 | trans = btrfs_attach_transaction(root); |
2777 | if (IS_ERR(trans)) { | |
2778 | if (PTR_ERR(trans) == -ENOENT) | |
2779 | return 0; | |
7132a262 AJ |
2780 | ret = PTR_ERR(trans); |
2781 | trans = NULL; | |
2782 | goto error_sysfs; | |
671415b7 | 2783 | } |
3a45bb20 | 2784 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2785 | } |
c9e9f97b | 2786 | |
7f551d96 AJ |
2787 | /* |
2788 | * Now that we have written a new super block to this device, check all | |
2789 | * other fs_devices list if device_path alienates any other scanned | |
2790 | * device. | |
2791 | * We can ignore the return value as it typically returns -EINVAL and | |
2792 | * only succeeds if the device was an alien. | |
2793 | */ | |
4889bc05 | 2794 | btrfs_forget_devices(device->devt); |
7f551d96 AJ |
2795 | |
2796 | /* Update ctime/mtime for blkid or udev */ | |
54fde91f | 2797 | update_dev_time(device_path); |
7f551d96 | 2798 | |
2b82032c | 2799 | return ret; |
79787eaa | 2800 | |
d31c32f6 | 2801 | error_sysfs: |
53f8a74c | 2802 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2803 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2804 | mutex_lock(&fs_info->chunk_mutex); | |
2805 | list_del_rcu(&device->dev_list); | |
2806 | list_del(&device->dev_alloc_list); | |
2807 | fs_info->fs_devices->num_devices--; | |
2808 | fs_info->fs_devices->open_devices--; | |
2809 | fs_info->fs_devices->rw_devices--; | |
2810 | fs_info->fs_devices->total_devices--; | |
2811 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2812 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2813 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2814 | orig_super_total_bytes); | |
2815 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2816 | orig_super_num_devices); | |
2817 | mutex_unlock(&fs_info->chunk_mutex); | |
2818 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2819 | error_trans: |
0af2c4bf | 2820 | if (seeding_dev) |
a0a1db70 | 2821 | btrfs_set_sb_rdonly(sb); |
7132a262 AJ |
2822 | if (trans) |
2823 | btrfs_end_transaction(trans); | |
5b316468 NA |
2824 | error_free_zone: |
2825 | btrfs_destroy_dev_zone_info(device); | |
5c4cf6c9 | 2826 | error_free_device: |
a425f9d4 | 2827 | btrfs_free_device(device); |
2b82032c | 2828 | error: |
e525fd89 | 2829 | blkdev_put(bdev, FMODE_EXCL); |
44cab9ba | 2830 | if (locked) { |
2b82032c YZ |
2831 | mutex_unlock(&uuid_mutex); |
2832 | up_write(&sb->s_umount); | |
2833 | } | |
c9e9f97b | 2834 | return ret; |
788f20eb CM |
2835 | } |
2836 | ||
d397712b CM |
2837 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2838 | struct btrfs_device *device) | |
0b86a832 CM |
2839 | { |
2840 | int ret; | |
2841 | struct btrfs_path *path; | |
0b246afa | 2842 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2843 | struct btrfs_dev_item *dev_item; |
2844 | struct extent_buffer *leaf; | |
2845 | struct btrfs_key key; | |
2846 | ||
0b86a832 CM |
2847 | path = btrfs_alloc_path(); |
2848 | if (!path) | |
2849 | return -ENOMEM; | |
2850 | ||
2851 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2852 | key.type = BTRFS_DEV_ITEM_KEY; | |
2853 | key.offset = device->devid; | |
2854 | ||
2855 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2856 | if (ret < 0) | |
2857 | goto out; | |
2858 | ||
2859 | if (ret > 0) { | |
2860 | ret = -ENOENT; | |
2861 | goto out; | |
2862 | } | |
2863 | ||
2864 | leaf = path->nodes[0]; | |
2865 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2866 | ||
2867 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2868 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2869 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2870 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2871 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2872 | btrfs_set_device_total_bytes(leaf, dev_item, |
2873 | btrfs_device_get_disk_total_bytes(device)); | |
2874 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2875 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2876 | btrfs_mark_buffer_dirty(leaf); |
2877 | ||
2878 | out: | |
2879 | btrfs_free_path(path); | |
2880 | return ret; | |
2881 | } | |
2882 | ||
2196d6e8 | 2883 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2884 | struct btrfs_device *device, u64 new_size) |
2885 | { | |
0b246afa JM |
2886 | struct btrfs_fs_info *fs_info = device->fs_info; |
2887 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2888 | u64 old_total; |
2889 | u64 diff; | |
2bb2e00e | 2890 | int ret; |
8f18cf13 | 2891 | |
ebbede42 | 2892 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2893 | return -EACCES; |
2196d6e8 | 2894 | |
7dfb8be1 NB |
2895 | new_size = round_down(new_size, fs_info->sectorsize); |
2896 | ||
34441361 | 2897 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2898 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2899 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2900 | |
63a212ab | 2901 | if (new_size <= device->total_bytes || |
401e29c1 | 2902 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2903 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2904 | return -EINVAL; |
2196d6e8 | 2905 | } |
2b82032c | 2906 | |
7dfb8be1 NB |
2907 | btrfs_set_super_total_bytes(super_copy, |
2908 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2909 | device->fs_devices->total_rw_bytes += diff; |
2910 | ||
7cc8e58d MX |
2911 | btrfs_device_set_total_bytes(device, new_size); |
2912 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2913 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2914 | if (list_empty(&device->post_commit_list)) |
2915 | list_add_tail(&device->post_commit_list, | |
2916 | &trans->transaction->dev_update_list); | |
34441361 | 2917 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2918 | |
2bb2e00e FM |
2919 | btrfs_reserve_chunk_metadata(trans, false); |
2920 | ret = btrfs_update_device(trans, device); | |
2921 | btrfs_trans_release_chunk_metadata(trans); | |
2922 | ||
2923 | return ret; | |
8f18cf13 CM |
2924 | } |
2925 | ||
f4208794 | 2926 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2927 | { |
f4208794 | 2928 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2929 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2930 | int ret; |
2931 | struct btrfs_path *path; | |
2932 | struct btrfs_key key; | |
2933 | ||
8f18cf13 CM |
2934 | path = btrfs_alloc_path(); |
2935 | if (!path) | |
2936 | return -ENOMEM; | |
2937 | ||
408fbf19 | 2938 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2939 | key.offset = chunk_offset; |
2940 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2941 | ||
2942 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2943 | if (ret < 0) |
2944 | goto out; | |
2945 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2946 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2947 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2948 | ret = -ENOENT; |
2949 | goto out; | |
2950 | } | |
8f18cf13 CM |
2951 | |
2952 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2953 | if (ret < 0) |
0b246afa JM |
2954 | btrfs_handle_fs_error(fs_info, ret, |
2955 | "Failed to delete chunk item."); | |
79787eaa | 2956 | out: |
8f18cf13 | 2957 | btrfs_free_path(path); |
65a246c5 | 2958 | return ret; |
8f18cf13 CM |
2959 | } |
2960 | ||
408fbf19 | 2961 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2962 | { |
0b246afa | 2963 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2964 | struct btrfs_disk_key *disk_key; |
2965 | struct btrfs_chunk *chunk; | |
2966 | u8 *ptr; | |
2967 | int ret = 0; | |
2968 | u32 num_stripes; | |
2969 | u32 array_size; | |
2970 | u32 len = 0; | |
2971 | u32 cur; | |
2972 | struct btrfs_key key; | |
2973 | ||
79bd3712 | 2974 | lockdep_assert_held(&fs_info->chunk_mutex); |
8f18cf13 CM |
2975 | array_size = btrfs_super_sys_array_size(super_copy); |
2976 | ||
2977 | ptr = super_copy->sys_chunk_array; | |
2978 | cur = 0; | |
2979 | ||
2980 | while (cur < array_size) { | |
2981 | disk_key = (struct btrfs_disk_key *)ptr; | |
2982 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2983 | ||
2984 | len = sizeof(*disk_key); | |
2985 | ||
2986 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2987 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2988 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2989 | len += btrfs_chunk_item_size(num_stripes); | |
2990 | } else { | |
2991 | ret = -EIO; | |
2992 | break; | |
2993 | } | |
408fbf19 | 2994 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
2995 | key.offset == chunk_offset) { |
2996 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2997 | array_size -= len; | |
2998 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2999 | } else { | |
3000 | ptr += len; | |
3001 | cur += len; | |
3002 | } | |
3003 | } | |
3004 | return ret; | |
3005 | } | |
3006 | ||
60ca842e OS |
3007 | /* |
3008 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
3009 | * @logical: Logical block offset in bytes. | |
3010 | * @length: Length of extent in bytes. | |
3011 | * | |
3012 | * Return: Chunk mapping or ERR_PTR. | |
3013 | */ | |
3014 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
3015 | u64 logical, u64 length) | |
592d92ee LB |
3016 | { |
3017 | struct extent_map_tree *em_tree; | |
3018 | struct extent_map *em; | |
3019 | ||
c8bf1b67 | 3020 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
3021 | read_lock(&em_tree->lock); |
3022 | em = lookup_extent_mapping(em_tree, logical, length); | |
3023 | read_unlock(&em_tree->lock); | |
3024 | ||
3025 | if (!em) { | |
3026 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
3027 | logical, length); | |
3028 | return ERR_PTR(-EINVAL); | |
3029 | } | |
3030 | ||
3031 | if (em->start > logical || em->start + em->len < logical) { | |
3032 | btrfs_crit(fs_info, | |
3033 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
3034 | logical, length, em->start, em->start + em->len); | |
3035 | free_extent_map(em); | |
3036 | return ERR_PTR(-EINVAL); | |
3037 | } | |
3038 | ||
3039 | /* callers are responsible for dropping em's ref. */ | |
3040 | return em; | |
3041 | } | |
3042 | ||
79bd3712 FM |
3043 | static int remove_chunk_item(struct btrfs_trans_handle *trans, |
3044 | struct map_lookup *map, u64 chunk_offset) | |
3045 | { | |
3046 | int i; | |
3047 | ||
3048 | /* | |
3049 | * Removing chunk items and updating the device items in the chunks btree | |
3050 | * requires holding the chunk_mutex. | |
3051 | * See the comment at btrfs_chunk_alloc() for the details. | |
3052 | */ | |
3053 | lockdep_assert_held(&trans->fs_info->chunk_mutex); | |
3054 | ||
3055 | for (i = 0; i < map->num_stripes; i++) { | |
3056 | int ret; | |
3057 | ||
3058 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
3059 | if (ret) | |
3060 | return ret; | |
3061 | } | |
3062 | ||
3063 | return btrfs_free_chunk(trans, chunk_offset); | |
3064 | } | |
3065 | ||
97aff912 | 3066 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3067 | { |
97aff912 | 3068 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3069 | struct extent_map *em; |
3070 | struct map_lookup *map; | |
2196d6e8 | 3071 | u64 dev_extent_len = 0; |
47ab2a6c | 3072 | int i, ret = 0; |
0b246afa | 3073 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3074 | |
60ca842e | 3075 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3076 | if (IS_ERR(em)) { |
47ab2a6c JB |
3077 | /* |
3078 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3079 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3080 | * do anything we still error out. |
3081 | */ | |
3082 | ASSERT(0); | |
592d92ee | 3083 | return PTR_ERR(em); |
47ab2a6c | 3084 | } |
95617d69 | 3085 | map = em->map_lookup; |
8f18cf13 | 3086 | |
57ba4cb8 | 3087 | /* |
79bd3712 FM |
3088 | * First delete the device extent items from the devices btree. |
3089 | * We take the device_list_mutex to avoid racing with the finishing phase | |
3090 | * of a device replace operation. See the comment below before acquiring | |
3091 | * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex | |
3092 | * because that can result in a deadlock when deleting the device extent | |
3093 | * items from the devices btree - COWing an extent buffer from the btree | |
3094 | * may result in allocating a new metadata chunk, which would attempt to | |
3095 | * lock again fs_info->chunk_mutex. | |
57ba4cb8 FM |
3096 | */ |
3097 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3098 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3099 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3100 | ret = btrfs_free_dev_extent(trans, device, |
3101 | map->stripes[i].physical, | |
3102 | &dev_extent_len); | |
47ab2a6c | 3103 | if (ret) { |
57ba4cb8 | 3104 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3105 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3106 | goto out; |
3107 | } | |
a061fc8d | 3108 | |
2196d6e8 | 3109 | if (device->bytes_used > 0) { |
34441361 | 3110 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3111 | btrfs_device_set_bytes_used(device, |
3112 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3113 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3114 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3115 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3116 | } |
79bd3712 FM |
3117 | } |
3118 | mutex_unlock(&fs_devices->device_list_mutex); | |
a061fc8d | 3119 | |
79bd3712 FM |
3120 | /* |
3121 | * We acquire fs_info->chunk_mutex for 2 reasons: | |
3122 | * | |
3123 | * 1) Just like with the first phase of the chunk allocation, we must | |
3124 | * reserve system space, do all chunk btree updates and deletions, and | |
3125 | * update the system chunk array in the superblock while holding this | |
3126 | * mutex. This is for similar reasons as explained on the comment at | |
3127 | * the top of btrfs_chunk_alloc(); | |
3128 | * | |
3129 | * 2) Prevent races with the final phase of a device replace operation | |
3130 | * that replaces the device object associated with the map's stripes, | |
3131 | * because the device object's id can change at any time during that | |
3132 | * final phase of the device replace operation | |
3133 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
3134 | * replaced device and then see it with an ID of | |
3135 | * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating | |
3136 | * the device item, which does not exists on the chunk btree. | |
3137 | * The finishing phase of device replace acquires both the | |
3138 | * device_list_mutex and the chunk_mutex, in that order, so we are | |
3139 | * safe by just acquiring the chunk_mutex. | |
3140 | */ | |
3141 | trans->removing_chunk = true; | |
3142 | mutex_lock(&fs_info->chunk_mutex); | |
3143 | ||
3144 | check_system_chunk(trans, map->type); | |
3145 | ||
3146 | ret = remove_chunk_item(trans, map, chunk_offset); | |
3147 | /* | |
3148 | * Normally we should not get -ENOSPC since we reserved space before | |
3149 | * through the call to check_system_chunk(). | |
3150 | * | |
3151 | * Despite our system space_info having enough free space, we may not | |
3152 | * be able to allocate extents from its block groups, because all have | |
3153 | * an incompatible profile, which will force us to allocate a new system | |
3154 | * block group with the right profile, or right after we called | |
3155 | * check_system_space() above, a scrub turned the only system block group | |
3156 | * with enough free space into RO mode. | |
3157 | * This is explained with more detail at do_chunk_alloc(). | |
3158 | * | |
3159 | * So if we get -ENOSPC, allocate a new system chunk and retry once. | |
3160 | */ | |
3161 | if (ret == -ENOSPC) { | |
3162 | const u64 sys_flags = btrfs_system_alloc_profile(fs_info); | |
3163 | struct btrfs_block_group *sys_bg; | |
3164 | ||
f6f39f7a | 3165 | sys_bg = btrfs_create_chunk(trans, sys_flags); |
79bd3712 FM |
3166 | if (IS_ERR(sys_bg)) { |
3167 | ret = PTR_ERR(sys_bg); | |
3168 | btrfs_abort_transaction(trans, ret); | |
3169 | goto out; | |
3170 | } | |
3171 | ||
3172 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
64bc6c2a | 3173 | if (ret) { |
64bc6c2a NB |
3174 | btrfs_abort_transaction(trans, ret); |
3175 | goto out; | |
dfe25020 | 3176 | } |
57ba4cb8 | 3177 | |
79bd3712 FM |
3178 | ret = remove_chunk_item(trans, map, chunk_offset); |
3179 | if (ret) { | |
3180 | btrfs_abort_transaction(trans, ret); | |
3181 | goto out; | |
3182 | } | |
3183 | } else if (ret) { | |
66642832 | 3184 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3185 | goto out; |
3186 | } | |
8f18cf13 | 3187 | |
6bccf3ab | 3188 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3189 | |
8f18cf13 | 3190 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3191 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3192 | if (ret) { |
66642832 | 3193 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3194 | goto out; |
3195 | } | |
8f18cf13 CM |
3196 | } |
3197 | ||
79bd3712 FM |
3198 | mutex_unlock(&fs_info->chunk_mutex); |
3199 | trans->removing_chunk = false; | |
3200 | ||
3201 | /* | |
3202 | * We are done with chunk btree updates and deletions, so release the | |
3203 | * system space we previously reserved (with check_system_chunk()). | |
3204 | */ | |
3205 | btrfs_trans_release_chunk_metadata(trans); | |
3206 | ||
5a98ec01 | 3207 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3208 | if (ret) { |
66642832 | 3209 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3210 | goto out; |
3211 | } | |
2b82032c | 3212 | |
47ab2a6c | 3213 | out: |
79bd3712 FM |
3214 | if (trans->removing_chunk) { |
3215 | mutex_unlock(&fs_info->chunk_mutex); | |
3216 | trans->removing_chunk = false; | |
3217 | } | |
2b82032c YZ |
3218 | /* once for us */ |
3219 | free_extent_map(em); | |
47ab2a6c JB |
3220 | return ret; |
3221 | } | |
2b82032c | 3222 | |
18bb8bbf | 3223 | int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3224 | { |
5b4aacef | 3225 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3226 | struct btrfs_trans_handle *trans; |
b0643e59 | 3227 | struct btrfs_block_group *block_group; |
01e86008 | 3228 | u64 length; |
47ab2a6c | 3229 | int ret; |
2b82032c | 3230 | |
4b349253 JB |
3231 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
3232 | btrfs_err(fs_info, | |
3233 | "relocate: not supported on extent tree v2 yet"); | |
3234 | return -EINVAL; | |
3235 | } | |
3236 | ||
67c5e7d4 FM |
3237 | /* |
3238 | * Prevent races with automatic removal of unused block groups. | |
3239 | * After we relocate and before we remove the chunk with offset | |
3240 | * chunk_offset, automatic removal of the block group can kick in, | |
3241 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3242 | * | |
3243 | * Make sure to acquire this mutex before doing a tree search (dev | |
3244 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3245 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3246 | * we release the path used to search the chunk/dev tree and before | |
3247 | * the current task acquires this mutex and calls us. | |
3248 | */ | |
f3372065 | 3249 | lockdep_assert_held(&fs_info->reclaim_bgs_lock); |
67c5e7d4 | 3250 | |
47ab2a6c | 3251 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3252 | btrfs_scrub_pause(fs_info); |
0b246afa | 3253 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3254 | btrfs_scrub_continue(fs_info); |
47ab2a6c JB |
3255 | if (ret) |
3256 | return ret; | |
3257 | ||
b0643e59 DZ |
3258 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3259 | if (!block_group) | |
3260 | return -ENOENT; | |
3261 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
01e86008 | 3262 | length = block_group->length; |
b0643e59 DZ |
3263 | btrfs_put_block_group(block_group); |
3264 | ||
01e86008 JT |
3265 | /* |
3266 | * On a zoned file system, discard the whole block group, this will | |
3267 | * trigger a REQ_OP_ZONE_RESET operation on the device zone. If | |
3268 | * resetting the zone fails, don't treat it as a fatal problem from the | |
3269 | * filesystem's point of view. | |
3270 | */ | |
3271 | if (btrfs_is_zoned(fs_info)) { | |
3272 | ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); | |
3273 | if (ret) | |
3274 | btrfs_info(fs_info, | |
3275 | "failed to reset zone %llu after relocation", | |
3276 | chunk_offset); | |
3277 | } | |
3278 | ||
19c4d2f9 CM |
3279 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3280 | chunk_offset); | |
3281 | if (IS_ERR(trans)) { | |
3282 | ret = PTR_ERR(trans); | |
3283 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3284 | return ret; | |
3285 | } | |
3286 | ||
47ab2a6c | 3287 | /* |
19c4d2f9 CM |
3288 | * step two, delete the device extents and the |
3289 | * chunk tree entries | |
47ab2a6c | 3290 | */ |
97aff912 | 3291 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3292 | btrfs_end_transaction(trans); |
19c4d2f9 | 3293 | return ret; |
2b82032c YZ |
3294 | } |
3295 | ||
2ff7e61e | 3296 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3297 | { |
0b246afa | 3298 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3299 | struct btrfs_path *path; |
3300 | struct extent_buffer *leaf; | |
3301 | struct btrfs_chunk *chunk; | |
3302 | struct btrfs_key key; | |
3303 | struct btrfs_key found_key; | |
2b82032c | 3304 | u64 chunk_type; |
ba1bf481 JB |
3305 | bool retried = false; |
3306 | int failed = 0; | |
2b82032c YZ |
3307 | int ret; |
3308 | ||
3309 | path = btrfs_alloc_path(); | |
3310 | if (!path) | |
3311 | return -ENOMEM; | |
3312 | ||
ba1bf481 | 3313 | again: |
2b82032c YZ |
3314 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3315 | key.offset = (u64)-1; | |
3316 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3317 | ||
3318 | while (1) { | |
f3372065 | 3319 | mutex_lock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3320 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3321 | if (ret < 0) { |
f3372065 | 3322 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3323 | goto error; |
67c5e7d4 | 3324 | } |
79787eaa | 3325 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3326 | |
3327 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3328 | key.type); | |
67c5e7d4 | 3329 | if (ret) |
f3372065 | 3330 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c YZ |
3331 | if (ret < 0) |
3332 | goto error; | |
3333 | if (ret > 0) | |
3334 | break; | |
1a40e23b | 3335 | |
2b82032c YZ |
3336 | leaf = path->nodes[0]; |
3337 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3338 | |
2b82032c YZ |
3339 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3340 | struct btrfs_chunk); | |
3341 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3342 | btrfs_release_path(path); |
8f18cf13 | 3343 | |
2b82032c | 3344 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3345 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3346 | if (ret == -ENOSPC) |
3347 | failed++; | |
14586651 HS |
3348 | else |
3349 | BUG_ON(ret); | |
2b82032c | 3350 | } |
f3372065 | 3351 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 3352 | |
2b82032c YZ |
3353 | if (found_key.offset == 0) |
3354 | break; | |
3355 | key.offset = found_key.offset - 1; | |
3356 | } | |
3357 | ret = 0; | |
ba1bf481 JB |
3358 | if (failed && !retried) { |
3359 | failed = 0; | |
3360 | retried = true; | |
3361 | goto again; | |
fae7f21c | 3362 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3363 | ret = -ENOSPC; |
3364 | } | |
2b82032c YZ |
3365 | error: |
3366 | btrfs_free_path(path); | |
3367 | return ret; | |
8f18cf13 CM |
3368 | } |
3369 | ||
a6f93c71 LB |
3370 | /* |
3371 | * return 1 : allocate a data chunk successfully, | |
3372 | * return <0: errors during allocating a data chunk, | |
3373 | * return 0 : no need to allocate a data chunk. | |
3374 | */ | |
3375 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3376 | u64 chunk_offset) | |
3377 | { | |
32da5386 | 3378 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3379 | u64 bytes_used; |
3380 | u64 chunk_type; | |
3381 | ||
3382 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3383 | ASSERT(cache); | |
3384 | chunk_type = cache->flags; | |
3385 | btrfs_put_block_group(cache); | |
3386 | ||
5ae21692 JT |
3387 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3388 | return 0; | |
3389 | ||
3390 | spin_lock(&fs_info->data_sinfo->lock); | |
3391 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3392 | spin_unlock(&fs_info->data_sinfo->lock); | |
3393 | ||
3394 | if (!bytes_used) { | |
3395 | struct btrfs_trans_handle *trans; | |
3396 | int ret; | |
3397 | ||
3398 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3399 | if (IS_ERR(trans)) | |
3400 | return PTR_ERR(trans); | |
3401 | ||
3402 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3403 | btrfs_end_transaction(trans); | |
3404 | if (ret < 0) | |
3405 | return ret; | |
3406 | return 1; | |
a6f93c71 | 3407 | } |
5ae21692 | 3408 | |
a6f93c71 LB |
3409 | return 0; |
3410 | } | |
3411 | ||
6bccf3ab | 3412 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3413 | struct btrfs_balance_control *bctl) |
3414 | { | |
6bccf3ab | 3415 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3416 | struct btrfs_trans_handle *trans; |
3417 | struct btrfs_balance_item *item; | |
3418 | struct btrfs_disk_balance_args disk_bargs; | |
3419 | struct btrfs_path *path; | |
3420 | struct extent_buffer *leaf; | |
3421 | struct btrfs_key key; | |
3422 | int ret, err; | |
3423 | ||
3424 | path = btrfs_alloc_path(); | |
3425 | if (!path) | |
3426 | return -ENOMEM; | |
3427 | ||
3428 | trans = btrfs_start_transaction(root, 0); | |
3429 | if (IS_ERR(trans)) { | |
3430 | btrfs_free_path(path); | |
3431 | return PTR_ERR(trans); | |
3432 | } | |
3433 | ||
3434 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3435 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3436 | key.offset = 0; |
3437 | ||
3438 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3439 | sizeof(*item)); | |
3440 | if (ret) | |
3441 | goto out; | |
3442 | ||
3443 | leaf = path->nodes[0]; | |
3444 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3445 | ||
b159fa28 | 3446 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3447 | |
3448 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3449 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3450 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3451 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3452 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3453 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3454 | ||
3455 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3456 | ||
3457 | btrfs_mark_buffer_dirty(leaf); | |
3458 | out: | |
3459 | btrfs_free_path(path); | |
3a45bb20 | 3460 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3461 | if (err && !ret) |
3462 | ret = err; | |
3463 | return ret; | |
3464 | } | |
3465 | ||
6bccf3ab | 3466 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3467 | { |
6bccf3ab | 3468 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3469 | struct btrfs_trans_handle *trans; |
3470 | struct btrfs_path *path; | |
3471 | struct btrfs_key key; | |
3472 | int ret, err; | |
3473 | ||
3474 | path = btrfs_alloc_path(); | |
3475 | if (!path) | |
3476 | return -ENOMEM; | |
3477 | ||
3502a8c0 | 3478 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3479 | if (IS_ERR(trans)) { |
3480 | btrfs_free_path(path); | |
3481 | return PTR_ERR(trans); | |
3482 | } | |
3483 | ||
3484 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3485 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3486 | key.offset = 0; |
3487 | ||
3488 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3489 | if (ret < 0) | |
3490 | goto out; | |
3491 | if (ret > 0) { | |
3492 | ret = -ENOENT; | |
3493 | goto out; | |
3494 | } | |
3495 | ||
3496 | ret = btrfs_del_item(trans, root, path); | |
3497 | out: | |
3498 | btrfs_free_path(path); | |
3a45bb20 | 3499 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3500 | if (err && !ret) |
3501 | ret = err; | |
3502 | return ret; | |
3503 | } | |
3504 | ||
59641015 ID |
3505 | /* |
3506 | * This is a heuristic used to reduce the number of chunks balanced on | |
3507 | * resume after balance was interrupted. | |
3508 | */ | |
3509 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3510 | { | |
3511 | /* | |
3512 | * Turn on soft mode for chunk types that were being converted. | |
3513 | */ | |
3514 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3515 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3516 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3517 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3518 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3519 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3520 | ||
3521 | /* | |
3522 | * Turn on usage filter if is not already used. The idea is | |
3523 | * that chunks that we have already balanced should be | |
3524 | * reasonably full. Don't do it for chunks that are being | |
3525 | * converted - that will keep us from relocating unconverted | |
3526 | * (albeit full) chunks. | |
3527 | */ | |
3528 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3529 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3530 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3531 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3532 | bctl->data.usage = 90; | |
3533 | } | |
3534 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3535 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3536 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3537 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3538 | bctl->sys.usage = 90; | |
3539 | } | |
3540 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3541 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3542 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3543 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3544 | bctl->meta.usage = 90; | |
3545 | } | |
3546 | } | |
3547 | ||
149196a2 DS |
3548 | /* |
3549 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3550 | */ | |
3551 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3552 | { |
3553 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3554 | int ret; |
c9e9f97b ID |
3555 | |
3556 | BUG_ON(!fs_info->balance_ctl); | |
3557 | ||
3558 | spin_lock(&fs_info->balance_lock); | |
3559 | fs_info->balance_ctl = NULL; | |
3560 | spin_unlock(&fs_info->balance_lock); | |
3561 | ||
3562 | kfree(bctl); | |
149196a2 DS |
3563 | ret = del_balance_item(fs_info); |
3564 | if (ret) | |
3565 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3566 | } |
3567 | ||
ed25e9b2 ID |
3568 | /* |
3569 | * Balance filters. Return 1 if chunk should be filtered out | |
3570 | * (should not be balanced). | |
3571 | */ | |
899c81ea | 3572 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3573 | struct btrfs_balance_args *bargs) |
3574 | { | |
899c81ea ID |
3575 | chunk_type = chunk_to_extended(chunk_type) & |
3576 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3577 | |
899c81ea | 3578 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3579 | return 0; |
3580 | ||
3581 | return 1; | |
3582 | } | |
3583 | ||
dba72cb3 | 3584 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3585 | struct btrfs_balance_args *bargs) |
bc309467 | 3586 | { |
32da5386 | 3587 | struct btrfs_block_group *cache; |
bc309467 DS |
3588 | u64 chunk_used; |
3589 | u64 user_thresh_min; | |
3590 | u64 user_thresh_max; | |
3591 | int ret = 1; | |
3592 | ||
3593 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3594 | chunk_used = cache->used; |
bc309467 DS |
3595 | |
3596 | if (bargs->usage_min == 0) | |
3597 | user_thresh_min = 0; | |
3598 | else | |
428c8e03 | 3599 | user_thresh_min = mult_perc(cache->length, bargs->usage_min); |
bc309467 DS |
3600 | |
3601 | if (bargs->usage_max == 0) | |
3602 | user_thresh_max = 1; | |
3603 | else if (bargs->usage_max > 100) | |
b3470b5d | 3604 | user_thresh_max = cache->length; |
bc309467 | 3605 | else |
428c8e03 | 3606 | user_thresh_max = mult_perc(cache->length, bargs->usage_max); |
bc309467 DS |
3607 | |
3608 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3609 | ret = 0; | |
3610 | ||
3611 | btrfs_put_block_group(cache); | |
3612 | return ret; | |
3613 | } | |
3614 | ||
dba72cb3 | 3615 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3616 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3617 | { |
32da5386 | 3618 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3619 | u64 chunk_used, user_thresh; |
3620 | int ret = 1; | |
3621 | ||
3622 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3623 | chunk_used = cache->used; |
5ce5b3c0 | 3624 | |
bc309467 | 3625 | if (bargs->usage_min == 0) |
3e39cea6 | 3626 | user_thresh = 1; |
a105bb88 | 3627 | else if (bargs->usage > 100) |
b3470b5d | 3628 | user_thresh = cache->length; |
a105bb88 | 3629 | else |
428c8e03 | 3630 | user_thresh = mult_perc(cache->length, bargs->usage); |
a105bb88 | 3631 | |
5ce5b3c0 ID |
3632 | if (chunk_used < user_thresh) |
3633 | ret = 0; | |
3634 | ||
3635 | btrfs_put_block_group(cache); | |
3636 | return ret; | |
3637 | } | |
3638 | ||
409d404b ID |
3639 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3640 | struct btrfs_chunk *chunk, | |
3641 | struct btrfs_balance_args *bargs) | |
3642 | { | |
3643 | struct btrfs_stripe *stripe; | |
3644 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3645 | int i; | |
3646 | ||
3647 | for (i = 0; i < num_stripes; i++) { | |
3648 | stripe = btrfs_stripe_nr(chunk, i); | |
3649 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3650 | return 0; | |
3651 | } | |
3652 | ||
3653 | return 1; | |
3654 | } | |
3655 | ||
946c9256 DS |
3656 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3657 | { | |
3658 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3659 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3660 | const int nparity = btrfs_raid_array[index].nparity; | |
3661 | ||
d58ede8d | 3662 | return (num_stripes - nparity) / ncopies; |
946c9256 DS |
3663 | } |
3664 | ||
94e60d5a ID |
3665 | /* [pstart, pend) */ |
3666 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3667 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3668 | struct btrfs_balance_args *bargs) |
3669 | { | |
3670 | struct btrfs_stripe *stripe; | |
3671 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3672 | u64 stripe_offset; | |
3673 | u64 stripe_length; | |
946c9256 | 3674 | u64 type; |
94e60d5a ID |
3675 | int factor; |
3676 | int i; | |
3677 | ||
3678 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3679 | return 0; | |
3680 | ||
946c9256 DS |
3681 | type = btrfs_chunk_type(leaf, chunk); |
3682 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3683 | |
3684 | for (i = 0; i < num_stripes; i++) { | |
3685 | stripe = btrfs_stripe_nr(chunk, i); | |
3686 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3687 | continue; | |
3688 | ||
3689 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3690 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3691 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3692 | |
3693 | if (stripe_offset < bargs->pend && | |
3694 | stripe_offset + stripe_length > bargs->pstart) | |
3695 | return 0; | |
3696 | } | |
3697 | ||
3698 | return 1; | |
3699 | } | |
3700 | ||
ea67176a ID |
3701 | /* [vstart, vend) */ |
3702 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3703 | struct btrfs_chunk *chunk, | |
3704 | u64 chunk_offset, | |
3705 | struct btrfs_balance_args *bargs) | |
3706 | { | |
3707 | if (chunk_offset < bargs->vend && | |
3708 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3709 | /* at least part of the chunk is inside this vrange */ | |
3710 | return 0; | |
3711 | ||
3712 | return 1; | |
3713 | } | |
3714 | ||
dee32d0a GAP |
3715 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3716 | struct btrfs_chunk *chunk, | |
3717 | struct btrfs_balance_args *bargs) | |
3718 | { | |
3719 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3720 | ||
3721 | if (bargs->stripes_min <= num_stripes | |
3722 | && num_stripes <= bargs->stripes_max) | |
3723 | return 0; | |
3724 | ||
3725 | return 1; | |
3726 | } | |
3727 | ||
899c81ea | 3728 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3729 | struct btrfs_balance_args *bargs) |
3730 | { | |
3731 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3732 | return 0; | |
3733 | ||
899c81ea ID |
3734 | chunk_type = chunk_to_extended(chunk_type) & |
3735 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3736 | |
899c81ea | 3737 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3738 | return 1; |
3739 | ||
3740 | return 0; | |
3741 | } | |
3742 | ||
6ec0896c | 3743 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3744 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3745 | { | |
6ec0896c | 3746 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3747 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3748 | struct btrfs_balance_args *bargs = NULL; |
3749 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3750 | ||
3751 | /* type filter */ | |
3752 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3753 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3754 | return 0; | |
3755 | } | |
3756 | ||
3757 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3758 | bargs = &bctl->data; | |
3759 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3760 | bargs = &bctl->sys; | |
3761 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3762 | bargs = &bctl->meta; | |
3763 | ||
ed25e9b2 ID |
3764 | /* profiles filter */ |
3765 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3766 | chunk_profiles_filter(chunk_type, bargs)) { | |
3767 | return 0; | |
5ce5b3c0 ID |
3768 | } |
3769 | ||
3770 | /* usage filter */ | |
3771 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3772 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3773 | return 0; |
bc309467 | 3774 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3775 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3776 | return 0; |
409d404b ID |
3777 | } |
3778 | ||
3779 | /* devid filter */ | |
3780 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3781 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3782 | return 0; | |
94e60d5a ID |
3783 | } |
3784 | ||
3785 | /* drange filter, makes sense only with devid filter */ | |
3786 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3787 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3788 | return 0; |
ea67176a ID |
3789 | } |
3790 | ||
3791 | /* vrange filter */ | |
3792 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3793 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3794 | return 0; | |
ed25e9b2 ID |
3795 | } |
3796 | ||
dee32d0a GAP |
3797 | /* stripes filter */ |
3798 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3799 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3800 | return 0; | |
3801 | } | |
3802 | ||
cfa4c961 ID |
3803 | /* soft profile changing mode */ |
3804 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3805 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3806 | return 0; | |
3807 | } | |
3808 | ||
7d824b6f DS |
3809 | /* |
3810 | * limited by count, must be the last filter | |
3811 | */ | |
3812 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3813 | if (bargs->limit == 0) | |
3814 | return 0; | |
3815 | else | |
3816 | bargs->limit--; | |
12907fc7 DS |
3817 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3818 | /* | |
3819 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3820 | * determined here because we do not have the global information |
12907fc7 DS |
3821 | * about the count of all chunks that satisfy the filters. |
3822 | */ | |
3823 | if (bargs->limit_max == 0) | |
3824 | return 0; | |
3825 | else | |
3826 | bargs->limit_max--; | |
7d824b6f DS |
3827 | } |
3828 | ||
f43ffb60 ID |
3829 | return 1; |
3830 | } | |
3831 | ||
c9e9f97b | 3832 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3833 | { |
19a39dce | 3834 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3835 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3836 | u64 chunk_type; |
f43ffb60 | 3837 | struct btrfs_chunk *chunk; |
5a488b9d | 3838 | struct btrfs_path *path = NULL; |
ec44a35c | 3839 | struct btrfs_key key; |
ec44a35c | 3840 | struct btrfs_key found_key; |
f43ffb60 ID |
3841 | struct extent_buffer *leaf; |
3842 | int slot; | |
c9e9f97b ID |
3843 | int ret; |
3844 | int enospc_errors = 0; | |
19a39dce | 3845 | bool counting = true; |
12907fc7 | 3846 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3847 | u64 limit_data = bctl->data.limit; |
3848 | u64 limit_meta = bctl->meta.limit; | |
3849 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3850 | u32 count_data = 0; |
3851 | u32 count_meta = 0; | |
3852 | u32 count_sys = 0; | |
2c9fe835 | 3853 | int chunk_reserved = 0; |
ec44a35c | 3854 | |
ec44a35c | 3855 | path = btrfs_alloc_path(); |
17e9f796 MF |
3856 | if (!path) { |
3857 | ret = -ENOMEM; | |
3858 | goto error; | |
3859 | } | |
19a39dce ID |
3860 | |
3861 | /* zero out stat counters */ | |
3862 | spin_lock(&fs_info->balance_lock); | |
3863 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3864 | spin_unlock(&fs_info->balance_lock); | |
3865 | again: | |
7d824b6f | 3866 | if (!counting) { |
12907fc7 DS |
3867 | /* |
3868 | * The single value limit and min/max limits use the same bytes | |
3869 | * in the | |
3870 | */ | |
7d824b6f DS |
3871 | bctl->data.limit = limit_data; |
3872 | bctl->meta.limit = limit_meta; | |
3873 | bctl->sys.limit = limit_sys; | |
3874 | } | |
ec44a35c CM |
3875 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3876 | key.offset = (u64)-1; | |
3877 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3878 | ||
d397712b | 3879 | while (1) { |
19a39dce | 3880 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3881 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3882 | ret = -ECANCELED; |
3883 | goto error; | |
3884 | } | |
3885 | ||
f3372065 | 3886 | mutex_lock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3887 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3888 | if (ret < 0) { |
f3372065 | 3889 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3890 | goto error; |
67c5e7d4 | 3891 | } |
ec44a35c CM |
3892 | |
3893 | /* | |
3894 | * this shouldn't happen, it means the last relocate | |
3895 | * failed | |
3896 | */ | |
3897 | if (ret == 0) | |
c9e9f97b | 3898 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3899 | |
3900 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3901 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3902 | if (ret) { |
f3372065 | 3903 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
c9e9f97b | 3904 | ret = 0; |
ec44a35c | 3905 | break; |
c9e9f97b | 3906 | } |
7d9eb12c | 3907 | |
f43ffb60 ID |
3908 | leaf = path->nodes[0]; |
3909 | slot = path->slots[0]; | |
3910 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3911 | |
67c5e7d4 | 3912 | if (found_key.objectid != key.objectid) { |
f3372065 | 3913 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3914 | break; |
67c5e7d4 | 3915 | } |
7d9eb12c | 3916 | |
f43ffb60 | 3917 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3918 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3919 | |
19a39dce ID |
3920 | if (!counting) { |
3921 | spin_lock(&fs_info->balance_lock); | |
3922 | bctl->stat.considered++; | |
3923 | spin_unlock(&fs_info->balance_lock); | |
3924 | } | |
3925 | ||
6ec0896c | 3926 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3927 | |
b3b4aa74 | 3928 | btrfs_release_path(path); |
67c5e7d4 | 3929 | if (!ret) { |
f3372065 | 3930 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f43ffb60 | 3931 | goto loop; |
67c5e7d4 | 3932 | } |
f43ffb60 | 3933 | |
19a39dce | 3934 | if (counting) { |
f3372065 | 3935 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3936 | spin_lock(&fs_info->balance_lock); |
3937 | bctl->stat.expected++; | |
3938 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3939 | |
3940 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3941 | count_data++; | |
3942 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3943 | count_sys++; | |
3944 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3945 | count_meta++; | |
3946 | ||
3947 | goto loop; | |
3948 | } | |
3949 | ||
3950 | /* | |
3951 | * Apply limit_min filter, no need to check if the LIMITS | |
3952 | * filter is used, limit_min is 0 by default | |
3953 | */ | |
3954 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3955 | count_data < bctl->data.limit_min) | |
3956 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3957 | count_meta < bctl->meta.limit_min) | |
3958 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3959 | count_sys < bctl->sys.limit_min)) { | |
f3372065 | 3960 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3961 | goto loop; |
3962 | } | |
3963 | ||
a6f93c71 LB |
3964 | if (!chunk_reserved) { |
3965 | /* | |
3966 | * We may be relocating the only data chunk we have, | |
3967 | * which could potentially end up with losing data's | |
3968 | * raid profile, so lets allocate an empty one in | |
3969 | * advance. | |
3970 | */ | |
3971 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3972 | found_key.offset); | |
2c9fe835 | 3973 | if (ret < 0) { |
f3372065 | 3974 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2c9fe835 | 3975 | goto error; |
a6f93c71 LB |
3976 | } else if (ret == 1) { |
3977 | chunk_reserved = 1; | |
2c9fe835 | 3978 | } |
2c9fe835 ZL |
3979 | } |
3980 | ||
5b4aacef | 3981 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
f3372065 | 3982 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce | 3983 | if (ret == -ENOSPC) { |
c9e9f97b | 3984 | enospc_errors++; |
eede2bf3 OS |
3985 | } else if (ret == -ETXTBSY) { |
3986 | btrfs_info(fs_info, | |
3987 | "skipping relocation of block group %llu due to active swapfile", | |
3988 | found_key.offset); | |
3989 | ret = 0; | |
3990 | } else if (ret) { | |
3991 | goto error; | |
19a39dce ID |
3992 | } else { |
3993 | spin_lock(&fs_info->balance_lock); | |
3994 | bctl->stat.completed++; | |
3995 | spin_unlock(&fs_info->balance_lock); | |
3996 | } | |
f43ffb60 | 3997 | loop: |
795a3321 ID |
3998 | if (found_key.offset == 0) |
3999 | break; | |
ba1bf481 | 4000 | key.offset = found_key.offset - 1; |
ec44a35c | 4001 | } |
c9e9f97b | 4002 | |
19a39dce ID |
4003 | if (counting) { |
4004 | btrfs_release_path(path); | |
4005 | counting = false; | |
4006 | goto again; | |
4007 | } | |
ec44a35c CM |
4008 | error: |
4009 | btrfs_free_path(path); | |
c9e9f97b | 4010 | if (enospc_errors) { |
efe120a0 | 4011 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 4012 | enospc_errors); |
c9e9f97b ID |
4013 | if (!ret) |
4014 | ret = -ENOSPC; | |
4015 | } | |
4016 | ||
ec44a35c CM |
4017 | return ret; |
4018 | } | |
4019 | ||
43dd529a DS |
4020 | /* |
4021 | * See if a given profile is valid and reduced. | |
4022 | * | |
4023 | * @flags: profile to validate | |
4024 | * @extended: if true @flags is treated as an extended profile | |
0c460c0d ID |
4025 | */ |
4026 | static int alloc_profile_is_valid(u64 flags, int extended) | |
4027 | { | |
4028 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
4029 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
4030 | ||
4031 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4032 | ||
4033 | /* 1) check that all other bits are zeroed */ | |
4034 | if (flags & ~mask) | |
4035 | return 0; | |
4036 | ||
4037 | /* 2) see if profile is reduced */ | |
4038 | if (flags == 0) | |
4039 | return !extended; /* "0" is valid for usual profiles */ | |
4040 | ||
c1499166 | 4041 | return has_single_bit_set(flags); |
0c460c0d ID |
4042 | } |
4043 | ||
837d5b6e ID |
4044 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
4045 | { | |
a7e99c69 ID |
4046 | /* cancel requested || normal exit path */ |
4047 | return atomic_read(&fs_info->balance_cancel_req) || | |
4048 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
4049 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
4050 | } |
4051 | ||
5ba366c3 DS |
4052 | /* |
4053 | * Validate target profile against allowed profiles and return true if it's OK. | |
4054 | * Otherwise print the error message and return false. | |
4055 | */ | |
4056 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
4057 | const struct btrfs_balance_args *bargs, | |
4058 | u64 allowed, const char *type) | |
bdcd3c97 | 4059 | { |
5ba366c3 DS |
4060 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
4061 | return true; | |
4062 | ||
4063 | /* Profile is valid and does not have bits outside of the allowed set */ | |
4064 | if (alloc_profile_is_valid(bargs->target, 1) && | |
4065 | (bargs->target & ~allowed) == 0) | |
4066 | return true; | |
4067 | ||
4068 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
4069 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
4070 | return false; | |
bdcd3c97 AM |
4071 | } |
4072 | ||
56fc37d9 AJ |
4073 | /* |
4074 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
4075 | * @size_buf including the terminating null. The output may be trimmed if it | |
4076 | * does not fit into the provided buffer. | |
4077 | */ | |
4078 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
4079 | u32 size_buf) | |
4080 | { | |
4081 | int ret; | |
4082 | u32 size_bp = size_buf; | |
4083 | char *bp = buf; | |
4084 | u64 flags = bargs->flags; | |
4085 | char tmp_buf[128] = {'\0'}; | |
4086 | ||
4087 | if (!flags) | |
4088 | return; | |
4089 | ||
4090 | #define CHECK_APPEND_NOARG(a) \ | |
4091 | do { \ | |
4092 | ret = snprintf(bp, size_bp, (a)); \ | |
4093 | if (ret < 0 || ret >= size_bp) \ | |
4094 | goto out_overflow; \ | |
4095 | size_bp -= ret; \ | |
4096 | bp += ret; \ | |
4097 | } while (0) | |
4098 | ||
4099 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4100 | do { \ | |
4101 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4102 | if (ret < 0 || ret >= size_bp) \ | |
4103 | goto out_overflow; \ | |
4104 | size_bp -= ret; \ | |
4105 | bp += ret; \ | |
4106 | } while (0) | |
4107 | ||
4108 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
4109 | do { \ | |
4110 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
4111 | if (ret < 0 || ret >= size_bp) \ | |
4112 | goto out_overflow; \ | |
4113 | size_bp -= ret; \ | |
4114 | bp += ret; \ | |
4115 | } while (0) | |
4116 | ||
158da513 DS |
4117 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
4118 | CHECK_APPEND_1ARG("convert=%s,", | |
4119 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
4120 | |
4121 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
4122 | CHECK_APPEND_NOARG("soft,"); | |
4123 | ||
4124 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
4125 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
4126 | sizeof(tmp_buf)); | |
4127 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
4128 | } | |
4129 | ||
4130 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
4131 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
4132 | ||
4133 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
4134 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
4135 | bargs->usage_min, bargs->usage_max); | |
4136 | ||
4137 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
4138 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
4139 | ||
4140 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4141 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4142 | bargs->pstart, bargs->pend); | |
4143 | ||
4144 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4145 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4146 | bargs->vstart, bargs->vend); | |
4147 | ||
4148 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4149 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4150 | ||
4151 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4152 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4153 | bargs->limit_min, bargs->limit_max); | |
4154 | ||
4155 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4156 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4157 | bargs->stripes_min, bargs->stripes_max); | |
4158 | ||
4159 | #undef CHECK_APPEND_2ARG | |
4160 | #undef CHECK_APPEND_1ARG | |
4161 | #undef CHECK_APPEND_NOARG | |
4162 | ||
4163 | out_overflow: | |
4164 | ||
4165 | if (size_bp < size_buf) | |
4166 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4167 | else | |
4168 | buf[0] = '\0'; | |
4169 | } | |
4170 | ||
4171 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4172 | { | |
4173 | u32 size_buf = 1024; | |
4174 | char tmp_buf[192] = {'\0'}; | |
4175 | char *buf; | |
4176 | char *bp; | |
4177 | u32 size_bp = size_buf; | |
4178 | int ret; | |
4179 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4180 | ||
4181 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4182 | if (!buf) | |
4183 | return; | |
4184 | ||
4185 | bp = buf; | |
4186 | ||
4187 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4188 | do { \ | |
4189 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4190 | if (ret < 0 || ret >= size_bp) \ | |
4191 | goto out_overflow; \ | |
4192 | size_bp -= ret; \ | |
4193 | bp += ret; \ | |
4194 | } while (0) | |
4195 | ||
4196 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4197 | CHECK_APPEND_1ARG("%s", "-f "); | |
4198 | ||
4199 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4200 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4201 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4202 | } | |
4203 | ||
4204 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4205 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4206 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4207 | } | |
4208 | ||
4209 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4210 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4211 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4212 | } | |
4213 | ||
4214 | #undef CHECK_APPEND_1ARG | |
4215 | ||
4216 | out_overflow: | |
4217 | ||
4218 | if (size_bp < size_buf) | |
4219 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4220 | btrfs_info(fs_info, "balance: %s %s", | |
4221 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4222 | "resume" : "start", buf); | |
4223 | ||
4224 | kfree(buf); | |
4225 | } | |
4226 | ||
c9e9f97b | 4227 | /* |
dccdb07b | 4228 | * Should be called with balance mutexe held |
c9e9f97b | 4229 | */ |
6fcf6e2b DS |
4230 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4231 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4232 | struct btrfs_ioctl_balance_args *bargs) |
4233 | { | |
14506127 | 4234 | u64 meta_target, data_target; |
f43ffb60 | 4235 | u64 allowed; |
e4837f8f | 4236 | int mixed = 0; |
c9e9f97b | 4237 | int ret; |
8dabb742 | 4238 | u64 num_devices; |
de98ced9 | 4239 | unsigned seq; |
e62869be | 4240 | bool reducing_redundancy; |
081db89b | 4241 | int i; |
c9e9f97b | 4242 | |
837d5b6e | 4243 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4244 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4245 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4246 | ret = -EINVAL; |
4247 | goto out; | |
4248 | } | |
4249 | ||
e4837f8f ID |
4250 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4251 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4252 | mixed = 1; | |
4253 | ||
f43ffb60 ID |
4254 | /* |
4255 | * In case of mixed groups both data and meta should be picked, | |
4256 | * and identical options should be given for both of them. | |
4257 | */ | |
e4837f8f ID |
4258 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4259 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4260 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4261 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4262 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4263 | btrfs_err(fs_info, |
6dac13f8 | 4264 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4265 | ret = -EINVAL; |
4266 | goto out; | |
4267 | } | |
4268 | } | |
4269 | ||
b35cf1f0 JB |
4270 | /* |
4271 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4272 | * are exclusive |
b35cf1f0 JB |
4273 | */ |
4274 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4275 | |
4276 | /* | |
4277 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4278 | * special bit for it, to make it easier to distinguish. Thus we need | |
4279 | * to set it manually, or balance would refuse the profile. | |
4280 | */ | |
4281 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4282 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4283 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4284 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4285 | |
5ba366c3 DS |
4286 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4287 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4288 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4289 | ret = -EINVAL; |
4290 | goto out; | |
4291 | } | |
4292 | ||
6079e12c DS |
4293 | /* |
4294 | * Allow to reduce metadata or system integrity only if force set for | |
4295 | * profiles with redundancy (copies, parity) | |
4296 | */ | |
4297 | allowed = 0; | |
4298 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4299 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4300 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4301 | allowed |= btrfs_raid_array[i].bg_flag; | |
4302 | } | |
de98ced9 MX |
4303 | do { |
4304 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4305 | ||
4306 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4307 | (fs_info->avail_system_alloc_bits & allowed) && | |
4308 | !(bctl->sys.target & allowed)) || | |
4309 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4310 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4311 | !(bctl->meta.target & allowed))) |
e62869be | 4312 | reducing_redundancy = true; |
5a8067c0 | 4313 | else |
e62869be | 4314 | reducing_redundancy = false; |
5a8067c0 FM |
4315 | |
4316 | /* if we're not converting, the target field is uninitialized */ | |
4317 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4318 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4319 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4320 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4321 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4322 | |
e62869be | 4323 | if (reducing_redundancy) { |
5a8067c0 FM |
4324 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4325 | btrfs_info(fs_info, | |
e62869be | 4326 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4327 | } else { |
4328 | btrfs_err(fs_info, | |
e62869be | 4329 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4330 | ret = -EINVAL; |
4331 | goto out; | |
4332 | } | |
4333 | } | |
4334 | ||
14506127 AB |
4335 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4336 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4337 | btrfs_warn(fs_info, |
6dac13f8 | 4338 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4339 | btrfs_bg_type_to_raid_name(meta_target), |
4340 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4341 | } |
4342 | ||
6bccf3ab | 4343 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4344 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4345 | goto out; |
4346 | ||
59641015 ID |
4347 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4348 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4349 | BUG_ON(fs_info->balance_ctl); |
4350 | spin_lock(&fs_info->balance_lock); | |
4351 | fs_info->balance_ctl = bctl; | |
4352 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4353 | } else { |
4354 | BUG_ON(ret != -EEXIST); | |
4355 | spin_lock(&fs_info->balance_lock); | |
4356 | update_balance_args(bctl); | |
4357 | spin_unlock(&fs_info->balance_lock); | |
4358 | } | |
c9e9f97b | 4359 | |
3009a62f DS |
4360 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4361 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4362 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4363 | mutex_unlock(&fs_info->balance_mutex); |
4364 | ||
4365 | ret = __btrfs_balance(fs_info); | |
4366 | ||
4367 | mutex_lock(&fs_info->balance_mutex); | |
efc0e69c | 4368 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { |
7333bd02 | 4369 | btrfs_info(fs_info, "balance: paused"); |
efc0e69c NB |
4370 | btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); |
4371 | } | |
44d354ab QW |
4372 | /* |
4373 | * Balance can be canceled by: | |
4374 | * | |
4375 | * - Regular cancel request | |
4376 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4377 | * | |
4378 | * - Fatal signal to "btrfs" process | |
4379 | * Either the signal caught by wait_reserve_ticket() and callers | |
4380 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4381 | * got -ECANCELED. | |
4382 | * Either way, in this case balance_cancel_req = 0, and | |
4383 | * ret == -EINTR or ret == -ECANCELED. | |
4384 | * | |
4385 | * So here we only check the return value to catch canceled balance. | |
4386 | */ | |
4387 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4388 | btrfs_info(fs_info, "balance: canceled"); |
4389 | else | |
4390 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4391 | ||
3009a62f | 4392 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4393 | |
4394 | if (bargs) { | |
4395 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4396 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4397 | } |
4398 | ||
3a01aa7a ID |
4399 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4400 | balance_need_close(fs_info)) { | |
149196a2 | 4401 | reset_balance_state(fs_info); |
c3e1f96c | 4402 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4403 | } |
4404 | ||
837d5b6e | 4405 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4406 | |
4407 | return ret; | |
4408 | out: | |
59641015 | 4409 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4410 | reset_balance_state(fs_info); |
a17c95df | 4411 | else |
59641015 | 4412 | kfree(bctl); |
c3e1f96c | 4413 | btrfs_exclop_finish(fs_info); |
a17c95df | 4414 | |
59641015 ID |
4415 | return ret; |
4416 | } | |
4417 | ||
4418 | static int balance_kthread(void *data) | |
4419 | { | |
2b6ba629 | 4420 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4421 | int ret = 0; |
59641015 | 4422 | |
a690e5f2 | 4423 | sb_start_write(fs_info->sb); |
59641015 | 4424 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4425 | if (fs_info->balance_ctl) |
6fcf6e2b | 4426 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4427 | mutex_unlock(&fs_info->balance_mutex); |
a690e5f2 | 4428 | sb_end_write(fs_info->sb); |
2b6ba629 | 4429 | |
59641015 ID |
4430 | return ret; |
4431 | } | |
4432 | ||
2b6ba629 ID |
4433 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4434 | { | |
4435 | struct task_struct *tsk; | |
4436 | ||
1354e1a1 | 4437 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4438 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4439 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4440 | return 0; |
4441 | } | |
1354e1a1 | 4442 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4443 | |
3cdde224 | 4444 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4445 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4446 | return 0; |
4447 | } | |
4448 | ||
efc0e69c NB |
4449 | spin_lock(&fs_info->super_lock); |
4450 | ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); | |
4451 | fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; | |
4452 | spin_unlock(&fs_info->super_lock); | |
02ee654d AJ |
4453 | /* |
4454 | * A ro->rw remount sequence should continue with the paused balance | |
4455 | * regardless of who pauses it, system or the user as of now, so set | |
4456 | * the resume flag. | |
4457 | */ | |
4458 | spin_lock(&fs_info->balance_lock); | |
4459 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4460 | spin_unlock(&fs_info->balance_lock); | |
4461 | ||
2b6ba629 | 4462 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4463 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4464 | } |
4465 | ||
68310a5e | 4466 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4467 | { |
59641015 ID |
4468 | struct btrfs_balance_control *bctl; |
4469 | struct btrfs_balance_item *item; | |
4470 | struct btrfs_disk_balance_args disk_bargs; | |
4471 | struct btrfs_path *path; | |
4472 | struct extent_buffer *leaf; | |
4473 | struct btrfs_key key; | |
4474 | int ret; | |
4475 | ||
4476 | path = btrfs_alloc_path(); | |
4477 | if (!path) | |
4478 | return -ENOMEM; | |
4479 | ||
59641015 | 4480 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4481 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4482 | key.offset = 0; |
4483 | ||
68310a5e | 4484 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4485 | if (ret < 0) |
68310a5e | 4486 | goto out; |
59641015 ID |
4487 | if (ret > 0) { /* ret = -ENOENT; */ |
4488 | ret = 0; | |
68310a5e ID |
4489 | goto out; |
4490 | } | |
4491 | ||
4492 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4493 | if (!bctl) { | |
4494 | ret = -ENOMEM; | |
4495 | goto out; | |
59641015 ID |
4496 | } |
4497 | ||
4498 | leaf = path->nodes[0]; | |
4499 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4500 | ||
68310a5e ID |
4501 | bctl->flags = btrfs_balance_flags(leaf, item); |
4502 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4503 | |
4504 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4505 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4506 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4507 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4508 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4509 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4510 | ||
eee95e3f DS |
4511 | /* |
4512 | * This should never happen, as the paused balance state is recovered | |
4513 | * during mount without any chance of other exclusive ops to collide. | |
4514 | * | |
4515 | * This gives the exclusive op status to balance and keeps in paused | |
4516 | * state until user intervention (cancel or umount). If the ownership | |
4517 | * cannot be assigned, show a message but do not fail. The balance | |
4518 | * is in a paused state and must have fs_info::balance_ctl properly | |
4519 | * set up. | |
4520 | */ | |
efc0e69c | 4521 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) |
eee95e3f | 4522 | btrfs_warn(fs_info, |
6dac13f8 | 4523 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4524 | |
fb286100 JB |
4525 | btrfs_release_path(path); |
4526 | ||
68310a5e | 4527 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4528 | BUG_ON(fs_info->balance_ctl); |
4529 | spin_lock(&fs_info->balance_lock); | |
4530 | fs_info->balance_ctl = bctl; | |
4531 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4532 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4533 | out: |
4534 | btrfs_free_path(path); | |
ec44a35c CM |
4535 | return ret; |
4536 | } | |
4537 | ||
837d5b6e ID |
4538 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4539 | { | |
4540 | int ret = 0; | |
4541 | ||
4542 | mutex_lock(&fs_info->balance_mutex); | |
4543 | if (!fs_info->balance_ctl) { | |
4544 | mutex_unlock(&fs_info->balance_mutex); | |
4545 | return -ENOTCONN; | |
4546 | } | |
4547 | ||
3009a62f | 4548 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4549 | atomic_inc(&fs_info->balance_pause_req); |
4550 | mutex_unlock(&fs_info->balance_mutex); | |
4551 | ||
4552 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4553 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4554 | |
4555 | mutex_lock(&fs_info->balance_mutex); | |
4556 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4557 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4558 | atomic_dec(&fs_info->balance_pause_req); |
4559 | } else { | |
4560 | ret = -ENOTCONN; | |
4561 | } | |
4562 | ||
4563 | mutex_unlock(&fs_info->balance_mutex); | |
4564 | return ret; | |
4565 | } | |
4566 | ||
a7e99c69 ID |
4567 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4568 | { | |
4569 | mutex_lock(&fs_info->balance_mutex); | |
4570 | if (!fs_info->balance_ctl) { | |
4571 | mutex_unlock(&fs_info->balance_mutex); | |
4572 | return -ENOTCONN; | |
4573 | } | |
4574 | ||
cf7d20f4 DS |
4575 | /* |
4576 | * A paused balance with the item stored on disk can be resumed at | |
4577 | * mount time if the mount is read-write. Otherwise it's still paused | |
4578 | * and we must not allow cancelling as it deletes the item. | |
4579 | */ | |
4580 | if (sb_rdonly(fs_info->sb)) { | |
4581 | mutex_unlock(&fs_info->balance_mutex); | |
4582 | return -EROFS; | |
4583 | } | |
4584 | ||
a7e99c69 ID |
4585 | atomic_inc(&fs_info->balance_cancel_req); |
4586 | /* | |
4587 | * if we are running just wait and return, balance item is | |
4588 | * deleted in btrfs_balance in this case | |
4589 | */ | |
3009a62f | 4590 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4591 | mutex_unlock(&fs_info->balance_mutex); |
4592 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4593 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4594 | mutex_lock(&fs_info->balance_mutex); |
4595 | } else { | |
a7e99c69 | 4596 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4597 | /* |
4598 | * Lock released to allow other waiters to continue, we'll | |
4599 | * reexamine the status again. | |
4600 | */ | |
a7e99c69 ID |
4601 | mutex_lock(&fs_info->balance_mutex); |
4602 | ||
a17c95df | 4603 | if (fs_info->balance_ctl) { |
149196a2 | 4604 | reset_balance_state(fs_info); |
c3e1f96c | 4605 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4606 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4607 | } |
a7e99c69 ID |
4608 | } |
4609 | ||
3009a62f DS |
4610 | BUG_ON(fs_info->balance_ctl || |
4611 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4612 | atomic_dec(&fs_info->balance_cancel_req); |
4613 | mutex_unlock(&fs_info->balance_mutex); | |
4614 | return 0; | |
4615 | } | |
4616 | ||
97f4dd09 | 4617 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4618 | { |
4619 | struct btrfs_fs_info *fs_info = data; | |
4620 | struct btrfs_root *root = fs_info->tree_root; | |
4621 | struct btrfs_key key; | |
803b2f54 SB |
4622 | struct btrfs_path *path = NULL; |
4623 | int ret = 0; | |
4624 | struct extent_buffer *eb; | |
4625 | int slot; | |
4626 | struct btrfs_root_item root_item; | |
4627 | u32 item_size; | |
f45388f3 | 4628 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4629 | bool closing = false; |
803b2f54 SB |
4630 | |
4631 | path = btrfs_alloc_path(); | |
4632 | if (!path) { | |
4633 | ret = -ENOMEM; | |
4634 | goto out; | |
4635 | } | |
4636 | ||
4637 | key.objectid = 0; | |
4638 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4639 | key.offset = 0; | |
4640 | ||
803b2f54 | 4641 | while (1) { |
c94bec2c JB |
4642 | if (btrfs_fs_closing(fs_info)) { |
4643 | closing = true; | |
4644 | break; | |
4645 | } | |
7c829b72 AJ |
4646 | ret = btrfs_search_forward(root, &key, path, |
4647 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4648 | if (ret) { |
4649 | if (ret > 0) | |
4650 | ret = 0; | |
4651 | break; | |
4652 | } | |
4653 | ||
4654 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4655 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4656 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4657 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4658 | goto skip; | |
4659 | ||
4660 | eb = path->nodes[0]; | |
4661 | slot = path->slots[0]; | |
3212fa14 | 4662 | item_size = btrfs_item_size(eb, slot); |
803b2f54 SB |
4663 | if (item_size < sizeof(root_item)) |
4664 | goto skip; | |
4665 | ||
803b2f54 SB |
4666 | read_extent_buffer(eb, &root_item, |
4667 | btrfs_item_ptr_offset(eb, slot), | |
4668 | (int)sizeof(root_item)); | |
4669 | if (btrfs_root_refs(&root_item) == 0) | |
4670 | goto skip; | |
f45388f3 FDBM |
4671 | |
4672 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4673 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4674 | if (trans) | |
4675 | goto update_tree; | |
4676 | ||
4677 | btrfs_release_path(path); | |
803b2f54 SB |
4678 | /* |
4679 | * 1 - subvol uuid item | |
4680 | * 1 - received_subvol uuid item | |
4681 | */ | |
4682 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4683 | if (IS_ERR(trans)) { | |
4684 | ret = PTR_ERR(trans); | |
4685 | break; | |
4686 | } | |
f45388f3 FDBM |
4687 | continue; |
4688 | } else { | |
4689 | goto skip; | |
4690 | } | |
4691 | update_tree: | |
9771a5cf | 4692 | btrfs_release_path(path); |
f45388f3 | 4693 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4694 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4695 | BTRFS_UUID_KEY_SUBVOL, |
4696 | key.objectid); | |
4697 | if (ret < 0) { | |
efe120a0 | 4698 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4699 | ret); |
803b2f54 SB |
4700 | break; |
4701 | } | |
4702 | } | |
4703 | ||
4704 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4705 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4706 | root_item.received_uuid, |
4707 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4708 | key.objectid); | |
4709 | if (ret < 0) { | |
efe120a0 | 4710 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4711 | ret); |
803b2f54 SB |
4712 | break; |
4713 | } | |
4714 | } | |
4715 | ||
f45388f3 | 4716 | skip: |
9771a5cf | 4717 | btrfs_release_path(path); |
803b2f54 | 4718 | if (trans) { |
3a45bb20 | 4719 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4720 | trans = NULL; |
803b2f54 SB |
4721 | if (ret) |
4722 | break; | |
4723 | } | |
4724 | ||
803b2f54 SB |
4725 | if (key.offset < (u64)-1) { |
4726 | key.offset++; | |
4727 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4728 | key.offset = 0; | |
4729 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4730 | } else if (key.objectid < (u64)-1) { | |
4731 | key.offset = 0; | |
4732 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4733 | key.objectid++; | |
4734 | } else { | |
4735 | break; | |
4736 | } | |
4737 | cond_resched(); | |
4738 | } | |
4739 | ||
4740 | out: | |
4741 | btrfs_free_path(path); | |
f45388f3 | 4742 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4743 | btrfs_end_transaction(trans); |
803b2f54 | 4744 | if (ret) |
efe120a0 | 4745 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4746 | else if (!closing) |
afcdd129 | 4747 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4748 | up(&fs_info->uuid_tree_rescan_sem); |
4749 | return 0; | |
4750 | } | |
4751 | ||
f7a81ea4 SB |
4752 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4753 | { | |
4754 | struct btrfs_trans_handle *trans; | |
4755 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4756 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4757 | struct task_struct *task; |
4758 | int ret; | |
f7a81ea4 SB |
4759 | |
4760 | /* | |
4761 | * 1 - root node | |
4762 | * 1 - root item | |
4763 | */ | |
4764 | trans = btrfs_start_transaction(tree_root, 2); | |
4765 | if (IS_ERR(trans)) | |
4766 | return PTR_ERR(trans); | |
4767 | ||
9b7a2440 | 4768 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4769 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4770 | ret = PTR_ERR(uuid_root); |
66642832 | 4771 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4772 | btrfs_end_transaction(trans); |
6d13f549 | 4773 | return ret; |
f7a81ea4 SB |
4774 | } |
4775 | ||
4776 | fs_info->uuid_root = uuid_root; | |
4777 | ||
3a45bb20 | 4778 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4779 | if (ret) |
4780 | return ret; | |
4781 | ||
4782 | down(&fs_info->uuid_tree_rescan_sem); | |
4783 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4784 | if (IS_ERR(task)) { | |
70f80175 | 4785 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4786 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4787 | up(&fs_info->uuid_tree_rescan_sem); |
4788 | return PTR_ERR(task); | |
4789 | } | |
4790 | ||
4791 | return 0; | |
f7a81ea4 | 4792 | } |
803b2f54 | 4793 | |
8f18cf13 CM |
4794 | /* |
4795 | * shrinking a device means finding all of the device extents past | |
4796 | * the new size, and then following the back refs to the chunks. | |
4797 | * The chunk relocation code actually frees the device extent | |
4798 | */ | |
4799 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4800 | { | |
0b246afa JM |
4801 | struct btrfs_fs_info *fs_info = device->fs_info; |
4802 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4803 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4804 | struct btrfs_dev_extent *dev_extent = NULL; |
4805 | struct btrfs_path *path; | |
4806 | u64 length; | |
8f18cf13 CM |
4807 | u64 chunk_offset; |
4808 | int ret; | |
4809 | int slot; | |
ba1bf481 JB |
4810 | int failed = 0; |
4811 | bool retried = false; | |
8f18cf13 CM |
4812 | struct extent_buffer *l; |
4813 | struct btrfs_key key; | |
0b246afa | 4814 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4815 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4816 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4817 | u64 diff; |
61d0d0d2 | 4818 | u64 start; |
7dfb8be1 NB |
4819 | |
4820 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4821 | start = new_size; |
0e4324a4 | 4822 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4823 | |
401e29c1 | 4824 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4825 | return -EINVAL; |
4826 | ||
8f18cf13 CM |
4827 | path = btrfs_alloc_path(); |
4828 | if (!path) | |
4829 | return -ENOMEM; | |
4830 | ||
0338dff6 | 4831 | path->reada = READA_BACK; |
8f18cf13 | 4832 | |
61d0d0d2 NB |
4833 | trans = btrfs_start_transaction(root, 0); |
4834 | if (IS_ERR(trans)) { | |
4835 | btrfs_free_path(path); | |
4836 | return PTR_ERR(trans); | |
4837 | } | |
4838 | ||
34441361 | 4839 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4840 | |
7cc8e58d | 4841 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4842 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4843 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4844 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4845 | } |
61d0d0d2 NB |
4846 | |
4847 | /* | |
4848 | * Once the device's size has been set to the new size, ensure all | |
4849 | * in-memory chunks are synced to disk so that the loop below sees them | |
4850 | * and relocates them accordingly. | |
4851 | */ | |
1c11b63e | 4852 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4853 | mutex_unlock(&fs_info->chunk_mutex); |
4854 | ret = btrfs_commit_transaction(trans); | |
4855 | if (ret) | |
4856 | goto done; | |
4857 | } else { | |
4858 | mutex_unlock(&fs_info->chunk_mutex); | |
4859 | btrfs_end_transaction(trans); | |
4860 | } | |
8f18cf13 | 4861 | |
ba1bf481 | 4862 | again: |
8f18cf13 CM |
4863 | key.objectid = device->devid; |
4864 | key.offset = (u64)-1; | |
4865 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4866 | ||
213e64da | 4867 | do { |
f3372065 | 4868 | mutex_lock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4869 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4870 | if (ret < 0) { |
f3372065 | 4871 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4872 | goto done; |
67c5e7d4 | 4873 | } |
8f18cf13 CM |
4874 | |
4875 | ret = btrfs_previous_item(root, path, 0, key.type); | |
8f18cf13 | 4876 | if (ret) { |
f3372065 | 4877 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
7056bf69 NB |
4878 | if (ret < 0) |
4879 | goto done; | |
8f18cf13 | 4880 | ret = 0; |
b3b4aa74 | 4881 | btrfs_release_path(path); |
bf1fb512 | 4882 | break; |
8f18cf13 CM |
4883 | } |
4884 | ||
4885 | l = path->nodes[0]; | |
4886 | slot = path->slots[0]; | |
4887 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4888 | ||
ba1bf481 | 4889 | if (key.objectid != device->devid) { |
f3372065 | 4890 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4891 | btrfs_release_path(path); |
bf1fb512 | 4892 | break; |
ba1bf481 | 4893 | } |
8f18cf13 CM |
4894 | |
4895 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4896 | length = btrfs_dev_extent_length(l, dev_extent); | |
4897 | ||
ba1bf481 | 4898 | if (key.offset + length <= new_size) { |
f3372065 | 4899 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4900 | btrfs_release_path(path); |
d6397bae | 4901 | break; |
ba1bf481 | 4902 | } |
8f18cf13 | 4903 | |
8f18cf13 | 4904 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4905 | btrfs_release_path(path); |
8f18cf13 | 4906 | |
a6f93c71 LB |
4907 | /* |
4908 | * We may be relocating the only data chunk we have, | |
4909 | * which could potentially end up with losing data's | |
4910 | * raid profile, so lets allocate an empty one in | |
4911 | * advance. | |
4912 | */ | |
4913 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4914 | if (ret < 0) { | |
f3372065 | 4915 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
a6f93c71 LB |
4916 | goto done; |
4917 | } | |
4918 | ||
0b246afa | 4919 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
f3372065 | 4920 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
eede2bf3 | 4921 | if (ret == -ENOSPC) { |
ba1bf481 | 4922 | failed++; |
eede2bf3 OS |
4923 | } else if (ret) { |
4924 | if (ret == -ETXTBSY) { | |
4925 | btrfs_warn(fs_info, | |
4926 | "could not shrink block group %llu due to active swapfile", | |
4927 | chunk_offset); | |
4928 | } | |
4929 | goto done; | |
4930 | } | |
213e64da | 4931 | } while (key.offset-- > 0); |
ba1bf481 JB |
4932 | |
4933 | if (failed && !retried) { | |
4934 | failed = 0; | |
4935 | retried = true; | |
4936 | goto again; | |
4937 | } else if (failed && retried) { | |
4938 | ret = -ENOSPC; | |
ba1bf481 | 4939 | goto done; |
8f18cf13 CM |
4940 | } |
4941 | ||
d6397bae | 4942 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4943 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4944 | if (IS_ERR(trans)) { |
4945 | ret = PTR_ERR(trans); | |
4946 | goto done; | |
4947 | } | |
4948 | ||
34441361 | 4949 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4950 | /* Clear all state bits beyond the shrunk device size */ |
4951 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4952 | CHUNK_STATE_MASK); | |
4953 | ||
7cc8e58d | 4954 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4955 | if (list_empty(&device->post_commit_list)) |
4956 | list_add_tail(&device->post_commit_list, | |
4957 | &trans->transaction->dev_update_list); | |
d6397bae | 4958 | |
d6397bae | 4959 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4960 | btrfs_set_super_total_bytes(super_copy, |
4961 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4962 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 4963 | |
2bb2e00e | 4964 | btrfs_reserve_chunk_metadata(trans, false); |
2196d6e8 MX |
4965 | /* Now btrfs_update_device() will change the on-disk size. */ |
4966 | ret = btrfs_update_device(trans, device); | |
2bb2e00e | 4967 | btrfs_trans_release_chunk_metadata(trans); |
801660b0 AJ |
4968 | if (ret < 0) { |
4969 | btrfs_abort_transaction(trans, ret); | |
4970 | btrfs_end_transaction(trans); | |
4971 | } else { | |
4972 | ret = btrfs_commit_transaction(trans); | |
4973 | } | |
8f18cf13 CM |
4974 | done: |
4975 | btrfs_free_path(path); | |
53e489bc | 4976 | if (ret) { |
34441361 | 4977 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 4978 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 4979 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 4980 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 4981 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 4982 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 4983 | } |
8f18cf13 CM |
4984 | return ret; |
4985 | } | |
4986 | ||
2ff7e61e | 4987 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
4988 | struct btrfs_key *key, |
4989 | struct btrfs_chunk *chunk, int item_size) | |
4990 | { | |
0b246afa | 4991 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
4992 | struct btrfs_disk_key disk_key; |
4993 | u32 array_size; | |
4994 | u8 *ptr; | |
4995 | ||
79bd3712 FM |
4996 | lockdep_assert_held(&fs_info->chunk_mutex); |
4997 | ||
0b86a832 | 4998 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4999 | if (array_size + item_size + sizeof(disk_key) |
79bd3712 | 5000 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) |
0b86a832 CM |
5001 | return -EFBIG; |
5002 | ||
5003 | ptr = super_copy->sys_chunk_array + array_size; | |
5004 | btrfs_cpu_key_to_disk(&disk_key, key); | |
5005 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
5006 | ptr += sizeof(disk_key); | |
5007 | memcpy(ptr, chunk, item_size); | |
5008 | item_size += sizeof(disk_key); | |
5009 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 | 5010 | |
0b86a832 CM |
5011 | return 0; |
5012 | } | |
5013 | ||
73c5de00 AJ |
5014 | /* |
5015 | * sort the devices in descending order by max_avail, total_avail | |
5016 | */ | |
5017 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 5018 | { |
73c5de00 AJ |
5019 | const struct btrfs_device_info *di_a = a; |
5020 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 5021 | |
73c5de00 | 5022 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 5023 | return -1; |
73c5de00 | 5024 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 5025 | return 1; |
73c5de00 AJ |
5026 | if (di_a->total_avail > di_b->total_avail) |
5027 | return -1; | |
5028 | if (di_a->total_avail < di_b->total_avail) | |
5029 | return 1; | |
5030 | return 0; | |
b2117a39 | 5031 | } |
0b86a832 | 5032 | |
53b381b3 DW |
5033 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5034 | { | |
ffe2d203 | 5035 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
5036 | return; |
5037 | ||
ceda0864 | 5038 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
5039 | } |
5040 | ||
cfbb825c DS |
5041 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5042 | { | |
5043 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
5044 | return; | |
5045 | ||
5046 | btrfs_set_fs_incompat(info, RAID1C34); | |
5047 | } | |
5048 | ||
4f2bafe8 | 5049 | /* |
f6f39f7a | 5050 | * Structure used internally for btrfs_create_chunk() function. |
4f2bafe8 NA |
5051 | * Wraps needed parameters. |
5052 | */ | |
5053 | struct alloc_chunk_ctl { | |
5054 | u64 start; | |
5055 | u64 type; | |
5056 | /* Total number of stripes to allocate */ | |
5057 | int num_stripes; | |
5058 | /* sub_stripes info for map */ | |
5059 | int sub_stripes; | |
5060 | /* Stripes per device */ | |
5061 | int dev_stripes; | |
5062 | /* Maximum number of devices to use */ | |
5063 | int devs_max; | |
5064 | /* Minimum number of devices to use */ | |
5065 | int devs_min; | |
5066 | /* ndevs has to be a multiple of this */ | |
5067 | int devs_increment; | |
5068 | /* Number of copies */ | |
5069 | int ncopies; | |
5070 | /* Number of stripes worth of bytes to store parity information */ | |
5071 | int nparity; | |
5072 | u64 max_stripe_size; | |
5073 | u64 max_chunk_size; | |
6aafb303 | 5074 | u64 dev_extent_min; |
4f2bafe8 NA |
5075 | u64 stripe_size; |
5076 | u64 chunk_size; | |
5077 | int ndevs; | |
5078 | }; | |
5079 | ||
27c314d5 NA |
5080 | static void init_alloc_chunk_ctl_policy_regular( |
5081 | struct btrfs_fs_devices *fs_devices, | |
5082 | struct alloc_chunk_ctl *ctl) | |
5083 | { | |
f6fca391 | 5084 | struct btrfs_space_info *space_info; |
27c314d5 | 5085 | |
f6fca391 SR |
5086 | space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); |
5087 | ASSERT(space_info); | |
5088 | ||
5089 | ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); | |
5090 | ctl->max_stripe_size = ctl->max_chunk_size; | |
5091 | ||
5092 | if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) | |
5093 | ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); | |
27c314d5 NA |
5094 | |
5095 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5096 | ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), |
27c314d5 | 5097 | ctl->max_chunk_size); |
6aafb303 | 5098 | ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes; |
27c314d5 NA |
5099 | } |
5100 | ||
1cd6121f NA |
5101 | static void init_alloc_chunk_ctl_policy_zoned( |
5102 | struct btrfs_fs_devices *fs_devices, | |
5103 | struct alloc_chunk_ctl *ctl) | |
5104 | { | |
5105 | u64 zone_size = fs_devices->fs_info->zone_size; | |
5106 | u64 limit; | |
5107 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; | |
5108 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; | |
5109 | u64 min_chunk_size = min_data_stripes * zone_size; | |
5110 | u64 type = ctl->type; | |
5111 | ||
5112 | ctl->max_stripe_size = zone_size; | |
5113 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
5114 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, | |
5115 | zone_size); | |
5116 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
5117 | ctl->max_chunk_size = ctl->max_stripe_size; | |
5118 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
5119 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
5120 | ctl->devs_max = min_t(int, ctl->devs_max, | |
5121 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
bb05b298 AB |
5122 | } else { |
5123 | BUG(); | |
1cd6121f NA |
5124 | } |
5125 | ||
5126 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5127 | limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), |
1cd6121f NA |
5128 | zone_size), |
5129 | min_chunk_size); | |
5130 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); | |
5131 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; | |
5132 | } | |
5133 | ||
27c314d5 NA |
5134 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
5135 | struct alloc_chunk_ctl *ctl) | |
5136 | { | |
5137 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
5138 | ||
5139 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
5140 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5141 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
5142 | if (!ctl->devs_max) | |
5143 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
5144 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
5145 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
5146 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
5147 | ctl->nparity = btrfs_raid_array[index].nparity; | |
5148 | ctl->ndevs = 0; | |
5149 | ||
5150 | switch (fs_devices->chunk_alloc_policy) { | |
5151 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5152 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
5153 | break; | |
1cd6121f NA |
5154 | case BTRFS_CHUNK_ALLOC_ZONED: |
5155 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); | |
5156 | break; | |
27c314d5 NA |
5157 | default: |
5158 | BUG(); | |
5159 | } | |
5160 | } | |
5161 | ||
560156cb NA |
5162 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
5163 | struct alloc_chunk_ctl *ctl, | |
5164 | struct btrfs_device_info *devices_info) | |
b2117a39 | 5165 | { |
560156cb | 5166 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 5167 | struct btrfs_device *device; |
73c5de00 | 5168 | u64 total_avail; |
560156cb | 5169 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 5170 | int ret; |
560156cb NA |
5171 | int ndevs = 0; |
5172 | u64 max_avail; | |
5173 | u64 dev_offset; | |
0cad8a11 | 5174 | |
9f680ce0 | 5175 | /* |
73c5de00 AJ |
5176 | * in the first pass through the devices list, we gather information |
5177 | * about the available holes on each device. | |
9f680ce0 | 5178 | */ |
ebcc9301 | 5179 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 5180 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5181 | WARN(1, KERN_ERR |
efe120a0 | 5182 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5183 | continue; |
5184 | } | |
b2117a39 | 5185 | |
e12c9621 AJ |
5186 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5187 | &device->dev_state) || | |
401e29c1 | 5188 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5189 | continue; |
b2117a39 | 5190 | |
73c5de00 AJ |
5191 | if (device->total_bytes > device->bytes_used) |
5192 | total_avail = device->total_bytes - device->bytes_used; | |
5193 | else | |
5194 | total_avail = 0; | |
38c01b96 | 5195 | |
5196 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 5197 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 5198 | continue; |
b2117a39 | 5199 | |
560156cb NA |
5200 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
5201 | &max_avail); | |
73c5de00 | 5202 | if (ret && ret != -ENOSPC) |
560156cb | 5203 | return ret; |
b2117a39 | 5204 | |
73c5de00 | 5205 | if (ret == 0) |
560156cb | 5206 | max_avail = dev_extent_want; |
b2117a39 | 5207 | |
6aafb303 | 5208 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
5209 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
5210 | btrfs_debug(info, | |
560156cb | 5211 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 5212 | __func__, device->devid, max_avail, |
6aafb303 | 5213 | ctl->dev_extent_min); |
73c5de00 | 5214 | continue; |
4117f207 | 5215 | } |
b2117a39 | 5216 | |
063d006f ES |
5217 | if (ndevs == fs_devices->rw_devices) { |
5218 | WARN(1, "%s: found more than %llu devices\n", | |
5219 | __func__, fs_devices->rw_devices); | |
5220 | break; | |
5221 | } | |
73c5de00 AJ |
5222 | devices_info[ndevs].dev_offset = dev_offset; |
5223 | devices_info[ndevs].max_avail = max_avail; | |
5224 | devices_info[ndevs].total_avail = total_avail; | |
5225 | devices_info[ndevs].dev = device; | |
5226 | ++ndevs; | |
5227 | } | |
560156cb | 5228 | ctl->ndevs = ndevs; |
b2117a39 | 5229 | |
73c5de00 AJ |
5230 | /* |
5231 | * now sort the devices by hole size / available space | |
5232 | */ | |
560156cb | 5233 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 5234 | btrfs_cmp_device_info, NULL); |
b2117a39 | 5235 | |
560156cb NA |
5236 | return 0; |
5237 | } | |
5238 | ||
5badf512 NA |
5239 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
5240 | struct btrfs_device_info *devices_info) | |
5241 | { | |
5242 | /* Number of stripes that count for block group size */ | |
5243 | int data_stripes; | |
5244 | ||
5245 | /* | |
5246 | * The primary goal is to maximize the number of stripes, so use as | |
5247 | * many devices as possible, even if the stripes are not maximum sized. | |
5248 | * | |
5249 | * The DUP profile stores more than one stripe per device, the | |
5250 | * max_avail is the total size so we have to adjust. | |
5251 | */ | |
5252 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5253 | ctl->dev_stripes); | |
5254 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5255 | ||
5256 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5257 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5258 | ||
5259 | /* | |
5260 | * Use the number of data stripes to figure out how big this chunk is | |
5261 | * really going to be in terms of logical address space, and compare | |
5262 | * that answer with the max chunk size. If it's higher, we try to | |
5263 | * reduce stripe_size. | |
5264 | */ | |
5265 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5266 | /* | |
5267 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5268 | * then use it, unless it ends up being even bigger than the | |
5269 | * previous value we had already. | |
5270 | */ | |
5271 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5272 | data_stripes), SZ_16M), | |
5273 | ctl->stripe_size); | |
5274 | } | |
5275 | ||
5da431b7 QW |
5276 | /* Stripe size should not go beyond 1G. */ |
5277 | ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); | |
5278 | ||
5badf512 NA |
5279 | /* Align to BTRFS_STRIPE_LEN */ |
5280 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5281 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5282 | ||
5283 | return 0; | |
5284 | } | |
5285 | ||
1cd6121f NA |
5286 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
5287 | struct btrfs_device_info *devices_info) | |
5288 | { | |
5289 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; | |
5290 | /* Number of stripes that count for block group size */ | |
5291 | int data_stripes; | |
5292 | ||
5293 | /* | |
5294 | * It should hold because: | |
5295 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes | |
5296 | */ | |
5297 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); | |
5298 | ||
5299 | ctl->stripe_size = zone_size; | |
5300 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5301 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5302 | ||
5303 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ | |
5304 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5305 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, | |
5306 | ctl->stripe_size) + ctl->nparity, | |
5307 | ctl->dev_stripes); | |
5308 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5309 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5310 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); | |
5311 | } | |
5312 | ||
5313 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5314 | ||
5315 | return 0; | |
5316 | } | |
5317 | ||
5badf512 NA |
5318 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
5319 | struct alloc_chunk_ctl *ctl, | |
5320 | struct btrfs_device_info *devices_info) | |
5321 | { | |
5322 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5323 | ||
5324 | /* | |
5325 | * Round down to number of usable stripes, devs_increment can be any | |
5326 | * number so we can't use round_down() that requires power of 2, while | |
5327 | * rounddown is safe. | |
5328 | */ | |
5329 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5330 | ||
5331 | if (ctl->ndevs < ctl->devs_min) { | |
5332 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5333 | btrfs_debug(info, | |
5334 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5335 | __func__, ctl->ndevs, ctl->devs_min); | |
5336 | } | |
5337 | return -ENOSPC; | |
5338 | } | |
5339 | ||
5340 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5341 | ||
5342 | switch (fs_devices->chunk_alloc_policy) { | |
5343 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5344 | return decide_stripe_size_regular(ctl, devices_info); | |
1cd6121f NA |
5345 | case BTRFS_CHUNK_ALLOC_ZONED: |
5346 | return decide_stripe_size_zoned(ctl, devices_info); | |
5badf512 NA |
5347 | default: |
5348 | BUG(); | |
5349 | } | |
5350 | } | |
5351 | ||
79bd3712 | 5352 | static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, |
dce580ca NA |
5353 | struct alloc_chunk_ctl *ctl, |
5354 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5355 | { |
5356 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5357 | struct map_lookup *map = NULL; |
5358 | struct extent_map_tree *em_tree; | |
79bd3712 | 5359 | struct btrfs_block_group *block_group; |
560156cb | 5360 | struct extent_map *em; |
dce580ca NA |
5361 | u64 start = ctl->start; |
5362 | u64 type = ctl->type; | |
560156cb NA |
5363 | int ret; |
5364 | int i; | |
5365 | int j; | |
5366 | ||
dce580ca NA |
5367 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5368 | if (!map) | |
79bd3712 | 5369 | return ERR_PTR(-ENOMEM); |
dce580ca | 5370 | map->num_stripes = ctl->num_stripes; |
560156cb | 5371 | |
dce580ca NA |
5372 | for (i = 0; i < ctl->ndevs; ++i) { |
5373 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5374 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5375 | map->stripes[s].dev = devices_info[i].dev; |
5376 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5377 | j * ctl->stripe_size; |
6324fbf3 | 5378 | } |
6324fbf3 | 5379 | } |
500ceed8 NB |
5380 | map->stripe_len = BTRFS_STRIPE_LEN; |
5381 | map->io_align = BTRFS_STRIPE_LEN; | |
5382 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5383 | map->type = type; |
dce580ca | 5384 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5385 | |
dce580ca | 5386 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5387 | |
172ddd60 | 5388 | em = alloc_extent_map(); |
2b82032c | 5389 | if (!em) { |
298a8f9c | 5390 | kfree(map); |
79bd3712 | 5391 | return ERR_PTR(-ENOMEM); |
593060d7 | 5392 | } |
298a8f9c | 5393 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5394 | em->map_lookup = map; |
2b82032c | 5395 | em->start = start; |
dce580ca | 5396 | em->len = ctl->chunk_size; |
2b82032c YZ |
5397 | em->block_start = 0; |
5398 | em->block_len = em->len; | |
dce580ca | 5399 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5400 | |
c8bf1b67 | 5401 | em_tree = &info->mapping_tree; |
890871be | 5402 | write_lock(&em_tree->lock); |
09a2a8f9 | 5403 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5404 | if (ret) { |
1efb72a3 | 5405 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5406 | free_extent_map(em); |
79bd3712 | 5407 | return ERR_PTR(ret); |
0f5d42b2 | 5408 | } |
1efb72a3 NB |
5409 | write_unlock(&em_tree->lock); |
5410 | ||
79bd3712 FM |
5411 | block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size); |
5412 | if (IS_ERR(block_group)) | |
6df9a95e | 5413 | goto error_del_extent; |
2b82032c | 5414 | |
bbbf7243 NB |
5415 | for (i = 0; i < map->num_stripes; i++) { |
5416 | struct btrfs_device *dev = map->stripes[i].dev; | |
5417 | ||
4f2bafe8 | 5418 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5419 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5420 | if (list_empty(&dev->post_commit_list)) |
5421 | list_add_tail(&dev->post_commit_list, | |
5422 | &trans->transaction->dev_update_list); | |
5423 | } | |
43530c46 | 5424 | |
dce580ca | 5425 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5426 | &info->free_chunk_space); |
1c116187 | 5427 | |
0f5d42b2 | 5428 | free_extent_map(em); |
0b246afa | 5429 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5430 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5431 | |
79bd3712 | 5432 | return block_group; |
b2117a39 | 5433 | |
6df9a95e | 5434 | error_del_extent: |
0f5d42b2 JB |
5435 | write_lock(&em_tree->lock); |
5436 | remove_extent_mapping(em_tree, em); | |
5437 | write_unlock(&em_tree->lock); | |
5438 | ||
5439 | /* One for our allocation */ | |
5440 | free_extent_map(em); | |
5441 | /* One for the tree reference */ | |
5442 | free_extent_map(em); | |
dce580ca | 5443 | |
79bd3712 | 5444 | return block_group; |
dce580ca NA |
5445 | } |
5446 | ||
f6f39f7a | 5447 | struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, |
79bd3712 | 5448 | u64 type) |
dce580ca NA |
5449 | { |
5450 | struct btrfs_fs_info *info = trans->fs_info; | |
5451 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5452 | struct btrfs_device_info *devices_info = NULL; | |
5453 | struct alloc_chunk_ctl ctl; | |
79bd3712 | 5454 | struct btrfs_block_group *block_group; |
dce580ca NA |
5455 | int ret; |
5456 | ||
11c67b1a NB |
5457 | lockdep_assert_held(&info->chunk_mutex); |
5458 | ||
dce580ca NA |
5459 | if (!alloc_profile_is_valid(type, 0)) { |
5460 | ASSERT(0); | |
79bd3712 | 5461 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5462 | } |
5463 | ||
5464 | if (list_empty(&fs_devices->alloc_list)) { | |
5465 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5466 | btrfs_debug(info, "%s: no writable device", __func__); | |
79bd3712 | 5467 | return ERR_PTR(-ENOSPC); |
dce580ca NA |
5468 | } |
5469 | ||
5470 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5471 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5472 | ASSERT(0); | |
79bd3712 | 5473 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5474 | } |
5475 | ||
11c67b1a | 5476 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5477 | ctl.type = type; |
5478 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5479 | ||
5480 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5481 | GFP_NOFS); | |
5482 | if (!devices_info) | |
79bd3712 | 5483 | return ERR_PTR(-ENOMEM); |
dce580ca NA |
5484 | |
5485 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5486 | if (ret < 0) { |
5487 | block_group = ERR_PTR(ret); | |
dce580ca | 5488 | goto out; |
79bd3712 | 5489 | } |
dce580ca NA |
5490 | |
5491 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5492 | if (ret < 0) { |
5493 | block_group = ERR_PTR(ret); | |
dce580ca | 5494 | goto out; |
79bd3712 | 5495 | } |
dce580ca | 5496 | |
79bd3712 | 5497 | block_group = create_chunk(trans, &ctl, devices_info); |
dce580ca NA |
5498 | |
5499 | out: | |
b2117a39 | 5500 | kfree(devices_info); |
79bd3712 | 5501 | return block_group; |
2b82032c YZ |
5502 | } |
5503 | ||
79bd3712 FM |
5504 | /* |
5505 | * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the | |
5506 | * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system | |
5507 | * chunks. | |
5508 | * | |
5509 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
5510 | * phases. | |
5511 | */ | |
5512 | int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, | |
5513 | struct btrfs_block_group *bg) | |
5514 | { | |
5515 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
79bd3712 FM |
5516 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
5517 | struct btrfs_key key; | |
5518 | struct btrfs_chunk *chunk; | |
5519 | struct btrfs_stripe *stripe; | |
5520 | struct extent_map *em; | |
5521 | struct map_lookup *map; | |
5522 | size_t item_size; | |
5523 | int i; | |
5524 | int ret; | |
5525 | ||
5526 | /* | |
5527 | * We take the chunk_mutex for 2 reasons: | |
5528 | * | |
5529 | * 1) Updates and insertions in the chunk btree must be done while holding | |
5530 | * the chunk_mutex, as well as updating the system chunk array in the | |
5531 | * superblock. See the comment on top of btrfs_chunk_alloc() for the | |
5532 | * details; | |
5533 | * | |
5534 | * 2) To prevent races with the final phase of a device replace operation | |
5535 | * that replaces the device object associated with the map's stripes, | |
5536 | * because the device object's id can change at any time during that | |
5537 | * final phase of the device replace operation | |
5538 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
5539 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
5540 | * which would cause a failure when updating the device item, which does | |
5541 | * not exists, or persisting a stripe of the chunk item with such ID. | |
5542 | * Here we can't use the device_list_mutex because our caller already | |
5543 | * has locked the chunk_mutex, and the final phase of device replace | |
5544 | * acquires both mutexes - first the device_list_mutex and then the | |
5545 | * chunk_mutex. Using any of those two mutexes protects us from a | |
5546 | * concurrent device replace. | |
5547 | */ | |
5548 | lockdep_assert_held(&fs_info->chunk_mutex); | |
5549 | ||
5550 | em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); | |
5551 | if (IS_ERR(em)) { | |
5552 | ret = PTR_ERR(em); | |
5553 | btrfs_abort_transaction(trans, ret); | |
5554 | return ret; | |
5555 | } | |
5556 | ||
5557 | map = em->map_lookup; | |
5558 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
5559 | ||
5560 | chunk = kzalloc(item_size, GFP_NOFS); | |
5561 | if (!chunk) { | |
5562 | ret = -ENOMEM; | |
5563 | btrfs_abort_transaction(trans, ret); | |
50460e37 | 5564 | goto out; |
2b82032c YZ |
5565 | } |
5566 | ||
79bd3712 FM |
5567 | for (i = 0; i < map->num_stripes; i++) { |
5568 | struct btrfs_device *device = map->stripes[i].dev; | |
5569 | ||
5570 | ret = btrfs_update_device(trans, device); | |
5571 | if (ret) | |
5572 | goto out; | |
5573 | } | |
5574 | ||
2b82032c | 5575 | stripe = &chunk->stripe; |
6df9a95e | 5576 | for (i = 0; i < map->num_stripes; i++) { |
79bd3712 FM |
5577 | struct btrfs_device *device = map->stripes[i].dev; |
5578 | const u64 dev_offset = map->stripes[i].physical; | |
0b86a832 | 5579 | |
e17cade2 CM |
5580 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5581 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5582 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5583 | stripe++; |
0b86a832 CM |
5584 | } |
5585 | ||
79bd3712 | 5586 | btrfs_set_stack_chunk_length(chunk, bg->length); |
fd51eb2f | 5587 | btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); |
2b82032c YZ |
5588 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
5589 | btrfs_set_stack_chunk_type(chunk, map->type); | |
5590 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
5591 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
5592 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 5593 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5594 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5595 | |
2b82032c YZ |
5596 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5597 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
79bd3712 | 5598 | key.offset = bg->start; |
0b86a832 | 5599 | |
2b82032c | 5600 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
79bd3712 FM |
5601 | if (ret) |
5602 | goto out; | |
5603 | ||
3349b57f | 5604 | set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); |
79bd3712 FM |
5605 | |
5606 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2ff7e61e | 5607 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
79bd3712 FM |
5608 | if (ret) |
5609 | goto out; | |
8f18cf13 | 5610 | } |
1abe9b8a | 5611 | |
6df9a95e | 5612 | out: |
0b86a832 | 5613 | kfree(chunk); |
6df9a95e | 5614 | free_extent_map(em); |
4ed1d16e | 5615 | return ret; |
2b82032c | 5616 | } |
0b86a832 | 5617 | |
6f8e0fc7 | 5618 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5619 | { |
6f8e0fc7 | 5620 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5621 | u64 alloc_profile; |
79bd3712 FM |
5622 | struct btrfs_block_group *meta_bg; |
5623 | struct btrfs_block_group *sys_bg; | |
5624 | ||
5625 | /* | |
5626 | * When adding a new device for sprouting, the seed device is read-only | |
5627 | * so we must first allocate a metadata and a system chunk. But before | |
5628 | * adding the block group items to the extent, device and chunk btrees, | |
5629 | * we must first: | |
5630 | * | |
5631 | * 1) Create both chunks without doing any changes to the btrees, as | |
5632 | * otherwise we would get -ENOSPC since the block groups from the | |
5633 | * seed device are read-only; | |
5634 | * | |
5635 | * 2) Add the device item for the new sprout device - finishing the setup | |
5636 | * of a new block group requires updating the device item in the chunk | |
5637 | * btree, so it must exist when we attempt to do it. The previous step | |
5638 | * ensures this does not fail with -ENOSPC. | |
5639 | * | |
5640 | * After that we can add the block group items to their btrees: | |
5641 | * update existing device item in the chunk btree, add a new block group | |
5642 | * item to the extent btree, add a new chunk item to the chunk btree and | |
5643 | * finally add the new device extent items to the devices btree. | |
5644 | */ | |
2b82032c | 5645 | |
1b86826d | 5646 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
f6f39f7a | 5647 | meta_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5648 | if (IS_ERR(meta_bg)) |
5649 | return PTR_ERR(meta_bg); | |
2b82032c | 5650 | |
1b86826d | 5651 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
f6f39f7a | 5652 | sys_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5653 | if (IS_ERR(sys_bg)) |
5654 | return PTR_ERR(sys_bg); | |
5655 | ||
5656 | return 0; | |
2b82032c YZ |
5657 | } |
5658 | ||
d20983b4 MX |
5659 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5660 | { | |
fc9a2ac7 | 5661 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5662 | |
fc9a2ac7 | 5663 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5664 | } |
5665 | ||
a09f23c3 | 5666 | bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5667 | { |
5668 | struct extent_map *em; | |
5669 | struct map_lookup *map; | |
d20983b4 | 5670 | int miss_ndevs = 0; |
2b82032c | 5671 | int i; |
a09f23c3 | 5672 | bool ret = true; |
2b82032c | 5673 | |
60ca842e | 5674 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5675 | if (IS_ERR(em)) |
a09f23c3 | 5676 | return false; |
2b82032c | 5677 | |
95617d69 | 5678 | map = em->map_lookup; |
2b82032c | 5679 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5680 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5681 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5682 | miss_ndevs++; |
5683 | continue; | |
5684 | } | |
ebbede42 AJ |
5685 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5686 | &map->stripes[i].dev->dev_state)) { | |
a09f23c3 | 5687 | ret = false; |
d20983b4 | 5688 | goto end; |
2b82032c YZ |
5689 | } |
5690 | } | |
d20983b4 MX |
5691 | |
5692 | /* | |
a09f23c3 AJ |
5693 | * If the number of missing devices is larger than max errors, we can |
5694 | * not write the data into that chunk successfully. | |
d20983b4 MX |
5695 | */ |
5696 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
a09f23c3 | 5697 | ret = false; |
d20983b4 | 5698 | end: |
0b86a832 | 5699 | free_extent_map(em); |
a09f23c3 | 5700 | return ret; |
0b86a832 CM |
5701 | } |
5702 | ||
c8bf1b67 | 5703 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5704 | { |
5705 | struct extent_map *em; | |
5706 | ||
d397712b | 5707 | while (1) { |
c8bf1b67 DS |
5708 | write_lock(&tree->lock); |
5709 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5710 | if (em) |
c8bf1b67 DS |
5711 | remove_extent_mapping(tree, em); |
5712 | write_unlock(&tree->lock); | |
0b86a832 CM |
5713 | if (!em) |
5714 | break; | |
0b86a832 CM |
5715 | /* once for us */ |
5716 | free_extent_map(em); | |
5717 | /* once for the tree */ | |
5718 | free_extent_map(em); | |
5719 | } | |
5720 | } | |
5721 | ||
5d964051 | 5722 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5723 | { |
5724 | struct extent_map *em; | |
5725 | struct map_lookup *map; | |
6d322b48 QW |
5726 | enum btrfs_raid_types index; |
5727 | int ret = 1; | |
f188591e | 5728 | |
60ca842e | 5729 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5730 | if (IS_ERR(em)) |
5731 | /* | |
5732 | * We could return errors for these cases, but that could get | |
5733 | * ugly and we'd probably do the same thing which is just not do | |
5734 | * anything else and exit, so return 1 so the callers don't try | |
5735 | * to use other copies. | |
5736 | */ | |
fb7669b5 | 5737 | return 1; |
fb7669b5 | 5738 | |
95617d69 | 5739 | map = em->map_lookup; |
6d322b48 QW |
5740 | index = btrfs_bg_flags_to_raid_index(map->type); |
5741 | ||
5742 | /* Non-RAID56, use their ncopies from btrfs_raid_array. */ | |
5743 | if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
5744 | ret = btrfs_raid_array[index].ncopies; | |
53b381b3 DW |
5745 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5746 | ret = 2; | |
5747 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5748 | /* |
5749 | * There could be two corrupted data stripes, we need | |
5750 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5751 | * |
8810f751 LB |
5752 | * Fail a stripe at a time on every retry except the |
5753 | * stripe under reconstruction. | |
5754 | */ | |
5755 | ret = map->num_stripes; | |
f188591e | 5756 | free_extent_map(em); |
ad6d620e | 5757 | |
cb5583dd | 5758 | down_read(&fs_info->dev_replace.rwsem); |
6fad823f LB |
5759 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) && |
5760 | fs_info->dev_replace.tgtdev) | |
ad6d620e | 5761 | ret++; |
cb5583dd | 5762 | up_read(&fs_info->dev_replace.rwsem); |
ad6d620e | 5763 | |
f188591e CM |
5764 | return ret; |
5765 | } | |
5766 | ||
2ff7e61e | 5767 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5768 | u64 logical) |
5769 | { | |
5770 | struct extent_map *em; | |
5771 | struct map_lookup *map; | |
0b246afa | 5772 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5773 | |
b036f479 QW |
5774 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
5775 | return len; | |
5776 | ||
60ca842e | 5777 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5778 | |
69f03f13 NB |
5779 | if (!WARN_ON(IS_ERR(em))) { |
5780 | map = em->map_lookup; | |
5781 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5782 | len = map->stripe_len * nr_data_stripes(map); | |
5783 | free_extent_map(em); | |
5784 | } | |
53b381b3 DW |
5785 | return len; |
5786 | } | |
5787 | ||
e4ff5fb5 | 5788 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5789 | { |
5790 | struct extent_map *em; | |
5791 | struct map_lookup *map; | |
53b381b3 DW |
5792 | int ret = 0; |
5793 | ||
b036f479 QW |
5794 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
5795 | return 0; | |
5796 | ||
60ca842e | 5797 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5798 | |
69f03f13 NB |
5799 | if(!WARN_ON(IS_ERR(em))) { |
5800 | map = em->map_lookup; | |
5801 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5802 | ret = 1; | |
5803 | free_extent_map(em); | |
5804 | } | |
53b381b3 DW |
5805 | return ret; |
5806 | } | |
5807 | ||
30d9861f | 5808 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5809 | struct map_lookup *map, int first, |
8ba0ae78 | 5810 | int dev_replace_is_ongoing) |
dfe25020 CM |
5811 | { |
5812 | int i; | |
99f92a7c | 5813 | int num_stripes; |
8ba0ae78 | 5814 | int preferred_mirror; |
30d9861f SB |
5815 | int tolerance; |
5816 | struct btrfs_device *srcdev; | |
5817 | ||
99f92a7c | 5818 | ASSERT((map->type & |
c7369b3f | 5819 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5820 | |
5821 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5822 | num_stripes = map->sub_stripes; | |
5823 | else | |
5824 | num_stripes = map->num_stripes; | |
5825 | ||
33fd2f71 AJ |
5826 | switch (fs_info->fs_devices->read_policy) { |
5827 | default: | |
5828 | /* Shouldn't happen, just warn and use pid instead of failing */ | |
5829 | btrfs_warn_rl(fs_info, | |
5830 | "unknown read_policy type %u, reset to pid", | |
5831 | fs_info->fs_devices->read_policy); | |
5832 | fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; | |
5833 | fallthrough; | |
5834 | case BTRFS_READ_POLICY_PID: | |
5835 | preferred_mirror = first + (current->pid % num_stripes); | |
5836 | break; | |
5837 | } | |
8ba0ae78 | 5838 | |
30d9861f SB |
5839 | if (dev_replace_is_ongoing && |
5840 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5841 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5842 | srcdev = fs_info->dev_replace.srcdev; | |
5843 | else | |
5844 | srcdev = NULL; | |
5845 | ||
5846 | /* | |
5847 | * try to avoid the drive that is the source drive for a | |
5848 | * dev-replace procedure, only choose it if no other non-missing | |
5849 | * mirror is available | |
5850 | */ | |
5851 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5852 | if (map->stripes[preferred_mirror].dev->bdev && |
5853 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5854 | return preferred_mirror; | |
99f92a7c | 5855 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5856 | if (map->stripes[i].dev->bdev && |
5857 | (tolerance || map->stripes[i].dev != srcdev)) | |
5858 | return i; | |
5859 | } | |
dfe25020 | 5860 | } |
30d9861f | 5861 | |
dfe25020 CM |
5862 | /* we couldn't find one that doesn't fail. Just return something |
5863 | * and the io error handling code will clean up eventually | |
5864 | */ | |
8ba0ae78 | 5865 | return preferred_mirror; |
dfe25020 CM |
5866 | } |
5867 | ||
53b381b3 | 5868 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ |
4c664611 | 5869 | static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes) |
53b381b3 | 5870 | { |
53b381b3 | 5871 | int i; |
53b381b3 DW |
5872 | int again = 1; |
5873 | ||
5874 | while (again) { | |
5875 | again = 0; | |
cc7539ed | 5876 | for (i = 0; i < num_stripes - 1; i++) { |
eeb6f172 | 5877 | /* Swap if parity is on a smaller index */ |
4c664611 QW |
5878 | if (bioc->raid_map[i] > bioc->raid_map[i + 1]) { |
5879 | swap(bioc->stripes[i], bioc->stripes[i + 1]); | |
5880 | swap(bioc->raid_map[i], bioc->raid_map[i + 1]); | |
53b381b3 DW |
5881 | again = 1; |
5882 | } | |
5883 | } | |
5884 | } | |
5885 | } | |
5886 | ||
731ccf15 QW |
5887 | static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, |
5888 | int total_stripes, | |
4c664611 | 5889 | int real_stripes) |
6e9606d2 | 5890 | { |
4c664611 QW |
5891 | struct btrfs_io_context *bioc = kzalloc( |
5892 | /* The size of btrfs_io_context */ | |
5893 | sizeof(struct btrfs_io_context) + | |
5894 | /* Plus the variable array for the stripes */ | |
5895 | sizeof(struct btrfs_io_stripe) * (total_stripes) + | |
5896 | /* Plus the variable array for the tgt dev */ | |
6e9606d2 | 5897 | sizeof(int) * (real_stripes) + |
e57cf21e | 5898 | /* |
4c664611 QW |
5899 | * Plus the raid_map, which includes both the tgt dev |
5900 | * and the stripes. | |
e57cf21e CM |
5901 | */ |
5902 | sizeof(u64) * (total_stripes), | |
9f0eac07 L |
5903 | GFP_NOFS); |
5904 | ||
5905 | if (!bioc) | |
5906 | return NULL; | |
6e9606d2 | 5907 | |
4c664611 | 5908 | refcount_set(&bioc->refs, 1); |
6e9606d2 | 5909 | |
731ccf15 | 5910 | bioc->fs_info = fs_info; |
4c664611 QW |
5911 | bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes); |
5912 | bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes); | |
608769a4 | 5913 | |
4c664611 | 5914 | return bioc; |
6e9606d2 ZL |
5915 | } |
5916 | ||
4c664611 | 5917 | void btrfs_get_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 5918 | { |
4c664611 QW |
5919 | WARN_ON(!refcount_read(&bioc->refs)); |
5920 | refcount_inc(&bioc->refs); | |
6e9606d2 ZL |
5921 | } |
5922 | ||
4c664611 | 5923 | void btrfs_put_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 5924 | { |
4c664611 | 5925 | if (!bioc) |
6e9606d2 | 5926 | return; |
4c664611 QW |
5927 | if (refcount_dec_and_test(&bioc->refs)) |
5928 | kfree(bioc); | |
6e9606d2 ZL |
5929 | } |
5930 | ||
0b3d4cd3 LB |
5931 | /* |
5932 | * Please note that, discard won't be sent to target device of device | |
5933 | * replace. | |
5934 | */ | |
a4012f06 CH |
5935 | struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, |
5936 | u64 logical, u64 *length_ret, | |
5937 | u32 *num_stripes) | |
0b3d4cd3 LB |
5938 | { |
5939 | struct extent_map *em; | |
5940 | struct map_lookup *map; | |
a4012f06 | 5941 | struct btrfs_discard_stripe *stripes; |
6b7faadd | 5942 | u64 length = *length_ret; |
0b3d4cd3 LB |
5943 | u64 offset; |
5944 | u64 stripe_nr; | |
5945 | u64 stripe_nr_end; | |
5946 | u64 stripe_end_offset; | |
5947 | u64 stripe_cnt; | |
5948 | u64 stripe_len; | |
5949 | u64 stripe_offset; | |
0b3d4cd3 LB |
5950 | u32 stripe_index; |
5951 | u32 factor = 0; | |
5952 | u32 sub_stripes = 0; | |
5953 | u64 stripes_per_dev = 0; | |
5954 | u32 remaining_stripes = 0; | |
5955 | u32 last_stripe = 0; | |
a4012f06 | 5956 | int ret; |
0b3d4cd3 LB |
5957 | int i; |
5958 | ||
60ca842e | 5959 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 | 5960 | if (IS_ERR(em)) |
a4012f06 | 5961 | return ERR_CAST(em); |
0b3d4cd3 LB |
5962 | |
5963 | map = em->map_lookup; | |
a4012f06 | 5964 | |
0b3d4cd3 LB |
5965 | /* we don't discard raid56 yet */ |
5966 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5967 | ret = -EOPNOTSUPP; | |
a4012f06 CH |
5968 | goto out_free_map; |
5969 | } | |
0b3d4cd3 LB |
5970 | |
5971 | offset = logical - em->start; | |
2d974619 | 5972 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5973 | *length_ret = length; |
0b3d4cd3 LB |
5974 | |
5975 | stripe_len = map->stripe_len; | |
5976 | /* | |
5977 | * stripe_nr counts the total number of stripes we have to stride | |
5978 | * to get to this block | |
5979 | */ | |
5980 | stripe_nr = div64_u64(offset, stripe_len); | |
5981 | ||
5982 | /* stripe_offset is the offset of this block in its stripe */ | |
5983 | stripe_offset = offset - stripe_nr * stripe_len; | |
5984 | ||
5985 | stripe_nr_end = round_up(offset + length, map->stripe_len); | |
42c61ab6 | 5986 | stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len); |
0b3d4cd3 LB |
5987 | stripe_cnt = stripe_nr_end - stripe_nr; |
5988 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5989 | (offset + length); | |
5990 | /* | |
5991 | * after this, stripe_nr is the number of stripes on this | |
5992 | * device we have to walk to find the data, and stripe_index is | |
5993 | * the number of our device in the stripe array | |
5994 | */ | |
a4012f06 | 5995 | *num_stripes = 1; |
0b3d4cd3 LB |
5996 | stripe_index = 0; |
5997 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5998 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5999 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
6000 | sub_stripes = 1; | |
6001 | else | |
6002 | sub_stripes = map->sub_stripes; | |
6003 | ||
6004 | factor = map->num_stripes / sub_stripes; | |
a4012f06 | 6005 | *num_stripes = min_t(u64, map->num_stripes, |
0b3d4cd3 LB |
6006 | sub_stripes * stripe_cnt); |
6007 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
6008 | stripe_index *= sub_stripes; | |
6009 | stripes_per_dev = div_u64_rem(stripe_cnt, factor, | |
6010 | &remaining_stripes); | |
6011 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
6012 | last_stripe *= sub_stripes; | |
c7369b3f | 6013 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 | 6014 | BTRFS_BLOCK_GROUP_DUP)) { |
a4012f06 | 6015 | *num_stripes = map->num_stripes; |
0b3d4cd3 LB |
6016 | } else { |
6017 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
6018 | &stripe_index); | |
6019 | } | |
6020 | ||
a4012f06 CH |
6021 | stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); |
6022 | if (!stripes) { | |
0b3d4cd3 | 6023 | ret = -ENOMEM; |
a4012f06 | 6024 | goto out_free_map; |
0b3d4cd3 LB |
6025 | } |
6026 | ||
a4012f06 CH |
6027 | for (i = 0; i < *num_stripes; i++) { |
6028 | stripes[i].physical = | |
0b3d4cd3 LB |
6029 | map->stripes[stripe_index].physical + |
6030 | stripe_offset + stripe_nr * map->stripe_len; | |
a4012f06 | 6031 | stripes[i].dev = map->stripes[stripe_index].dev; |
0b3d4cd3 LB |
6032 | |
6033 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
6034 | BTRFS_BLOCK_GROUP_RAID10)) { | |
a4012f06 | 6035 | stripes[i].length = stripes_per_dev * map->stripe_len; |
0b3d4cd3 LB |
6036 | |
6037 | if (i / sub_stripes < remaining_stripes) | |
a4012f06 | 6038 | stripes[i].length += map->stripe_len; |
0b3d4cd3 LB |
6039 | |
6040 | /* | |
6041 | * Special for the first stripe and | |
6042 | * the last stripe: | |
6043 | * | |
6044 | * |-------|...|-------| | |
6045 | * |----------| | |
6046 | * off end_off | |
6047 | */ | |
6048 | if (i < sub_stripes) | |
a4012f06 | 6049 | stripes[i].length -= stripe_offset; |
0b3d4cd3 LB |
6050 | |
6051 | if (stripe_index >= last_stripe && | |
6052 | stripe_index <= (last_stripe + | |
6053 | sub_stripes - 1)) | |
a4012f06 | 6054 | stripes[i].length -= stripe_end_offset; |
0b3d4cd3 LB |
6055 | |
6056 | if (i == sub_stripes - 1) | |
6057 | stripe_offset = 0; | |
6058 | } else { | |
a4012f06 | 6059 | stripes[i].length = length; |
0b3d4cd3 LB |
6060 | } |
6061 | ||
6062 | stripe_index++; | |
6063 | if (stripe_index == map->num_stripes) { | |
6064 | stripe_index = 0; | |
6065 | stripe_nr++; | |
6066 | } | |
6067 | } | |
6068 | ||
0b3d4cd3 | 6069 | free_extent_map(em); |
a4012f06 CH |
6070 | return stripes; |
6071 | out_free_map: | |
6072 | free_extent_map(em); | |
6073 | return ERR_PTR(ret); | |
0b3d4cd3 LB |
6074 | } |
6075 | ||
5ab56090 LB |
6076 | /* |
6077 | * In dev-replace case, for repair case (that's the only case where the mirror | |
6078 | * is selected explicitly when calling btrfs_map_block), blocks left of the | |
6079 | * left cursor can also be read from the target drive. | |
6080 | * | |
6081 | * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the | |
6082 | * array of stripes. | |
6083 | * For READ, it also needs to be supported using the same mirror number. | |
6084 | * | |
6085 | * If the requested block is not left of the left cursor, EIO is returned. This | |
6086 | * can happen because btrfs_num_copies() returns one more in the dev-replace | |
6087 | * case. | |
6088 | */ | |
6089 | static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info, | |
6090 | u64 logical, u64 length, | |
6091 | u64 srcdev_devid, int *mirror_num, | |
6092 | u64 *physical) | |
6093 | { | |
4c664611 | 6094 | struct btrfs_io_context *bioc = NULL; |
5ab56090 LB |
6095 | int num_stripes; |
6096 | int index_srcdev = 0; | |
6097 | int found = 0; | |
6098 | u64 physical_of_found = 0; | |
6099 | int i; | |
6100 | int ret = 0; | |
6101 | ||
6102 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, | |
03793cbb | 6103 | logical, &length, &bioc, NULL, NULL, 0); |
5ab56090 | 6104 | if (ret) { |
4c664611 | 6105 | ASSERT(bioc == NULL); |
5ab56090 LB |
6106 | return ret; |
6107 | } | |
6108 | ||
4c664611 | 6109 | num_stripes = bioc->num_stripes; |
5ab56090 LB |
6110 | if (*mirror_num > num_stripes) { |
6111 | /* | |
6112 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror, | |
6113 | * that means that the requested area is not left of the left | |
6114 | * cursor | |
6115 | */ | |
4c664611 | 6116 | btrfs_put_bioc(bioc); |
5ab56090 LB |
6117 | return -EIO; |
6118 | } | |
6119 | ||
6120 | /* | |
6121 | * process the rest of the function using the mirror_num of the source | |
6122 | * drive. Therefore look it up first. At the end, patch the device | |
6123 | * pointer to the one of the target drive. | |
6124 | */ | |
6125 | for (i = 0; i < num_stripes; i++) { | |
4c664611 | 6126 | if (bioc->stripes[i].dev->devid != srcdev_devid) |
5ab56090 LB |
6127 | continue; |
6128 | ||
6129 | /* | |
6130 | * In case of DUP, in order to keep it simple, only add the | |
6131 | * mirror with the lowest physical address | |
6132 | */ | |
6133 | if (found && | |
4c664611 | 6134 | physical_of_found <= bioc->stripes[i].physical) |
5ab56090 LB |
6135 | continue; |
6136 | ||
6137 | index_srcdev = i; | |
6138 | found = 1; | |
4c664611 | 6139 | physical_of_found = bioc->stripes[i].physical; |
5ab56090 LB |
6140 | } |
6141 | ||
4c664611 | 6142 | btrfs_put_bioc(bioc); |
5ab56090 LB |
6143 | |
6144 | ASSERT(found); | |
6145 | if (!found) | |
6146 | return -EIO; | |
6147 | ||
6148 | *mirror_num = index_srcdev + 1; | |
6149 | *physical = physical_of_found; | |
6150 | return ret; | |
6151 | } | |
6152 | ||
6143c23c NA |
6153 | static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) |
6154 | { | |
6155 | struct btrfs_block_group *cache; | |
6156 | bool ret; | |
6157 | ||
de17addc | 6158 | /* Non zoned filesystem does not use "to_copy" flag */ |
6143c23c NA |
6159 | if (!btrfs_is_zoned(fs_info)) |
6160 | return false; | |
6161 | ||
6162 | cache = btrfs_lookup_block_group(fs_info, logical); | |
6163 | ||
3349b57f | 6164 | ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); |
6143c23c NA |
6165 | |
6166 | btrfs_put_block_group(cache); | |
6167 | return ret; | |
6168 | } | |
6169 | ||
73c0f228 | 6170 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
4c664611 | 6171 | struct btrfs_io_context **bioc_ret, |
73c0f228 | 6172 | struct btrfs_dev_replace *dev_replace, |
6143c23c | 6173 | u64 logical, |
73c0f228 LB |
6174 | int *num_stripes_ret, int *max_errors_ret) |
6175 | { | |
4c664611 | 6176 | struct btrfs_io_context *bioc = *bioc_ret; |
73c0f228 LB |
6177 | u64 srcdev_devid = dev_replace->srcdev->devid; |
6178 | int tgtdev_indexes = 0; | |
6179 | int num_stripes = *num_stripes_ret; | |
6180 | int max_errors = *max_errors_ret; | |
6181 | int i; | |
6182 | ||
6183 | if (op == BTRFS_MAP_WRITE) { | |
6184 | int index_where_to_add; | |
6185 | ||
6143c23c NA |
6186 | /* |
6187 | * A block group which have "to_copy" set will eventually | |
6188 | * copied by dev-replace process. We can avoid cloning IO here. | |
6189 | */ | |
6190 | if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) | |
6191 | return; | |
6192 | ||
73c0f228 LB |
6193 | /* |
6194 | * duplicate the write operations while the dev replace | |
6195 | * procedure is running. Since the copying of the old disk to | |
6196 | * the new disk takes place at run time while the filesystem is | |
6197 | * mounted writable, the regular write operations to the old | |
6198 | * disk have to be duplicated to go to the new disk as well. | |
6199 | * | |
6200 | * Note that device->missing is handled by the caller, and that | |
6201 | * the write to the old disk is already set up in the stripes | |
6202 | * array. | |
6203 | */ | |
6204 | index_where_to_add = num_stripes; | |
6205 | for (i = 0; i < num_stripes; i++) { | |
4c664611 | 6206 | if (bioc->stripes[i].dev->devid == srcdev_devid) { |
73c0f228 | 6207 | /* write to new disk, too */ |
4c664611 QW |
6208 | struct btrfs_io_stripe *new = |
6209 | bioc->stripes + index_where_to_add; | |
6210 | struct btrfs_io_stripe *old = | |
6211 | bioc->stripes + i; | |
73c0f228 LB |
6212 | |
6213 | new->physical = old->physical; | |
73c0f228 | 6214 | new->dev = dev_replace->tgtdev; |
4c664611 | 6215 | bioc->tgtdev_map[i] = index_where_to_add; |
73c0f228 LB |
6216 | index_where_to_add++; |
6217 | max_errors++; | |
6218 | tgtdev_indexes++; | |
6219 | } | |
6220 | } | |
6221 | num_stripes = index_where_to_add; | |
6222 | } else if (op == BTRFS_MAP_GET_READ_MIRRORS) { | |
6223 | int index_srcdev = 0; | |
6224 | int found = 0; | |
6225 | u64 physical_of_found = 0; | |
6226 | ||
6227 | /* | |
6228 | * During the dev-replace procedure, the target drive can also | |
6229 | * be used to read data in case it is needed to repair a corrupt | |
6230 | * block elsewhere. This is possible if the requested area is | |
6231 | * left of the left cursor. In this area, the target drive is a | |
6232 | * full copy of the source drive. | |
6233 | */ | |
6234 | for (i = 0; i < num_stripes; i++) { | |
4c664611 | 6235 | if (bioc->stripes[i].dev->devid == srcdev_devid) { |
73c0f228 LB |
6236 | /* |
6237 | * In case of DUP, in order to keep it simple, | |
6238 | * only add the mirror with the lowest physical | |
6239 | * address | |
6240 | */ | |
6241 | if (found && | |
4c664611 | 6242 | physical_of_found <= bioc->stripes[i].physical) |
73c0f228 LB |
6243 | continue; |
6244 | index_srcdev = i; | |
6245 | found = 1; | |
4c664611 | 6246 | physical_of_found = bioc->stripes[i].physical; |
73c0f228 LB |
6247 | } |
6248 | } | |
6249 | if (found) { | |
4c664611 QW |
6250 | struct btrfs_io_stripe *tgtdev_stripe = |
6251 | bioc->stripes + num_stripes; | |
73c0f228 LB |
6252 | |
6253 | tgtdev_stripe->physical = physical_of_found; | |
73c0f228 | 6254 | tgtdev_stripe->dev = dev_replace->tgtdev; |
4c664611 | 6255 | bioc->tgtdev_map[index_srcdev] = num_stripes; |
73c0f228 LB |
6256 | |
6257 | tgtdev_indexes++; | |
6258 | num_stripes++; | |
6259 | } | |
6260 | } | |
6261 | ||
6262 | *num_stripes_ret = num_stripes; | |
6263 | *max_errors_ret = max_errors; | |
4c664611 QW |
6264 | bioc->num_tgtdevs = tgtdev_indexes; |
6265 | *bioc_ret = bioc; | |
73c0f228 LB |
6266 | } |
6267 | ||
2b19a1fe LB |
6268 | static bool need_full_stripe(enum btrfs_map_op op) |
6269 | { | |
6270 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
6271 | } | |
6272 | ||
5f141126 | 6273 | /* |
42034313 MR |
6274 | * Calculate the geometry of a particular (address, len) tuple. This |
6275 | * information is used to calculate how big a particular bio can get before it | |
6276 | * straddles a stripe. | |
5f141126 | 6277 | * |
42034313 MR |
6278 | * @fs_info: the filesystem |
6279 | * @em: mapping containing the logical extent | |
6280 | * @op: type of operation - write or read | |
6281 | * @logical: address that we want to figure out the geometry of | |
42034313 | 6282 | * @io_geom: pointer used to return values |
5f141126 NB |
6283 | * |
6284 | * Returns < 0 in case a chunk for the given logical address cannot be found, | |
6285 | * usually shouldn't happen unless @logical is corrupted, 0 otherwise. | |
6286 | */ | |
42034313 | 6287 | int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em, |
43c0d1a5 | 6288 | enum btrfs_map_op op, u64 logical, |
42034313 | 6289 | struct btrfs_io_geometry *io_geom) |
5f141126 | 6290 | { |
5f141126 | 6291 | struct map_lookup *map; |
43c0d1a5 | 6292 | u64 len; |
5f141126 NB |
6293 | u64 offset; |
6294 | u64 stripe_offset; | |
6295 | u64 stripe_nr; | |
cc353a8b | 6296 | u32 stripe_len; |
5f141126 NB |
6297 | u64 raid56_full_stripe_start = (u64)-1; |
6298 | int data_stripes; | |
6299 | ||
6300 | ASSERT(op != BTRFS_MAP_DISCARD); | |
6301 | ||
5f141126 NB |
6302 | map = em->map_lookup; |
6303 | /* Offset of this logical address in the chunk */ | |
6304 | offset = logical - em->start; | |
6305 | /* Len of a stripe in a chunk */ | |
6306 | stripe_len = map->stripe_len; | |
cc353a8b QW |
6307 | /* |
6308 | * Stripe_nr is where this block falls in | |
6309 | * stripe_offset is the offset of this block in its stripe. | |
6310 | */ | |
6311 | stripe_nr = div64_u64_rem(offset, stripe_len, &stripe_offset); | |
6312 | ASSERT(stripe_offset < U32_MAX); | |
5f141126 | 6313 | |
5f141126 NB |
6314 | data_stripes = nr_data_stripes(map); |
6315 | ||
bf08387f QW |
6316 | /* Only stripe based profiles needs to check against stripe length. */ |
6317 | if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) { | |
5f141126 NB |
6318 | u64 max_len = stripe_len - stripe_offset; |
6319 | ||
6320 | /* | |
6321 | * In case of raid56, we need to know the stripe aligned start | |
6322 | */ | |
6323 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
6324 | unsigned long full_stripe_len = stripe_len * data_stripes; | |
6325 | raid56_full_stripe_start = offset; | |
6326 | ||
6327 | /* | |
6328 | * Allow a write of a full stripe, but make sure we | |
6329 | * don't allow straddling of stripes | |
6330 | */ | |
6331 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
6332 | full_stripe_len); | |
6333 | raid56_full_stripe_start *= full_stripe_len; | |
6334 | ||
6335 | /* | |
6336 | * For writes to RAID[56], allow a full stripeset across | |
6337 | * all disks. For other RAID types and for RAID[56] | |
6338 | * reads, just allow a single stripe (on a single disk). | |
6339 | */ | |
6340 | if (op == BTRFS_MAP_WRITE) { | |
6341 | max_len = stripe_len * data_stripes - | |
6342 | (offset - raid56_full_stripe_start); | |
6343 | } | |
6344 | } | |
6345 | len = min_t(u64, em->len - offset, max_len); | |
6346 | } else { | |
6347 | len = em->len - offset; | |
6348 | } | |
6349 | ||
6350 | io_geom->len = len; | |
6351 | io_geom->offset = offset; | |
6352 | io_geom->stripe_len = stripe_len; | |
6353 | io_geom->stripe_nr = stripe_nr; | |
6354 | io_geom->stripe_offset = stripe_offset; | |
6355 | io_geom->raid56_stripe_offset = raid56_full_stripe_start; | |
6356 | ||
42034313 | 6357 | return 0; |
5f141126 NB |
6358 | } |
6359 | ||
03793cbb CH |
6360 | static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, |
6361 | u32 stripe_index, u64 stripe_offset, u64 stripe_nr) | |
6362 | { | |
6363 | dst->dev = map->stripes[stripe_index].dev; | |
6364 | dst->physical = map->stripes[stripe_index].physical + | |
6365 | stripe_offset + stripe_nr * map->stripe_len; | |
6366 | } | |
6367 | ||
103c1972 CH |
6368 | int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
6369 | u64 logical, u64 *length, | |
6370 | struct btrfs_io_context **bioc_ret, | |
6371 | struct btrfs_io_stripe *smap, int *mirror_num_ret, | |
6372 | int need_raid_map) | |
0b86a832 CM |
6373 | { |
6374 | struct extent_map *em; | |
6375 | struct map_lookup *map; | |
593060d7 CM |
6376 | u64 stripe_offset; |
6377 | u64 stripe_nr; | |
53b381b3 | 6378 | u64 stripe_len; |
9d644a62 | 6379 | u32 stripe_index; |
cff82672 | 6380 | int data_stripes; |
cea9e445 | 6381 | int i; |
de11cc12 | 6382 | int ret = 0; |
03793cbb | 6383 | int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); |
f2d8d74d | 6384 | int num_stripes; |
a236aed1 | 6385 | int max_errors = 0; |
2c8cdd6e | 6386 | int tgtdev_indexes = 0; |
4c664611 | 6387 | struct btrfs_io_context *bioc = NULL; |
472262f3 SB |
6388 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6389 | int dev_replace_is_ongoing = 0; | |
6390 | int num_alloc_stripes; | |
ad6d620e SB |
6391 | int patch_the_first_stripe_for_dev_replace = 0; |
6392 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 6393 | u64 raid56_full_stripe_start = (u64)-1; |
89b798ad NB |
6394 | struct btrfs_io_geometry geom; |
6395 | ||
4c664611 | 6396 | ASSERT(bioc_ret); |
75fb2e9e | 6397 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6398 | |
42034313 MR |
6399 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
6400 | ASSERT(!IS_ERR(em)); | |
6401 | ||
43c0d1a5 | 6402 | ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom); |
89b798ad NB |
6403 | if (ret < 0) |
6404 | return ret; | |
0b86a832 | 6405 | |
95617d69 | 6406 | map = em->map_lookup; |
593060d7 | 6407 | |
89b798ad | 6408 | *length = geom.len; |
89b798ad NB |
6409 | stripe_len = geom.stripe_len; |
6410 | stripe_nr = geom.stripe_nr; | |
6411 | stripe_offset = geom.stripe_offset; | |
6412 | raid56_full_stripe_start = geom.raid56_stripe_offset; | |
cff82672 | 6413 | data_stripes = nr_data_stripes(map); |
593060d7 | 6414 | |
cb5583dd | 6415 | down_read(&dev_replace->rwsem); |
472262f3 | 6416 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6417 | /* |
6418 | * Hold the semaphore for read during the whole operation, write is | |
6419 | * requested at commit time but must wait. | |
6420 | */ | |
472262f3 | 6421 | if (!dev_replace_is_ongoing) |
cb5583dd | 6422 | up_read(&dev_replace->rwsem); |
472262f3 | 6423 | |
ad6d620e | 6424 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
2b19a1fe | 6425 | !need_full_stripe(op) && dev_replace->tgtdev != NULL) { |
5ab56090 LB |
6426 | ret = get_extra_mirror_from_replace(fs_info, logical, *length, |
6427 | dev_replace->srcdev->devid, | |
6428 | &mirror_num, | |
6429 | &physical_to_patch_in_first_stripe); | |
6430 | if (ret) | |
ad6d620e | 6431 | goto out; |
5ab56090 LB |
6432 | else |
6433 | patch_the_first_stripe_for_dev_replace = 1; | |
ad6d620e SB |
6434 | } else if (mirror_num > map->num_stripes) { |
6435 | mirror_num = 0; | |
6436 | } | |
6437 | ||
f2d8d74d | 6438 | num_stripes = 1; |
cea9e445 | 6439 | stripe_index = 0; |
fce3bb9a | 6440 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
47c5713f DS |
6441 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6442 | &stripe_index); | |
de483734 | 6443 | if (!need_full_stripe(op)) |
28e1cc7d | 6444 | mirror_num = 1; |
c7369b3f | 6445 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6446 | if (need_full_stripe(op)) |
f2d8d74d | 6447 | num_stripes = map->num_stripes; |
2fff734f | 6448 | else if (mirror_num) |
f188591e | 6449 | stripe_index = mirror_num - 1; |
dfe25020 | 6450 | else { |
30d9861f | 6451 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6452 | dev_replace_is_ongoing); |
a1d3c478 | 6453 | mirror_num = stripe_index + 1; |
dfe25020 | 6454 | } |
2fff734f | 6455 | |
611f0e00 | 6456 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6457 | if (need_full_stripe(op)) { |
f2d8d74d | 6458 | num_stripes = map->num_stripes; |
a1d3c478 | 6459 | } else if (mirror_num) { |
f188591e | 6460 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6461 | } else { |
6462 | mirror_num = 1; | |
6463 | } | |
2fff734f | 6464 | |
321aecc6 | 6465 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6466 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6467 | |
47c5713f | 6468 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
6469 | stripe_index *= map->sub_stripes; |
6470 | ||
de483734 | 6471 | if (need_full_stripe(op)) |
f2d8d74d | 6472 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6473 | else if (mirror_num) |
6474 | stripe_index += mirror_num - 1; | |
dfe25020 | 6475 | else { |
3e74317a | 6476 | int old_stripe_index = stripe_index; |
30d9861f SB |
6477 | stripe_index = find_live_mirror(fs_info, map, |
6478 | stripe_index, | |
30d9861f | 6479 | dev_replace_is_ongoing); |
3e74317a | 6480 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6481 | } |
53b381b3 | 6482 | |
ffe2d203 | 6483 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
ff18a4af | 6484 | ASSERT(map->stripe_len == BTRFS_STRIPE_LEN); |
de483734 | 6485 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
53b381b3 | 6486 | /* push stripe_nr back to the start of the full stripe */ |
42c61ab6 | 6487 | stripe_nr = div64_u64(raid56_full_stripe_start, |
cff82672 | 6488 | stripe_len * data_stripes); |
53b381b3 DW |
6489 | |
6490 | /* RAID[56] write or recovery. Return all stripes */ | |
6491 | num_stripes = map->num_stripes; | |
6dead96c | 6492 | max_errors = btrfs_chunk_max_errors(map); |
53b381b3 | 6493 | |
462b0b2a QW |
6494 | /* Return the length to the full stripe end */ |
6495 | *length = min(logical + *length, | |
6496 | raid56_full_stripe_start + em->start + | |
6497 | data_stripes * stripe_len) - logical; | |
53b381b3 DW |
6498 | stripe_index = 0; |
6499 | stripe_offset = 0; | |
6500 | } else { | |
6501 | /* | |
6502 | * Mirror #0 or #1 means the original data block. | |
6503 | * Mirror #2 is RAID5 parity block. | |
6504 | * Mirror #3 is RAID6 Q block. | |
6505 | */ | |
47c5713f | 6506 | stripe_nr = div_u64_rem(stripe_nr, |
cff82672 | 6507 | data_stripes, &stripe_index); |
53b381b3 | 6508 | if (mirror_num > 1) |
cff82672 | 6509 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6510 | |
6511 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
6512 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
6513 | &stripe_index); | |
de483734 | 6514 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6515 | mirror_num = 1; |
53b381b3 | 6516 | } |
8790d502 CM |
6517 | } else { |
6518 | /* | |
47c5713f DS |
6519 | * after this, stripe_nr is the number of stripes on this |
6520 | * device we have to walk to find the data, and stripe_index is | |
6521 | * the number of our device in the stripe array | |
8790d502 | 6522 | */ |
47c5713f DS |
6523 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
6524 | &stripe_index); | |
a1d3c478 | 6525 | mirror_num = stripe_index + 1; |
8790d502 | 6526 | } |
e042d1ec | 6527 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6528 | btrfs_crit(fs_info, |
6529 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6530 | stripe_index, map->num_stripes); |
6531 | ret = -EINVAL; | |
6532 | goto out; | |
6533 | } | |
cea9e445 | 6534 | |
472262f3 | 6535 | num_alloc_stripes = num_stripes; |
6fad823f | 6536 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) { |
0b3d4cd3 | 6537 | if (op == BTRFS_MAP_WRITE) |
ad6d620e | 6538 | num_alloc_stripes <<= 1; |
cf8cddd3 | 6539 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 6540 | num_alloc_stripes++; |
2c8cdd6e | 6541 | tgtdev_indexes = num_stripes; |
ad6d620e | 6542 | } |
2c8cdd6e | 6543 | |
03793cbb CH |
6544 | /* |
6545 | * If this I/O maps to a single device, try to return the device and | |
6546 | * physical block information on the stack instead of allocating an | |
6547 | * I/O context structure. | |
6548 | */ | |
6549 | if (smap && num_alloc_stripes == 1 && | |
6550 | !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && | |
6551 | (!need_full_stripe(op) || !dev_replace_is_ongoing || | |
6552 | !dev_replace->tgtdev)) { | |
6553 | if (patch_the_first_stripe_for_dev_replace) { | |
6554 | smap->dev = dev_replace->tgtdev; | |
6555 | smap->physical = physical_to_patch_in_first_stripe; | |
6556 | *mirror_num_ret = map->num_stripes + 1; | |
6557 | } else { | |
6558 | set_io_stripe(smap, map, stripe_index, stripe_offset, | |
6559 | stripe_nr); | |
6560 | *mirror_num_ret = mirror_num; | |
6561 | } | |
6562 | *bioc_ret = NULL; | |
6563 | ret = 0; | |
6564 | goto out; | |
6565 | } | |
6566 | ||
731ccf15 | 6567 | bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes); |
4c664611 | 6568 | if (!bioc) { |
de11cc12 LZ |
6569 | ret = -ENOMEM; |
6570 | goto out; | |
6571 | } | |
608769a4 NB |
6572 | |
6573 | for (i = 0; i < num_stripes; i++) { | |
03793cbb CH |
6574 | set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset, |
6575 | stripe_nr); | |
608769a4 NB |
6576 | stripe_index++; |
6577 | } | |
de11cc12 | 6578 | |
4c664611 | 6579 | /* Build raid_map */ |
2b19a1fe LB |
6580 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6581 | (need_full_stripe(op) || mirror_num > 1)) { | |
8e5cfb55 | 6582 | u64 tmp; |
9d644a62 | 6583 | unsigned rot; |
8e5cfb55 | 6584 | |
8e5cfb55 | 6585 | /* Work out the disk rotation on this stripe-set */ |
47c5713f | 6586 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
6587 | |
6588 | /* Fill in the logical address of each stripe */ | |
cff82672 DS |
6589 | tmp = stripe_nr * data_stripes; |
6590 | for (i = 0; i < data_stripes; i++) | |
4c664611 | 6591 | bioc->raid_map[(i + rot) % num_stripes] = |
8e5cfb55 ZL |
6592 | em->start + (tmp + i) * map->stripe_len; |
6593 | ||
4c664611 | 6594 | bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE; |
8e5cfb55 | 6595 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) |
4c664611 | 6596 | bioc->raid_map[(i + rot + 1) % num_stripes] = |
8e5cfb55 | 6597 | RAID6_Q_STRIPE; |
8e5cfb55 | 6598 | |
4c664611 | 6599 | sort_parity_stripes(bioc, num_stripes); |
593060d7 | 6600 | } |
de11cc12 | 6601 | |
2b19a1fe | 6602 | if (need_full_stripe(op)) |
d20983b4 | 6603 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6604 | |
73c0f228 | 6605 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6606 | need_full_stripe(op)) { |
4c664611 | 6607 | handle_ops_on_dev_replace(op, &bioc, dev_replace, logical, |
6143c23c | 6608 | &num_stripes, &max_errors); |
472262f3 SB |
6609 | } |
6610 | ||
4c664611 QW |
6611 | *bioc_ret = bioc; |
6612 | bioc->map_type = map->type; | |
6613 | bioc->num_stripes = num_stripes; | |
6614 | bioc->max_errors = max_errors; | |
6615 | bioc->mirror_num = mirror_num; | |
ad6d620e SB |
6616 | |
6617 | /* | |
6618 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
6619 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
6620 | * available as a mirror | |
6621 | */ | |
6622 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
6623 | WARN_ON(num_stripes > 1); | |
4c664611 QW |
6624 | bioc->stripes[0].dev = dev_replace->tgtdev; |
6625 | bioc->stripes[0].physical = physical_to_patch_in_first_stripe; | |
6626 | bioc->mirror_num = map->num_stripes + 1; | |
ad6d620e | 6627 | } |
cea9e445 | 6628 | out: |
73beece9 | 6629 | if (dev_replace_is_ongoing) { |
53176dde DS |
6630 | lockdep_assert_held(&dev_replace->rwsem); |
6631 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6632 | up_read(&dev_replace->rwsem); |
73beece9 | 6633 | } |
0b86a832 | 6634 | free_extent_map(em); |
de11cc12 | 6635 | return ret; |
0b86a832 CM |
6636 | } |
6637 | ||
cf8cddd3 | 6638 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6639 | u64 logical, u64 *length, |
4c664611 | 6640 | struct btrfs_io_context **bioc_ret, int mirror_num) |
f2d8d74d | 6641 | { |
4c664611 | 6642 | return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, |
03793cbb | 6643 | NULL, &mirror_num, 0); |
f2d8d74d CM |
6644 | } |
6645 | ||
af8e2d1d | 6646 | /* For Scrub/replace */ |
cf8cddd3 | 6647 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6648 | u64 logical, u64 *length, |
4c664611 | 6649 | struct btrfs_io_context **bioc_ret) |
af8e2d1d | 6650 | { |
03793cbb CH |
6651 | return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, |
6652 | NULL, NULL, 1); | |
af8e2d1d MX |
6653 | } |
6654 | ||
562d7b15 JB |
6655 | static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, |
6656 | const struct btrfs_fs_devices *fs_devices) | |
6657 | { | |
6658 | if (args->fsid == NULL) | |
6659 | return true; | |
6660 | if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) | |
6661 | return true; | |
6662 | return false; | |
6663 | } | |
6664 | ||
6665 | static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, | |
6666 | const struct btrfs_device *device) | |
6667 | { | |
0fca385d LS |
6668 | if (args->missing) { |
6669 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && | |
6670 | !device->bdev) | |
6671 | return true; | |
6672 | return false; | |
6673 | } | |
562d7b15 | 6674 | |
0fca385d | 6675 | if (device->devid != args->devid) |
562d7b15 JB |
6676 | return false; |
6677 | if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) | |
6678 | return false; | |
0fca385d | 6679 | return true; |
562d7b15 JB |
6680 | } |
6681 | ||
09ba3bc9 AJ |
6682 | /* |
6683 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6684 | * return NULL. | |
6685 | * | |
6686 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6687 | * only devid is used. | |
09ba3bc9 | 6688 | */ |
562d7b15 JB |
6689 | struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, |
6690 | const struct btrfs_dev_lookup_args *args) | |
0b86a832 | 6691 | { |
2b82032c | 6692 | struct btrfs_device *device; |
944d3f9f NB |
6693 | struct btrfs_fs_devices *seed_devs; |
6694 | ||
562d7b15 | 6695 | if (dev_args_match_fs_devices(args, fs_devices)) { |
944d3f9f | 6696 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
562d7b15 | 6697 | if (dev_args_match_device(args, device)) |
944d3f9f NB |
6698 | return device; |
6699 | } | |
6700 | } | |
2b82032c | 6701 | |
944d3f9f | 6702 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
562d7b15 JB |
6703 | if (!dev_args_match_fs_devices(args, seed_devs)) |
6704 | continue; | |
6705 | list_for_each_entry(device, &seed_devs->devices, dev_list) { | |
6706 | if (dev_args_match_device(args, device)) | |
6707 | return device; | |
2b82032c | 6708 | } |
2b82032c | 6709 | } |
944d3f9f | 6710 | |
2b82032c | 6711 | return NULL; |
0b86a832 CM |
6712 | } |
6713 | ||
2ff7e61e | 6714 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6715 | u64 devid, u8 *dev_uuid) |
6716 | { | |
6717 | struct btrfs_device *device; | |
fccc0007 | 6718 | unsigned int nofs_flag; |
dfe25020 | 6719 | |
fccc0007 JB |
6720 | /* |
6721 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6722 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6723 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6724 | * places. | |
6725 | */ | |
bb21e302 | 6726 | |
fccc0007 | 6727 | nofs_flag = memalloc_nofs_save(); |
bb21e302 | 6728 | device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); |
fccc0007 | 6729 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6730 | if (IS_ERR(device)) |
adfb69af | 6731 | return device; |
12bd2fc0 ID |
6732 | |
6733 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6734 | device->fs_devices = fs_devices; |
dfe25020 | 6735 | fs_devices->num_devices++; |
12bd2fc0 | 6736 | |
e6e674bd | 6737 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6738 | fs_devices->missing_devices++; |
12bd2fc0 | 6739 | |
dfe25020 CM |
6740 | return device; |
6741 | } | |
6742 | ||
43dd529a DS |
6743 | /* |
6744 | * Allocate new device struct, set up devid and UUID. | |
6745 | * | |
12bd2fc0 ID |
6746 | * @fs_info: used only for generating a new devid, can be NULL if |
6747 | * devid is provided (i.e. @devid != NULL). | |
6748 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6749 | * is generated. | |
6750 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6751 | * is generated. | |
bb21e302 | 6752 | * @path: a pointer to device path if available, NULL otherwise. |
12bd2fc0 ID |
6753 | * |
6754 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6755 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6756 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6757 | */ |
6758 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
bb21e302 AJ |
6759 | const u64 *devid, const u8 *uuid, |
6760 | const char *path) | |
12bd2fc0 ID |
6761 | { |
6762 | struct btrfs_device *dev; | |
6763 | u64 tmp; | |
6764 | ||
fae7f21c | 6765 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6766 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6767 | |
fe4f46d4 DS |
6768 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
6769 | if (!dev) | |
6770 | return ERR_PTR(-ENOMEM); | |
6771 | ||
fe4f46d4 DS |
6772 | INIT_LIST_HEAD(&dev->dev_list); |
6773 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
6774 | INIT_LIST_HEAD(&dev->post_commit_list); | |
6775 | ||
fe4f46d4 DS |
6776 | atomic_set(&dev->dev_stats_ccnt, 0); |
6777 | btrfs_device_data_ordered_init(dev); | |
35da5a7e | 6778 | extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); |
12bd2fc0 ID |
6779 | |
6780 | if (devid) | |
6781 | tmp = *devid; | |
6782 | else { | |
6783 | int ret; | |
6784 | ||
6785 | ret = find_next_devid(fs_info, &tmp); | |
6786 | if (ret) { | |
a425f9d4 | 6787 | btrfs_free_device(dev); |
12bd2fc0 ID |
6788 | return ERR_PTR(ret); |
6789 | } | |
6790 | } | |
6791 | dev->devid = tmp; | |
6792 | ||
6793 | if (uuid) | |
6794 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6795 | else | |
6796 | generate_random_uuid(dev->uuid); | |
6797 | ||
bb21e302 AJ |
6798 | if (path) { |
6799 | struct rcu_string *name; | |
6800 | ||
6801 | name = rcu_string_strdup(path, GFP_KERNEL); | |
6802 | if (!name) { | |
6803 | btrfs_free_device(dev); | |
6804 | return ERR_PTR(-ENOMEM); | |
6805 | } | |
6806 | rcu_assign_pointer(dev->name, name); | |
6807 | } | |
6808 | ||
12bd2fc0 ID |
6809 | return dev; |
6810 | } | |
6811 | ||
5a2b8e60 | 6812 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6813 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6814 | { |
2b902dfc AJ |
6815 | if (error) |
6816 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6817 | devid, uuid); | |
6818 | else | |
6819 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6820 | devid, uuid); | |
5a2b8e60 AJ |
6821 | } |
6822 | ||
bc88b486 | 6823 | u64 btrfs_calc_stripe_length(const struct extent_map *em) |
39e264a4 | 6824 | { |
bc88b486 QW |
6825 | const struct map_lookup *map = em->map_lookup; |
6826 | const int data_stripes = calc_data_stripes(map->type, map->num_stripes); | |
e4f6c6be | 6827 | |
bc88b486 | 6828 | return div_u64(em->len, data_stripes); |
39e264a4 NB |
6829 | } |
6830 | ||
e9306ad4 QW |
6831 | #if BITS_PER_LONG == 32 |
6832 | /* | |
6833 | * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE | |
6834 | * can't be accessed on 32bit systems. | |
6835 | * | |
6836 | * This function do mount time check to reject the fs if it already has | |
6837 | * metadata chunk beyond that limit. | |
6838 | */ | |
6839 | static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6840 | u64 logical, u64 length, u64 type) | |
6841 | { | |
6842 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6843 | return 0; | |
6844 | ||
6845 | if (logical + length < MAX_LFS_FILESIZE) | |
6846 | return 0; | |
6847 | ||
6848 | btrfs_err_32bit_limit(fs_info); | |
6849 | return -EOVERFLOW; | |
6850 | } | |
6851 | ||
6852 | /* | |
6853 | * This is to give early warning for any metadata chunk reaching | |
6854 | * BTRFS_32BIT_EARLY_WARN_THRESHOLD. | |
6855 | * Although we can still access the metadata, it's not going to be possible | |
6856 | * once the limit is reached. | |
6857 | */ | |
6858 | static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6859 | u64 logical, u64 length, u64 type) | |
6860 | { | |
6861 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6862 | return; | |
6863 | ||
6864 | if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6865 | return; | |
6866 | ||
6867 | btrfs_warn_32bit_limit(fs_info); | |
6868 | } | |
6869 | #endif | |
6870 | ||
ff37c89f NB |
6871 | static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, |
6872 | u64 devid, u8 *uuid) | |
6873 | { | |
6874 | struct btrfs_device *dev; | |
6875 | ||
6876 | if (!btrfs_test_opt(fs_info, DEGRADED)) { | |
6877 | btrfs_report_missing_device(fs_info, devid, uuid, true); | |
6878 | return ERR_PTR(-ENOENT); | |
6879 | } | |
6880 | ||
6881 | dev = add_missing_dev(fs_info->fs_devices, devid, uuid); | |
6882 | if (IS_ERR(dev)) { | |
6883 | btrfs_err(fs_info, "failed to init missing device %llu: %ld", | |
6884 | devid, PTR_ERR(dev)); | |
6885 | return dev; | |
6886 | } | |
6887 | btrfs_report_missing_device(fs_info, devid, uuid, false); | |
6888 | ||
6889 | return dev; | |
6890 | } | |
6891 | ||
9690ac09 | 6892 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6893 | struct btrfs_chunk *chunk) |
6894 | { | |
562d7b15 | 6895 | BTRFS_DEV_LOOKUP_ARGS(args); |
9690ac09 | 6896 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6897 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6898 | struct map_lookup *map; |
6899 | struct extent_map *em; | |
6900 | u64 logical; | |
6901 | u64 length; | |
e06cd3dd | 6902 | u64 devid; |
e9306ad4 | 6903 | u64 type; |
e06cd3dd | 6904 | u8 uuid[BTRFS_UUID_SIZE]; |
76a66ba1 | 6905 | int index; |
e06cd3dd LB |
6906 | int num_stripes; |
6907 | int ret; | |
6908 | int i; | |
6909 | ||
6910 | logical = key->offset; | |
6911 | length = btrfs_chunk_length(leaf, chunk); | |
e9306ad4 | 6912 | type = btrfs_chunk_type(leaf, chunk); |
76a66ba1 | 6913 | index = btrfs_bg_flags_to_raid_index(type); |
e06cd3dd LB |
6914 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6915 | ||
e9306ad4 QW |
6916 | #if BITS_PER_LONG == 32 |
6917 | ret = check_32bit_meta_chunk(fs_info, logical, length, type); | |
6918 | if (ret < 0) | |
6919 | return ret; | |
6920 | warn_32bit_meta_chunk(fs_info, logical, length, type); | |
6921 | #endif | |
6922 | ||
075cb3c7 QW |
6923 | /* |
6924 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6925 | * as chunk item in tree block is already verified by tree-checker. | |
6926 | */ | |
6927 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6928 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6929 | if (ret) |
6930 | return ret; | |
6931 | } | |
a061fc8d | 6932 | |
c8bf1b67 DS |
6933 | read_lock(&map_tree->lock); |
6934 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6935 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6936 | |
6937 | /* already mapped? */ | |
6938 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6939 | free_extent_map(em); | |
0b86a832 CM |
6940 | return 0; |
6941 | } else if (em) { | |
6942 | free_extent_map(em); | |
6943 | } | |
0b86a832 | 6944 | |
172ddd60 | 6945 | em = alloc_extent_map(); |
0b86a832 CM |
6946 | if (!em) |
6947 | return -ENOMEM; | |
593060d7 | 6948 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6949 | if (!map) { |
6950 | free_extent_map(em); | |
6951 | return -ENOMEM; | |
6952 | } | |
6953 | ||
298a8f9c | 6954 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6955 | em->map_lookup = map; |
0b86a832 CM |
6956 | em->start = logical; |
6957 | em->len = length; | |
70c8a91c | 6958 | em->orig_start = 0; |
0b86a832 | 6959 | em->block_start = 0; |
c8b97818 | 6960 | em->block_len = em->len; |
0b86a832 | 6961 | |
593060d7 CM |
6962 | map->num_stripes = num_stripes; |
6963 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6964 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
593060d7 | 6965 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
e9306ad4 | 6966 | map->type = type; |
76a66ba1 QW |
6967 | /* |
6968 | * We can't use the sub_stripes value, as for profiles other than | |
6969 | * RAID10, they may have 0 as sub_stripes for filesystems created by | |
6970 | * older mkfs (<v5.4). | |
6971 | * In that case, it can cause divide-by-zero errors later. | |
6972 | * Since currently sub_stripes is fixed for each profile, let's | |
6973 | * use the trusted value instead. | |
6974 | */ | |
6975 | map->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
cf90d884 | 6976 | map->verified_stripes = 0; |
bc88b486 | 6977 | em->orig_block_len = btrfs_calc_stripe_length(em); |
593060d7 CM |
6978 | for (i = 0; i < num_stripes; i++) { |
6979 | map->stripes[i].physical = | |
6980 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6981 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
562d7b15 | 6982 | args.devid = devid; |
a443755f CM |
6983 | read_extent_buffer(leaf, uuid, (unsigned long) |
6984 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6985 | BTRFS_UUID_SIZE); | |
562d7b15 JB |
6986 | args.uuid = uuid; |
6987 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); | |
dfe25020 | 6988 | if (!map->stripes[i].dev) { |
ff37c89f NB |
6989 | map->stripes[i].dev = handle_missing_device(fs_info, |
6990 | devid, uuid); | |
adfb69af | 6991 | if (IS_ERR(map->stripes[i].dev)) { |
1742e1c9 | 6992 | ret = PTR_ERR(map->stripes[i].dev); |
dfe25020 | 6993 | free_extent_map(em); |
1742e1c9 | 6994 | return ret; |
dfe25020 CM |
6995 | } |
6996 | } | |
ff37c89f | 6997 | |
e12c9621 AJ |
6998 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6999 | &(map->stripes[i].dev->dev_state)); | |
0b86a832 CM |
7000 | } |
7001 | ||
c8bf1b67 DS |
7002 | write_lock(&map_tree->lock); |
7003 | ret = add_extent_mapping(map_tree, em, 0); | |
7004 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
7005 | if (ret < 0) { |
7006 | btrfs_err(fs_info, | |
7007 | "failed to add chunk map, start=%llu len=%llu: %d", | |
7008 | em->start, em->len, ret); | |
7009 | } | |
0b86a832 CM |
7010 | free_extent_map(em); |
7011 | ||
64f64f43 | 7012 | return ret; |
0b86a832 CM |
7013 | } |
7014 | ||
143bede5 | 7015 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
7016 | struct btrfs_dev_item *dev_item, |
7017 | struct btrfs_device *device) | |
7018 | { | |
7019 | unsigned long ptr; | |
0b86a832 CM |
7020 | |
7021 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
7022 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
7023 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 7024 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 7025 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 7026 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
7027 | device->type = btrfs_device_type(leaf, dev_item); |
7028 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
7029 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
7030 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 7031 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 7032 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 7033 | |
410ba3a2 | 7034 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 7035 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
7036 | } |
7037 | ||
2ff7e61e | 7038 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 7039 | u8 *fsid) |
2b82032c YZ |
7040 | { |
7041 | struct btrfs_fs_devices *fs_devices; | |
7042 | int ret; | |
7043 | ||
a32bf9a3 | 7044 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 7045 | ASSERT(fsid); |
2b82032c | 7046 | |
427c8fdd | 7047 | /* This will match only for multi-device seed fs */ |
944d3f9f | 7048 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 7049 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
7050 | return fs_devices; |
7051 | ||
2b82032c | 7052 | |
7239ff4b | 7053 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 7054 | if (!fs_devices) { |
0b246afa | 7055 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
7056 | return ERR_PTR(-ENOENT); |
7057 | ||
7239ff4b | 7058 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
7059 | if (IS_ERR(fs_devices)) |
7060 | return fs_devices; | |
7061 | ||
0395d84f | 7062 | fs_devices->seeding = true; |
5f375835 MX |
7063 | fs_devices->opened = 1; |
7064 | return fs_devices; | |
2b82032c | 7065 | } |
e4404d6e | 7066 | |
427c8fdd NB |
7067 | /* |
7068 | * Upon first call for a seed fs fsid, just create a private copy of the | |
7069 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
7070 | */ | |
e4404d6e | 7071 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
7072 | if (IS_ERR(fs_devices)) |
7073 | return fs_devices; | |
2b82032c | 7074 | |
897fb573 | 7075 | ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder); |
48d28232 JL |
7076 | if (ret) { |
7077 | free_fs_devices(fs_devices); | |
c83b60c0 | 7078 | return ERR_PTR(ret); |
48d28232 | 7079 | } |
2b82032c YZ |
7080 | |
7081 | if (!fs_devices->seeding) { | |
0226e0eb | 7082 | close_fs_devices(fs_devices); |
e4404d6e | 7083 | free_fs_devices(fs_devices); |
c83b60c0 | 7084 | return ERR_PTR(-EINVAL); |
2b82032c YZ |
7085 | } |
7086 | ||
944d3f9f | 7087 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
c83b60c0 | 7088 | |
5f375835 | 7089 | return fs_devices; |
2b82032c YZ |
7090 | } |
7091 | ||
17850759 | 7092 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
7093 | struct btrfs_dev_item *dev_item) |
7094 | { | |
562d7b15 | 7095 | BTRFS_DEV_LOOKUP_ARGS(args); |
17850759 | 7096 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 7097 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
7098 | struct btrfs_device *device; |
7099 | u64 devid; | |
7100 | int ret; | |
44880fdc | 7101 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
7102 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
7103 | ||
c1867eb3 DS |
7104 | devid = btrfs_device_id(leaf, dev_item); |
7105 | args.devid = devid; | |
410ba3a2 | 7106 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 7107 | BTRFS_UUID_SIZE); |
1473b24e | 7108 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 7109 | BTRFS_FSID_SIZE); |
562d7b15 JB |
7110 | args.uuid = dev_uuid; |
7111 | args.fsid = fs_uuid; | |
2b82032c | 7112 | |
de37aa51 | 7113 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 7114 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
7115 | if (IS_ERR(fs_devices)) |
7116 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
7117 | } |
7118 | ||
562d7b15 | 7119 | device = btrfs_find_device(fs_info->fs_devices, &args); |
5f375835 | 7120 | if (!device) { |
c5502451 | 7121 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
7122 | btrfs_report_missing_device(fs_info, devid, |
7123 | dev_uuid, true); | |
45dbdbc9 | 7124 | return -ENOENT; |
c5502451 | 7125 | } |
2b82032c | 7126 | |
2ff7e61e | 7127 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
7128 | if (IS_ERR(device)) { |
7129 | btrfs_err(fs_info, | |
7130 | "failed to add missing dev %llu: %ld", | |
7131 | devid, PTR_ERR(device)); | |
7132 | return PTR_ERR(device); | |
7133 | } | |
2b902dfc | 7134 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 7135 | } else { |
c5502451 | 7136 | if (!device->bdev) { |
2b902dfc AJ |
7137 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
7138 | btrfs_report_missing_device(fs_info, | |
7139 | devid, dev_uuid, true); | |
45dbdbc9 | 7140 | return -ENOENT; |
2b902dfc AJ |
7141 | } |
7142 | btrfs_report_missing_device(fs_info, devid, | |
7143 | dev_uuid, false); | |
c5502451 | 7144 | } |
5f375835 | 7145 | |
e6e674bd AJ |
7146 | if (!device->bdev && |
7147 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
7148 | /* |
7149 | * this happens when a device that was properly setup | |
7150 | * in the device info lists suddenly goes bad. | |
7151 | * device->bdev is NULL, and so we have to set | |
7152 | * device->missing to one here | |
7153 | */ | |
5f375835 | 7154 | device->fs_devices->missing_devices++; |
e6e674bd | 7155 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 7156 | } |
5f375835 MX |
7157 | |
7158 | /* Move the device to its own fs_devices */ | |
7159 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
7160 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
7161 | &device->dev_state)); | |
5f375835 MX |
7162 | |
7163 | list_move(&device->dev_list, &fs_devices->devices); | |
7164 | device->fs_devices->num_devices--; | |
7165 | fs_devices->num_devices++; | |
7166 | ||
7167 | device->fs_devices->missing_devices--; | |
7168 | fs_devices->missing_devices++; | |
7169 | ||
7170 | device->fs_devices = fs_devices; | |
7171 | } | |
2b82032c YZ |
7172 | } |
7173 | ||
0b246afa | 7174 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7175 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7176 | if (device->generation != |
7177 | btrfs_device_generation(leaf, dev_item)) | |
7178 | return -EINVAL; | |
6324fbf3 | 7179 | } |
0b86a832 CM |
7180 | |
7181 | fill_device_from_item(leaf, dev_item, device); | |
3a160a93 | 7182 | if (device->bdev) { |
cda00eba | 7183 | u64 max_total_bytes = bdev_nr_bytes(device->bdev); |
3a160a93 AJ |
7184 | |
7185 | if (device->total_bytes > max_total_bytes) { | |
7186 | btrfs_err(fs_info, | |
7187 | "device total_bytes should be at most %llu but found %llu", | |
7188 | max_total_bytes, device->total_bytes); | |
7189 | return -EINVAL; | |
7190 | } | |
7191 | } | |
e12c9621 | 7192 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7193 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7194 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7195 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7196 | atomic64_add(device->total_bytes - device->bytes_used, |
7197 | &fs_info->free_chunk_space); | |
2bf64758 | 7198 | } |
0b86a832 | 7199 | ret = 0; |
0b86a832 CM |
7200 | return ret; |
7201 | } | |
7202 | ||
6bccf3ab | 7203 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7204 | { |
ab8d0fc4 | 7205 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7206 | struct extent_buffer *sb; |
0b86a832 | 7207 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7208 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7209 | u8 *array_ptr; |
7210 | unsigned long sb_array_offset; | |
84eed90f | 7211 | int ret = 0; |
0b86a832 CM |
7212 | u32 num_stripes; |
7213 | u32 array_size; | |
7214 | u32 len = 0; | |
1ffb22cf | 7215 | u32 cur_offset; |
e06cd3dd | 7216 | u64 type; |
84eed90f | 7217 | struct btrfs_key key; |
0b86a832 | 7218 | |
0b246afa | 7219 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
e959d3c1 | 7220 | |
a83fffb7 | 7221 | /* |
e959d3c1 QW |
7222 | * We allocated a dummy extent, just to use extent buffer accessors. |
7223 | * There will be unused space after BTRFS_SUPER_INFO_SIZE, but | |
7224 | * that's fine, we will not go beyond system chunk array anyway. | |
a83fffb7 | 7225 | */ |
e959d3c1 QW |
7226 | sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); |
7227 | if (!sb) | |
7228 | return -ENOMEM; | |
4db8c528 | 7229 | set_extent_buffer_uptodate(sb); |
4008c04a | 7230 | |
a061fc8d | 7231 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7232 | array_size = btrfs_super_sys_array_size(super_copy); |
7233 | ||
1ffb22cf DS |
7234 | array_ptr = super_copy->sys_chunk_array; |
7235 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7236 | cur_offset = 0; | |
0b86a832 | 7237 | |
1ffb22cf DS |
7238 | while (cur_offset < array_size) { |
7239 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7240 | len = sizeof(*disk_key); |
7241 | if (cur_offset + len > array_size) | |
7242 | goto out_short_read; | |
7243 | ||
0b86a832 CM |
7244 | btrfs_disk_key_to_cpu(&key, disk_key); |
7245 | ||
1ffb22cf DS |
7246 | array_ptr += len; |
7247 | sb_array_offset += len; | |
7248 | cur_offset += len; | |
0b86a832 | 7249 | |
32ab3d1b JT |
7250 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
7251 | btrfs_err(fs_info, | |
7252 | "unexpected item type %u in sys_array at offset %u", | |
7253 | (u32)key.type, cur_offset); | |
7254 | ret = -EIO; | |
7255 | break; | |
7256 | } | |
f5cdedd7 | 7257 | |
32ab3d1b JT |
7258 | chunk = (struct btrfs_chunk *)sb_array_offset; |
7259 | /* | |
7260 | * At least one btrfs_chunk with one stripe must be present, | |
7261 | * exact stripe count check comes afterwards | |
7262 | */ | |
7263 | len = btrfs_chunk_item_size(1); | |
7264 | if (cur_offset + len > array_size) | |
7265 | goto out_short_read; | |
e06cd3dd | 7266 | |
32ab3d1b JT |
7267 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
7268 | if (!num_stripes) { | |
7269 | btrfs_err(fs_info, | |
7270 | "invalid number of stripes %u in sys_array at offset %u", | |
7271 | num_stripes, cur_offset); | |
7272 | ret = -EIO; | |
7273 | break; | |
7274 | } | |
e3540eab | 7275 | |
32ab3d1b JT |
7276 | type = btrfs_chunk_type(sb, chunk); |
7277 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7278 | btrfs_err(fs_info, |
32ab3d1b JT |
7279 | "invalid chunk type %llu in sys_array at offset %u", |
7280 | type, cur_offset); | |
84eed90f CM |
7281 | ret = -EIO; |
7282 | break; | |
0b86a832 | 7283 | } |
32ab3d1b JT |
7284 | |
7285 | len = btrfs_chunk_item_size(num_stripes); | |
7286 | if (cur_offset + len > array_size) | |
7287 | goto out_short_read; | |
7288 | ||
7289 | ret = read_one_chunk(&key, sb, chunk); | |
7290 | if (ret) | |
7291 | break; | |
7292 | ||
1ffb22cf DS |
7293 | array_ptr += len; |
7294 | sb_array_offset += len; | |
7295 | cur_offset += len; | |
0b86a832 | 7296 | } |
d865177a | 7297 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7298 | free_extent_buffer_stale(sb); |
84eed90f | 7299 | return ret; |
e3540eab DS |
7300 | |
7301 | out_short_read: | |
ab8d0fc4 | 7302 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7303 | len, cur_offset); |
d865177a | 7304 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7305 | free_extent_buffer_stale(sb); |
e3540eab | 7306 | return -EIO; |
0b86a832 CM |
7307 | } |
7308 | ||
21634a19 QW |
7309 | /* |
7310 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7311 | * | |
6528b99d AJ |
7312 | * If the @failing_dev is specified, it's accounted as missing. |
7313 | * | |
21634a19 QW |
7314 | * Return true if all chunks meet the minimal RW mount requirements. |
7315 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7316 | */ | |
6528b99d AJ |
7317 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7318 | struct btrfs_device *failing_dev) | |
21634a19 | 7319 | { |
c8bf1b67 | 7320 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7321 | struct extent_map *em; |
7322 | u64 next_start = 0; | |
7323 | bool ret = true; | |
7324 | ||
c8bf1b67 DS |
7325 | read_lock(&map_tree->lock); |
7326 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7327 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7328 | /* No chunk at all? Return false anyway */ |
7329 | if (!em) { | |
7330 | ret = false; | |
7331 | goto out; | |
7332 | } | |
7333 | while (em) { | |
7334 | struct map_lookup *map; | |
7335 | int missing = 0; | |
7336 | int max_tolerated; | |
7337 | int i; | |
7338 | ||
7339 | map = em->map_lookup; | |
7340 | max_tolerated = | |
7341 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7342 | map->type); | |
7343 | for (i = 0; i < map->num_stripes; i++) { | |
7344 | struct btrfs_device *dev = map->stripes[i].dev; | |
7345 | ||
e6e674bd AJ |
7346 | if (!dev || !dev->bdev || |
7347 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7348 | dev->last_flush_error) |
7349 | missing++; | |
6528b99d AJ |
7350 | else if (failing_dev && failing_dev == dev) |
7351 | missing++; | |
21634a19 QW |
7352 | } |
7353 | if (missing > max_tolerated) { | |
6528b99d AJ |
7354 | if (!failing_dev) |
7355 | btrfs_warn(fs_info, | |
52042d8e | 7356 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7357 | em->start, missing, max_tolerated); |
7358 | free_extent_map(em); | |
7359 | ret = false; | |
7360 | goto out; | |
7361 | } | |
7362 | next_start = extent_map_end(em); | |
7363 | free_extent_map(em); | |
7364 | ||
c8bf1b67 DS |
7365 | read_lock(&map_tree->lock); |
7366 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7367 | (u64)(-1) - next_start); |
c8bf1b67 | 7368 | read_unlock(&map_tree->lock); |
21634a19 QW |
7369 | } |
7370 | out: | |
7371 | return ret; | |
7372 | } | |
7373 | ||
d85327b1 DS |
7374 | static void readahead_tree_node_children(struct extent_buffer *node) |
7375 | { | |
7376 | int i; | |
7377 | const int nr_items = btrfs_header_nritems(node); | |
7378 | ||
bfb484d9 JB |
7379 | for (i = 0; i < nr_items; i++) |
7380 | btrfs_readahead_node_child(node, i); | |
d85327b1 DS |
7381 | } |
7382 | ||
5b4aacef | 7383 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7384 | { |
5b4aacef | 7385 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7386 | struct btrfs_path *path; |
7387 | struct extent_buffer *leaf; | |
7388 | struct btrfs_key key; | |
7389 | struct btrfs_key found_key; | |
7390 | int ret; | |
7391 | int slot; | |
43cb1478 | 7392 | int iter_ret = 0; |
99e3ecfc | 7393 | u64 total_dev = 0; |
d85327b1 | 7394 | u64 last_ra_node = 0; |
0b86a832 | 7395 | |
0b86a832 CM |
7396 | path = btrfs_alloc_path(); |
7397 | if (!path) | |
7398 | return -ENOMEM; | |
7399 | ||
3dd0f7a3 AJ |
7400 | /* |
7401 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7402 | * otherwise we don't need it. | |
7403 | */ | |
b367e47f | 7404 | mutex_lock(&uuid_mutex); |
b367e47f | 7405 | |
48cfa61b BB |
7406 | /* |
7407 | * It is possible for mount and umount to race in such a way that | |
7408 | * we execute this code path, but open_fs_devices failed to clear | |
7409 | * total_rw_bytes. We certainly want it cleared before reading the | |
7410 | * device items, so clear it here. | |
7411 | */ | |
7412 | fs_info->fs_devices->total_rw_bytes = 0; | |
7413 | ||
4d9380e0 FM |
7414 | /* |
7415 | * Lockdep complains about possible circular locking dependency between | |
7416 | * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores | |
7417 | * used for freeze procection of a fs (struct super_block.s_writers), | |
7418 | * which we take when starting a transaction, and extent buffers of the | |
7419 | * chunk tree if we call read_one_dev() while holding a lock on an | |
7420 | * extent buffer of the chunk tree. Since we are mounting the filesystem | |
7421 | * and at this point there can't be any concurrent task modifying the | |
7422 | * chunk tree, to keep it simple, just skip locking on the chunk tree. | |
7423 | */ | |
7424 | ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); | |
7425 | path->skip_locking = 1; | |
7426 | ||
395927a9 FDBM |
7427 | /* |
7428 | * Read all device items, and then all the chunk items. All | |
7429 | * device items are found before any chunk item (their object id | |
7430 | * is smaller than the lowest possible object id for a chunk | |
7431 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7432 | */ |
7433 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7434 | key.offset = 0; | |
7435 | key.type = 0; | |
43cb1478 GN |
7436 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
7437 | struct extent_buffer *node = path->nodes[1]; | |
d85327b1 | 7438 | |
0b86a832 CM |
7439 | leaf = path->nodes[0]; |
7440 | slot = path->slots[0]; | |
43cb1478 | 7441 | |
d85327b1 DS |
7442 | if (node) { |
7443 | if (last_ra_node != node->start) { | |
7444 | readahead_tree_node_children(node); | |
7445 | last_ra_node = node->start; | |
7446 | } | |
7447 | } | |
395927a9 FDBM |
7448 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7449 | struct btrfs_dev_item *dev_item; | |
7450 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7451 | struct btrfs_dev_item); |
17850759 | 7452 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7453 | if (ret) |
7454 | goto error; | |
99e3ecfc | 7455 | total_dev++; |
0b86a832 CM |
7456 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7457 | struct btrfs_chunk *chunk; | |
79bd3712 FM |
7458 | |
7459 | /* | |
7460 | * We are only called at mount time, so no need to take | |
7461 | * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, | |
7462 | * we always lock first fs_info->chunk_mutex before | |
7463 | * acquiring any locks on the chunk tree. This is a | |
7464 | * requirement for chunk allocation, see the comment on | |
7465 | * top of btrfs_chunk_alloc() for details. | |
7466 | */ | |
0b86a832 | 7467 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
9690ac09 | 7468 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7469 | if (ret) |
7470 | goto error; | |
0b86a832 | 7471 | } |
43cb1478 GN |
7472 | } |
7473 | /* Catch error found during iteration */ | |
7474 | if (iter_ret < 0) { | |
7475 | ret = iter_ret; | |
7476 | goto error; | |
0b86a832 | 7477 | } |
99e3ecfc LB |
7478 | |
7479 | /* | |
7480 | * After loading chunk tree, we've got all device information, | |
7481 | * do another round of validation checks. | |
7482 | */ | |
0b246afa | 7483 | if (total_dev != fs_info->fs_devices->total_devices) { |
d201238c QW |
7484 | btrfs_warn(fs_info, |
7485 | "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", | |
0b246afa | 7486 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc | 7487 | total_dev); |
d201238c QW |
7488 | fs_info->fs_devices->total_devices = total_dev; |
7489 | btrfs_set_super_num_devices(fs_info->super_copy, total_dev); | |
99e3ecfc | 7490 | } |
0b246afa JM |
7491 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7492 | fs_info->fs_devices->total_rw_bytes) { | |
7493 | btrfs_err(fs_info, | |
99e3ecfc | 7494 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7495 | btrfs_super_total_bytes(fs_info->super_copy), |
7496 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7497 | ret = -EINVAL; |
7498 | goto error; | |
7499 | } | |
0b86a832 CM |
7500 | ret = 0; |
7501 | error: | |
b367e47f LZ |
7502 | mutex_unlock(&uuid_mutex); |
7503 | ||
2b82032c | 7504 | btrfs_free_path(path); |
0b86a832 CM |
7505 | return ret; |
7506 | } | |
442a4f63 | 7507 | |
a8d1b164 | 7508 | int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
cb517eab | 7509 | { |
944d3f9f | 7510 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab | 7511 | struct btrfs_device *device; |
a8d1b164 | 7512 | int ret = 0; |
cb517eab | 7513 | |
944d3f9f NB |
7514 | fs_devices->fs_info = fs_info; |
7515 | ||
7516 | mutex_lock(&fs_devices->device_list_mutex); | |
7517 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7518 | device->fs_info = fs_info; | |
944d3f9f NB |
7519 | |
7520 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
a8d1b164 | 7521 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
fb456252 | 7522 | device->fs_info = fs_info; |
a8d1b164 JT |
7523 | ret = btrfs_get_dev_zone_info(device, false); |
7524 | if (ret) | |
7525 | break; | |
7526 | } | |
29cc83f6 | 7527 | |
944d3f9f | 7528 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7529 | } |
e17125b5 | 7530 | mutex_unlock(&fs_devices->device_list_mutex); |
a8d1b164 JT |
7531 | |
7532 | return ret; | |
cb517eab MX |
7533 | } |
7534 | ||
1dc990df DS |
7535 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7536 | const struct btrfs_dev_stats_item *ptr, | |
7537 | int index) | |
7538 | { | |
7539 | u64 val; | |
7540 | ||
7541 | read_extent_buffer(eb, &val, | |
7542 | offsetof(struct btrfs_dev_stats_item, values) + | |
7543 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7544 | sizeof(val)); | |
7545 | return val; | |
7546 | } | |
7547 | ||
7548 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7549 | struct btrfs_dev_stats_item *ptr, | |
7550 | int index, u64 val) | |
7551 | { | |
7552 | write_extent_buffer(eb, &val, | |
7553 | offsetof(struct btrfs_dev_stats_item, values) + | |
7554 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7555 | sizeof(val)); | |
7556 | } | |
7557 | ||
92e26df4 JB |
7558 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
7559 | struct btrfs_path *path) | |
733f4fbb | 7560 | { |
124604eb | 7561 | struct btrfs_dev_stats_item *ptr; |
733f4fbb | 7562 | struct extent_buffer *eb; |
124604eb JB |
7563 | struct btrfs_key key; |
7564 | int item_size; | |
7565 | int i, ret, slot; | |
7566 | ||
82d62d06 JB |
7567 | if (!device->fs_info->dev_root) |
7568 | return 0; | |
7569 | ||
124604eb JB |
7570 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7571 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
7572 | key.offset = device->devid; | |
7573 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); | |
7574 | if (ret) { | |
7575 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7576 | btrfs_dev_stat_set(device, i, 0); | |
7577 | device->dev_stats_valid = 1; | |
7578 | btrfs_release_path(path); | |
92e26df4 | 7579 | return ret < 0 ? ret : 0; |
124604eb JB |
7580 | } |
7581 | slot = path->slots[0]; | |
7582 | eb = path->nodes[0]; | |
3212fa14 | 7583 | item_size = btrfs_item_size(eb, slot); |
124604eb JB |
7584 | |
7585 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); | |
7586 | ||
7587 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7588 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7589 | btrfs_dev_stat_set(device, i, | |
7590 | btrfs_dev_stats_value(eb, ptr, i)); | |
7591 | else | |
7592 | btrfs_dev_stat_set(device, i, 0); | |
7593 | } | |
7594 | ||
7595 | device->dev_stats_valid = 1; | |
7596 | btrfs_dev_stat_print_on_load(device); | |
7597 | btrfs_release_path(path); | |
92e26df4 JB |
7598 | |
7599 | return 0; | |
124604eb JB |
7600 | } |
7601 | ||
7602 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7603 | { | |
7604 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
733f4fbb SB |
7605 | struct btrfs_device *device; |
7606 | struct btrfs_path *path = NULL; | |
92e26df4 | 7607 | int ret = 0; |
733f4fbb SB |
7608 | |
7609 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7610 | if (!path) |
7611 | return -ENOMEM; | |
733f4fbb SB |
7612 | |
7613 | mutex_lock(&fs_devices->device_list_mutex); | |
92e26df4 JB |
7614 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
7615 | ret = btrfs_device_init_dev_stats(device, path); | |
7616 | if (ret) | |
7617 | goto out; | |
7618 | } | |
124604eb | 7619 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
92e26df4 JB |
7620 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
7621 | ret = btrfs_device_init_dev_stats(device, path); | |
7622 | if (ret) | |
7623 | goto out; | |
7624 | } | |
733f4fbb | 7625 | } |
92e26df4 | 7626 | out: |
733f4fbb SB |
7627 | mutex_unlock(&fs_devices->device_list_mutex); |
7628 | ||
733f4fbb | 7629 | btrfs_free_path(path); |
92e26df4 | 7630 | return ret; |
733f4fbb SB |
7631 | } |
7632 | ||
7633 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7634 | struct btrfs_device *device) |
7635 | { | |
5495f195 | 7636 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7637 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7638 | struct btrfs_path *path; |
7639 | struct btrfs_key key; | |
7640 | struct extent_buffer *eb; | |
7641 | struct btrfs_dev_stats_item *ptr; | |
7642 | int ret; | |
7643 | int i; | |
7644 | ||
242e2956 DS |
7645 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7646 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7647 | key.offset = device->devid; |
7648 | ||
7649 | path = btrfs_alloc_path(); | |
fa252992 DS |
7650 | if (!path) |
7651 | return -ENOMEM; | |
733f4fbb SB |
7652 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7653 | if (ret < 0) { | |
0b246afa | 7654 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7655 | "error %d while searching for dev_stats item for device %s", |
cb3e217b | 7656 | ret, btrfs_dev_name(device)); |
733f4fbb SB |
7657 | goto out; |
7658 | } | |
7659 | ||
7660 | if (ret == 0 && | |
3212fa14 | 7661 | btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { |
733f4fbb SB |
7662 | /* need to delete old one and insert a new one */ |
7663 | ret = btrfs_del_item(trans, dev_root, path); | |
7664 | if (ret != 0) { | |
0b246afa | 7665 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7666 | "delete too small dev_stats item for device %s failed %d", |
cb3e217b | 7667 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7668 | goto out; |
7669 | } | |
7670 | ret = 1; | |
7671 | } | |
7672 | ||
7673 | if (ret == 1) { | |
7674 | /* need to insert a new item */ | |
7675 | btrfs_release_path(path); | |
7676 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7677 | &key, sizeof(*ptr)); | |
7678 | if (ret < 0) { | |
0b246afa | 7679 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7680 | "insert dev_stats item for device %s failed %d", |
cb3e217b | 7681 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7682 | goto out; |
7683 | } | |
7684 | } | |
7685 | ||
7686 | eb = path->nodes[0]; | |
7687 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7688 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7689 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7690 | btrfs_dev_stat_read(device, i)); | |
7691 | btrfs_mark_buffer_dirty(eb); | |
7692 | ||
7693 | out: | |
7694 | btrfs_free_path(path); | |
7695 | return ret; | |
7696 | } | |
7697 | ||
7698 | /* | |
7699 | * called from commit_transaction. Writes all changed device stats to disk. | |
7700 | */ | |
196c9d8d | 7701 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7702 | { |
196c9d8d | 7703 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7704 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7705 | struct btrfs_device *device; | |
addc3fa7 | 7706 | int stats_cnt; |
733f4fbb SB |
7707 | int ret = 0; |
7708 | ||
7709 | mutex_lock(&fs_devices->device_list_mutex); | |
7710 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7711 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7712 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7713 | continue; |
7714 | ||
9deae968 NB |
7715 | |
7716 | /* | |
7717 | * There is a LOAD-LOAD control dependency between the value of | |
7718 | * dev_stats_ccnt and updating the on-disk values which requires | |
7719 | * reading the in-memory counters. Such control dependencies | |
7720 | * require explicit read memory barriers. | |
7721 | * | |
7722 | * This memory barriers pairs with smp_mb__before_atomic in | |
7723 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7724 | * barrier implied by atomic_xchg in | |
7725 | * btrfs_dev_stats_read_and_reset | |
7726 | */ | |
7727 | smp_rmb(); | |
7728 | ||
5495f195 | 7729 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7730 | if (!ret) |
addc3fa7 | 7731 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7732 | } |
7733 | mutex_unlock(&fs_devices->device_list_mutex); | |
7734 | ||
7735 | return ret; | |
7736 | } | |
7737 | ||
442a4f63 SB |
7738 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7739 | { | |
7740 | btrfs_dev_stat_inc(dev, index); | |
442a4f63 | 7741 | |
733f4fbb SB |
7742 | if (!dev->dev_stats_valid) |
7743 | return; | |
fb456252 | 7744 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7745 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7746 | btrfs_dev_name(dev), |
442a4f63 SB |
7747 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7748 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7749 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7750 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7751 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7752 | } |
c11d2c23 | 7753 | |
733f4fbb SB |
7754 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7755 | { | |
a98cdb85 SB |
7756 | int i; |
7757 | ||
7758 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7759 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7760 | break; | |
7761 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7762 | return; /* all values == 0, suppress message */ | |
7763 | ||
fb456252 | 7764 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7765 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7766 | btrfs_dev_name(dev), |
733f4fbb SB |
7767 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7768 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7769 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7770 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7771 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7772 | } | |
7773 | ||
2ff7e61e | 7774 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7775 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 | 7776 | { |
562d7b15 | 7777 | BTRFS_DEV_LOOKUP_ARGS(args); |
c11d2c23 | 7778 | struct btrfs_device *dev; |
0b246afa | 7779 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7780 | int i; |
7781 | ||
7782 | mutex_lock(&fs_devices->device_list_mutex); | |
562d7b15 JB |
7783 | args.devid = stats->devid; |
7784 | dev = btrfs_find_device(fs_info->fs_devices, &args); | |
c11d2c23 SB |
7785 | mutex_unlock(&fs_devices->device_list_mutex); |
7786 | ||
7787 | if (!dev) { | |
0b246afa | 7788 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7789 | return -ENODEV; |
733f4fbb | 7790 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7791 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7792 | return -ENODEV; |
b27f7c0c | 7793 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7794 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7795 | if (stats->nr_items > i) | |
7796 | stats->values[i] = | |
7797 | btrfs_dev_stat_read_and_reset(dev, i); | |
7798 | else | |
4e411a7d | 7799 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7800 | } |
a69976bc AJ |
7801 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7802 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7803 | } else { |
7804 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7805 | if (stats->nr_items > i) | |
7806 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7807 | } | |
7808 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7809 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7810 | return 0; | |
7811 | } | |
a8a6dab7 | 7812 | |
935e5cc9 | 7813 | /* |
bbbf7243 NB |
7814 | * Update the size and bytes used for each device where it changed. This is |
7815 | * delayed since we would otherwise get errors while writing out the | |
7816 | * superblocks. | |
7817 | * | |
7818 | * Must be invoked during transaction commit. | |
935e5cc9 | 7819 | */ |
bbbf7243 | 7820 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7821 | { |
935e5cc9 MX |
7822 | struct btrfs_device *curr, *next; |
7823 | ||
bbbf7243 | 7824 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7825 | |
bbbf7243 | 7826 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7827 | return; |
7828 | ||
bbbf7243 NB |
7829 | /* |
7830 | * We don't need the device_list_mutex here. This list is owned by the | |
7831 | * transaction and the transaction must complete before the device is | |
7832 | * released. | |
7833 | */ | |
7834 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7835 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7836 | post_commit_list) { | |
7837 | list_del_init(&curr->post_commit_list); | |
7838 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7839 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7840 | } |
bbbf7243 | 7841 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7842 | } |
5a13f430 | 7843 | |
46df06b8 DS |
7844 | /* |
7845 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7846 | */ | |
7847 | int btrfs_bg_type_to_factor(u64 flags) | |
7848 | { | |
44b28ada DS |
7849 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7850 | ||
7851 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7852 | } |
cf90d884 QW |
7853 | |
7854 | ||
cf90d884 QW |
7855 | |
7856 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7857 | u64 chunk_offset, u64 devid, | |
7858 | u64 physical_offset, u64 physical_len) | |
7859 | { | |
562d7b15 | 7860 | struct btrfs_dev_lookup_args args = { .devid = devid }; |
c8bf1b67 | 7861 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7862 | struct extent_map *em; |
7863 | struct map_lookup *map; | |
05a37c48 | 7864 | struct btrfs_device *dev; |
cf90d884 QW |
7865 | u64 stripe_len; |
7866 | bool found = false; | |
7867 | int ret = 0; | |
7868 | int i; | |
7869 | ||
7870 | read_lock(&em_tree->lock); | |
7871 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7872 | read_unlock(&em_tree->lock); | |
7873 | ||
7874 | if (!em) { | |
7875 | btrfs_err(fs_info, | |
7876 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7877 | physical_offset, devid); | |
7878 | ret = -EUCLEAN; | |
7879 | goto out; | |
7880 | } | |
7881 | ||
7882 | map = em->map_lookup; | |
bc88b486 | 7883 | stripe_len = btrfs_calc_stripe_length(em); |
cf90d884 QW |
7884 | if (physical_len != stripe_len) { |
7885 | btrfs_err(fs_info, | |
7886 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7887 | physical_offset, devid, em->start, physical_len, | |
7888 | stripe_len); | |
7889 | ret = -EUCLEAN; | |
7890 | goto out; | |
7891 | } | |
7892 | ||
3613249a QW |
7893 | /* |
7894 | * Very old mkfs.btrfs (before v4.1) will not respect the reserved | |
7895 | * space. Although kernel can handle it without problem, better to warn | |
7896 | * the users. | |
7897 | */ | |
7898 | if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) | |
7899 | btrfs_warn(fs_info, | |
7900 | "devid %llu physical %llu len %llu inside the reserved space", | |
7901 | devid, physical_offset, physical_len); | |
7902 | ||
cf90d884 QW |
7903 | for (i = 0; i < map->num_stripes; i++) { |
7904 | if (map->stripes[i].dev->devid == devid && | |
7905 | map->stripes[i].physical == physical_offset) { | |
7906 | found = true; | |
7907 | if (map->verified_stripes >= map->num_stripes) { | |
7908 | btrfs_err(fs_info, | |
7909 | "too many dev extents for chunk %llu found", | |
7910 | em->start); | |
7911 | ret = -EUCLEAN; | |
7912 | goto out; | |
7913 | } | |
7914 | map->verified_stripes++; | |
7915 | break; | |
7916 | } | |
7917 | } | |
7918 | if (!found) { | |
7919 | btrfs_err(fs_info, | |
7920 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7921 | physical_offset, devid); | |
7922 | ret = -EUCLEAN; | |
7923 | } | |
05a37c48 | 7924 | |
1a9fd417 | 7925 | /* Make sure no dev extent is beyond device boundary */ |
562d7b15 | 7926 | dev = btrfs_find_device(fs_info->fs_devices, &args); |
05a37c48 QW |
7927 | if (!dev) { |
7928 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7929 | ret = -EUCLEAN; | |
7930 | goto out; | |
7931 | } | |
1b3922a8 | 7932 | |
05a37c48 QW |
7933 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7934 | btrfs_err(fs_info, | |
7935 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7936 | devid, physical_offset, physical_len, | |
7937 | dev->disk_total_bytes); | |
7938 | ret = -EUCLEAN; | |
7939 | goto out; | |
7940 | } | |
381a696e NA |
7941 | |
7942 | if (dev->zone_info) { | |
7943 | u64 zone_size = dev->zone_info->zone_size; | |
7944 | ||
7945 | if (!IS_ALIGNED(physical_offset, zone_size) || | |
7946 | !IS_ALIGNED(physical_len, zone_size)) { | |
7947 | btrfs_err(fs_info, | |
7948 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", | |
7949 | devid, physical_offset, physical_len); | |
7950 | ret = -EUCLEAN; | |
7951 | goto out; | |
7952 | } | |
7953 | } | |
7954 | ||
cf90d884 QW |
7955 | out: |
7956 | free_extent_map(em); | |
7957 | return ret; | |
7958 | } | |
7959 | ||
7960 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7961 | { | |
c8bf1b67 | 7962 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7963 | struct extent_map *em; |
7964 | struct rb_node *node; | |
7965 | int ret = 0; | |
7966 | ||
7967 | read_lock(&em_tree->lock); | |
07e1ce09 | 7968 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7969 | em = rb_entry(node, struct extent_map, rb_node); |
7970 | if (em->map_lookup->num_stripes != | |
7971 | em->map_lookup->verified_stripes) { | |
7972 | btrfs_err(fs_info, | |
7973 | "chunk %llu has missing dev extent, have %d expect %d", | |
7974 | em->start, em->map_lookup->verified_stripes, | |
7975 | em->map_lookup->num_stripes); | |
7976 | ret = -EUCLEAN; | |
7977 | goto out; | |
7978 | } | |
7979 | } | |
7980 | out: | |
7981 | read_unlock(&em_tree->lock); | |
7982 | return ret; | |
7983 | } | |
7984 | ||
7985 | /* | |
7986 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7987 | * later chunk allocation/free would cause unexpected behavior. | |
7988 | * | |
7989 | * NOTE: This will iterate through the whole device tree, which should be of | |
7990 | * the same size level as the chunk tree. This slightly increases mount time. | |
7991 | */ | |
7992 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7993 | { | |
7994 | struct btrfs_path *path; | |
7995 | struct btrfs_root *root = fs_info->dev_root; | |
7996 | struct btrfs_key key; | |
5eb19381 QW |
7997 | u64 prev_devid = 0; |
7998 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7999 | int ret = 0; |
8000 | ||
42437a63 JB |
8001 | /* |
8002 | * We don't have a dev_root because we mounted with ignorebadroots and | |
8003 | * failed to load the root, so we want to skip the verification in this | |
8004 | * case for sure. | |
8005 | * | |
8006 | * However if the dev root is fine, but the tree itself is corrupted | |
8007 | * we'd still fail to mount. This verification is only to make sure | |
8008 | * writes can happen safely, so instead just bypass this check | |
8009 | * completely in the case of IGNOREBADROOTS. | |
8010 | */ | |
8011 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) | |
8012 | return 0; | |
8013 | ||
cf90d884 QW |
8014 | key.objectid = 1; |
8015 | key.type = BTRFS_DEV_EXTENT_KEY; | |
8016 | key.offset = 0; | |
8017 | ||
8018 | path = btrfs_alloc_path(); | |
8019 | if (!path) | |
8020 | return -ENOMEM; | |
8021 | ||
8022 | path->reada = READA_FORWARD; | |
8023 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
8024 | if (ret < 0) | |
8025 | goto out; | |
8026 | ||
8027 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 8028 | ret = btrfs_next_leaf(root, path); |
cf90d884 QW |
8029 | if (ret < 0) |
8030 | goto out; | |
8031 | /* No dev extents at all? Not good */ | |
8032 | if (ret > 0) { | |
8033 | ret = -EUCLEAN; | |
8034 | goto out; | |
8035 | } | |
8036 | } | |
8037 | while (1) { | |
8038 | struct extent_buffer *leaf = path->nodes[0]; | |
8039 | struct btrfs_dev_extent *dext; | |
8040 | int slot = path->slots[0]; | |
8041 | u64 chunk_offset; | |
8042 | u64 physical_offset; | |
8043 | u64 physical_len; | |
8044 | u64 devid; | |
8045 | ||
8046 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
8047 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
8048 | break; | |
8049 | devid = key.objectid; | |
8050 | physical_offset = key.offset; | |
8051 | ||
8052 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
8053 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
8054 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
8055 | ||
5eb19381 QW |
8056 | /* Check if this dev extent overlaps with the previous one */ |
8057 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
8058 | btrfs_err(fs_info, | |
8059 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
8060 | devid, physical_offset, prev_dev_ext_end); | |
8061 | ret = -EUCLEAN; | |
8062 | goto out; | |
8063 | } | |
8064 | ||
cf90d884 QW |
8065 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
8066 | physical_offset, physical_len); | |
8067 | if (ret < 0) | |
8068 | goto out; | |
5eb19381 QW |
8069 | prev_devid = devid; |
8070 | prev_dev_ext_end = physical_offset + physical_len; | |
8071 | ||
cf90d884 QW |
8072 | ret = btrfs_next_item(root, path); |
8073 | if (ret < 0) | |
8074 | goto out; | |
8075 | if (ret > 0) { | |
8076 | ret = 0; | |
8077 | break; | |
8078 | } | |
8079 | } | |
8080 | ||
8081 | /* Ensure all chunks have corresponding dev extents */ | |
8082 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
8083 | out: | |
8084 | btrfs_free_path(path); | |
8085 | return ret; | |
8086 | } | |
eede2bf3 OS |
8087 | |
8088 | /* | |
8089 | * Check whether the given block group or device is pinned by any inode being | |
8090 | * used as a swapfile. | |
8091 | */ | |
8092 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
8093 | { | |
8094 | struct btrfs_swapfile_pin *sp; | |
8095 | struct rb_node *node; | |
8096 | ||
8097 | spin_lock(&fs_info->swapfile_pins_lock); | |
8098 | node = fs_info->swapfile_pins.rb_node; | |
8099 | while (node) { | |
8100 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
8101 | if (ptr < sp->ptr) | |
8102 | node = node->rb_left; | |
8103 | else if (ptr > sp->ptr) | |
8104 | node = node->rb_right; | |
8105 | else | |
8106 | break; | |
8107 | } | |
8108 | spin_unlock(&fs_info->swapfile_pins_lock); | |
8109 | return node != NULL; | |
8110 | } | |
f7ef5287 NA |
8111 | |
8112 | static int relocating_repair_kthread(void *data) | |
8113 | { | |
0d031dc4 | 8114 | struct btrfs_block_group *cache = data; |
f7ef5287 NA |
8115 | struct btrfs_fs_info *fs_info = cache->fs_info; |
8116 | u64 target; | |
8117 | int ret = 0; | |
8118 | ||
8119 | target = cache->start; | |
8120 | btrfs_put_block_group(cache); | |
8121 | ||
ca5e4ea0 | 8122 | sb_start_write(fs_info->sb); |
f7ef5287 NA |
8123 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { |
8124 | btrfs_info(fs_info, | |
8125 | "zoned: skip relocating block group %llu to repair: EBUSY", | |
8126 | target); | |
ca5e4ea0 | 8127 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
8128 | return -EBUSY; |
8129 | } | |
8130 | ||
f3372065 | 8131 | mutex_lock(&fs_info->reclaim_bgs_lock); |
f7ef5287 NA |
8132 | |
8133 | /* Ensure block group still exists */ | |
8134 | cache = btrfs_lookup_block_group(fs_info, target); | |
8135 | if (!cache) | |
8136 | goto out; | |
8137 | ||
3349b57f | 8138 | if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) |
f7ef5287 NA |
8139 | goto out; |
8140 | ||
8141 | ret = btrfs_may_alloc_data_chunk(fs_info, target); | |
8142 | if (ret < 0) | |
8143 | goto out; | |
8144 | ||
8145 | btrfs_info(fs_info, | |
8146 | "zoned: relocating block group %llu to repair IO failure", | |
8147 | target); | |
8148 | ret = btrfs_relocate_chunk(fs_info, target); | |
8149 | ||
8150 | out: | |
8151 | if (cache) | |
8152 | btrfs_put_block_group(cache); | |
f3372065 | 8153 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f7ef5287 | 8154 | btrfs_exclop_finish(fs_info); |
ca5e4ea0 | 8155 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
8156 | |
8157 | return ret; | |
8158 | } | |
8159 | ||
554aed7d | 8160 | bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) |
f7ef5287 NA |
8161 | { |
8162 | struct btrfs_block_group *cache; | |
8163 | ||
554aed7d JT |
8164 | if (!btrfs_is_zoned(fs_info)) |
8165 | return false; | |
8166 | ||
f7ef5287 NA |
8167 | /* Do not attempt to repair in degraded state */ |
8168 | if (btrfs_test_opt(fs_info, DEGRADED)) | |
554aed7d | 8169 | return true; |
f7ef5287 NA |
8170 | |
8171 | cache = btrfs_lookup_block_group(fs_info, logical); | |
8172 | if (!cache) | |
554aed7d | 8173 | return true; |
f7ef5287 | 8174 | |
3349b57f | 8175 | if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { |
f7ef5287 | 8176 | btrfs_put_block_group(cache); |
554aed7d | 8177 | return true; |
f7ef5287 | 8178 | } |
f7ef5287 NA |
8179 | |
8180 | kthread_run(relocating_repair_kthread, cache, | |
8181 | "btrfs-relocating-repair"); | |
8182 | ||
554aed7d | 8183 | return true; |
f7ef5287 | 8184 | } |