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); |
611ccc58 | 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; | |
5f58d783 | 406 | |
e4404d6e YZ |
407 | WARN_ON(fs_devices->opened); |
408 | while (!list_empty(&fs_devices->devices)) { | |
409 | device = list_entry(fs_devices->devices.next, | |
410 | struct btrfs_device, dev_list); | |
411 | list_del(&device->dev_list); | |
a425f9d4 | 412 | btrfs_free_device(device); |
e4404d6e YZ |
413 | } |
414 | kfree(fs_devices); | |
415 | } | |
416 | ||
ffc5a379 | 417 | void __exit btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
418 | { |
419 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 420 | |
2b82032c YZ |
421 | while (!list_empty(&fs_uuids)) { |
422 | fs_devices = list_entry(fs_uuids.next, | |
c4babc5e AJ |
423 | struct btrfs_fs_devices, fs_list); |
424 | list_del(&fs_devices->fs_list); | |
e4404d6e | 425 | free_fs_devices(fs_devices); |
8a4b83cc | 426 | } |
8a4b83cc CM |
427 | } |
428 | ||
7239ff4b NB |
429 | static noinline struct btrfs_fs_devices *find_fsid( |
430 | const u8 *fsid, const u8 *metadata_fsid) | |
8a4b83cc | 431 | { |
8a4b83cc CM |
432 | struct btrfs_fs_devices *fs_devices; |
433 | ||
7239ff4b NB |
434 | ASSERT(fsid); |
435 | ||
7a62d0f0 | 436 | /* Handle non-split brain cases */ |
c4babc5e | 437 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
7239ff4b NB |
438 | if (metadata_fsid) { |
439 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0 | |
440 | && memcmp(metadata_fsid, fs_devices->metadata_uuid, | |
441 | BTRFS_FSID_SIZE) == 0) | |
442 | return fs_devices; | |
443 | } else { | |
444 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
445 | return fs_devices; | |
446 | } | |
8a4b83cc CM |
447 | } |
448 | return NULL; | |
449 | } | |
450 | ||
c6730a0e SY |
451 | static struct btrfs_fs_devices *find_fsid_with_metadata_uuid( |
452 | struct btrfs_super_block *disk_super) | |
453 | { | |
454 | ||
455 | struct btrfs_fs_devices *fs_devices; | |
456 | ||
457 | /* | |
458 | * Handle scanned device having completed its fsid change but | |
459 | * belonging to a fs_devices that was created by first scanning | |
460 | * a device which didn't have its fsid/metadata_uuid changed | |
461 | * at all and the CHANGING_FSID_V2 flag set. | |
462 | */ | |
463 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
464 | if (fs_devices->fsid_change && | |
465 | memcmp(disk_super->metadata_uuid, fs_devices->fsid, | |
466 | BTRFS_FSID_SIZE) == 0 && | |
467 | memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
468 | BTRFS_FSID_SIZE) == 0) { | |
469 | return fs_devices; | |
470 | } | |
471 | } | |
472 | /* | |
473 | * Handle scanned device having completed its fsid change but | |
474 | * belonging to a fs_devices that was created by a device that | |
475 | * has an outdated pair of fsid/metadata_uuid and | |
476 | * CHANGING_FSID_V2 flag set. | |
477 | */ | |
478 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
479 | if (fs_devices->fsid_change && | |
480 | memcmp(fs_devices->metadata_uuid, | |
481 | fs_devices->fsid, BTRFS_FSID_SIZE) != 0 && | |
482 | memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid, | |
483 | BTRFS_FSID_SIZE) == 0) { | |
484 | return fs_devices; | |
485 | } | |
486 | } | |
487 | ||
488 | return find_fsid(disk_super->fsid, disk_super->metadata_uuid); | |
489 | } | |
490 | ||
491 | ||
beaf8ab3 | 492 | static int |
05bdb996 | 493 | btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder, |
beaf8ab3 | 494 | int flush, struct block_device **bdev, |
8f32380d | 495 | struct btrfs_super_block **disk_super) |
beaf8ab3 SB |
496 | { |
497 | int ret; | |
498 | ||
0718afd4 | 499 | *bdev = blkdev_get_by_path(device_path, flags, holder, NULL); |
beaf8ab3 SB |
500 | |
501 | if (IS_ERR(*bdev)) { | |
502 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
503 | goto error; |
504 | } | |
505 | ||
506 | if (flush) | |
1226dfff | 507 | sync_blockdev(*bdev); |
9f6d2510 | 508 | ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE); |
beaf8ab3 | 509 | if (ret) { |
2736e8ee | 510 | blkdev_put(*bdev, holder); |
beaf8ab3 SB |
511 | goto error; |
512 | } | |
513 | invalidate_bdev(*bdev); | |
8f32380d JT |
514 | *disk_super = btrfs_read_dev_super(*bdev); |
515 | if (IS_ERR(*disk_super)) { | |
516 | ret = PTR_ERR(*disk_super); | |
2736e8ee | 517 | blkdev_put(*bdev, holder); |
beaf8ab3 SB |
518 | goto error; |
519 | } | |
520 | ||
521 | return 0; | |
522 | ||
523 | error: | |
524 | *bdev = NULL; | |
beaf8ab3 SB |
525 | return ret; |
526 | } | |
527 | ||
43dd529a DS |
528 | /* |
529 | * Search and remove all stale devices (which are not mounted). When both | |
530 | * inputs are NULL, it will search and release all stale devices. | |
16cab91a | 531 | * |
43dd529a DS |
532 | * @devt: Optional. When provided will it release all unmounted devices |
533 | * matching this devt only. | |
16cab91a | 534 | * @skip_device: Optional. Will skip this device when searching for the stale |
43dd529a | 535 | * devices. |
16cab91a AJ |
536 | * |
537 | * Return: 0 for success or if @devt is 0. | |
538 | * -EBUSY if @devt is a mounted device. | |
539 | * -ENOENT if @devt does not match any device in the list. | |
d8367db3 | 540 | */ |
16cab91a | 541 | static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) |
4fde46f0 | 542 | { |
fa6d2ae5 AJ |
543 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
544 | struct btrfs_device *device, *tmp_device; | |
70bc7088 AJ |
545 | int ret = 0; |
546 | ||
c1247069 AJ |
547 | lockdep_assert_held(&uuid_mutex); |
548 | ||
16cab91a | 549 | if (devt) |
70bc7088 | 550 | ret = -ENOENT; |
4fde46f0 | 551 | |
fa6d2ae5 | 552 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
4fde46f0 | 553 | |
70bc7088 | 554 | mutex_lock(&fs_devices->device_list_mutex); |
fa6d2ae5 AJ |
555 | list_for_each_entry_safe(device, tmp_device, |
556 | &fs_devices->devices, dev_list) { | |
fa6d2ae5 | 557 | if (skip_device && skip_device == device) |
d8367db3 | 558 | continue; |
330a5bf4 | 559 | if (devt && devt != device->devt) |
38cf665d | 560 | continue; |
70bc7088 AJ |
561 | if (fs_devices->opened) { |
562 | /* for an already deleted device return 0 */ | |
16cab91a | 563 | if (devt && ret != 0) |
70bc7088 AJ |
564 | ret = -EBUSY; |
565 | break; | |
566 | } | |
4fde46f0 | 567 | |
4fde46f0 | 568 | /* delete the stale device */ |
7bcb8164 AJ |
569 | fs_devices->num_devices--; |
570 | list_del(&device->dev_list); | |
571 | btrfs_free_device(device); | |
572 | ||
70bc7088 | 573 | ret = 0; |
7bcb8164 AJ |
574 | } |
575 | mutex_unlock(&fs_devices->device_list_mutex); | |
70bc7088 | 576 | |
7bcb8164 AJ |
577 | if (fs_devices->num_devices == 0) { |
578 | btrfs_sysfs_remove_fsid(fs_devices); | |
579 | list_del(&fs_devices->fs_list); | |
580 | free_fs_devices(fs_devices); | |
4fde46f0 AJ |
581 | } |
582 | } | |
70bc7088 AJ |
583 | |
584 | return ret; | |
4fde46f0 AJ |
585 | } |
586 | ||
18c850fd JB |
587 | /* |
588 | * This is only used on mount, and we are protected from competing things | |
589 | * messing with our fs_devices by the uuid_mutex, thus we do not need the | |
590 | * fs_devices->device_list_mutex here. | |
591 | */ | |
0fb08bcc | 592 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 593 | struct btrfs_device *device, blk_mode_t flags, |
0fb08bcc AJ |
594 | void *holder) |
595 | { | |
0fb08bcc | 596 | struct block_device *bdev; |
0fb08bcc AJ |
597 | struct btrfs_super_block *disk_super; |
598 | u64 devid; | |
599 | int ret; | |
600 | ||
601 | if (device->bdev) | |
602 | return -EINVAL; | |
603 | if (!device->name) | |
604 | return -EINVAL; | |
605 | ||
606 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
8f32380d | 607 | &bdev, &disk_super); |
0fb08bcc AJ |
608 | if (ret) |
609 | return ret; | |
610 | ||
0fb08bcc AJ |
611 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
612 | if (devid != device->devid) | |
8f32380d | 613 | goto error_free_page; |
0fb08bcc AJ |
614 | |
615 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) | |
8f32380d | 616 | goto error_free_page; |
0fb08bcc AJ |
617 | |
618 | device->generation = btrfs_super_generation(disk_super); | |
619 | ||
620 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
7239ff4b NB |
621 | if (btrfs_super_incompat_flags(disk_super) & |
622 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { | |
623 | pr_err( | |
624 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); | |
8f32380d | 625 | goto error_free_page; |
7239ff4b NB |
626 | } |
627 | ||
ebbede42 | 628 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
0395d84f | 629 | fs_devices->seeding = true; |
0fb08bcc | 630 | } else { |
ebbede42 AJ |
631 | if (bdev_read_only(bdev)) |
632 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
633 | else | |
634 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); | |
0fb08bcc AJ |
635 | } |
636 | ||
10f0d2a5 | 637 | if (!bdev_nonrot(bdev)) |
7f0432d0 | 638 | fs_devices->rotating = true; |
0fb08bcc | 639 | |
63a7cb13 DS |
640 | if (bdev_max_discard_sectors(bdev)) |
641 | fs_devices->discardable = true; | |
642 | ||
0fb08bcc | 643 | device->bdev = bdev; |
e12c9621 | 644 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
2736e8ee | 645 | device->holder = holder; |
0fb08bcc AJ |
646 | |
647 | fs_devices->open_devices++; | |
ebbede42 AJ |
648 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
649 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
0fb08bcc | 650 | fs_devices->rw_devices++; |
b1b8e386 | 651 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
0fb08bcc | 652 | } |
8f32380d | 653 | btrfs_release_disk_super(disk_super); |
0fb08bcc AJ |
654 | |
655 | return 0; | |
656 | ||
8f32380d JT |
657 | error_free_page: |
658 | btrfs_release_disk_super(disk_super); | |
2736e8ee | 659 | blkdev_put(bdev, holder); |
0fb08bcc AJ |
660 | |
661 | return -EINVAL; | |
662 | } | |
663 | ||
7a62d0f0 NB |
664 | /* |
665 | * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices | |
c0d81c7c SY |
666 | * being created with a disk that has already completed its fsid change. Such |
667 | * disk can belong to an fs which has its FSID changed or to one which doesn't. | |
668 | * Handle both cases here. | |
7a62d0f0 NB |
669 | */ |
670 | static struct btrfs_fs_devices *find_fsid_inprogress( | |
671 | struct btrfs_super_block *disk_super) | |
672 | { | |
673 | struct btrfs_fs_devices *fs_devices; | |
674 | ||
675 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
676 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
677 | BTRFS_FSID_SIZE) != 0 && | |
678 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
679 | BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) { | |
680 | return fs_devices; | |
681 | } | |
682 | } | |
683 | ||
c0d81c7c | 684 | return find_fsid(disk_super->fsid, NULL); |
7a62d0f0 NB |
685 | } |
686 | ||
cc5de4e7 NB |
687 | |
688 | static struct btrfs_fs_devices *find_fsid_changed( | |
689 | struct btrfs_super_block *disk_super) | |
690 | { | |
691 | struct btrfs_fs_devices *fs_devices; | |
692 | ||
693 | /* | |
694 | * Handles the case where scanned device is part of an fs that had | |
1a9fd417 | 695 | * multiple successful changes of FSID but currently device didn't |
05840710 NB |
696 | * observe it. Meaning our fsid will be different than theirs. We need |
697 | * to handle two subcases : | |
698 | * 1 - The fs still continues to have different METADATA/FSID uuids. | |
699 | * 2 - The fs is switched back to its original FSID (METADATA/FSID | |
700 | * are equal). | |
cc5de4e7 NB |
701 | */ |
702 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
05840710 | 703 | /* Changed UUIDs */ |
cc5de4e7 NB |
704 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, |
705 | BTRFS_FSID_SIZE) != 0 && | |
706 | memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid, | |
707 | BTRFS_FSID_SIZE) == 0 && | |
708 | memcmp(fs_devices->fsid, disk_super->fsid, | |
05840710 NB |
709 | BTRFS_FSID_SIZE) != 0) |
710 | return fs_devices; | |
711 | ||
712 | /* Unchanged UUIDs */ | |
713 | if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid, | |
714 | BTRFS_FSID_SIZE) == 0 && | |
715 | memcmp(fs_devices->fsid, disk_super->metadata_uuid, | |
716 | BTRFS_FSID_SIZE) == 0) | |
cc5de4e7 | 717 | return fs_devices; |
cc5de4e7 NB |
718 | } |
719 | ||
720 | return NULL; | |
721 | } | |
1362089d NB |
722 | |
723 | static struct btrfs_fs_devices *find_fsid_reverted_metadata( | |
724 | struct btrfs_super_block *disk_super) | |
725 | { | |
726 | struct btrfs_fs_devices *fs_devices; | |
727 | ||
728 | /* | |
729 | * Handle the case where the scanned device is part of an fs whose last | |
730 | * metadata UUID change reverted it to the original FSID. At the same | |
67da05b3 | 731 | * time fs_devices was first created by another constituent device |
1362089d NB |
732 | * which didn't fully observe the operation. This results in an |
733 | * btrfs_fs_devices created with metadata/fsid different AND | |
734 | * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the | |
735 | * fs_devices equal to the FSID of the disk. | |
736 | */ | |
737 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { | |
738 | if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid, | |
739 | BTRFS_FSID_SIZE) != 0 && | |
740 | memcmp(fs_devices->metadata_uuid, disk_super->fsid, | |
741 | BTRFS_FSID_SIZE) == 0 && | |
742 | fs_devices->fsid_change) | |
743 | return fs_devices; | |
744 | } | |
745 | ||
746 | return NULL; | |
747 | } | |
60999ca4 DS |
748 | /* |
749 | * Add new device to list of registered devices | |
750 | * | |
751 | * Returns: | |
e124ece5 AJ |
752 | * device pointer which was just added or updated when successful |
753 | * error pointer when failed | |
60999ca4 | 754 | */ |
e124ece5 | 755 | static noinline struct btrfs_device *device_list_add(const char *path, |
4306a974 AJ |
756 | struct btrfs_super_block *disk_super, |
757 | bool *new_device_added) | |
8a4b83cc CM |
758 | { |
759 | struct btrfs_device *device; | |
7a62d0f0 | 760 | struct btrfs_fs_devices *fs_devices = NULL; |
606686ee | 761 | struct rcu_string *name; |
8a4b83cc | 762 | u64 found_transid = btrfs_super_generation(disk_super); |
3acbcbfc | 763 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
4889bc05 AJ |
764 | dev_t path_devt; |
765 | int error; | |
7239ff4b NB |
766 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
767 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); | |
d1a63002 NB |
768 | bool fsid_change_in_progress = (btrfs_super_flags(disk_super) & |
769 | BTRFS_SUPER_FLAG_CHANGING_FSID_V2); | |
7239ff4b | 770 | |
4889bc05 | 771 | error = lookup_bdev(path, &path_devt); |
ed02363f QW |
772 | if (error) { |
773 | btrfs_err(NULL, "failed to lookup block device for path %s: %d", | |
774 | path, error); | |
4889bc05 | 775 | return ERR_PTR(error); |
ed02363f | 776 | } |
4889bc05 | 777 | |
cc5de4e7 | 778 | if (fsid_change_in_progress) { |
c0d81c7c | 779 | if (!has_metadata_uuid) |
cc5de4e7 | 780 | fs_devices = find_fsid_inprogress(disk_super); |
c0d81c7c | 781 | else |
cc5de4e7 | 782 | fs_devices = find_fsid_changed(disk_super); |
7a62d0f0 | 783 | } else if (has_metadata_uuid) { |
c6730a0e | 784 | fs_devices = find_fsid_with_metadata_uuid(disk_super); |
7a62d0f0 | 785 | } else { |
1362089d NB |
786 | fs_devices = find_fsid_reverted_metadata(disk_super); |
787 | if (!fs_devices) | |
788 | fs_devices = find_fsid(disk_super->fsid, NULL); | |
7a62d0f0 NB |
789 | } |
790 | ||
8a4b83cc | 791 | |
8a4b83cc | 792 | if (!fs_devices) { |
7239ff4b NB |
793 | if (has_metadata_uuid) |
794 | fs_devices = alloc_fs_devices(disk_super->fsid, | |
795 | disk_super->metadata_uuid); | |
796 | else | |
797 | fs_devices = alloc_fs_devices(disk_super->fsid, NULL); | |
798 | ||
2208a378 | 799 | if (IS_ERR(fs_devices)) |
e124ece5 | 800 | return ERR_CAST(fs_devices); |
2208a378 | 801 | |
92900e51 AV |
802 | fs_devices->fsid_change = fsid_change_in_progress; |
803 | ||
9c6d173e | 804 | mutex_lock(&fs_devices->device_list_mutex); |
c4babc5e | 805 | list_add(&fs_devices->fs_list, &fs_uuids); |
2208a378 | 806 | |
8a4b83cc CM |
807 | device = NULL; |
808 | } else { | |
562d7b15 JB |
809 | struct btrfs_dev_lookup_args args = { |
810 | .devid = devid, | |
811 | .uuid = disk_super->dev_item.uuid, | |
812 | }; | |
813 | ||
9c6d173e | 814 | mutex_lock(&fs_devices->device_list_mutex); |
562d7b15 | 815 | device = btrfs_find_device(fs_devices, &args); |
7a62d0f0 NB |
816 | |
817 | /* | |
818 | * If this disk has been pulled into an fs devices created by | |
819 | * a device which had the CHANGING_FSID_V2 flag then replace the | |
820 | * metadata_uuid/fsid values of the fs_devices. | |
821 | */ | |
1362089d | 822 | if (fs_devices->fsid_change && |
7a62d0f0 NB |
823 | found_transid > fs_devices->latest_generation) { |
824 | memcpy(fs_devices->fsid, disk_super->fsid, | |
825 | BTRFS_FSID_SIZE); | |
1362089d NB |
826 | |
827 | if (has_metadata_uuid) | |
828 | memcpy(fs_devices->metadata_uuid, | |
829 | disk_super->metadata_uuid, | |
830 | BTRFS_FSID_SIZE); | |
831 | else | |
832 | memcpy(fs_devices->metadata_uuid, | |
833 | disk_super->fsid, BTRFS_FSID_SIZE); | |
7a62d0f0 NB |
834 | |
835 | fs_devices->fsid_change = false; | |
836 | } | |
8a4b83cc | 837 | } |
443f24fe | 838 | |
8a4b83cc | 839 | if (!device) { |
bb21e302 AJ |
840 | unsigned int nofs_flag; |
841 | ||
9c6d173e | 842 | if (fs_devices->opened) { |
ed02363f QW |
843 | btrfs_err(NULL, |
844 | "device %s belongs to fsid %pU, and the fs is already mounted", | |
845 | path, fs_devices->fsid); | |
9c6d173e | 846 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 847 | return ERR_PTR(-EBUSY); |
9c6d173e | 848 | } |
2b82032c | 849 | |
bb21e302 | 850 | nofs_flag = memalloc_nofs_save(); |
12bd2fc0 | 851 | device = btrfs_alloc_device(NULL, &devid, |
bb21e302 AJ |
852 | disk_super->dev_item.uuid, path); |
853 | memalloc_nofs_restore(nofs_flag); | |
12bd2fc0 | 854 | if (IS_ERR(device)) { |
9c6d173e | 855 | mutex_unlock(&fs_devices->device_list_mutex); |
8a4b83cc | 856 | /* we can safely leave the fs_devices entry around */ |
e124ece5 | 857 | return device; |
8a4b83cc | 858 | } |
606686ee | 859 | |
4889bc05 | 860 | device->devt = path_devt; |
90519d66 | 861 | |
1f78160c | 862 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 863 | fs_devices->num_devices++; |
e5e9a520 | 864 | |
2b82032c | 865 | device->fs_devices = fs_devices; |
4306a974 | 866 | *new_device_added = true; |
327f18cc AJ |
867 | |
868 | if (disk_super->label[0]) | |
aa6c0df7 AJ |
869 | pr_info( |
870 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", | |
871 | disk_super->label, devid, found_transid, path, | |
872 | current->comm, task_pid_nr(current)); | |
327f18cc | 873 | else |
aa6c0df7 AJ |
874 | pr_info( |
875 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", | |
876 | disk_super->fsid, devid, found_transid, path, | |
877 | current->comm, task_pid_nr(current)); | |
327f18cc | 878 | |
606686ee | 879 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
880 | /* |
881 | * When FS is already mounted. | |
882 | * 1. If you are here and if the device->name is NULL that | |
883 | * means this device was missing at time of FS mount. | |
884 | * 2. If you are here and if the device->name is different | |
885 | * from 'path' that means either | |
886 | * a. The same device disappeared and reappeared with | |
887 | * different name. or | |
888 | * b. The missing-disk-which-was-replaced, has | |
889 | * reappeared now. | |
890 | * | |
891 | * We must allow 1 and 2a above. But 2b would be a spurious | |
892 | * and unintentional. | |
893 | * | |
894 | * Further in case of 1 and 2a above, the disk at 'path' | |
895 | * would have missed some transaction when it was away and | |
896 | * in case of 2a the stale bdev has to be updated as well. | |
897 | * 2b must not be allowed at all time. | |
898 | */ | |
899 | ||
900 | /* | |
0f23ae74 CM |
901 | * For now, we do allow update to btrfs_fs_device through the |
902 | * btrfs dev scan cli after FS has been mounted. We're still | |
903 | * tracking a problem where systems fail mount by subvolume id | |
904 | * when we reject replacement on a mounted FS. | |
b96de000 | 905 | */ |
0f23ae74 | 906 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
907 | /* |
908 | * That is if the FS is _not_ mounted and if you | |
909 | * are here, that means there is more than one | |
910 | * disk with same uuid and devid.We keep the one | |
911 | * with larger generation number or the last-in if | |
912 | * generation are equal. | |
913 | */ | |
9c6d173e | 914 | mutex_unlock(&fs_devices->device_list_mutex); |
ed02363f QW |
915 | btrfs_err(NULL, |
916 | "device %s already registered with a higher generation, found %llu expect %llu", | |
917 | path, found_transid, device->generation); | |
e124ece5 | 918 | return ERR_PTR(-EEXIST); |
77bdae4d | 919 | } |
b96de000 | 920 | |
a9261d41 AJ |
921 | /* |
922 | * We are going to replace the device path for a given devid, | |
923 | * make sure it's the same device if the device is mounted | |
79c9234b DM |
924 | * |
925 | * NOTE: the device->fs_info may not be reliable here so pass | |
926 | * in a NULL to message helpers instead. This avoids a possible | |
927 | * use-after-free when the fs_info and fs_info->sb are already | |
928 | * torn down. | |
a9261d41 AJ |
929 | */ |
930 | if (device->bdev) { | |
4889bc05 | 931 | if (device->devt != path_devt) { |
a9261d41 | 932 | mutex_unlock(&fs_devices->device_list_mutex); |
0697d9a6 | 933 | btrfs_warn_in_rcu(NULL, |
79dae17d AJ |
934 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
935 | path, devid, found_transid, | |
936 | current->comm, | |
937 | task_pid_nr(current)); | |
a9261d41 AJ |
938 | return ERR_PTR(-EEXIST); |
939 | } | |
79c9234b | 940 | btrfs_info_in_rcu(NULL, |
79dae17d | 941 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
cb3e217b | 942 | devid, btrfs_dev_name(device), |
79dae17d AJ |
943 | path, current->comm, |
944 | task_pid_nr(current)); | |
a9261d41 AJ |
945 | } |
946 | ||
606686ee | 947 | name = rcu_string_strdup(path, GFP_NOFS); |
9c6d173e AJ |
948 | if (!name) { |
949 | mutex_unlock(&fs_devices->device_list_mutex); | |
e124ece5 | 950 | return ERR_PTR(-ENOMEM); |
9c6d173e | 951 | } |
606686ee JB |
952 | rcu_string_free(device->name); |
953 | rcu_assign_pointer(device->name, name); | |
e6e674bd | 954 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
cd02dca5 | 955 | fs_devices->missing_devices--; |
e6e674bd | 956 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 957 | } |
4889bc05 | 958 | device->devt = path_devt; |
8a4b83cc CM |
959 | } |
960 | ||
77bdae4d AJ |
961 | /* |
962 | * Unmount does not free the btrfs_device struct but would zero | |
963 | * generation along with most of the other members. So just update | |
964 | * it back. We need it to pick the disk with largest generation | |
965 | * (as above). | |
966 | */ | |
d1a63002 | 967 | if (!fs_devices->opened) { |
77bdae4d | 968 | device->generation = found_transid; |
d1a63002 NB |
969 | fs_devices->latest_generation = max_t(u64, found_transid, |
970 | fs_devices->latest_generation); | |
971 | } | |
77bdae4d | 972 | |
f2788d2f AJ |
973 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
974 | ||
9c6d173e | 975 | mutex_unlock(&fs_devices->device_list_mutex); |
e124ece5 | 976 | return device; |
8a4b83cc CM |
977 | } |
978 | ||
e4404d6e YZ |
979 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
980 | { | |
981 | struct btrfs_fs_devices *fs_devices; | |
982 | struct btrfs_device *device; | |
983 | struct btrfs_device *orig_dev; | |
d2979aa2 | 984 | int ret = 0; |
e4404d6e | 985 | |
c1247069 AJ |
986 | lockdep_assert_held(&uuid_mutex); |
987 | ||
7239ff4b | 988 | fs_devices = alloc_fs_devices(orig->fsid, NULL); |
2208a378 ID |
989 | if (IS_ERR(fs_devices)) |
990 | return fs_devices; | |
e4404d6e | 991 | |
02db0844 | 992 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
993 | |
994 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
bb21e302 AJ |
995 | const char *dev_path = NULL; |
996 | ||
997 | /* | |
998 | * This is ok to do without RCU read locked because we hold the | |
999 | * uuid mutex so nothing we touch in here is going to disappear. | |
1000 | */ | |
1001 | if (orig_dev->name) | |
1002 | dev_path = orig_dev->name->str; | |
606686ee | 1003 | |
12bd2fc0 | 1004 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
bb21e302 | 1005 | orig_dev->uuid, dev_path); |
d2979aa2 AJ |
1006 | if (IS_ERR(device)) { |
1007 | ret = PTR_ERR(device); | |
e4404d6e | 1008 | goto error; |
d2979aa2 | 1009 | } |
e4404d6e | 1010 | |
21e61ec6 JT |
1011 | if (orig_dev->zone_info) { |
1012 | struct btrfs_zoned_device_info *zone_info; | |
1013 | ||
1014 | zone_info = btrfs_clone_dev_zone_info(orig_dev); | |
1015 | if (!zone_info) { | |
1016 | btrfs_free_device(device); | |
1017 | ret = -ENOMEM; | |
1018 | goto error; | |
1019 | } | |
1020 | device->zone_info = zone_info; | |
1021 | } | |
1022 | ||
e4404d6e YZ |
1023 | list_add(&device->dev_list, &fs_devices->devices); |
1024 | device->fs_devices = fs_devices; | |
1025 | fs_devices->num_devices++; | |
1026 | } | |
1027 | return fs_devices; | |
1028 | error: | |
1029 | free_fs_devices(fs_devices); | |
d2979aa2 | 1030 | return ERR_PTR(ret); |
e4404d6e YZ |
1031 | } |
1032 | ||
3712ccb7 | 1033 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
bacce86a | 1034 | struct btrfs_device **latest_dev) |
dfe25020 | 1035 | { |
c6e30871 | 1036 | struct btrfs_device *device, *next; |
a6b0d5c8 | 1037 | |
46224705 | 1038 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 1039 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
3712ccb7 | 1040 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
401e29c1 | 1041 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
3712ccb7 | 1042 | &device->dev_state) && |
998a0671 AJ |
1043 | !test_bit(BTRFS_DEV_STATE_MISSING, |
1044 | &device->dev_state) && | |
3712ccb7 NB |
1045 | (!*latest_dev || |
1046 | device->generation > (*latest_dev)->generation)) { | |
1047 | *latest_dev = device; | |
a6b0d5c8 | 1048 | } |
2b82032c | 1049 | continue; |
a6b0d5c8 | 1050 | } |
2b82032c | 1051 | |
cf89af14 AJ |
1052 | /* |
1053 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, | |
1054 | * in btrfs_init_dev_replace() so just continue. | |
1055 | */ | |
1056 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) | |
1057 | continue; | |
1058 | ||
2b82032c | 1059 | if (device->bdev) { |
2736e8ee | 1060 | blkdev_put(device->bdev, device->holder); |
2b82032c YZ |
1061 | device->bdev = NULL; |
1062 | fs_devices->open_devices--; | |
1063 | } | |
ebbede42 | 1064 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 1065 | list_del_init(&device->dev_alloc_list); |
ebbede42 | 1066 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
b2a61667 | 1067 | fs_devices->rw_devices--; |
2b82032c | 1068 | } |
e4404d6e YZ |
1069 | list_del_init(&device->dev_list); |
1070 | fs_devices->num_devices--; | |
a425f9d4 | 1071 | btrfs_free_device(device); |
dfe25020 | 1072 | } |
2b82032c | 1073 | |
3712ccb7 NB |
1074 | } |
1075 | ||
1076 | /* | |
1077 | * After we have read the system tree and know devids belonging to this | |
1078 | * filesystem, remove the device which does not belong there. | |
1079 | */ | |
bacce86a | 1080 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
3712ccb7 NB |
1081 | { |
1082 | struct btrfs_device *latest_dev = NULL; | |
944d3f9f | 1083 | struct btrfs_fs_devices *seed_dev; |
3712ccb7 NB |
1084 | |
1085 | mutex_lock(&uuid_mutex); | |
bacce86a | 1086 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
944d3f9f NB |
1087 | |
1088 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) | |
bacce86a | 1089 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
2b82032c | 1090 | |
d24fa5c1 | 1091 | fs_devices->latest_dev = latest_dev; |
a6b0d5c8 | 1092 | |
dfe25020 | 1093 | mutex_unlock(&uuid_mutex); |
dfe25020 | 1094 | } |
a0af469b | 1095 | |
14238819 AJ |
1096 | static void btrfs_close_bdev(struct btrfs_device *device) |
1097 | { | |
08ffcae8 DS |
1098 | if (!device->bdev) |
1099 | return; | |
1100 | ||
ebbede42 | 1101 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
14238819 AJ |
1102 | sync_blockdev(device->bdev); |
1103 | invalidate_bdev(device->bdev); | |
1104 | } | |
1105 | ||
2736e8ee | 1106 | blkdev_put(device->bdev, device->holder); |
14238819 AJ |
1107 | } |
1108 | ||
959b1c04 | 1109 | static void btrfs_close_one_device(struct btrfs_device *device) |
f448341a AJ |
1110 | { |
1111 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
f448341a | 1112 | |
ebbede42 | 1113 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
f448341a AJ |
1114 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
1115 | list_del_init(&device->dev_alloc_list); | |
1116 | fs_devices->rw_devices--; | |
1117 | } | |
1118 | ||
0d977e0e DCZX |
1119 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
1120 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); | |
1121 | ||
5d03dbeb LZ |
1122 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
1123 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); | |
f448341a | 1124 | fs_devices->missing_devices--; |
5d03dbeb | 1125 | } |
f448341a | 1126 | |
959b1c04 | 1127 | btrfs_close_bdev(device); |
321f69f8 | 1128 | if (device->bdev) { |
3fff3975 | 1129 | fs_devices->open_devices--; |
321f69f8 | 1130 | device->bdev = NULL; |
f448341a | 1131 | } |
321f69f8 | 1132 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
5b316468 | 1133 | btrfs_destroy_dev_zone_info(device); |
f448341a | 1134 | |
321f69f8 JT |
1135 | device->fs_info = NULL; |
1136 | atomic_set(&device->dev_stats_ccnt, 0); | |
1137 | extent_io_tree_release(&device->alloc_state); | |
959b1c04 | 1138 | |
6b225baa FM |
1139 | /* |
1140 | * Reset the flush error record. We might have a transient flush error | |
1141 | * in this mount, and if so we aborted the current transaction and set | |
1142 | * the fs to an error state, guaranteeing no super blocks can be further | |
1143 | * committed. However that error might be transient and if we unmount the | |
1144 | * filesystem and mount it again, we should allow the mount to succeed | |
1145 | * (btrfs_check_rw_degradable() should not fail) - if after mounting the | |
1146 | * filesystem again we still get flush errors, then we will again abort | |
1147 | * any transaction and set the error state, guaranteeing no commits of | |
1148 | * unsafe super blocks. | |
1149 | */ | |
1150 | device->last_flush_error = 0; | |
1151 | ||
321f69f8 | 1152 | /* Verify the device is back in a pristine state */ |
1f16033c AJ |
1153 | WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); |
1154 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); | |
1155 | WARN_ON(!list_empty(&device->dev_alloc_list)); | |
1156 | WARN_ON(!list_empty(&device->post_commit_list)); | |
f448341a AJ |
1157 | } |
1158 | ||
54eed6ae | 1159 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 1160 | { |
2037a093 | 1161 | struct btrfs_device *device, *tmp; |
e4404d6e | 1162 | |
425c6ed6 JB |
1163 | lockdep_assert_held(&uuid_mutex); |
1164 | ||
2b82032c | 1165 | if (--fs_devices->opened > 0) |
54eed6ae | 1166 | return; |
8a4b83cc | 1167 | |
425c6ed6 | 1168 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
959b1c04 | 1169 | btrfs_close_one_device(device); |
c9513edb | 1170 | |
e4404d6e YZ |
1171 | WARN_ON(fs_devices->open_devices); |
1172 | WARN_ON(fs_devices->rw_devices); | |
2b82032c | 1173 | fs_devices->opened = 0; |
0395d84f | 1174 | fs_devices->seeding = false; |
c4989c2f | 1175 | fs_devices->fs_info = NULL; |
8a4b83cc CM |
1176 | } |
1177 | ||
54eed6ae | 1178 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
2b82032c | 1179 | { |
944d3f9f NB |
1180 | LIST_HEAD(list); |
1181 | struct btrfs_fs_devices *tmp; | |
2b82032c YZ |
1182 | |
1183 | mutex_lock(&uuid_mutex); | |
54eed6ae | 1184 | close_fs_devices(fs_devices); |
5f58d783 | 1185 | if (!fs_devices->opened) { |
944d3f9f | 1186 | list_splice_init(&fs_devices->seed_list, &list); |
e4404d6e | 1187 | |
5f58d783 AJ |
1188 | /* |
1189 | * If the struct btrfs_fs_devices is not assembled with any | |
1190 | * other device, it can be re-initialized during the next mount | |
1191 | * without the needing device-scan step. Therefore, it can be | |
1192 | * fully freed. | |
1193 | */ | |
1194 | if (fs_devices->num_devices == 1) { | |
1195 | list_del(&fs_devices->fs_list); | |
1196 | free_fs_devices(fs_devices); | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | ||
944d3f9f | 1201 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
0226e0eb | 1202 | close_fs_devices(fs_devices); |
944d3f9f | 1203 | list_del(&fs_devices->seed_list); |
e4404d6e YZ |
1204 | free_fs_devices(fs_devices); |
1205 | } | |
425c6ed6 | 1206 | mutex_unlock(&uuid_mutex); |
2b82032c YZ |
1207 | } |
1208 | ||
897fb573 | 1209 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 1210 | blk_mode_t flags, void *holder) |
8a4b83cc | 1211 | { |
8a4b83cc | 1212 | struct btrfs_device *device; |
443f24fe | 1213 | struct btrfs_device *latest_dev = NULL; |
96c2e067 | 1214 | struct btrfs_device *tmp_device; |
8a4b83cc | 1215 | |
96c2e067 AJ |
1216 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
1217 | dev_list) { | |
1218 | int ret; | |
a0af469b | 1219 | |
96c2e067 AJ |
1220 | ret = btrfs_open_one_device(fs_devices, device, flags, holder); |
1221 | if (ret == 0 && | |
1222 | (!latest_dev || device->generation > latest_dev->generation)) { | |
9f050db4 | 1223 | latest_dev = device; |
96c2e067 AJ |
1224 | } else if (ret == -ENODATA) { |
1225 | fs_devices->num_devices--; | |
1226 | list_del(&device->dev_list); | |
1227 | btrfs_free_device(device); | |
1228 | } | |
8a4b83cc | 1229 | } |
1ed802c9 AJ |
1230 | if (fs_devices->open_devices == 0) |
1231 | return -EINVAL; | |
1232 | ||
2b82032c | 1233 | fs_devices->opened = 1; |
d24fa5c1 | 1234 | fs_devices->latest_dev = latest_dev; |
2b82032c | 1235 | fs_devices->total_rw_bytes = 0; |
c4a816c6 | 1236 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
33fd2f71 | 1237 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
1ed802c9 AJ |
1238 | |
1239 | return 0; | |
2b82032c YZ |
1240 | } |
1241 | ||
4f0f586b ST |
1242 | static int devid_cmp(void *priv, const struct list_head *a, |
1243 | const struct list_head *b) | |
f8e10cd3 | 1244 | { |
214cc184 | 1245 | const struct btrfs_device *dev1, *dev2; |
f8e10cd3 AJ |
1246 | |
1247 | dev1 = list_entry(a, struct btrfs_device, dev_list); | |
1248 | dev2 = list_entry(b, struct btrfs_device, dev_list); | |
1249 | ||
1250 | if (dev1->devid < dev2->devid) | |
1251 | return -1; | |
1252 | else if (dev1->devid > dev2->devid) | |
1253 | return 1; | |
1254 | return 0; | |
1255 | } | |
1256 | ||
2b82032c | 1257 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
05bdb996 | 1258 | blk_mode_t flags, void *holder) |
2b82032c YZ |
1259 | { |
1260 | int ret; | |
1261 | ||
f5194e34 | 1262 | lockdep_assert_held(&uuid_mutex); |
18c850fd JB |
1263 | /* |
1264 | * The device_list_mutex cannot be taken here in case opening the | |
a8698707 | 1265 | * underlying device takes further locks like open_mutex. |
18c850fd JB |
1266 | * |
1267 | * We also don't need the lock here as this is called during mount and | |
1268 | * exclusion is provided by uuid_mutex | |
1269 | */ | |
f5194e34 | 1270 | |
2b82032c | 1271 | if (fs_devices->opened) { |
e4404d6e YZ |
1272 | fs_devices->opened++; |
1273 | ret = 0; | |
2b82032c | 1274 | } else { |
f8e10cd3 | 1275 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
897fb573 | 1276 | ret = open_fs_devices(fs_devices, flags, holder); |
2b82032c | 1277 | } |
542c5908 | 1278 | |
8a4b83cc CM |
1279 | return ret; |
1280 | } | |
1281 | ||
8f32380d | 1282 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
6cf86a00 | 1283 | { |
8f32380d JT |
1284 | struct page *page = virt_to_page(super); |
1285 | ||
6cf86a00 AJ |
1286 | put_page(page); |
1287 | } | |
1288 | ||
b335eab8 | 1289 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
12659251 | 1290 | u64 bytenr, u64 bytenr_orig) |
6cf86a00 | 1291 | { |
b335eab8 NB |
1292 | struct btrfs_super_block *disk_super; |
1293 | struct page *page; | |
6cf86a00 AJ |
1294 | void *p; |
1295 | pgoff_t index; | |
1296 | ||
1297 | /* make sure our super fits in the device */ | |
cda00eba | 1298 | if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) |
b335eab8 | 1299 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1300 | |
1301 | /* make sure our super fits in the page */ | |
b335eab8 NB |
1302 | if (sizeof(*disk_super) > PAGE_SIZE) |
1303 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1304 | |
1305 | /* make sure our super doesn't straddle pages on disk */ | |
1306 | index = bytenr >> PAGE_SHIFT; | |
b335eab8 NB |
1307 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
1308 | return ERR_PTR(-EINVAL); | |
6cf86a00 AJ |
1309 | |
1310 | /* pull in the page with our super */ | |
b335eab8 | 1311 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
6cf86a00 | 1312 | |
b335eab8 NB |
1313 | if (IS_ERR(page)) |
1314 | return ERR_CAST(page); | |
6cf86a00 | 1315 | |
b335eab8 | 1316 | p = page_address(page); |
6cf86a00 AJ |
1317 | |
1318 | /* align our pointer to the offset of the super block */ | |
b335eab8 | 1319 | disk_super = p + offset_in_page(bytenr); |
6cf86a00 | 1320 | |
12659251 | 1321 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
b335eab8 | 1322 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
8f32380d | 1323 | btrfs_release_disk_super(p); |
b335eab8 | 1324 | return ERR_PTR(-EINVAL); |
6cf86a00 AJ |
1325 | } |
1326 | ||
b335eab8 NB |
1327 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
1328 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; | |
6cf86a00 | 1329 | |
b335eab8 | 1330 | return disk_super; |
6cf86a00 AJ |
1331 | } |
1332 | ||
16cab91a | 1333 | int btrfs_forget_devices(dev_t devt) |
228a73ab AJ |
1334 | { |
1335 | int ret; | |
1336 | ||
1337 | mutex_lock(&uuid_mutex); | |
16cab91a | 1338 | ret = btrfs_free_stale_devices(devt, NULL); |
228a73ab AJ |
1339 | mutex_unlock(&uuid_mutex); |
1340 | ||
1341 | return ret; | |
1342 | } | |
1343 | ||
6f60cbd3 DS |
1344 | /* |
1345 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1346 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1347 | * is read via pagecache | |
1348 | */ | |
05bdb996 | 1349 | struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags) |
8a4b83cc CM |
1350 | { |
1351 | struct btrfs_super_block *disk_super; | |
4306a974 | 1352 | bool new_device_added = false; |
36350e95 | 1353 | struct btrfs_device *device = NULL; |
8a4b83cc | 1354 | struct block_device *bdev; |
12659251 NA |
1355 | u64 bytenr, bytenr_orig; |
1356 | int ret; | |
8a4b83cc | 1357 | |
899f9307 DS |
1358 | lockdep_assert_held(&uuid_mutex); |
1359 | ||
6f60cbd3 DS |
1360 | /* |
1361 | * we would like to check all the supers, but that would make | |
1362 | * a btrfs mount succeed after a mkfs from a different FS. | |
1363 | * So, we need to add a special mount option to scan for | |
1364 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1365 | */ | |
6f60cbd3 | 1366 | |
50d281fc | 1367 | /* |
2ef78928 CH |
1368 | * Avoid an exclusive open here, as the systemd-udev may initiate the |
1369 | * device scan which may race with the user's mount or mkfs command, | |
1370 | * resulting in failure. | |
1371 | * Since the device scan is solely for reading purposes, there is no | |
1372 | * need for an exclusive open. Additionally, the devices are read again | |
50d281fc AJ |
1373 | * during the mount process. It is ok to get some inconsistent |
1374 | * values temporarily, as the device paths of the fsid are the only | |
1375 | * required information for assembling the volume. | |
1376 | */ | |
2ef78928 | 1377 | bdev = blkdev_get_by_path(path, flags, NULL, NULL); |
b6ed73bc | 1378 | if (IS_ERR(bdev)) |
36350e95 | 1379 | return ERR_CAST(bdev); |
6f60cbd3 | 1380 | |
12659251 NA |
1381 | bytenr_orig = btrfs_sb_offset(0); |
1382 | ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr); | |
4989d4a0 SK |
1383 | if (ret) { |
1384 | device = ERR_PTR(ret); | |
1385 | goto error_bdev_put; | |
1386 | } | |
12659251 NA |
1387 | |
1388 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig); | |
b335eab8 NB |
1389 | if (IS_ERR(disk_super)) { |
1390 | device = ERR_CAST(disk_super); | |
6f60cbd3 | 1391 | goto error_bdev_put; |
05a5c55d | 1392 | } |
6f60cbd3 | 1393 | |
4306a974 | 1394 | device = device_list_add(path, disk_super, &new_device_added); |
4889bc05 AJ |
1395 | if (!IS_ERR(device) && new_device_added) |
1396 | btrfs_free_stale_devices(device->devt, device); | |
6f60cbd3 | 1397 | |
8f32380d | 1398 | btrfs_release_disk_super(disk_super); |
6f60cbd3 DS |
1399 | |
1400 | error_bdev_put: | |
2736e8ee | 1401 | blkdev_put(bdev, NULL); |
b6ed73bc | 1402 | |
36350e95 | 1403 | return device; |
8a4b83cc | 1404 | } |
0b86a832 | 1405 | |
1c11b63e JM |
1406 | /* |
1407 | * Try to find a chunk that intersects [start, start + len] range and when one | |
1408 | * such is found, record the end of it in *start | |
1409 | */ | |
1c11b63e JM |
1410 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
1411 | u64 len) | |
6df9a95e | 1412 | { |
1c11b63e | 1413 | u64 physical_start, physical_end; |
6df9a95e | 1414 | |
1c11b63e | 1415 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
6df9a95e | 1416 | |
1c11b63e JM |
1417 | if (!find_first_extent_bit(&device->alloc_state, *start, |
1418 | &physical_start, &physical_end, | |
1419 | CHUNK_ALLOCATED, NULL)) { | |
c152b63e | 1420 | |
1c11b63e JM |
1421 | if (in_range(physical_start, *start, len) || |
1422 | in_range(*start, physical_start, | |
1423 | physical_end - physical_start)) { | |
1424 | *start = physical_end + 1; | |
1425 | return true; | |
6df9a95e JB |
1426 | } |
1427 | } | |
1c11b63e | 1428 | return false; |
6df9a95e JB |
1429 | } |
1430 | ||
3b4ffa40 NA |
1431 | static u64 dev_extent_search_start(struct btrfs_device *device, u64 start) |
1432 | { | |
1433 | switch (device->fs_devices->chunk_alloc_policy) { | |
1434 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
37f85ec3 | 1435 | return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED); |
1cd6121f NA |
1436 | case BTRFS_CHUNK_ALLOC_ZONED: |
1437 | /* | |
1438 | * We don't care about the starting region like regular | |
1439 | * allocator, because we anyway use/reserve the first two zones | |
1440 | * for superblock logging. | |
1441 | */ | |
1442 | return ALIGN(start, device->zone_info->zone_size); | |
3b4ffa40 NA |
1443 | default: |
1444 | BUG(); | |
1445 | } | |
1446 | } | |
1447 | ||
1cd6121f NA |
1448 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
1449 | u64 *hole_start, u64 *hole_size, | |
1450 | u64 num_bytes) | |
1451 | { | |
1452 | u64 zone_size = device->zone_info->zone_size; | |
1453 | u64 pos; | |
1454 | int ret; | |
1455 | bool changed = false; | |
1456 | ||
1457 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); | |
1458 | ||
1459 | while (*hole_size > 0) { | |
1460 | pos = btrfs_find_allocatable_zones(device, *hole_start, | |
1461 | *hole_start + *hole_size, | |
1462 | num_bytes); | |
1463 | if (pos != *hole_start) { | |
1464 | *hole_size = *hole_start + *hole_size - pos; | |
1465 | *hole_start = pos; | |
1466 | changed = true; | |
1467 | if (*hole_size < num_bytes) | |
1468 | break; | |
1469 | } | |
1470 | ||
1471 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); | |
1472 | ||
1473 | /* Range is ensured to be empty */ | |
1474 | if (!ret) | |
1475 | return changed; | |
1476 | ||
1477 | /* Given hole range was invalid (outside of device) */ | |
1478 | if (ret == -ERANGE) { | |
1479 | *hole_start += *hole_size; | |
d6f67afb | 1480 | *hole_size = 0; |
7000babd | 1481 | return true; |
1cd6121f NA |
1482 | } |
1483 | ||
1484 | *hole_start += zone_size; | |
1485 | *hole_size -= zone_size; | |
1486 | changed = true; | |
1487 | } | |
1488 | ||
1489 | return changed; | |
1490 | } | |
1491 | ||
43dd529a DS |
1492 | /* |
1493 | * Check if specified hole is suitable for allocation. | |
1494 | * | |
3b4ffa40 NA |
1495 | * @device: the device which we have the hole |
1496 | * @hole_start: starting position of the hole | |
1497 | * @hole_size: the size of the hole | |
1498 | * @num_bytes: the size of the free space that we need | |
1499 | * | |
1cd6121f | 1500 | * This function may modify @hole_start and @hole_size to reflect the suitable |
3b4ffa40 NA |
1501 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
1502 | */ | |
1503 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, | |
1504 | u64 *hole_size, u64 num_bytes) | |
1505 | { | |
1506 | bool changed = false; | |
1507 | u64 hole_end = *hole_start + *hole_size; | |
1508 | ||
1cd6121f NA |
1509 | for (;;) { |
1510 | /* | |
1511 | * Check before we set max_hole_start, otherwise we could end up | |
1512 | * sending back this offset anyway. | |
1513 | */ | |
1514 | if (contains_pending_extent(device, hole_start, *hole_size)) { | |
1515 | if (hole_end >= *hole_start) | |
1516 | *hole_size = hole_end - *hole_start; | |
1517 | else | |
1518 | *hole_size = 0; | |
1519 | changed = true; | |
1520 | } | |
1521 | ||
1522 | switch (device->fs_devices->chunk_alloc_policy) { | |
1523 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
1524 | /* No extra check */ | |
1525 | break; | |
1526 | case BTRFS_CHUNK_ALLOC_ZONED: | |
1527 | if (dev_extent_hole_check_zoned(device, hole_start, | |
1528 | hole_size, num_bytes)) { | |
1529 | changed = true; | |
1530 | /* | |
1531 | * The changed hole can contain pending extent. | |
1532 | * Loop again to check that. | |
1533 | */ | |
1534 | continue; | |
1535 | } | |
1536 | break; | |
1537 | default: | |
1538 | BUG(); | |
1539 | } | |
3b4ffa40 | 1540 | |
3b4ffa40 | 1541 | break; |
3b4ffa40 NA |
1542 | } |
1543 | ||
1544 | return changed; | |
1545 | } | |
6df9a95e | 1546 | |
0b86a832 | 1547 | /* |
43dd529a DS |
1548 | * Find free space in the specified device. |
1549 | * | |
499f377f JM |
1550 | * @device: the device which we search the free space in |
1551 | * @num_bytes: the size of the free space that we need | |
1552 | * @search_start: the position from which to begin the search | |
1553 | * @start: store the start of the free space. | |
1554 | * @len: the size of the free space. that we find, or the size | |
1555 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1556 | * |
43dd529a DS |
1557 | * This does a pretty simple search, the expectation is that it is called very |
1558 | * infrequently and that a given device has a small number of extents. | |
7bfc837d MX |
1559 | * |
1560 | * @start is used to store the start of the free space if we find. But if we | |
1561 | * don't find suitable free space, it will be used to store the start position | |
1562 | * of the max free space. | |
1563 | * | |
1564 | * @len is used to store the size of the free space that we find. | |
1565 | * But if we don't find suitable free space, it is used to store the size of | |
1566 | * the max free space. | |
135da976 QW |
1567 | * |
1568 | * NOTE: This function will search *commit* root of device tree, and does extra | |
1569 | * check to ensure dev extents are not double allocated. | |
1570 | * This makes the function safe to allocate dev extents but may not report | |
1571 | * correct usable device space, as device extent freed in current transaction | |
1a9fd417 | 1572 | * is not reported as available. |
0b86a832 | 1573 | */ |
9e3246a5 QW |
1574 | static int find_free_dev_extent_start(struct btrfs_device *device, |
1575 | u64 num_bytes, u64 search_start, u64 *start, | |
1576 | u64 *len) | |
0b86a832 | 1577 | { |
0b246afa JM |
1578 | struct btrfs_fs_info *fs_info = device->fs_info; |
1579 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1580 | struct btrfs_key key; |
7bfc837d | 1581 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1582 | struct btrfs_path *path; |
7bfc837d MX |
1583 | u64 hole_size; |
1584 | u64 max_hole_start; | |
1585 | u64 max_hole_size; | |
1586 | u64 extent_end; | |
0b86a832 CM |
1587 | u64 search_end = device->total_bytes; |
1588 | int ret; | |
7bfc837d | 1589 | int slot; |
0b86a832 | 1590 | struct extent_buffer *l; |
8cdc7c5b | 1591 | |
3b4ffa40 | 1592 | search_start = dev_extent_search_start(device, search_start); |
0b86a832 | 1593 | |
1cd6121f NA |
1594 | WARN_ON(device->zone_info && |
1595 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); | |
1596 | ||
6df9a95e JB |
1597 | path = btrfs_alloc_path(); |
1598 | if (!path) | |
1599 | return -ENOMEM; | |
f2ab7618 | 1600 | |
7bfc837d MX |
1601 | max_hole_start = search_start; |
1602 | max_hole_size = 0; | |
1603 | ||
f2ab7618 | 1604 | again: |
401e29c1 AJ |
1605 | if (search_start >= search_end || |
1606 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { | |
7bfc837d | 1607 | ret = -ENOSPC; |
6df9a95e | 1608 | goto out; |
7bfc837d MX |
1609 | } |
1610 | ||
e4058b54 | 1611 | path->reada = READA_FORWARD; |
6df9a95e JB |
1612 | path->search_commit_root = 1; |
1613 | path->skip_locking = 1; | |
7bfc837d | 1614 | |
0b86a832 CM |
1615 | key.objectid = device->devid; |
1616 | key.offset = search_start; | |
1617 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1618 | |
0ff40a91 | 1619 | ret = btrfs_search_backwards(root, &key, path); |
0b86a832 | 1620 | if (ret < 0) |
7bfc837d | 1621 | goto out; |
7bfc837d | 1622 | |
3c538de0 | 1623 | while (search_start < search_end) { |
0b86a832 CM |
1624 | l = path->nodes[0]; |
1625 | slot = path->slots[0]; | |
1626 | if (slot >= btrfs_header_nritems(l)) { | |
1627 | ret = btrfs_next_leaf(root, path); | |
1628 | if (ret == 0) | |
1629 | continue; | |
1630 | if (ret < 0) | |
7bfc837d MX |
1631 | goto out; |
1632 | ||
1633 | break; | |
0b86a832 CM |
1634 | } |
1635 | btrfs_item_key_to_cpu(l, &key, slot); | |
1636 | ||
1637 | if (key.objectid < device->devid) | |
1638 | goto next; | |
1639 | ||
1640 | if (key.objectid > device->devid) | |
7bfc837d | 1641 | break; |
0b86a832 | 1642 | |
962a298f | 1643 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1644 | goto next; |
9779b72f | 1645 | |
3c538de0 JB |
1646 | if (key.offset > search_end) |
1647 | break; | |
1648 | ||
7bfc837d MX |
1649 | if (key.offset > search_start) { |
1650 | hole_size = key.offset - search_start; | |
3b4ffa40 NA |
1651 | dev_extent_hole_check(device, &search_start, &hole_size, |
1652 | num_bytes); | |
6df9a95e | 1653 | |
7bfc837d MX |
1654 | if (hole_size > max_hole_size) { |
1655 | max_hole_start = search_start; | |
1656 | max_hole_size = hole_size; | |
1657 | } | |
9779b72f | 1658 | |
7bfc837d MX |
1659 | /* |
1660 | * If this free space is greater than which we need, | |
1661 | * it must be the max free space that we have found | |
1662 | * until now, so max_hole_start must point to the start | |
1663 | * of this free space and the length of this free space | |
1664 | * is stored in max_hole_size. Thus, we return | |
1665 | * max_hole_start and max_hole_size and go back to the | |
1666 | * caller. | |
1667 | */ | |
1668 | if (hole_size >= num_bytes) { | |
1669 | ret = 0; | |
1670 | goto out; | |
0b86a832 CM |
1671 | } |
1672 | } | |
0b86a832 | 1673 | |
0b86a832 | 1674 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1675 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1676 | dev_extent); | |
1677 | if (extent_end > search_start) | |
1678 | search_start = extent_end; | |
0b86a832 CM |
1679 | next: |
1680 | path->slots[0]++; | |
1681 | cond_resched(); | |
1682 | } | |
0b86a832 | 1683 | |
38c01b96 | 1684 | /* |
1685 | * At this point, search_start should be the end of | |
1686 | * allocated dev extents, and when shrinking the device, | |
1687 | * search_end may be smaller than search_start. | |
1688 | */ | |
f2ab7618 | 1689 | if (search_end > search_start) { |
38c01b96 | 1690 | hole_size = search_end - search_start; |
3b4ffa40 NA |
1691 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
1692 | num_bytes)) { | |
f2ab7618 ZL |
1693 | btrfs_release_path(path); |
1694 | goto again; | |
1695 | } | |
0b86a832 | 1696 | |
f2ab7618 ZL |
1697 | if (hole_size > max_hole_size) { |
1698 | max_hole_start = search_start; | |
1699 | max_hole_size = hole_size; | |
1700 | } | |
6df9a95e JB |
1701 | } |
1702 | ||
7bfc837d | 1703 | /* See above. */ |
f2ab7618 | 1704 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1705 | ret = -ENOSPC; |
1706 | else | |
1707 | ret = 0; | |
1708 | ||
3c538de0 | 1709 | ASSERT(max_hole_start + max_hole_size <= search_end); |
7bfc837d | 1710 | out: |
2b82032c | 1711 | btrfs_free_path(path); |
7bfc837d | 1712 | *start = max_hole_start; |
b2117a39 | 1713 | if (len) |
7bfc837d | 1714 | *len = max_hole_size; |
0b86a832 CM |
1715 | return ret; |
1716 | } | |
1717 | ||
60dfdf25 | 1718 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
499f377f JM |
1719 | u64 *start, u64 *len) |
1720 | { | |
499f377f | 1721 | /* FIXME use last free of some kind */ |
60dfdf25 | 1722 | return find_free_dev_extent_start(device, num_bytes, 0, start, len); |
499f377f JM |
1723 | } |
1724 | ||
b2950863 | 1725 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1726 | struct btrfs_device *device, |
2196d6e8 | 1727 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1728 | { |
0b246afa JM |
1729 | struct btrfs_fs_info *fs_info = device->fs_info; |
1730 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1731 | int ret; |
1732 | struct btrfs_path *path; | |
8f18cf13 | 1733 | struct btrfs_key key; |
a061fc8d CM |
1734 | struct btrfs_key found_key; |
1735 | struct extent_buffer *leaf = NULL; | |
1736 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1737 | |
1738 | path = btrfs_alloc_path(); | |
1739 | if (!path) | |
1740 | return -ENOMEM; | |
1741 | ||
1742 | key.objectid = device->devid; | |
1743 | key.offset = start; | |
1744 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1745 | again: |
8f18cf13 | 1746 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1747 | if (ret > 0) { |
1748 | ret = btrfs_previous_item(root, path, key.objectid, | |
1749 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1750 | if (ret) |
1751 | goto out; | |
a061fc8d CM |
1752 | leaf = path->nodes[0]; |
1753 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1754 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1755 | struct btrfs_dev_extent); | |
1756 | BUG_ON(found_key.offset > start || found_key.offset + | |
1757 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1758 | key = found_key; |
1759 | btrfs_release_path(path); | |
1760 | goto again; | |
a061fc8d CM |
1761 | } else if (ret == 0) { |
1762 | leaf = path->nodes[0]; | |
1763 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1764 | struct btrfs_dev_extent); | |
79787eaa | 1765 | } else { |
79787eaa | 1766 | goto out; |
a061fc8d | 1767 | } |
8f18cf13 | 1768 | |
2196d6e8 MX |
1769 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1770 | ||
8f18cf13 | 1771 | ret = btrfs_del_item(trans, root, path); |
79bd3712 | 1772 | if (ret == 0) |
3204d33c | 1773 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
b0b802d7 | 1774 | out: |
8f18cf13 CM |
1775 | btrfs_free_path(path); |
1776 | return ret; | |
1777 | } | |
1778 | ||
6df9a95e | 1779 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1780 | { |
6df9a95e JB |
1781 | struct extent_map_tree *em_tree; |
1782 | struct extent_map *em; | |
1783 | struct rb_node *n; | |
1784 | u64 ret = 0; | |
0b86a832 | 1785 | |
c8bf1b67 | 1786 | em_tree = &fs_info->mapping_tree; |
6df9a95e | 1787 | read_lock(&em_tree->lock); |
07e1ce09 | 1788 | n = rb_last(&em_tree->map.rb_root); |
6df9a95e JB |
1789 | if (n) { |
1790 | em = rb_entry(n, struct extent_map, rb_node); | |
1791 | ret = em->start + em->len; | |
0b86a832 | 1792 | } |
6df9a95e JB |
1793 | read_unlock(&em_tree->lock); |
1794 | ||
0b86a832 CM |
1795 | return ret; |
1796 | } | |
1797 | ||
53f10659 ID |
1798 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1799 | u64 *devid_ret) | |
0b86a832 CM |
1800 | { |
1801 | int ret; | |
1802 | struct btrfs_key key; | |
1803 | struct btrfs_key found_key; | |
2b82032c YZ |
1804 | struct btrfs_path *path; |
1805 | ||
2b82032c YZ |
1806 | path = btrfs_alloc_path(); |
1807 | if (!path) | |
1808 | return -ENOMEM; | |
0b86a832 CM |
1809 | |
1810 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1811 | key.type = BTRFS_DEV_ITEM_KEY; | |
1812 | key.offset = (u64)-1; | |
1813 | ||
53f10659 | 1814 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1815 | if (ret < 0) |
1816 | goto error; | |
1817 | ||
a06dee4d AJ |
1818 | if (ret == 0) { |
1819 | /* Corruption */ | |
1820 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); | |
1821 | ret = -EUCLEAN; | |
1822 | goto error; | |
1823 | } | |
0b86a832 | 1824 | |
53f10659 ID |
1825 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1826 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1827 | BTRFS_DEV_ITEM_KEY); |
1828 | if (ret) { | |
53f10659 | 1829 | *devid_ret = 1; |
0b86a832 CM |
1830 | } else { |
1831 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1832 | path->slots[0]); | |
53f10659 | 1833 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1834 | } |
1835 | ret = 0; | |
1836 | error: | |
2b82032c | 1837 | btrfs_free_path(path); |
0b86a832 CM |
1838 | return ret; |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * the device information is stored in the chunk root | |
1843 | * the btrfs_device struct should be fully filled in | |
1844 | */ | |
c74a0b02 | 1845 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
48a3b636 | 1846 | struct btrfs_device *device) |
0b86a832 CM |
1847 | { |
1848 | int ret; | |
1849 | struct btrfs_path *path; | |
1850 | struct btrfs_dev_item *dev_item; | |
1851 | struct extent_buffer *leaf; | |
1852 | struct btrfs_key key; | |
1853 | unsigned long ptr; | |
0b86a832 | 1854 | |
0b86a832 CM |
1855 | path = btrfs_alloc_path(); |
1856 | if (!path) | |
1857 | return -ENOMEM; | |
1858 | ||
0b86a832 CM |
1859 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1860 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1861 | key.offset = device->devid; |
0b86a832 | 1862 | |
2bb2e00e | 1863 | btrfs_reserve_chunk_metadata(trans, true); |
8e87e856 NB |
1864 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
1865 | &key, sizeof(*dev_item)); | |
2bb2e00e | 1866 | btrfs_trans_release_chunk_metadata(trans); |
0b86a832 CM |
1867 | if (ret) |
1868 | goto out; | |
1869 | ||
1870 | leaf = path->nodes[0]; | |
1871 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1872 | ||
1873 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1874 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1875 | btrfs_set_device_type(leaf, dev_item, device->type); |
1876 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1877 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1878 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1879 | btrfs_set_device_total_bytes(leaf, dev_item, |
1880 | btrfs_device_get_disk_total_bytes(device)); | |
1881 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1882 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1883 | btrfs_set_device_group(leaf, dev_item, 0); |
1884 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1885 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1886 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1887 | |
410ba3a2 | 1888 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1889 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1890 | ptr = btrfs_device_fsid(dev_item); |
de37aa51 NB |
1891 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
1892 | ptr, BTRFS_FSID_SIZE); | |
0b86a832 | 1893 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1894 | |
2b82032c | 1895 | ret = 0; |
0b86a832 CM |
1896 | out: |
1897 | btrfs_free_path(path); | |
1898 | return ret; | |
1899 | } | |
8f18cf13 | 1900 | |
5a1972bd QW |
1901 | /* |
1902 | * Function to update ctime/mtime for a given device path. | |
1903 | * Mainly used for ctime/mtime based probe like libblkid. | |
54fde91f JB |
1904 | * |
1905 | * We don't care about errors here, this is just to be kind to userspace. | |
5a1972bd | 1906 | */ |
54fde91f | 1907 | static void update_dev_time(const char *device_path) |
5a1972bd | 1908 | { |
54fde91f | 1909 | struct path path; |
8f96a5bf | 1910 | struct timespec64 now; |
54fde91f | 1911 | int ret; |
5a1972bd | 1912 | |
54fde91f JB |
1913 | ret = kern_path(device_path, LOOKUP_FOLLOW, &path); |
1914 | if (ret) | |
5a1972bd | 1915 | return; |
8f96a5bf | 1916 | |
54fde91f JB |
1917 | now = current_time(d_inode(path.dentry)); |
1918 | inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME); | |
1919 | path_put(&path); | |
5a1972bd QW |
1920 | } |
1921 | ||
bbac5869 QW |
1922 | static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, |
1923 | struct btrfs_device *device) | |
a061fc8d | 1924 | { |
f331a952 | 1925 | struct btrfs_root *root = device->fs_info->chunk_root; |
a061fc8d CM |
1926 | int ret; |
1927 | struct btrfs_path *path; | |
a061fc8d | 1928 | struct btrfs_key key; |
a061fc8d | 1929 | |
a061fc8d CM |
1930 | path = btrfs_alloc_path(); |
1931 | if (!path) | |
1932 | return -ENOMEM; | |
1933 | ||
a061fc8d CM |
1934 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1935 | key.type = BTRFS_DEV_ITEM_KEY; | |
1936 | key.offset = device->devid; | |
1937 | ||
2bb2e00e | 1938 | btrfs_reserve_chunk_metadata(trans, false); |
a061fc8d | 1939 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
2bb2e00e | 1940 | btrfs_trans_release_chunk_metadata(trans); |
5e9f2ad5 NB |
1941 | if (ret) { |
1942 | if (ret > 0) | |
1943 | ret = -ENOENT; | |
a061fc8d CM |
1944 | goto out; |
1945 | } | |
1946 | ||
1947 | ret = btrfs_del_item(trans, root, path); | |
a061fc8d CM |
1948 | out: |
1949 | btrfs_free_path(path); | |
a061fc8d CM |
1950 | return ret; |
1951 | } | |
1952 | ||
3cc31a0d DS |
1953 | /* |
1954 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1955 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1956 | * replace. | |
1957 | */ | |
1958 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1959 | u64 num_devices) | |
a061fc8d | 1960 | { |
a061fc8d | 1961 | u64 all_avail; |
de98ced9 | 1962 | unsigned seq; |
418775a2 | 1963 | int i; |
a061fc8d | 1964 | |
de98ced9 | 1965 | do { |
bd45ffbc | 1966 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1967 | |
bd45ffbc AJ |
1968 | all_avail = fs_info->avail_data_alloc_bits | |
1969 | fs_info->avail_system_alloc_bits | | |
1970 | fs_info->avail_metadata_alloc_bits; | |
1971 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1972 | |
418775a2 | 1973 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
41a6e891 | 1974 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
418775a2 | 1975 | continue; |
a061fc8d | 1976 | |
efc222f8 AJ |
1977 | if (num_devices < btrfs_raid_array[i].devs_min) |
1978 | return btrfs_raid_array[i].mindev_error; | |
53b381b3 DW |
1979 | } |
1980 | ||
bd45ffbc | 1981 | return 0; |
f1fa7f26 AJ |
1982 | } |
1983 | ||
c9162bdf OS |
1984 | static struct btrfs_device * btrfs_find_next_active_device( |
1985 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) | |
a061fc8d | 1986 | { |
2b82032c | 1987 | struct btrfs_device *next_device; |
88acff64 AJ |
1988 | |
1989 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1990 | if (next_device != device && | |
e6e674bd AJ |
1991 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
1992 | && next_device->bdev) | |
88acff64 AJ |
1993 | return next_device; |
1994 | } | |
1995 | ||
1996 | return NULL; | |
1997 | } | |
1998 | ||
1999 | /* | |
d24fa5c1 | 2000 | * Helper function to check if the given device is part of s_bdev / latest_dev |
88acff64 AJ |
2001 | * and replace it with the provided or the next active device, in the context |
2002 | * where this function called, there should be always be another device (or | |
2003 | * this_dev) which is active. | |
2004 | */ | |
b105e927 | 2005 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
e493e8f9 | 2006 | struct btrfs_device *next_device) |
88acff64 | 2007 | { |
d6507cf1 | 2008 | struct btrfs_fs_info *fs_info = device->fs_info; |
88acff64 | 2009 | |
e493e8f9 | 2010 | if (!next_device) |
88acff64 | 2011 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
e493e8f9 | 2012 | device); |
88acff64 AJ |
2013 | ASSERT(next_device); |
2014 | ||
2015 | if (fs_info->sb->s_bdev && | |
2016 | (fs_info->sb->s_bdev == device->bdev)) | |
2017 | fs_info->sb->s_bdev = next_device->bdev; | |
2018 | ||
d24fa5c1 AJ |
2019 | if (fs_info->fs_devices->latest_dev->bdev == device->bdev) |
2020 | fs_info->fs_devices->latest_dev = next_device; | |
88acff64 AJ |
2021 | } |
2022 | ||
1da73967 AJ |
2023 | /* |
2024 | * Return btrfs_fs_devices::num_devices excluding the device that's being | |
2025 | * currently replaced. | |
2026 | */ | |
2027 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) | |
2028 | { | |
2029 | u64 num_devices = fs_info->fs_devices->num_devices; | |
2030 | ||
cb5583dd | 2031 | down_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2032 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
2033 | ASSERT(num_devices > 1); | |
2034 | num_devices--; | |
2035 | } | |
cb5583dd | 2036 | up_read(&fs_info->dev_replace.rwsem); |
1da73967 AJ |
2037 | |
2038 | return num_devices; | |
2039 | } | |
2040 | ||
0e0078f7 CH |
2041 | static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, |
2042 | struct block_device *bdev, int copy_num) | |
2043 | { | |
2044 | struct btrfs_super_block *disk_super; | |
26ecf243 CH |
2045 | const size_t len = sizeof(disk_super->magic); |
2046 | const u64 bytenr = btrfs_sb_offset(copy_num); | |
0e0078f7 CH |
2047 | int ret; |
2048 | ||
26ecf243 | 2049 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); |
0e0078f7 CH |
2050 | if (IS_ERR(disk_super)) |
2051 | return; | |
2052 | ||
26ecf243 CH |
2053 | memset(&disk_super->magic, 0, len); |
2054 | folio_mark_dirty(virt_to_folio(disk_super)); | |
2055 | btrfs_release_disk_super(disk_super); | |
2056 | ||
2057 | ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); | |
0e0078f7 CH |
2058 | if (ret) |
2059 | btrfs_warn(fs_info, "error clearing superblock number %d (%d)", | |
2060 | copy_num, ret); | |
0e0078f7 CH |
2061 | } |
2062 | ||
313b0858 JB |
2063 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
2064 | struct block_device *bdev, | |
2065 | const char *device_path) | |
6fbceb9f | 2066 | { |
6fbceb9f JT |
2067 | int copy_num; |
2068 | ||
2069 | if (!bdev) | |
2070 | return; | |
2071 | ||
2072 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { | |
0e0078f7 | 2073 | if (bdev_is_zoned(bdev)) |
12659251 | 2074 | btrfs_reset_sb_log_zones(bdev, copy_num); |
0e0078f7 CH |
2075 | else |
2076 | btrfs_scratch_superblock(fs_info, bdev, copy_num); | |
6fbceb9f JT |
2077 | } |
2078 | ||
2079 | /* Notify udev that device has changed */ | |
2080 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
2081 | ||
2082 | /* Update ctime/mtime for device path for libblkid */ | |
54fde91f | 2083 | update_dev_time(device_path); |
6fbceb9f JT |
2084 | } |
2085 | ||
1a15eb72 JB |
2086 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, |
2087 | struct btrfs_dev_lookup_args *args, | |
2736e8ee | 2088 | struct block_device **bdev, void **holder) |
f1fa7f26 | 2089 | { |
bbac5869 | 2090 | struct btrfs_trans_handle *trans; |
f1fa7f26 | 2091 | struct btrfs_device *device; |
1f78160c | 2092 | struct btrfs_fs_devices *cur_devices; |
b5185197 | 2093 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2094 | u64 num_devices; |
a061fc8d CM |
2095 | int ret = 0; |
2096 | ||
914a519b JB |
2097 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
2098 | btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); | |
2099 | return -EINVAL; | |
2100 | } | |
2101 | ||
8ef9dc0f JB |
2102 | /* |
2103 | * The device list in fs_devices is accessed without locks (neither | |
2104 | * uuid_mutex nor device_list_mutex) as it won't change on a mounted | |
2105 | * filesystem and another device rm cannot run. | |
2106 | */ | |
1da73967 | 2107 | num_devices = btrfs_num_devices(fs_info); |
8dabb742 | 2108 | |
0b246afa | 2109 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 2110 | if (ret) |
bbac5869 | 2111 | return ret; |
a061fc8d | 2112 | |
1a15eb72 JB |
2113 | device = btrfs_find_device(fs_info->fs_devices, args); |
2114 | if (!device) { | |
2115 | if (args->missing) | |
a27a94c2 NB |
2116 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
2117 | else | |
1a15eb72 | 2118 | ret = -ENOENT; |
bbac5869 | 2119 | return ret; |
a27a94c2 | 2120 | } |
dfe25020 | 2121 | |
eede2bf3 OS |
2122 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
2123 | btrfs_warn_in_rcu(fs_info, | |
2124 | "cannot remove device %s (devid %llu) due to active swapfile", | |
cb3e217b | 2125 | btrfs_dev_name(device), device->devid); |
bbac5869 | 2126 | return -ETXTBSY; |
eede2bf3 OS |
2127 | } |
2128 | ||
bbac5869 QW |
2129 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
2130 | return BTRFS_ERROR_DEV_TGT_REPLACE; | |
63a212ab | 2131 | |
ebbede42 | 2132 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
bbac5869 QW |
2133 | fs_info->fs_devices->rw_devices == 1) |
2134 | return BTRFS_ERROR_DEV_ONLY_WRITABLE; | |
2b82032c | 2135 | |
ebbede42 | 2136 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2137 | mutex_lock(&fs_info->chunk_mutex); |
2b82032c | 2138 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 2139 | device->fs_devices->rw_devices--; |
34441361 | 2140 | mutex_unlock(&fs_info->chunk_mutex); |
dfe25020 | 2141 | } |
a061fc8d CM |
2142 | |
2143 | ret = btrfs_shrink_device(device, 0); | |
2144 | if (ret) | |
9b3517e9 | 2145 | goto error_undo; |
a061fc8d | 2146 | |
bbac5869 QW |
2147 | trans = btrfs_start_transaction(fs_info->chunk_root, 0); |
2148 | if (IS_ERR(trans)) { | |
2149 | ret = PTR_ERR(trans); | |
9b3517e9 | 2150 | goto error_undo; |
bbac5869 QW |
2151 | } |
2152 | ||
2153 | ret = btrfs_rm_dev_item(trans, device); | |
2154 | if (ret) { | |
2155 | /* Any error in dev item removal is critical */ | |
2156 | btrfs_crit(fs_info, | |
2157 | "failed to remove device item for devid %llu: %d", | |
2158 | device->devid, ret); | |
2159 | btrfs_abort_transaction(trans, ret); | |
2160 | btrfs_end_transaction(trans); | |
2161 | return ret; | |
2162 | } | |
a061fc8d | 2163 | |
e12c9621 | 2164 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
163e97ee | 2165 | btrfs_scrub_cancel_dev(device); |
e5e9a520 CM |
2166 | |
2167 | /* | |
2168 | * the device list mutex makes sure that we don't change | |
2169 | * the device list while someone else is writing out all | |
d7306801 FDBM |
2170 | * the device supers. Whoever is writing all supers, should |
2171 | * lock the device list mutex before getting the number of | |
2172 | * devices in the super block (super_copy). Conversely, | |
2173 | * whoever updates the number of devices in the super block | |
2174 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 2175 | */ |
1f78160c | 2176 | |
41a52a0f AJ |
2177 | /* |
2178 | * In normal cases the cur_devices == fs_devices. But in case | |
2179 | * of deleting a seed device, the cur_devices should point to | |
9675ea8c | 2180 | * its own fs_devices listed under the fs_devices->seed_list. |
41a52a0f | 2181 | */ |
1f78160c | 2182 | cur_devices = device->fs_devices; |
b5185197 | 2183 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 2184 | list_del_rcu(&device->dev_list); |
e5e9a520 | 2185 | |
41a52a0f AJ |
2186 | cur_devices->num_devices--; |
2187 | cur_devices->total_devices--; | |
b4993e64 AJ |
2188 | /* Update total_devices of the parent fs_devices if it's seed */ |
2189 | if (cur_devices != fs_devices) | |
2190 | fs_devices->total_devices--; | |
2b82032c | 2191 | |
e6e674bd | 2192 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
41a52a0f | 2193 | cur_devices->missing_devices--; |
cd02dca5 | 2194 | |
d6507cf1 | 2195 | btrfs_assign_next_active_device(device, NULL); |
2b82032c | 2196 | |
0bfaa9c5 | 2197 | if (device->bdev) { |
41a52a0f | 2198 | cur_devices->open_devices--; |
0bfaa9c5 | 2199 | /* remove sysfs entry */ |
53f8a74c | 2200 | btrfs_sysfs_remove_device(device); |
0bfaa9c5 | 2201 | } |
99994cde | 2202 | |
0b246afa JM |
2203 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
2204 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
b5185197 | 2205 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c | 2206 | |
cea67ab9 | 2207 | /* |
3fa421de JB |
2208 | * At this point, the device is zero sized and detached from the |
2209 | * devices list. All that's left is to zero out the old supers and | |
2210 | * free the device. | |
2211 | * | |
2212 | * We cannot call btrfs_close_bdev() here because we're holding the sb | |
2213 | * write lock, and blkdev_put() will pull in the ->open_mutex on the | |
2214 | * block device and it's dependencies. Instead just flush the device | |
2215 | * and let the caller do the final blkdev_put. | |
cea67ab9 | 2216 | */ |
3fa421de | 2217 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
8f32380d JT |
2218 | btrfs_scratch_superblocks(fs_info, device->bdev, |
2219 | device->name->str); | |
3fa421de JB |
2220 | if (device->bdev) { |
2221 | sync_blockdev(device->bdev); | |
2222 | invalidate_bdev(device->bdev); | |
2223 | } | |
2224 | } | |
cea67ab9 | 2225 | |
3fa421de | 2226 | *bdev = device->bdev; |
2736e8ee | 2227 | *holder = device->holder; |
8e75fd89 NB |
2228 | synchronize_rcu(); |
2229 | btrfs_free_device(device); | |
cea67ab9 | 2230 | |
8b41393f JB |
2231 | /* |
2232 | * This can happen if cur_devices is the private seed devices list. We | |
2233 | * cannot call close_fs_devices() here because it expects the uuid_mutex | |
2234 | * to be held, but in fact we don't need that for the private | |
2235 | * seed_devices, we can simply decrement cur_devices->opened and then | |
2236 | * remove it from our list and free the fs_devices. | |
2237 | */ | |
8e906945 | 2238 | if (cur_devices->num_devices == 0) { |
944d3f9f | 2239 | list_del_init(&cur_devices->seed_list); |
8b41393f JB |
2240 | ASSERT(cur_devices->opened == 1); |
2241 | cur_devices->opened--; | |
1f78160c | 2242 | free_fs_devices(cur_devices); |
2b82032c YZ |
2243 | } |
2244 | ||
bbac5869 QW |
2245 | ret = btrfs_commit_transaction(trans); |
2246 | ||
a061fc8d | 2247 | return ret; |
24fc572f | 2248 | |
9b3517e9 | 2249 | error_undo: |
ebbede42 | 2250 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
34441361 | 2251 | mutex_lock(&fs_info->chunk_mutex); |
9b3517e9 | 2252 | list_add(&device->dev_alloc_list, |
b5185197 | 2253 | &fs_devices->alloc_list); |
c3929c36 | 2254 | device->fs_devices->rw_devices++; |
34441361 | 2255 | mutex_unlock(&fs_info->chunk_mutex); |
9b3517e9 | 2256 | } |
bbac5869 | 2257 | return ret; |
a061fc8d CM |
2258 | } |
2259 | ||
68a9db5f | 2260 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
e93c89c1 | 2261 | { |
d51908ce AJ |
2262 | struct btrfs_fs_devices *fs_devices; |
2263 | ||
68a9db5f | 2264 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
1357272f | 2265 | |
25e8e911 AJ |
2266 | /* |
2267 | * in case of fs with no seed, srcdev->fs_devices will point | |
2268 | * to fs_devices of fs_info. However when the dev being replaced is | |
2269 | * a seed dev it will point to the seed's local fs_devices. In short | |
2270 | * srcdev will have its correct fs_devices in both the cases. | |
2271 | */ | |
2272 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2273 | |
e93c89c1 | 2274 | list_del_rcu(&srcdev->dev_list); |
619c47f3 | 2275 | list_del(&srcdev->dev_alloc_list); |
d51908ce | 2276 | fs_devices->num_devices--; |
e6e674bd | 2277 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
d51908ce | 2278 | fs_devices->missing_devices--; |
e93c89c1 | 2279 | |
ebbede42 | 2280 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
82372bc8 | 2281 | fs_devices->rw_devices--; |
1357272f | 2282 | |
82372bc8 | 2283 | if (srcdev->bdev) |
d51908ce | 2284 | fs_devices->open_devices--; |
084b6e7c QW |
2285 | } |
2286 | ||
65237ee3 | 2287 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
084b6e7c QW |
2288 | { |
2289 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2290 | |
a466c85e JB |
2291 | mutex_lock(&uuid_mutex); |
2292 | ||
14238819 | 2293 | btrfs_close_bdev(srcdev); |
8e75fd89 NB |
2294 | synchronize_rcu(); |
2295 | btrfs_free_device(srcdev); | |
94d5f0c2 | 2296 | |
94d5f0c2 AJ |
2297 | /* if this is no devs we rather delete the fs_devices */ |
2298 | if (!fs_devices->num_devices) { | |
6dd38f81 AJ |
2299 | /* |
2300 | * On a mounted FS, num_devices can't be zero unless it's a | |
2301 | * seed. In case of a seed device being replaced, the replace | |
2302 | * target added to the sprout FS, so there will be no more | |
2303 | * device left under the seed FS. | |
2304 | */ | |
2305 | ASSERT(fs_devices->seeding); | |
2306 | ||
944d3f9f | 2307 | list_del_init(&fs_devices->seed_list); |
0226e0eb | 2308 | close_fs_devices(fs_devices); |
8bef8401 | 2309 | free_fs_devices(fs_devices); |
94d5f0c2 | 2310 | } |
a466c85e | 2311 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
2312 | } |
2313 | ||
4f5ad7bd | 2314 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
e93c89c1 | 2315 | { |
4f5ad7bd | 2316 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
d9a071f0 | 2317 | |
d9a071f0 | 2318 | mutex_lock(&fs_devices->device_list_mutex); |
d2ff1b20 | 2319 | |
53f8a74c | 2320 | btrfs_sysfs_remove_device(tgtdev); |
d2ff1b20 | 2321 | |
779bf3fe | 2322 | if (tgtdev->bdev) |
d9a071f0 | 2323 | fs_devices->open_devices--; |
779bf3fe | 2324 | |
d9a071f0 | 2325 | fs_devices->num_devices--; |
e93c89c1 | 2326 | |
d6507cf1 | 2327 | btrfs_assign_next_active_device(tgtdev, NULL); |
e93c89c1 | 2328 | |
e93c89c1 | 2329 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 | 2330 | |
d9a071f0 | 2331 | mutex_unlock(&fs_devices->device_list_mutex); |
779bf3fe | 2332 | |
8f32380d JT |
2333 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
2334 | tgtdev->name->str); | |
14238819 AJ |
2335 | |
2336 | btrfs_close_bdev(tgtdev); | |
8e75fd89 NB |
2337 | synchronize_rcu(); |
2338 | btrfs_free_device(tgtdev); | |
e93c89c1 SB |
2339 | } |
2340 | ||
43dd529a DS |
2341 | /* |
2342 | * Populate args from device at path. | |
faa775c4 JB |
2343 | * |
2344 | * @fs_info: the filesystem | |
2345 | * @args: the args to populate | |
2346 | * @path: the path to the device | |
2347 | * | |
2348 | * This will read the super block of the device at @path and populate @args with | |
2349 | * the devid, fsid, and uuid. This is meant to be used for ioctls that need to | |
2350 | * lookup a device to operate on, but need to do it before we take any locks. | |
2351 | * This properly handles the special case of "missing" that a user may pass in, | |
2352 | * and does some basic sanity checks. The caller must make sure that @path is | |
2353 | * properly NUL terminated before calling in, and must call | |
2354 | * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and | |
2355 | * uuid buffers. | |
2356 | * | |
2357 | * Return: 0 for success, -errno for failure | |
2358 | */ | |
2359 | int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, | |
2360 | struct btrfs_dev_lookup_args *args, | |
2361 | const char *path) | |
7ba15b7d | 2362 | { |
7ba15b7d | 2363 | struct btrfs_super_block *disk_super; |
7ba15b7d | 2364 | struct block_device *bdev; |
faa775c4 | 2365 | int ret; |
7ba15b7d | 2366 | |
faa775c4 JB |
2367 | if (!path || !path[0]) |
2368 | return -EINVAL; | |
2369 | if (!strcmp(path, "missing")) { | |
2370 | args->missing = true; | |
2371 | return 0; | |
2372 | } | |
8f32380d | 2373 | |
faa775c4 JB |
2374 | args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); |
2375 | args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); | |
2376 | if (!args->uuid || !args->fsid) { | |
2377 | btrfs_put_dev_args_from_path(args); | |
2378 | return -ENOMEM; | |
2379 | } | |
8f32380d | 2380 | |
05bdb996 | 2381 | ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0, |
faa775c4 | 2382 | &bdev, &disk_super); |
9ea0106a ZF |
2383 | if (ret) { |
2384 | btrfs_put_dev_args_from_path(args); | |
faa775c4 | 2385 | return ret; |
9ea0106a ZF |
2386 | } |
2387 | ||
faa775c4 JB |
2388 | args->devid = btrfs_stack_device_id(&disk_super->dev_item); |
2389 | memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); | |
7239ff4b | 2390 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
faa775c4 | 2391 | memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); |
7239ff4b | 2392 | else |
faa775c4 | 2393 | memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); |
8f32380d | 2394 | btrfs_release_disk_super(disk_super); |
2736e8ee | 2395 | blkdev_put(bdev, NULL); |
faa775c4 | 2396 | return 0; |
7ba15b7d SB |
2397 | } |
2398 | ||
5c5c0df0 | 2399 | /* |
faa775c4 JB |
2400 | * Only use this jointly with btrfs_get_dev_args_from_path() because we will |
2401 | * allocate our ->uuid and ->fsid pointers, everybody else uses local variables | |
2402 | * that don't need to be freed. | |
5c5c0df0 | 2403 | */ |
faa775c4 JB |
2404 | void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) |
2405 | { | |
2406 | kfree(args->uuid); | |
2407 | kfree(args->fsid); | |
2408 | args->uuid = NULL; | |
2409 | args->fsid = NULL; | |
2410 | } | |
2411 | ||
a27a94c2 | 2412 | struct btrfs_device *btrfs_find_device_by_devspec( |
6e927ceb AJ |
2413 | struct btrfs_fs_info *fs_info, u64 devid, |
2414 | const char *device_path) | |
24e0474b | 2415 | { |
562d7b15 | 2416 | BTRFS_DEV_LOOKUP_ARGS(args); |
a27a94c2 | 2417 | struct btrfs_device *device; |
faa775c4 | 2418 | int ret; |
24e0474b | 2419 | |
5c5c0df0 | 2420 | if (devid) { |
562d7b15 JB |
2421 | args.devid = devid; |
2422 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
a27a94c2 NB |
2423 | if (!device) |
2424 | return ERR_PTR(-ENOENT); | |
6e927ceb AJ |
2425 | return device; |
2426 | } | |
2427 | ||
faa775c4 JB |
2428 | ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); |
2429 | if (ret) | |
2430 | return ERR_PTR(ret); | |
2431 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
2432 | btrfs_put_dev_args_from_path(&args); | |
2433 | if (!device) | |
6e927ceb | 2434 | return ERR_PTR(-ENOENT); |
faa775c4 | 2435 | return device; |
24e0474b AJ |
2436 | } |
2437 | ||
849eae5e | 2438 | static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) |
2b82032c | 2439 | { |
0b246afa | 2440 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
2b82032c | 2441 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2442 | struct btrfs_fs_devices *seed_devices; |
2b82032c | 2443 | |
a32bf9a3 | 2444 | lockdep_assert_held(&uuid_mutex); |
e4404d6e | 2445 | if (!fs_devices->seeding) |
849eae5e | 2446 | return ERR_PTR(-EINVAL); |
2b82032c | 2447 | |
427c8fdd NB |
2448 | /* |
2449 | * Private copy of the seed devices, anchored at | |
2450 | * fs_info->fs_devices->seed_list | |
2451 | */ | |
7239ff4b | 2452 | seed_devices = alloc_fs_devices(NULL, NULL); |
2208a378 | 2453 | if (IS_ERR(seed_devices)) |
849eae5e | 2454 | return seed_devices; |
2b82032c | 2455 | |
427c8fdd NB |
2456 | /* |
2457 | * It's necessary to retain a copy of the original seed fs_devices in | |
2458 | * fs_uuids so that filesystems which have been seeded can successfully | |
2459 | * reference the seed device from open_seed_devices. This also supports | |
2460 | * multiple fs seed. | |
2461 | */ | |
e4404d6e YZ |
2462 | old_devices = clone_fs_devices(fs_devices); |
2463 | if (IS_ERR(old_devices)) { | |
2464 | kfree(seed_devices); | |
849eae5e | 2465 | return old_devices; |
2b82032c | 2466 | } |
e4404d6e | 2467 | |
c4babc5e | 2468 | list_add(&old_devices->fs_list, &fs_uuids); |
2b82032c | 2469 | |
e4404d6e YZ |
2470 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2471 | seed_devices->opened = 1; | |
2472 | INIT_LIST_HEAD(&seed_devices->devices); | |
2473 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2474 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2475 | |
849eae5e AJ |
2476 | return seed_devices; |
2477 | } | |
2478 | ||
2479 | /* | |
2480 | * Splice seed devices into the sprout fs_devices. | |
2481 | * Generate a new fsid for the sprouted read-write filesystem. | |
2482 | */ | |
2483 | static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, | |
2484 | struct btrfs_fs_devices *seed_devices) | |
2485 | { | |
2486 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2487 | struct btrfs_super_block *disk_super = fs_info->super_copy; | |
2488 | struct btrfs_device *device; | |
2489 | u64 super_flags; | |
2490 | ||
2491 | /* | |
2492 | * We are updating the fsid, the thread leading to device_list_add() | |
2493 | * could race, so uuid_mutex is needed. | |
2494 | */ | |
2495 | lockdep_assert_held(&uuid_mutex); | |
2496 | ||
2497 | /* | |
2498 | * The threads listed below may traverse dev_list but can do that without | |
2499 | * device_list_mutex: | |
2500 | * - All device ops and balance - as we are in btrfs_exclop_start. | |
2501 | * - Various dev_list readers - are using RCU. | |
2502 | * - btrfs_ioctl_fitrim() - is using RCU. | |
2503 | * | |
2504 | * For-read threads as below are using device_list_mutex: | |
2505 | * - Readonly scrub btrfs_scrub_dev() | |
2506 | * - Readonly scrub btrfs_scrub_progress() | |
2507 | * - btrfs_get_dev_stats() | |
2508 | */ | |
2509 | lockdep_assert_held(&fs_devices->device_list_mutex); | |
2510 | ||
1f78160c XG |
2511 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2512 | synchronize_rcu); | |
2196d6e8 MX |
2513 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2514 | device->fs_devices = seed_devices; | |
c9513edb | 2515 | |
0395d84f | 2516 | fs_devices->seeding = false; |
2b82032c YZ |
2517 | fs_devices->num_devices = 0; |
2518 | fs_devices->open_devices = 0; | |
69611ac8 | 2519 | fs_devices->missing_devices = 0; |
7f0432d0 | 2520 | fs_devices->rotating = false; |
944d3f9f | 2521 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
2b82032c YZ |
2522 | |
2523 | generate_random_uuid(fs_devices->fsid); | |
7239ff4b | 2524 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2525 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
f7171750 | 2526 | |
2b82032c YZ |
2527 | super_flags = btrfs_super_flags(disk_super) & |
2528 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2529 | btrfs_set_super_flags(disk_super, super_flags); | |
2b82032c YZ |
2530 | } |
2531 | ||
2532 | /* | |
01327610 | 2533 | * Store the expected generation for seed devices in device items. |
2b82032c | 2534 | */ |
5c466629 | 2535 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
2b82032c | 2536 | { |
562d7b15 | 2537 | BTRFS_DEV_LOOKUP_ARGS(args); |
5c466629 | 2538 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2539 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2540 | struct btrfs_path *path; |
2541 | struct extent_buffer *leaf; | |
2542 | struct btrfs_dev_item *dev_item; | |
2543 | struct btrfs_device *device; | |
2544 | struct btrfs_key key; | |
44880fdc | 2545 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
2b82032c | 2546 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2b82032c YZ |
2547 | int ret; |
2548 | ||
2549 | path = btrfs_alloc_path(); | |
2550 | if (!path) | |
2551 | return -ENOMEM; | |
2552 | ||
2b82032c YZ |
2553 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2554 | key.offset = 0; | |
2555 | key.type = BTRFS_DEV_ITEM_KEY; | |
2556 | ||
2557 | while (1) { | |
2bb2e00e | 2558 | btrfs_reserve_chunk_metadata(trans, false); |
2b82032c | 2559 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
2bb2e00e | 2560 | btrfs_trans_release_chunk_metadata(trans); |
2b82032c YZ |
2561 | if (ret < 0) |
2562 | goto error; | |
2563 | ||
2564 | leaf = path->nodes[0]; | |
2565 | next_slot: | |
2566 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2567 | ret = btrfs_next_leaf(root, path); | |
2568 | if (ret > 0) | |
2569 | break; | |
2570 | if (ret < 0) | |
2571 | goto error; | |
2572 | leaf = path->nodes[0]; | |
2573 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2574 | btrfs_release_path(path); |
2b82032c YZ |
2575 | continue; |
2576 | } | |
2577 | ||
2578 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2579 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2580 | key.type != BTRFS_DEV_ITEM_KEY) | |
2581 | break; | |
2582 | ||
2583 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2584 | struct btrfs_dev_item); | |
562d7b15 | 2585 | args.devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 2586 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2587 | BTRFS_UUID_SIZE); |
1473b24e | 2588 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 2589 | BTRFS_FSID_SIZE); |
562d7b15 JB |
2590 | args.uuid = dev_uuid; |
2591 | args.fsid = fs_uuid; | |
2592 | device = btrfs_find_device(fs_info->fs_devices, &args); | |
79787eaa | 2593 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2594 | |
2595 | if (device->fs_devices->seeding) { | |
2596 | btrfs_set_device_generation(leaf, dev_item, | |
2597 | device->generation); | |
2598 | btrfs_mark_buffer_dirty(leaf); | |
2599 | } | |
2600 | ||
2601 | path->slots[0]++; | |
2602 | goto next_slot; | |
2603 | } | |
2604 | ret = 0; | |
2605 | error: | |
2606 | btrfs_free_path(path); | |
2607 | return ret; | |
2608 | } | |
2609 | ||
da353f6b | 2610 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
788f20eb | 2611 | { |
5112febb | 2612 | struct btrfs_root *root = fs_info->dev_root; |
788f20eb CM |
2613 | struct btrfs_trans_handle *trans; |
2614 | struct btrfs_device *device; | |
2615 | struct block_device *bdev; | |
0b246afa | 2616 | struct super_block *sb = fs_info->sb; |
5da54bc1 | 2617 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8ba7d5f5 | 2618 | struct btrfs_fs_devices *seed_devices = NULL; |
39379faa NA |
2619 | u64 orig_super_total_bytes; |
2620 | u64 orig_super_num_devices; | |
788f20eb | 2621 | int ret = 0; |
fd880809 | 2622 | bool seeding_dev = false; |
44cab9ba | 2623 | bool locked = false; |
788f20eb | 2624 | |
5da54bc1 | 2625 | if (sb_rdonly(sb) && !fs_devices->seeding) |
f8c5d0b4 | 2626 | return -EROFS; |
788f20eb | 2627 | |
05bdb996 | 2628 | bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE, |
0718afd4 | 2629 | fs_info->bdev_holder, NULL); |
7f59203a JB |
2630 | if (IS_ERR(bdev)) |
2631 | return PTR_ERR(bdev); | |
a2135011 | 2632 | |
b70f5097 NA |
2633 | if (!btrfs_check_device_zone_type(fs_info, bdev)) { |
2634 | ret = -EINVAL; | |
2635 | goto error; | |
2636 | } | |
2637 | ||
5da54bc1 | 2638 | if (fs_devices->seeding) { |
fd880809 | 2639 | seeding_dev = true; |
2b82032c YZ |
2640 | down_write(&sb->s_umount); |
2641 | mutex_lock(&uuid_mutex); | |
44cab9ba | 2642 | locked = true; |
2b82032c YZ |
2643 | } |
2644 | ||
b9ba017f | 2645 | sync_blockdev(bdev); |
a2135011 | 2646 | |
f4cfa9bd NB |
2647 | rcu_read_lock(); |
2648 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { | |
788f20eb CM |
2649 | if (device->bdev == bdev) { |
2650 | ret = -EEXIST; | |
f4cfa9bd | 2651 | rcu_read_unlock(); |
2b82032c | 2652 | goto error; |
788f20eb CM |
2653 | } |
2654 | } | |
f4cfa9bd | 2655 | rcu_read_unlock(); |
788f20eb | 2656 | |
bb21e302 | 2657 | device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); |
12bd2fc0 | 2658 | if (IS_ERR(device)) { |
788f20eb | 2659 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2660 | ret = PTR_ERR(device); |
2b82032c | 2661 | goto error; |
788f20eb CM |
2662 | } |
2663 | ||
5b316468 NA |
2664 | device->fs_info = fs_info; |
2665 | device->bdev = bdev; | |
4889bc05 AJ |
2666 | ret = lookup_bdev(device_path, &device->devt); |
2667 | if (ret) | |
2668 | goto error_free_device; | |
5b316468 | 2669 | |
16beac87 | 2670 | ret = btrfs_get_dev_zone_info(device, false); |
5b316468 NA |
2671 | if (ret) |
2672 | goto error_free_device; | |
2673 | ||
a22285a6 | 2674 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2675 | if (IS_ERR(trans)) { |
98d5dc13 | 2676 | ret = PTR_ERR(trans); |
5b316468 | 2677 | goto error_free_zone; |
98d5dc13 TI |
2678 | } |
2679 | ||
ebbede42 | 2680 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
2b82032c | 2681 | device->generation = trans->transid; |
0b246afa JM |
2682 | device->io_width = fs_info->sectorsize; |
2683 | device->io_align = fs_info->sectorsize; | |
2684 | device->sector_size = fs_info->sectorsize; | |
cda00eba CH |
2685 | device->total_bytes = |
2686 | round_down(bdev_nr_bytes(bdev), fs_info->sectorsize); | |
2cc3c559 | 2687 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2688 | device->commit_total_bytes = device->total_bytes; |
e12c9621 | 2689 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
401e29c1 | 2690 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
2736e8ee | 2691 | device->holder = fs_info->bdev_holder; |
27087f37 | 2692 | device->dev_stats_valid = 1; |
9f6d2510 | 2693 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
788f20eb | 2694 | |
2b82032c | 2695 | if (seeding_dev) { |
a0a1db70 | 2696 | btrfs_clear_sb_rdonly(sb); |
849eae5e AJ |
2697 | |
2698 | /* GFP_KERNEL allocation must not be under device_list_mutex */ | |
2699 | seed_devices = btrfs_init_sprout(fs_info); | |
2700 | if (IS_ERR(seed_devices)) { | |
2701 | ret = PTR_ERR(seed_devices); | |
d31c32f6 AJ |
2702 | btrfs_abort_transaction(trans, ret); |
2703 | goto error_trans; | |
2704 | } | |
849eae5e AJ |
2705 | } |
2706 | ||
2707 | mutex_lock(&fs_devices->device_list_mutex); | |
2708 | if (seeding_dev) { | |
2709 | btrfs_setup_sprout(fs_info, seed_devices); | |
b7cb29e6 AJ |
2710 | btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, |
2711 | device); | |
2b82032c | 2712 | } |
788f20eb | 2713 | |
5da54bc1 | 2714 | device->fs_devices = fs_devices; |
e5e9a520 | 2715 | |
34441361 | 2716 | mutex_lock(&fs_info->chunk_mutex); |
5da54bc1 AJ |
2717 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
2718 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); | |
2719 | fs_devices->num_devices++; | |
2720 | fs_devices->open_devices++; | |
2721 | fs_devices->rw_devices++; | |
2722 | fs_devices->total_devices++; | |
2723 | fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2724 | |
a5ed45f8 | 2725 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
2bf64758 | 2726 | |
10f0d2a5 | 2727 | if (!bdev_nonrot(bdev)) |
7f0432d0 | 2728 | fs_devices->rotating = true; |
c289811c | 2729 | |
39379faa | 2730 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
0b246afa | 2731 | btrfs_set_super_total_bytes(fs_info->super_copy, |
39379faa NA |
2732 | round_down(orig_super_total_bytes + device->total_bytes, |
2733 | fs_info->sectorsize)); | |
788f20eb | 2734 | |
39379faa NA |
2735 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
2736 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2737 | orig_super_num_devices + 1); | |
0d39376a | 2738 | |
2196d6e8 MX |
2739 | /* |
2740 | * we've got more storage, clear any full flags on the space | |
2741 | * infos | |
2742 | */ | |
0b246afa | 2743 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2744 | |
34441361 | 2745 | mutex_unlock(&fs_info->chunk_mutex); |
ca10845a JB |
2746 | |
2747 | /* Add sysfs device entry */ | |
cd36da2e | 2748 | btrfs_sysfs_add_device(device); |
ca10845a | 2749 | |
5da54bc1 | 2750 | mutex_unlock(&fs_devices->device_list_mutex); |
788f20eb | 2751 | |
2b82032c | 2752 | if (seeding_dev) { |
34441361 | 2753 | mutex_lock(&fs_info->chunk_mutex); |
6f8e0fc7 | 2754 | ret = init_first_rw_device(trans); |
34441361 | 2755 | mutex_unlock(&fs_info->chunk_mutex); |
005d6427 | 2756 | if (ret) { |
66642832 | 2757 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2758 | goto error_sysfs; |
005d6427 | 2759 | } |
2196d6e8 MX |
2760 | } |
2761 | ||
8e87e856 | 2762 | ret = btrfs_add_dev_item(trans, device); |
2196d6e8 | 2763 | if (ret) { |
66642832 | 2764 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2765 | goto error_sysfs; |
2196d6e8 MX |
2766 | } |
2767 | ||
2768 | if (seeding_dev) { | |
5c466629 | 2769 | ret = btrfs_finish_sprout(trans); |
005d6427 | 2770 | if (ret) { |
66642832 | 2771 | btrfs_abort_transaction(trans, ret); |
d31c32f6 | 2772 | goto error_sysfs; |
005d6427 | 2773 | } |
b2373f25 | 2774 | |
8e560081 NB |
2775 | /* |
2776 | * fs_devices now represents the newly sprouted filesystem and | |
849eae5e | 2777 | * its fsid has been changed by btrfs_sprout_splice(). |
8e560081 NB |
2778 | */ |
2779 | btrfs_sysfs_update_sprout_fsid(fs_devices); | |
2b82032c YZ |
2780 | } |
2781 | ||
3a45bb20 | 2782 | ret = btrfs_commit_transaction(trans); |
a2135011 | 2783 | |
2b82032c YZ |
2784 | if (seeding_dev) { |
2785 | mutex_unlock(&uuid_mutex); | |
2786 | up_write(&sb->s_umount); | |
44cab9ba | 2787 | locked = false; |
788f20eb | 2788 | |
79787eaa JM |
2789 | if (ret) /* transaction commit */ |
2790 | return ret; | |
2791 | ||
2ff7e61e | 2792 | ret = btrfs_relocate_sys_chunks(fs_info); |
79787eaa | 2793 | if (ret < 0) |
0b246afa | 2794 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2795 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2796 | trans = btrfs_attach_transaction(root); |
2797 | if (IS_ERR(trans)) { | |
2798 | if (PTR_ERR(trans) == -ENOENT) | |
2799 | return 0; | |
7132a262 AJ |
2800 | ret = PTR_ERR(trans); |
2801 | trans = NULL; | |
2802 | goto error_sysfs; | |
671415b7 | 2803 | } |
3a45bb20 | 2804 | ret = btrfs_commit_transaction(trans); |
2b82032c | 2805 | } |
c9e9f97b | 2806 | |
7f551d96 AJ |
2807 | /* |
2808 | * Now that we have written a new super block to this device, check all | |
2809 | * other fs_devices list if device_path alienates any other scanned | |
2810 | * device. | |
2811 | * We can ignore the return value as it typically returns -EINVAL and | |
2812 | * only succeeds if the device was an alien. | |
2813 | */ | |
4889bc05 | 2814 | btrfs_forget_devices(device->devt); |
7f551d96 AJ |
2815 | |
2816 | /* Update ctime/mtime for blkid or udev */ | |
54fde91f | 2817 | update_dev_time(device_path); |
7f551d96 | 2818 | |
2b82032c | 2819 | return ret; |
79787eaa | 2820 | |
d31c32f6 | 2821 | error_sysfs: |
53f8a74c | 2822 | btrfs_sysfs_remove_device(device); |
39379faa NA |
2823 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2824 | mutex_lock(&fs_info->chunk_mutex); | |
2825 | list_del_rcu(&device->dev_list); | |
2826 | list_del(&device->dev_alloc_list); | |
2827 | fs_info->fs_devices->num_devices--; | |
2828 | fs_info->fs_devices->open_devices--; | |
2829 | fs_info->fs_devices->rw_devices--; | |
2830 | fs_info->fs_devices->total_devices--; | |
2831 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; | |
2832 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); | |
2833 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
2834 | orig_super_total_bytes); | |
2835 | btrfs_set_super_num_devices(fs_info->super_copy, | |
2836 | orig_super_num_devices); | |
2837 | mutex_unlock(&fs_info->chunk_mutex); | |
2838 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
79787eaa | 2839 | error_trans: |
0af2c4bf | 2840 | if (seeding_dev) |
a0a1db70 | 2841 | btrfs_set_sb_rdonly(sb); |
7132a262 AJ |
2842 | if (trans) |
2843 | btrfs_end_transaction(trans); | |
5b316468 NA |
2844 | error_free_zone: |
2845 | btrfs_destroy_dev_zone_info(device); | |
5c4cf6c9 | 2846 | error_free_device: |
a425f9d4 | 2847 | btrfs_free_device(device); |
2b82032c | 2848 | error: |
2736e8ee | 2849 | blkdev_put(bdev, fs_info->bdev_holder); |
44cab9ba | 2850 | if (locked) { |
2b82032c YZ |
2851 | mutex_unlock(&uuid_mutex); |
2852 | up_write(&sb->s_umount); | |
2853 | } | |
c9e9f97b | 2854 | return ret; |
788f20eb CM |
2855 | } |
2856 | ||
d397712b CM |
2857 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2858 | struct btrfs_device *device) | |
0b86a832 CM |
2859 | { |
2860 | int ret; | |
2861 | struct btrfs_path *path; | |
0b246afa | 2862 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2863 | struct btrfs_dev_item *dev_item; |
2864 | struct extent_buffer *leaf; | |
2865 | struct btrfs_key key; | |
2866 | ||
0b86a832 CM |
2867 | path = btrfs_alloc_path(); |
2868 | if (!path) | |
2869 | return -ENOMEM; | |
2870 | ||
2871 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2872 | key.type = BTRFS_DEV_ITEM_KEY; | |
2873 | key.offset = device->devid; | |
2874 | ||
2875 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2876 | if (ret < 0) | |
2877 | goto out; | |
2878 | ||
2879 | if (ret > 0) { | |
2880 | ret = -ENOENT; | |
2881 | goto out; | |
2882 | } | |
2883 | ||
2884 | leaf = path->nodes[0]; | |
2885 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2886 | ||
2887 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2888 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2889 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2890 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2891 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2892 | btrfs_set_device_total_bytes(leaf, dev_item, |
2893 | btrfs_device_get_disk_total_bytes(device)); | |
2894 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2895 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2896 | btrfs_mark_buffer_dirty(leaf); |
2897 | ||
2898 | out: | |
2899 | btrfs_free_path(path); | |
2900 | return ret; | |
2901 | } | |
2902 | ||
2196d6e8 | 2903 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2904 | struct btrfs_device *device, u64 new_size) |
2905 | { | |
0b246afa JM |
2906 | struct btrfs_fs_info *fs_info = device->fs_info; |
2907 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
2196d6e8 MX |
2908 | u64 old_total; |
2909 | u64 diff; | |
2bb2e00e | 2910 | int ret; |
8f18cf13 | 2911 | |
ebbede42 | 2912 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
2b82032c | 2913 | return -EACCES; |
2196d6e8 | 2914 | |
7dfb8be1 NB |
2915 | new_size = round_down(new_size, fs_info->sectorsize); |
2916 | ||
34441361 | 2917 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 | 2918 | old_total = btrfs_super_total_bytes(super_copy); |
0e4324a4 | 2919 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
2196d6e8 | 2920 | |
63a212ab | 2921 | if (new_size <= device->total_bytes || |
401e29c1 | 2922 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
34441361 | 2923 | mutex_unlock(&fs_info->chunk_mutex); |
2b82032c | 2924 | return -EINVAL; |
2196d6e8 | 2925 | } |
2b82032c | 2926 | |
7dfb8be1 NB |
2927 | btrfs_set_super_total_bytes(super_copy, |
2928 | round_down(old_total + diff, fs_info->sectorsize)); | |
2b82032c YZ |
2929 | device->fs_devices->total_rw_bytes += diff; |
2930 | ||
7cc8e58d MX |
2931 | btrfs_device_set_total_bytes(device, new_size); |
2932 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2933 | btrfs_clear_space_info_full(device->fs_info); |
bbbf7243 NB |
2934 | if (list_empty(&device->post_commit_list)) |
2935 | list_add_tail(&device->post_commit_list, | |
2936 | &trans->transaction->dev_update_list); | |
34441361 | 2937 | mutex_unlock(&fs_info->chunk_mutex); |
4184ea7f | 2938 | |
2bb2e00e FM |
2939 | btrfs_reserve_chunk_metadata(trans, false); |
2940 | ret = btrfs_update_device(trans, device); | |
2941 | btrfs_trans_release_chunk_metadata(trans); | |
2942 | ||
2943 | return ret; | |
8f18cf13 CM |
2944 | } |
2945 | ||
f4208794 | 2946 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 2947 | { |
f4208794 | 2948 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5b4aacef | 2949 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2950 | int ret; |
2951 | struct btrfs_path *path; | |
2952 | struct btrfs_key key; | |
2953 | ||
8f18cf13 CM |
2954 | path = btrfs_alloc_path(); |
2955 | if (!path) | |
2956 | return -ENOMEM; | |
2957 | ||
408fbf19 | 2958 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
8f18cf13 CM |
2959 | key.offset = chunk_offset; |
2960 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2961 | ||
2962 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2963 | if (ret < 0) |
2964 | goto out; | |
2965 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2966 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2967 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2968 | ret = -ENOENT; |
2969 | goto out; | |
2970 | } | |
8f18cf13 CM |
2971 | |
2972 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2973 | if (ret < 0) |
0b246afa JM |
2974 | btrfs_handle_fs_error(fs_info, ret, |
2975 | "Failed to delete chunk item."); | |
79787eaa | 2976 | out: |
8f18cf13 | 2977 | btrfs_free_path(path); |
65a246c5 | 2978 | return ret; |
8f18cf13 CM |
2979 | } |
2980 | ||
408fbf19 | 2981 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 | 2982 | { |
0b246afa | 2983 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2984 | struct btrfs_disk_key *disk_key; |
2985 | struct btrfs_chunk *chunk; | |
2986 | u8 *ptr; | |
2987 | int ret = 0; | |
2988 | u32 num_stripes; | |
2989 | u32 array_size; | |
2990 | u32 len = 0; | |
2991 | u32 cur; | |
2992 | struct btrfs_key key; | |
2993 | ||
79bd3712 | 2994 | lockdep_assert_held(&fs_info->chunk_mutex); |
8f18cf13 CM |
2995 | array_size = btrfs_super_sys_array_size(super_copy); |
2996 | ||
2997 | ptr = super_copy->sys_chunk_array; | |
2998 | cur = 0; | |
2999 | ||
3000 | while (cur < array_size) { | |
3001 | disk_key = (struct btrfs_disk_key *)ptr; | |
3002 | btrfs_disk_key_to_cpu(&key, disk_key); | |
3003 | ||
3004 | len = sizeof(*disk_key); | |
3005 | ||
3006 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
3007 | chunk = (struct btrfs_chunk *)(ptr + len); | |
3008 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
3009 | len += btrfs_chunk_item_size(num_stripes); | |
3010 | } else { | |
3011 | ret = -EIO; | |
3012 | break; | |
3013 | } | |
408fbf19 | 3014 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
8f18cf13 CM |
3015 | key.offset == chunk_offset) { |
3016 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
3017 | array_size -= len; | |
3018 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
3019 | } else { | |
3020 | ptr += len; | |
3021 | cur += len; | |
3022 | } | |
3023 | } | |
3024 | return ret; | |
3025 | } | |
3026 | ||
60ca842e OS |
3027 | /* |
3028 | * btrfs_get_chunk_map() - Find the mapping containing the given logical extent. | |
3029 | * @logical: Logical block offset in bytes. | |
3030 | * @length: Length of extent in bytes. | |
3031 | * | |
3032 | * Return: Chunk mapping or ERR_PTR. | |
3033 | */ | |
3034 | struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, | |
3035 | u64 logical, u64 length) | |
592d92ee LB |
3036 | { |
3037 | struct extent_map_tree *em_tree; | |
3038 | struct extent_map *em; | |
3039 | ||
c8bf1b67 | 3040 | em_tree = &fs_info->mapping_tree; |
592d92ee LB |
3041 | read_lock(&em_tree->lock); |
3042 | em = lookup_extent_mapping(em_tree, logical, length); | |
3043 | read_unlock(&em_tree->lock); | |
3044 | ||
3045 | if (!em) { | |
3046 | btrfs_crit(fs_info, "unable to find logical %llu length %llu", | |
3047 | logical, length); | |
3048 | return ERR_PTR(-EINVAL); | |
3049 | } | |
3050 | ||
3051 | if (em->start > logical || em->start + em->len < logical) { | |
3052 | btrfs_crit(fs_info, | |
3053 | "found a bad mapping, wanted %llu-%llu, found %llu-%llu", | |
3054 | logical, length, em->start, em->start + em->len); | |
3055 | free_extent_map(em); | |
3056 | return ERR_PTR(-EINVAL); | |
3057 | } | |
3058 | ||
3059 | /* callers are responsible for dropping em's ref. */ | |
3060 | return em; | |
3061 | } | |
3062 | ||
79bd3712 FM |
3063 | static int remove_chunk_item(struct btrfs_trans_handle *trans, |
3064 | struct map_lookup *map, u64 chunk_offset) | |
3065 | { | |
3066 | int i; | |
3067 | ||
3068 | /* | |
3069 | * Removing chunk items and updating the device items in the chunks btree | |
3070 | * requires holding the chunk_mutex. | |
3071 | * See the comment at btrfs_chunk_alloc() for the details. | |
3072 | */ | |
3073 | lockdep_assert_held(&trans->fs_info->chunk_mutex); | |
3074 | ||
3075 | for (i = 0; i < map->num_stripes; i++) { | |
3076 | int ret; | |
3077 | ||
3078 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
3079 | if (ret) | |
3080 | return ret; | |
3081 | } | |
3082 | ||
3083 | return btrfs_free_chunk(trans, chunk_offset); | |
3084 | } | |
3085 | ||
97aff912 | 3086 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
8f18cf13 | 3087 | { |
97aff912 | 3088 | struct btrfs_fs_info *fs_info = trans->fs_info; |
8f18cf13 CM |
3089 | struct extent_map *em; |
3090 | struct map_lookup *map; | |
2196d6e8 | 3091 | u64 dev_extent_len = 0; |
47ab2a6c | 3092 | int i, ret = 0; |
0b246afa | 3093 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 3094 | |
60ca842e | 3095 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 3096 | if (IS_ERR(em)) { |
47ab2a6c JB |
3097 | /* |
3098 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 3099 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
3100 | * do anything we still error out. |
3101 | */ | |
3102 | ASSERT(0); | |
592d92ee | 3103 | return PTR_ERR(em); |
47ab2a6c | 3104 | } |
95617d69 | 3105 | map = em->map_lookup; |
8f18cf13 | 3106 | |
57ba4cb8 | 3107 | /* |
79bd3712 FM |
3108 | * First delete the device extent items from the devices btree. |
3109 | * We take the device_list_mutex to avoid racing with the finishing phase | |
3110 | * of a device replace operation. See the comment below before acquiring | |
3111 | * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex | |
3112 | * because that can result in a deadlock when deleting the device extent | |
3113 | * items from the devices btree - COWing an extent buffer from the btree | |
3114 | * may result in allocating a new metadata chunk, which would attempt to | |
3115 | * lock again fs_info->chunk_mutex. | |
57ba4cb8 FM |
3116 | */ |
3117 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 3118 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 3119 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
3120 | ret = btrfs_free_dev_extent(trans, device, |
3121 | map->stripes[i].physical, | |
3122 | &dev_extent_len); | |
47ab2a6c | 3123 | if (ret) { |
57ba4cb8 | 3124 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 3125 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3126 | goto out; |
3127 | } | |
a061fc8d | 3128 | |
2196d6e8 | 3129 | if (device->bytes_used > 0) { |
34441361 | 3130 | mutex_lock(&fs_info->chunk_mutex); |
2196d6e8 MX |
3131 | btrfs_device_set_bytes_used(device, |
3132 | device->bytes_used - dev_extent_len); | |
a5ed45f8 | 3133 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
0b246afa | 3134 | btrfs_clear_space_info_full(fs_info); |
34441361 | 3135 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 3136 | } |
79bd3712 FM |
3137 | } |
3138 | mutex_unlock(&fs_devices->device_list_mutex); | |
a061fc8d | 3139 | |
79bd3712 FM |
3140 | /* |
3141 | * We acquire fs_info->chunk_mutex for 2 reasons: | |
3142 | * | |
3143 | * 1) Just like with the first phase of the chunk allocation, we must | |
3144 | * reserve system space, do all chunk btree updates and deletions, and | |
3145 | * update the system chunk array in the superblock while holding this | |
3146 | * mutex. This is for similar reasons as explained on the comment at | |
3147 | * the top of btrfs_chunk_alloc(); | |
3148 | * | |
3149 | * 2) Prevent races with the final phase of a device replace operation | |
3150 | * that replaces the device object associated with the map's stripes, | |
3151 | * because the device object's id can change at any time during that | |
3152 | * final phase of the device replace operation | |
3153 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
3154 | * replaced device and then see it with an ID of | |
3155 | * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating | |
3156 | * the device item, which does not exists on the chunk btree. | |
3157 | * The finishing phase of device replace acquires both the | |
3158 | * device_list_mutex and the chunk_mutex, in that order, so we are | |
3159 | * safe by just acquiring the chunk_mutex. | |
3160 | */ | |
3161 | trans->removing_chunk = true; | |
3162 | mutex_lock(&fs_info->chunk_mutex); | |
3163 | ||
3164 | check_system_chunk(trans, map->type); | |
3165 | ||
3166 | ret = remove_chunk_item(trans, map, chunk_offset); | |
3167 | /* | |
3168 | * Normally we should not get -ENOSPC since we reserved space before | |
3169 | * through the call to check_system_chunk(). | |
3170 | * | |
3171 | * Despite our system space_info having enough free space, we may not | |
3172 | * be able to allocate extents from its block groups, because all have | |
3173 | * an incompatible profile, which will force us to allocate a new system | |
3174 | * block group with the right profile, or right after we called | |
3175 | * check_system_space() above, a scrub turned the only system block group | |
3176 | * with enough free space into RO mode. | |
3177 | * This is explained with more detail at do_chunk_alloc(). | |
3178 | * | |
3179 | * So if we get -ENOSPC, allocate a new system chunk and retry once. | |
3180 | */ | |
3181 | if (ret == -ENOSPC) { | |
3182 | const u64 sys_flags = btrfs_system_alloc_profile(fs_info); | |
3183 | struct btrfs_block_group *sys_bg; | |
3184 | ||
f6f39f7a | 3185 | sys_bg = btrfs_create_chunk(trans, sys_flags); |
79bd3712 FM |
3186 | if (IS_ERR(sys_bg)) { |
3187 | ret = PTR_ERR(sys_bg); | |
3188 | btrfs_abort_transaction(trans, ret); | |
3189 | goto out; | |
3190 | } | |
3191 | ||
3192 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); | |
64bc6c2a | 3193 | if (ret) { |
64bc6c2a NB |
3194 | btrfs_abort_transaction(trans, ret); |
3195 | goto out; | |
dfe25020 | 3196 | } |
57ba4cb8 | 3197 | |
79bd3712 FM |
3198 | ret = remove_chunk_item(trans, map, chunk_offset); |
3199 | if (ret) { | |
3200 | btrfs_abort_transaction(trans, ret); | |
3201 | goto out; | |
3202 | } | |
3203 | } else if (ret) { | |
66642832 | 3204 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3205 | goto out; |
3206 | } | |
8f18cf13 | 3207 | |
6bccf3ab | 3208 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 3209 | |
8f18cf13 | 3210 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
408fbf19 | 3211 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
47ab2a6c | 3212 | if (ret) { |
66642832 | 3213 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3214 | goto out; |
3215 | } | |
8f18cf13 CM |
3216 | } |
3217 | ||
79bd3712 FM |
3218 | mutex_unlock(&fs_info->chunk_mutex); |
3219 | trans->removing_chunk = false; | |
3220 | ||
3221 | /* | |
3222 | * We are done with chunk btree updates and deletions, so release the | |
3223 | * system space we previously reserved (with check_system_chunk()). | |
3224 | */ | |
3225 | btrfs_trans_release_chunk_metadata(trans); | |
3226 | ||
5a98ec01 | 3227 | ret = btrfs_remove_block_group(trans, chunk_offset, em); |
47ab2a6c | 3228 | if (ret) { |
66642832 | 3229 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
3230 | goto out; |
3231 | } | |
2b82032c | 3232 | |
47ab2a6c | 3233 | out: |
79bd3712 FM |
3234 | if (trans->removing_chunk) { |
3235 | mutex_unlock(&fs_info->chunk_mutex); | |
3236 | trans->removing_chunk = false; | |
3237 | } | |
2b82032c YZ |
3238 | /* once for us */ |
3239 | free_extent_map(em); | |
47ab2a6c JB |
3240 | return ret; |
3241 | } | |
2b82032c | 3242 | |
18bb8bbf | 3243 | int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 3244 | { |
5b4aacef | 3245 | struct btrfs_root *root = fs_info->chunk_root; |
19c4d2f9 | 3246 | struct btrfs_trans_handle *trans; |
b0643e59 | 3247 | struct btrfs_block_group *block_group; |
01e86008 | 3248 | u64 length; |
47ab2a6c | 3249 | int ret; |
2b82032c | 3250 | |
4b349253 JB |
3251 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
3252 | btrfs_err(fs_info, | |
3253 | "relocate: not supported on extent tree v2 yet"); | |
3254 | return -EINVAL; | |
3255 | } | |
3256 | ||
67c5e7d4 FM |
3257 | /* |
3258 | * Prevent races with automatic removal of unused block groups. | |
3259 | * After we relocate and before we remove the chunk with offset | |
3260 | * chunk_offset, automatic removal of the block group can kick in, | |
3261 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
3262 | * | |
3263 | * Make sure to acquire this mutex before doing a tree search (dev | |
3264 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
3265 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
3266 | * we release the path used to search the chunk/dev tree and before | |
3267 | * the current task acquires this mutex and calls us. | |
3268 | */ | |
f3372065 | 3269 | lockdep_assert_held(&fs_info->reclaim_bgs_lock); |
67c5e7d4 | 3270 | |
47ab2a6c | 3271 | /* step one, relocate all the extents inside this chunk */ |
2ff7e61e | 3272 | btrfs_scrub_pause(fs_info); |
0b246afa | 3273 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
2ff7e61e | 3274 | btrfs_scrub_continue(fs_info); |
2d82a40a FM |
3275 | if (ret) { |
3276 | /* | |
3277 | * If we had a transaction abort, stop all running scrubs. | |
3278 | * See transaction.c:cleanup_transaction() why we do it here. | |
3279 | */ | |
3280 | if (BTRFS_FS_ERROR(fs_info)) | |
3281 | btrfs_scrub_cancel(fs_info); | |
47ab2a6c | 3282 | return ret; |
2d82a40a | 3283 | } |
47ab2a6c | 3284 | |
b0643e59 DZ |
3285 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
3286 | if (!block_group) | |
3287 | return -ENOENT; | |
3288 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); | |
01e86008 | 3289 | length = block_group->length; |
b0643e59 DZ |
3290 | btrfs_put_block_group(block_group); |
3291 | ||
01e86008 JT |
3292 | /* |
3293 | * On a zoned file system, discard the whole block group, this will | |
3294 | * trigger a REQ_OP_ZONE_RESET operation on the device zone. If | |
3295 | * resetting the zone fails, don't treat it as a fatal problem from the | |
3296 | * filesystem's point of view. | |
3297 | */ | |
3298 | if (btrfs_is_zoned(fs_info)) { | |
3299 | ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); | |
3300 | if (ret) | |
3301 | btrfs_info(fs_info, | |
3302 | "failed to reset zone %llu after relocation", | |
3303 | chunk_offset); | |
3304 | } | |
3305 | ||
19c4d2f9 CM |
3306 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
3307 | chunk_offset); | |
3308 | if (IS_ERR(trans)) { | |
3309 | ret = PTR_ERR(trans); | |
3310 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
3311 | return ret; | |
3312 | } | |
3313 | ||
47ab2a6c | 3314 | /* |
19c4d2f9 CM |
3315 | * step two, delete the device extents and the |
3316 | * chunk tree entries | |
47ab2a6c | 3317 | */ |
97aff912 | 3318 | ret = btrfs_remove_chunk(trans, chunk_offset); |
3a45bb20 | 3319 | btrfs_end_transaction(trans); |
19c4d2f9 | 3320 | return ret; |
2b82032c YZ |
3321 | } |
3322 | ||
2ff7e61e | 3323 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
2b82032c | 3324 | { |
0b246afa | 3325 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2b82032c YZ |
3326 | struct btrfs_path *path; |
3327 | struct extent_buffer *leaf; | |
3328 | struct btrfs_chunk *chunk; | |
3329 | struct btrfs_key key; | |
3330 | struct btrfs_key found_key; | |
2b82032c | 3331 | u64 chunk_type; |
ba1bf481 JB |
3332 | bool retried = false; |
3333 | int failed = 0; | |
2b82032c YZ |
3334 | int ret; |
3335 | ||
3336 | path = btrfs_alloc_path(); | |
3337 | if (!path) | |
3338 | return -ENOMEM; | |
3339 | ||
ba1bf481 | 3340 | again: |
2b82032c YZ |
3341 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3342 | key.offset = (u64)-1; | |
3343 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3344 | ||
3345 | while (1) { | |
f3372065 | 3346 | mutex_lock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3347 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3348 | if (ret < 0) { |
f3372065 | 3349 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c | 3350 | goto error; |
67c5e7d4 | 3351 | } |
79787eaa | 3352 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
3353 | |
3354 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
3355 | key.type); | |
67c5e7d4 | 3356 | if (ret) |
f3372065 | 3357 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2b82032c YZ |
3358 | if (ret < 0) |
3359 | goto error; | |
3360 | if (ret > 0) | |
3361 | break; | |
1a40e23b | 3362 | |
2b82032c YZ |
3363 | leaf = path->nodes[0]; |
3364 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 3365 | |
2b82032c YZ |
3366 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3367 | struct btrfs_chunk); | |
3368 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3369 | btrfs_release_path(path); |
8f18cf13 | 3370 | |
2b82032c | 3371 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3372 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3373 | if (ret == -ENOSPC) |
3374 | failed++; | |
14586651 HS |
3375 | else |
3376 | BUG_ON(ret); | |
2b82032c | 3377 | } |
f3372065 | 3378 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 3379 | |
2b82032c YZ |
3380 | if (found_key.offset == 0) |
3381 | break; | |
3382 | key.offset = found_key.offset - 1; | |
3383 | } | |
3384 | ret = 0; | |
ba1bf481 JB |
3385 | if (failed && !retried) { |
3386 | failed = 0; | |
3387 | retried = true; | |
3388 | goto again; | |
fae7f21c | 3389 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3390 | ret = -ENOSPC; |
3391 | } | |
2b82032c YZ |
3392 | error: |
3393 | btrfs_free_path(path); | |
3394 | return ret; | |
8f18cf13 CM |
3395 | } |
3396 | ||
a6f93c71 LB |
3397 | /* |
3398 | * return 1 : allocate a data chunk successfully, | |
3399 | * return <0: errors during allocating a data chunk, | |
3400 | * return 0 : no need to allocate a data chunk. | |
3401 | */ | |
3402 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, | |
3403 | u64 chunk_offset) | |
3404 | { | |
32da5386 | 3405 | struct btrfs_block_group *cache; |
a6f93c71 LB |
3406 | u64 bytes_used; |
3407 | u64 chunk_type; | |
3408 | ||
3409 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3410 | ASSERT(cache); | |
3411 | chunk_type = cache->flags; | |
3412 | btrfs_put_block_group(cache); | |
3413 | ||
5ae21692 JT |
3414 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
3415 | return 0; | |
3416 | ||
3417 | spin_lock(&fs_info->data_sinfo->lock); | |
3418 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3419 | spin_unlock(&fs_info->data_sinfo->lock); | |
3420 | ||
3421 | if (!bytes_used) { | |
3422 | struct btrfs_trans_handle *trans; | |
3423 | int ret; | |
3424 | ||
3425 | trans = btrfs_join_transaction(fs_info->tree_root); | |
3426 | if (IS_ERR(trans)) | |
3427 | return PTR_ERR(trans); | |
3428 | ||
3429 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); | |
3430 | btrfs_end_transaction(trans); | |
3431 | if (ret < 0) | |
3432 | return ret; | |
3433 | return 1; | |
a6f93c71 | 3434 | } |
5ae21692 | 3435 | |
a6f93c71 LB |
3436 | return 0; |
3437 | } | |
3438 | ||
6bccf3ab | 3439 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3440 | struct btrfs_balance_control *bctl) |
3441 | { | |
6bccf3ab | 3442 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3443 | struct btrfs_trans_handle *trans; |
3444 | struct btrfs_balance_item *item; | |
3445 | struct btrfs_disk_balance_args disk_bargs; | |
3446 | struct btrfs_path *path; | |
3447 | struct extent_buffer *leaf; | |
3448 | struct btrfs_key key; | |
3449 | int ret, err; | |
3450 | ||
3451 | path = btrfs_alloc_path(); | |
3452 | if (!path) | |
3453 | return -ENOMEM; | |
3454 | ||
3455 | trans = btrfs_start_transaction(root, 0); | |
3456 | if (IS_ERR(trans)) { | |
3457 | btrfs_free_path(path); | |
3458 | return PTR_ERR(trans); | |
3459 | } | |
3460 | ||
3461 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3462 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3463 | key.offset = 0; |
3464 | ||
3465 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3466 | sizeof(*item)); | |
3467 | if (ret) | |
3468 | goto out; | |
3469 | ||
3470 | leaf = path->nodes[0]; | |
3471 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3472 | ||
b159fa28 | 3473 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3474 | |
3475 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3476 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3477 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3478 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3479 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3480 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3481 | ||
3482 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3483 | ||
3484 | btrfs_mark_buffer_dirty(leaf); | |
3485 | out: | |
3486 | btrfs_free_path(path); | |
3a45bb20 | 3487 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3488 | if (err && !ret) |
3489 | ret = err; | |
3490 | return ret; | |
3491 | } | |
3492 | ||
6bccf3ab | 3493 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3494 | { |
6bccf3ab | 3495 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3496 | struct btrfs_trans_handle *trans; |
3497 | struct btrfs_path *path; | |
3498 | struct btrfs_key key; | |
3499 | int ret, err; | |
3500 | ||
3501 | path = btrfs_alloc_path(); | |
3502 | if (!path) | |
3503 | return -ENOMEM; | |
3504 | ||
3502a8c0 | 3505 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
0940ebf6 ID |
3506 | if (IS_ERR(trans)) { |
3507 | btrfs_free_path(path); | |
3508 | return PTR_ERR(trans); | |
3509 | } | |
3510 | ||
3511 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3512 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3513 | key.offset = 0; |
3514 | ||
3515 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3516 | if (ret < 0) | |
3517 | goto out; | |
3518 | if (ret > 0) { | |
3519 | ret = -ENOENT; | |
3520 | goto out; | |
3521 | } | |
3522 | ||
3523 | ret = btrfs_del_item(trans, root, path); | |
3524 | out: | |
3525 | btrfs_free_path(path); | |
3a45bb20 | 3526 | err = btrfs_commit_transaction(trans); |
0940ebf6 ID |
3527 | if (err && !ret) |
3528 | ret = err; | |
3529 | return ret; | |
3530 | } | |
3531 | ||
59641015 ID |
3532 | /* |
3533 | * This is a heuristic used to reduce the number of chunks balanced on | |
3534 | * resume after balance was interrupted. | |
3535 | */ | |
3536 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3537 | { | |
3538 | /* | |
3539 | * Turn on soft mode for chunk types that were being converted. | |
3540 | */ | |
3541 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3542 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3543 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3544 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3545 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3546 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3547 | ||
3548 | /* | |
3549 | * Turn on usage filter if is not already used. The idea is | |
3550 | * that chunks that we have already balanced should be | |
3551 | * reasonably full. Don't do it for chunks that are being | |
3552 | * converted - that will keep us from relocating unconverted | |
3553 | * (albeit full) chunks. | |
3554 | */ | |
3555 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3556 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3557 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3558 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3559 | bctl->data.usage = 90; | |
3560 | } | |
3561 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3562 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3563 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3564 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3565 | bctl->sys.usage = 90; | |
3566 | } | |
3567 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3568 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3569 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3570 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3571 | bctl->meta.usage = 90; | |
3572 | } | |
3573 | } | |
3574 | ||
149196a2 DS |
3575 | /* |
3576 | * Clear the balance status in fs_info and delete the balance item from disk. | |
3577 | */ | |
3578 | static void reset_balance_state(struct btrfs_fs_info *fs_info) | |
c9e9f97b ID |
3579 | { |
3580 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
149196a2 | 3581 | int ret; |
c9e9f97b ID |
3582 | |
3583 | BUG_ON(!fs_info->balance_ctl); | |
3584 | ||
3585 | spin_lock(&fs_info->balance_lock); | |
3586 | fs_info->balance_ctl = NULL; | |
3587 | spin_unlock(&fs_info->balance_lock); | |
3588 | ||
3589 | kfree(bctl); | |
149196a2 DS |
3590 | ret = del_balance_item(fs_info); |
3591 | if (ret) | |
3592 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
c9e9f97b ID |
3593 | } |
3594 | ||
ed25e9b2 ID |
3595 | /* |
3596 | * Balance filters. Return 1 if chunk should be filtered out | |
3597 | * (should not be balanced). | |
3598 | */ | |
899c81ea | 3599 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3600 | struct btrfs_balance_args *bargs) |
3601 | { | |
899c81ea ID |
3602 | chunk_type = chunk_to_extended(chunk_type) & |
3603 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3604 | |
899c81ea | 3605 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3606 | return 0; |
3607 | ||
3608 | return 1; | |
3609 | } | |
3610 | ||
dba72cb3 | 3611 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3612 | struct btrfs_balance_args *bargs) |
bc309467 | 3613 | { |
32da5386 | 3614 | struct btrfs_block_group *cache; |
bc309467 DS |
3615 | u64 chunk_used; |
3616 | u64 user_thresh_min; | |
3617 | u64 user_thresh_max; | |
3618 | int ret = 1; | |
3619 | ||
3620 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3621 | chunk_used = cache->used; |
bc309467 DS |
3622 | |
3623 | if (bargs->usage_min == 0) | |
3624 | user_thresh_min = 0; | |
3625 | else | |
428c8e03 | 3626 | user_thresh_min = mult_perc(cache->length, bargs->usage_min); |
bc309467 DS |
3627 | |
3628 | if (bargs->usage_max == 0) | |
3629 | user_thresh_max = 1; | |
3630 | else if (bargs->usage_max > 100) | |
b3470b5d | 3631 | user_thresh_max = cache->length; |
bc309467 | 3632 | else |
428c8e03 | 3633 | user_thresh_max = mult_perc(cache->length, bargs->usage_max); |
bc309467 DS |
3634 | |
3635 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3636 | ret = 0; | |
3637 | ||
3638 | btrfs_put_block_group(cache); | |
3639 | return ret; | |
3640 | } | |
3641 | ||
dba72cb3 | 3642 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3643 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 | 3644 | { |
32da5386 | 3645 | struct btrfs_block_group *cache; |
5ce5b3c0 ID |
3646 | u64 chunk_used, user_thresh; |
3647 | int ret = 1; | |
3648 | ||
3649 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
bf38be65 | 3650 | chunk_used = cache->used; |
5ce5b3c0 | 3651 | |
bc309467 | 3652 | if (bargs->usage_min == 0) |
3e39cea6 | 3653 | user_thresh = 1; |
a105bb88 | 3654 | else if (bargs->usage > 100) |
b3470b5d | 3655 | user_thresh = cache->length; |
a105bb88 | 3656 | else |
428c8e03 | 3657 | user_thresh = mult_perc(cache->length, bargs->usage); |
a105bb88 | 3658 | |
5ce5b3c0 ID |
3659 | if (chunk_used < user_thresh) |
3660 | ret = 0; | |
3661 | ||
3662 | btrfs_put_block_group(cache); | |
3663 | return ret; | |
3664 | } | |
3665 | ||
409d404b ID |
3666 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3667 | struct btrfs_chunk *chunk, | |
3668 | struct btrfs_balance_args *bargs) | |
3669 | { | |
3670 | struct btrfs_stripe *stripe; | |
3671 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3672 | int i; | |
3673 | ||
3674 | for (i = 0; i < num_stripes; i++) { | |
3675 | stripe = btrfs_stripe_nr(chunk, i); | |
3676 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3677 | return 0; | |
3678 | } | |
3679 | ||
3680 | return 1; | |
3681 | } | |
3682 | ||
946c9256 DS |
3683 | static u64 calc_data_stripes(u64 type, int num_stripes) |
3684 | { | |
3685 | const int index = btrfs_bg_flags_to_raid_index(type); | |
3686 | const int ncopies = btrfs_raid_array[index].ncopies; | |
3687 | const int nparity = btrfs_raid_array[index].nparity; | |
3688 | ||
d58ede8d | 3689 | return (num_stripes - nparity) / ncopies; |
946c9256 DS |
3690 | } |
3691 | ||
94e60d5a ID |
3692 | /* [pstart, pend) */ |
3693 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3694 | struct btrfs_chunk *chunk, | |
94e60d5a ID |
3695 | struct btrfs_balance_args *bargs) |
3696 | { | |
3697 | struct btrfs_stripe *stripe; | |
3698 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3699 | u64 stripe_offset; | |
3700 | u64 stripe_length; | |
946c9256 | 3701 | u64 type; |
94e60d5a ID |
3702 | int factor; |
3703 | int i; | |
3704 | ||
3705 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3706 | return 0; | |
3707 | ||
946c9256 DS |
3708 | type = btrfs_chunk_type(leaf, chunk); |
3709 | factor = calc_data_stripes(type, num_stripes); | |
94e60d5a ID |
3710 | |
3711 | for (i = 0; i < num_stripes; i++) { | |
3712 | stripe = btrfs_stripe_nr(chunk, i); | |
3713 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3714 | continue; | |
3715 | ||
3716 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3717 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3718 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3719 | |
3720 | if (stripe_offset < bargs->pend && | |
3721 | stripe_offset + stripe_length > bargs->pstart) | |
3722 | return 0; | |
3723 | } | |
3724 | ||
3725 | return 1; | |
3726 | } | |
3727 | ||
ea67176a ID |
3728 | /* [vstart, vend) */ |
3729 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3730 | struct btrfs_chunk *chunk, | |
3731 | u64 chunk_offset, | |
3732 | struct btrfs_balance_args *bargs) | |
3733 | { | |
3734 | if (chunk_offset < bargs->vend && | |
3735 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3736 | /* at least part of the chunk is inside this vrange */ | |
3737 | return 0; | |
3738 | ||
3739 | return 1; | |
3740 | } | |
3741 | ||
dee32d0a GAP |
3742 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3743 | struct btrfs_chunk *chunk, | |
3744 | struct btrfs_balance_args *bargs) | |
3745 | { | |
3746 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3747 | ||
3748 | if (bargs->stripes_min <= num_stripes | |
3749 | && num_stripes <= bargs->stripes_max) | |
3750 | return 0; | |
3751 | ||
3752 | return 1; | |
3753 | } | |
3754 | ||
899c81ea | 3755 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3756 | struct btrfs_balance_args *bargs) |
3757 | { | |
3758 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3759 | return 0; | |
3760 | ||
899c81ea ID |
3761 | chunk_type = chunk_to_extended(chunk_type) & |
3762 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3763 | |
899c81ea | 3764 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3765 | return 1; |
3766 | ||
3767 | return 0; | |
3768 | } | |
3769 | ||
6ec0896c | 3770 | static int should_balance_chunk(struct extent_buffer *leaf, |
f43ffb60 ID |
3771 | struct btrfs_chunk *chunk, u64 chunk_offset) |
3772 | { | |
6ec0896c | 3773 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 3774 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
f43ffb60 ID |
3775 | struct btrfs_balance_args *bargs = NULL; |
3776 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3777 | ||
3778 | /* type filter */ | |
3779 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3780 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3781 | return 0; | |
3782 | } | |
3783 | ||
3784 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3785 | bargs = &bctl->data; | |
3786 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3787 | bargs = &bctl->sys; | |
3788 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3789 | bargs = &bctl->meta; | |
3790 | ||
ed25e9b2 ID |
3791 | /* profiles filter */ |
3792 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3793 | chunk_profiles_filter(chunk_type, bargs)) { | |
3794 | return 0; | |
5ce5b3c0 ID |
3795 | } |
3796 | ||
3797 | /* usage filter */ | |
3798 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3799 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3800 | return 0; |
bc309467 | 3801 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3802 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3803 | return 0; |
409d404b ID |
3804 | } |
3805 | ||
3806 | /* devid filter */ | |
3807 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3808 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3809 | return 0; | |
94e60d5a ID |
3810 | } |
3811 | ||
3812 | /* drange filter, makes sense only with devid filter */ | |
3813 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
e4ff5fb5 | 3814 | chunk_drange_filter(leaf, chunk, bargs)) { |
94e60d5a | 3815 | return 0; |
ea67176a ID |
3816 | } |
3817 | ||
3818 | /* vrange filter */ | |
3819 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3820 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3821 | return 0; | |
ed25e9b2 ID |
3822 | } |
3823 | ||
dee32d0a GAP |
3824 | /* stripes filter */ |
3825 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3826 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3827 | return 0; | |
3828 | } | |
3829 | ||
cfa4c961 ID |
3830 | /* soft profile changing mode */ |
3831 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3832 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3833 | return 0; | |
3834 | } | |
3835 | ||
7d824b6f DS |
3836 | /* |
3837 | * limited by count, must be the last filter | |
3838 | */ | |
3839 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3840 | if (bargs->limit == 0) | |
3841 | return 0; | |
3842 | else | |
3843 | bargs->limit--; | |
12907fc7 DS |
3844 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3845 | /* | |
3846 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3847 | * determined here because we do not have the global information |
12907fc7 DS |
3848 | * about the count of all chunks that satisfy the filters. |
3849 | */ | |
3850 | if (bargs->limit_max == 0) | |
3851 | return 0; | |
3852 | else | |
3853 | bargs->limit_max--; | |
7d824b6f DS |
3854 | } |
3855 | ||
f43ffb60 ID |
3856 | return 1; |
3857 | } | |
3858 | ||
c9e9f97b | 3859 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3860 | { |
19a39dce | 3861 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b | 3862 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
12907fc7 | 3863 | u64 chunk_type; |
f43ffb60 | 3864 | struct btrfs_chunk *chunk; |
5a488b9d | 3865 | struct btrfs_path *path = NULL; |
ec44a35c | 3866 | struct btrfs_key key; |
ec44a35c | 3867 | struct btrfs_key found_key; |
f43ffb60 ID |
3868 | struct extent_buffer *leaf; |
3869 | int slot; | |
c9e9f97b ID |
3870 | int ret; |
3871 | int enospc_errors = 0; | |
19a39dce | 3872 | bool counting = true; |
12907fc7 | 3873 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3874 | u64 limit_data = bctl->data.limit; |
3875 | u64 limit_meta = bctl->meta.limit; | |
3876 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3877 | u32 count_data = 0; |
3878 | u32 count_meta = 0; | |
3879 | u32 count_sys = 0; | |
2c9fe835 | 3880 | int chunk_reserved = 0; |
ec44a35c | 3881 | |
ec44a35c | 3882 | path = btrfs_alloc_path(); |
17e9f796 MF |
3883 | if (!path) { |
3884 | ret = -ENOMEM; | |
3885 | goto error; | |
3886 | } | |
19a39dce ID |
3887 | |
3888 | /* zero out stat counters */ | |
3889 | spin_lock(&fs_info->balance_lock); | |
3890 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3891 | spin_unlock(&fs_info->balance_lock); | |
3892 | again: | |
7d824b6f | 3893 | if (!counting) { |
12907fc7 DS |
3894 | /* |
3895 | * The single value limit and min/max limits use the same bytes | |
3896 | * in the | |
3897 | */ | |
7d824b6f DS |
3898 | bctl->data.limit = limit_data; |
3899 | bctl->meta.limit = limit_meta; | |
3900 | bctl->sys.limit = limit_sys; | |
3901 | } | |
ec44a35c CM |
3902 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3903 | key.offset = (u64)-1; | |
3904 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3905 | ||
d397712b | 3906 | while (1) { |
19a39dce | 3907 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3908 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3909 | ret = -ECANCELED; |
3910 | goto error; | |
3911 | } | |
3912 | ||
f3372065 | 3913 | mutex_lock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3914 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 3915 | if (ret < 0) { |
f3372065 | 3916 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3917 | goto error; |
67c5e7d4 | 3918 | } |
ec44a35c CM |
3919 | |
3920 | /* | |
3921 | * this shouldn't happen, it means the last relocate | |
3922 | * failed | |
3923 | */ | |
3924 | if (ret == 0) | |
c9e9f97b | 3925 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3926 | |
3927 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3928 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3929 | if (ret) { |
f3372065 | 3930 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
c9e9f97b | 3931 | ret = 0; |
ec44a35c | 3932 | break; |
c9e9f97b | 3933 | } |
7d9eb12c | 3934 | |
f43ffb60 ID |
3935 | leaf = path->nodes[0]; |
3936 | slot = path->slots[0]; | |
3937 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3938 | |
67c5e7d4 | 3939 | if (found_key.objectid != key.objectid) { |
f3372065 | 3940 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
ec44a35c | 3941 | break; |
67c5e7d4 | 3942 | } |
7d9eb12c | 3943 | |
f43ffb60 | 3944 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3945 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3946 | |
19a39dce ID |
3947 | if (!counting) { |
3948 | spin_lock(&fs_info->balance_lock); | |
3949 | bctl->stat.considered++; | |
3950 | spin_unlock(&fs_info->balance_lock); | |
3951 | } | |
3952 | ||
6ec0896c | 3953 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
2c9fe835 | 3954 | |
b3b4aa74 | 3955 | btrfs_release_path(path); |
67c5e7d4 | 3956 | if (!ret) { |
f3372065 | 3957 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f43ffb60 | 3958 | goto loop; |
67c5e7d4 | 3959 | } |
f43ffb60 | 3960 | |
19a39dce | 3961 | if (counting) { |
f3372065 | 3962 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3963 | spin_lock(&fs_info->balance_lock); |
3964 | bctl->stat.expected++; | |
3965 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3966 | |
3967 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3968 | count_data++; | |
3969 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3970 | count_sys++; | |
3971 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3972 | count_meta++; | |
3973 | ||
3974 | goto loop; | |
3975 | } | |
3976 | ||
3977 | /* | |
3978 | * Apply limit_min filter, no need to check if the LIMITS | |
3979 | * filter is used, limit_min is 0 by default | |
3980 | */ | |
3981 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3982 | count_data < bctl->data.limit_min) | |
3983 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3984 | count_meta < bctl->meta.limit_min) | |
3985 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3986 | count_sys < bctl->sys.limit_min)) { | |
f3372065 | 3987 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce ID |
3988 | goto loop; |
3989 | } | |
3990 | ||
a6f93c71 LB |
3991 | if (!chunk_reserved) { |
3992 | /* | |
3993 | * We may be relocating the only data chunk we have, | |
3994 | * which could potentially end up with losing data's | |
3995 | * raid profile, so lets allocate an empty one in | |
3996 | * advance. | |
3997 | */ | |
3998 | ret = btrfs_may_alloc_data_chunk(fs_info, | |
3999 | found_key.offset); | |
2c9fe835 | 4000 | if (ret < 0) { |
f3372065 | 4001 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
2c9fe835 | 4002 | goto error; |
a6f93c71 LB |
4003 | } else if (ret == 1) { |
4004 | chunk_reserved = 1; | |
2c9fe835 | 4005 | } |
2c9fe835 ZL |
4006 | } |
4007 | ||
5b4aacef | 4008 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
f3372065 | 4009 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
19a39dce | 4010 | if (ret == -ENOSPC) { |
c9e9f97b | 4011 | enospc_errors++; |
eede2bf3 OS |
4012 | } else if (ret == -ETXTBSY) { |
4013 | btrfs_info(fs_info, | |
4014 | "skipping relocation of block group %llu due to active swapfile", | |
4015 | found_key.offset); | |
4016 | ret = 0; | |
4017 | } else if (ret) { | |
4018 | goto error; | |
19a39dce ID |
4019 | } else { |
4020 | spin_lock(&fs_info->balance_lock); | |
4021 | bctl->stat.completed++; | |
4022 | spin_unlock(&fs_info->balance_lock); | |
4023 | } | |
f43ffb60 | 4024 | loop: |
795a3321 ID |
4025 | if (found_key.offset == 0) |
4026 | break; | |
ba1bf481 | 4027 | key.offset = found_key.offset - 1; |
ec44a35c | 4028 | } |
c9e9f97b | 4029 | |
19a39dce ID |
4030 | if (counting) { |
4031 | btrfs_release_path(path); | |
4032 | counting = false; | |
4033 | goto again; | |
4034 | } | |
ec44a35c CM |
4035 | error: |
4036 | btrfs_free_path(path); | |
c9e9f97b | 4037 | if (enospc_errors) { |
efe120a0 | 4038 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 4039 | enospc_errors); |
c9e9f97b ID |
4040 | if (!ret) |
4041 | ret = -ENOSPC; | |
4042 | } | |
4043 | ||
ec44a35c CM |
4044 | return ret; |
4045 | } | |
4046 | ||
43dd529a DS |
4047 | /* |
4048 | * See if a given profile is valid and reduced. | |
4049 | * | |
4050 | * @flags: profile to validate | |
4051 | * @extended: if true @flags is treated as an extended profile | |
0c460c0d ID |
4052 | */ |
4053 | static int alloc_profile_is_valid(u64 flags, int extended) | |
4054 | { | |
4055 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
4056 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
4057 | ||
4058 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4059 | ||
4060 | /* 1) check that all other bits are zeroed */ | |
4061 | if (flags & ~mask) | |
4062 | return 0; | |
4063 | ||
4064 | /* 2) see if profile is reduced */ | |
4065 | if (flags == 0) | |
4066 | return !extended; /* "0" is valid for usual profiles */ | |
4067 | ||
c1499166 | 4068 | return has_single_bit_set(flags); |
0c460c0d ID |
4069 | } |
4070 | ||
837d5b6e ID |
4071 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
4072 | { | |
a7e99c69 ID |
4073 | /* cancel requested || normal exit path */ |
4074 | return atomic_read(&fs_info->balance_cancel_req) || | |
4075 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
4076 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
4077 | } |
4078 | ||
5ba366c3 DS |
4079 | /* |
4080 | * Validate target profile against allowed profiles and return true if it's OK. | |
4081 | * Otherwise print the error message and return false. | |
4082 | */ | |
4083 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, | |
4084 | const struct btrfs_balance_args *bargs, | |
4085 | u64 allowed, const char *type) | |
bdcd3c97 | 4086 | { |
5ba366c3 DS |
4087 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
4088 | return true; | |
4089 | ||
4090 | /* Profile is valid and does not have bits outside of the allowed set */ | |
4091 | if (alloc_profile_is_valid(bargs->target, 1) && | |
4092 | (bargs->target & ~allowed) == 0) | |
4093 | return true; | |
4094 | ||
4095 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", | |
4096 | type, btrfs_bg_type_to_raid_name(bargs->target)); | |
4097 | return false; | |
bdcd3c97 AM |
4098 | } |
4099 | ||
56fc37d9 AJ |
4100 | /* |
4101 | * Fill @buf with textual description of balance filter flags @bargs, up to | |
4102 | * @size_buf including the terminating null. The output may be trimmed if it | |
4103 | * does not fit into the provided buffer. | |
4104 | */ | |
4105 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, | |
4106 | u32 size_buf) | |
4107 | { | |
4108 | int ret; | |
4109 | u32 size_bp = size_buf; | |
4110 | char *bp = buf; | |
4111 | u64 flags = bargs->flags; | |
4112 | char tmp_buf[128] = {'\0'}; | |
4113 | ||
4114 | if (!flags) | |
4115 | return; | |
4116 | ||
4117 | #define CHECK_APPEND_NOARG(a) \ | |
4118 | do { \ | |
4119 | ret = snprintf(bp, size_bp, (a)); \ | |
4120 | if (ret < 0 || ret >= size_bp) \ | |
4121 | goto out_overflow; \ | |
4122 | size_bp -= ret; \ | |
4123 | bp += ret; \ | |
4124 | } while (0) | |
4125 | ||
4126 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4127 | do { \ | |
4128 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4129 | if (ret < 0 || ret >= size_bp) \ | |
4130 | goto out_overflow; \ | |
4131 | size_bp -= ret; \ | |
4132 | bp += ret; \ | |
4133 | } while (0) | |
4134 | ||
4135 | #define CHECK_APPEND_2ARG(a, v1, v2) \ | |
4136 | do { \ | |
4137 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ | |
4138 | if (ret < 0 || ret >= size_bp) \ | |
4139 | goto out_overflow; \ | |
4140 | size_bp -= ret; \ | |
4141 | bp += ret; \ | |
4142 | } while (0) | |
4143 | ||
158da513 DS |
4144 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
4145 | CHECK_APPEND_1ARG("convert=%s,", | |
4146 | btrfs_bg_type_to_raid_name(bargs->target)); | |
56fc37d9 AJ |
4147 | |
4148 | if (flags & BTRFS_BALANCE_ARGS_SOFT) | |
4149 | CHECK_APPEND_NOARG("soft,"); | |
4150 | ||
4151 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { | |
4152 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, | |
4153 | sizeof(tmp_buf)); | |
4154 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); | |
4155 | } | |
4156 | ||
4157 | if (flags & BTRFS_BALANCE_ARGS_USAGE) | |
4158 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); | |
4159 | ||
4160 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) | |
4161 | CHECK_APPEND_2ARG("usage=%u..%u,", | |
4162 | bargs->usage_min, bargs->usage_max); | |
4163 | ||
4164 | if (flags & BTRFS_BALANCE_ARGS_DEVID) | |
4165 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); | |
4166 | ||
4167 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) | |
4168 | CHECK_APPEND_2ARG("drange=%llu..%llu,", | |
4169 | bargs->pstart, bargs->pend); | |
4170 | ||
4171 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) | |
4172 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", | |
4173 | bargs->vstart, bargs->vend); | |
4174 | ||
4175 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) | |
4176 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); | |
4177 | ||
4178 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) | |
4179 | CHECK_APPEND_2ARG("limit=%u..%u,", | |
4180 | bargs->limit_min, bargs->limit_max); | |
4181 | ||
4182 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) | |
4183 | CHECK_APPEND_2ARG("stripes=%u..%u,", | |
4184 | bargs->stripes_min, bargs->stripes_max); | |
4185 | ||
4186 | #undef CHECK_APPEND_2ARG | |
4187 | #undef CHECK_APPEND_1ARG | |
4188 | #undef CHECK_APPEND_NOARG | |
4189 | ||
4190 | out_overflow: | |
4191 | ||
4192 | if (size_bp < size_buf) | |
4193 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ | |
4194 | else | |
4195 | buf[0] = '\0'; | |
4196 | } | |
4197 | ||
4198 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) | |
4199 | { | |
4200 | u32 size_buf = 1024; | |
4201 | char tmp_buf[192] = {'\0'}; | |
4202 | char *buf; | |
4203 | char *bp; | |
4204 | u32 size_bp = size_buf; | |
4205 | int ret; | |
4206 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4207 | ||
4208 | buf = kzalloc(size_buf, GFP_KERNEL); | |
4209 | if (!buf) | |
4210 | return; | |
4211 | ||
4212 | bp = buf; | |
4213 | ||
4214 | #define CHECK_APPEND_1ARG(a, v1) \ | |
4215 | do { \ | |
4216 | ret = snprintf(bp, size_bp, (a), (v1)); \ | |
4217 | if (ret < 0 || ret >= size_bp) \ | |
4218 | goto out_overflow; \ | |
4219 | size_bp -= ret; \ | |
4220 | bp += ret; \ | |
4221 | } while (0) | |
4222 | ||
4223 | if (bctl->flags & BTRFS_BALANCE_FORCE) | |
4224 | CHECK_APPEND_1ARG("%s", "-f "); | |
4225 | ||
4226 | if (bctl->flags & BTRFS_BALANCE_DATA) { | |
4227 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); | |
4228 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); | |
4229 | } | |
4230 | ||
4231 | if (bctl->flags & BTRFS_BALANCE_METADATA) { | |
4232 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); | |
4233 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); | |
4234 | } | |
4235 | ||
4236 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { | |
4237 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); | |
4238 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); | |
4239 | } | |
4240 | ||
4241 | #undef CHECK_APPEND_1ARG | |
4242 | ||
4243 | out_overflow: | |
4244 | ||
4245 | if (size_bp < size_buf) | |
4246 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ | |
4247 | btrfs_info(fs_info, "balance: %s %s", | |
4248 | (bctl->flags & BTRFS_BALANCE_RESUME) ? | |
4249 | "resume" : "start", buf); | |
4250 | ||
4251 | kfree(buf); | |
4252 | } | |
4253 | ||
c9e9f97b | 4254 | /* |
dccdb07b | 4255 | * Should be called with balance mutexe held |
c9e9f97b | 4256 | */ |
6fcf6e2b DS |
4257 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
4258 | struct btrfs_balance_control *bctl, | |
c9e9f97b ID |
4259 | struct btrfs_ioctl_balance_args *bargs) |
4260 | { | |
14506127 | 4261 | u64 meta_target, data_target; |
f43ffb60 | 4262 | u64 allowed; |
e4837f8f | 4263 | int mixed = 0; |
c9e9f97b | 4264 | int ret; |
8dabb742 | 4265 | u64 num_devices; |
de98ced9 | 4266 | unsigned seq; |
e62869be | 4267 | bool reducing_redundancy; |
081db89b | 4268 | int i; |
c9e9f97b | 4269 | |
837d5b6e | 4270 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 | 4271 | atomic_read(&fs_info->balance_pause_req) || |
726a3421 | 4272 | btrfs_should_cancel_balance(fs_info)) { |
c9e9f97b ID |
4273 | ret = -EINVAL; |
4274 | goto out; | |
4275 | } | |
4276 | ||
e4837f8f ID |
4277 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
4278 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
4279 | mixed = 1; | |
4280 | ||
f43ffb60 ID |
4281 | /* |
4282 | * In case of mixed groups both data and meta should be picked, | |
4283 | * and identical options should be given for both of them. | |
4284 | */ | |
e4837f8f ID |
4285 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
4286 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
4287 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
4288 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
4289 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e | 4290 | btrfs_err(fs_info, |
6dac13f8 | 4291 | "balance: mixed groups data and metadata options must be the same"); |
f43ffb60 ID |
4292 | ret = -EINVAL; |
4293 | goto out; | |
4294 | } | |
4295 | } | |
4296 | ||
b35cf1f0 JB |
4297 | /* |
4298 | * rw_devices will not change at the moment, device add/delete/replace | |
c3e1f96c | 4299 | * are exclusive |
b35cf1f0 JB |
4300 | */ |
4301 | num_devices = fs_info->fs_devices->rw_devices; | |
fab27359 QW |
4302 | |
4303 | /* | |
4304 | * SINGLE profile on-disk has no profile bit, but in-memory we have a | |
4305 | * special bit for it, to make it easier to distinguish. Thus we need | |
4306 | * to set it manually, or balance would refuse the profile. | |
4307 | */ | |
4308 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
081db89b DS |
4309 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
4310 | if (num_devices >= btrfs_raid_array[i].devs_min) | |
4311 | allowed |= btrfs_raid_array[i].bg_flag; | |
1da73967 | 4312 | |
5ba366c3 DS |
4313 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
4314 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || | |
4315 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { | |
e4d8ec0f ID |
4316 | ret = -EINVAL; |
4317 | goto out; | |
4318 | } | |
4319 | ||
6079e12c DS |
4320 | /* |
4321 | * Allow to reduce metadata or system integrity only if force set for | |
4322 | * profiles with redundancy (copies, parity) | |
4323 | */ | |
4324 | allowed = 0; | |
4325 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { | |
4326 | if (btrfs_raid_array[i].ncopies >= 2 || | |
4327 | btrfs_raid_array[i].tolerated_failures >= 1) | |
4328 | allowed |= btrfs_raid_array[i].bg_flag; | |
4329 | } | |
de98ced9 MX |
4330 | do { |
4331 | seq = read_seqbegin(&fs_info->profiles_lock); | |
4332 | ||
4333 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4334 | (fs_info->avail_system_alloc_bits & allowed) && | |
4335 | !(bctl->sys.target & allowed)) || | |
4336 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
4337 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
5a8067c0 | 4338 | !(bctl->meta.target & allowed))) |
e62869be | 4339 | reducing_redundancy = true; |
5a8067c0 | 4340 | else |
e62869be | 4341 | reducing_redundancy = false; |
5a8067c0 FM |
4342 | |
4343 | /* if we're not converting, the target field is uninitialized */ | |
4344 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4345 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; | |
4346 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? | |
4347 | bctl->data.target : fs_info->avail_data_alloc_bits; | |
de98ced9 | 4348 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 4349 | |
e62869be | 4350 | if (reducing_redundancy) { |
5a8067c0 FM |
4351 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
4352 | btrfs_info(fs_info, | |
e62869be | 4353 | "balance: force reducing metadata redundancy"); |
5a8067c0 FM |
4354 | } else { |
4355 | btrfs_err(fs_info, | |
e62869be | 4356 | "balance: reduces metadata redundancy, use --force if you want this"); |
5a8067c0 FM |
4357 | ret = -EINVAL; |
4358 | goto out; | |
4359 | } | |
4360 | } | |
4361 | ||
14506127 AB |
4362 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
4363 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { | |
ee592d07 | 4364 | btrfs_warn(fs_info, |
6dac13f8 | 4365 | "balance: metadata profile %s has lower redundancy than data profile %s", |
158da513 DS |
4366 | btrfs_bg_type_to_raid_name(meta_target), |
4367 | btrfs_bg_type_to_raid_name(data_target)); | |
ee592d07 ST |
4368 | } |
4369 | ||
6bccf3ab | 4370 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 4371 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
4372 | goto out; |
4373 | ||
59641015 ID |
4374 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
4375 | BUG_ON(ret == -EEXIST); | |
833aae18 DS |
4376 | BUG_ON(fs_info->balance_ctl); |
4377 | spin_lock(&fs_info->balance_lock); | |
4378 | fs_info->balance_ctl = bctl; | |
4379 | spin_unlock(&fs_info->balance_lock); | |
59641015 ID |
4380 | } else { |
4381 | BUG_ON(ret != -EEXIST); | |
4382 | spin_lock(&fs_info->balance_lock); | |
4383 | update_balance_args(bctl); | |
4384 | spin_unlock(&fs_info->balance_lock); | |
4385 | } | |
c9e9f97b | 4386 | |
3009a62f DS |
4387 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
4388 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); | |
56fc37d9 | 4389 | describe_balance_start_or_resume(fs_info); |
c9e9f97b ID |
4390 | mutex_unlock(&fs_info->balance_mutex); |
4391 | ||
4392 | ret = __btrfs_balance(fs_info); | |
4393 | ||
4394 | mutex_lock(&fs_info->balance_mutex); | |
efc0e69c | 4395 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { |
7333bd02 | 4396 | btrfs_info(fs_info, "balance: paused"); |
efc0e69c NB |
4397 | btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); |
4398 | } | |
44d354ab QW |
4399 | /* |
4400 | * Balance can be canceled by: | |
4401 | * | |
4402 | * - Regular cancel request | |
4403 | * Then ret == -ECANCELED and balance_cancel_req > 0 | |
4404 | * | |
4405 | * - Fatal signal to "btrfs" process | |
4406 | * Either the signal caught by wait_reserve_ticket() and callers | |
4407 | * got -EINTR, or caught by btrfs_should_cancel_balance() and | |
4408 | * got -ECANCELED. | |
4409 | * Either way, in this case balance_cancel_req = 0, and | |
4410 | * ret == -EINTR or ret == -ECANCELED. | |
4411 | * | |
4412 | * So here we only check the return value to catch canceled balance. | |
4413 | */ | |
4414 | else if (ret == -ECANCELED || ret == -EINTR) | |
7333bd02 AJ |
4415 | btrfs_info(fs_info, "balance: canceled"); |
4416 | else | |
4417 | btrfs_info(fs_info, "balance: ended with status: %d", ret); | |
4418 | ||
3009a62f | 4419 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
c9e9f97b ID |
4420 | |
4421 | if (bargs) { | |
4422 | memset(bargs, 0, sizeof(*bargs)); | |
008ef096 | 4423 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
c9e9f97b ID |
4424 | } |
4425 | ||
3a01aa7a ID |
4426 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
4427 | balance_need_close(fs_info)) { | |
149196a2 | 4428 | reset_balance_state(fs_info); |
c3e1f96c | 4429 | btrfs_exclop_finish(fs_info); |
3a01aa7a ID |
4430 | } |
4431 | ||
837d5b6e | 4432 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
4433 | |
4434 | return ret; | |
4435 | out: | |
59641015 | 4436 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
149196a2 | 4437 | reset_balance_state(fs_info); |
a17c95df | 4438 | else |
59641015 | 4439 | kfree(bctl); |
c3e1f96c | 4440 | btrfs_exclop_finish(fs_info); |
a17c95df | 4441 | |
59641015 ID |
4442 | return ret; |
4443 | } | |
4444 | ||
4445 | static int balance_kthread(void *data) | |
4446 | { | |
2b6ba629 | 4447 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 4448 | int ret = 0; |
59641015 | 4449 | |
a690e5f2 | 4450 | sb_start_write(fs_info->sb); |
59641015 | 4451 | mutex_lock(&fs_info->balance_mutex); |
56fc37d9 | 4452 | if (fs_info->balance_ctl) |
6fcf6e2b | 4453 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
59641015 | 4454 | mutex_unlock(&fs_info->balance_mutex); |
a690e5f2 | 4455 | sb_end_write(fs_info->sb); |
2b6ba629 | 4456 | |
59641015 ID |
4457 | return ret; |
4458 | } | |
4459 | ||
2b6ba629 ID |
4460 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
4461 | { | |
4462 | struct task_struct *tsk; | |
4463 | ||
1354e1a1 | 4464 | mutex_lock(&fs_info->balance_mutex); |
2b6ba629 | 4465 | if (!fs_info->balance_ctl) { |
1354e1a1 | 4466 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 ID |
4467 | return 0; |
4468 | } | |
1354e1a1 | 4469 | mutex_unlock(&fs_info->balance_mutex); |
2b6ba629 | 4470 | |
3cdde224 | 4471 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
6dac13f8 | 4472 | btrfs_info(fs_info, "balance: resume skipped"); |
2b6ba629 ID |
4473 | return 0; |
4474 | } | |
4475 | ||
efc0e69c NB |
4476 | spin_lock(&fs_info->super_lock); |
4477 | ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); | |
4478 | fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; | |
4479 | spin_unlock(&fs_info->super_lock); | |
02ee654d AJ |
4480 | /* |
4481 | * A ro->rw remount sequence should continue with the paused balance | |
4482 | * regardless of who pauses it, system or the user as of now, so set | |
4483 | * the resume flag. | |
4484 | */ | |
4485 | spin_lock(&fs_info->balance_lock); | |
4486 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; | |
4487 | spin_unlock(&fs_info->balance_lock); | |
4488 | ||
2b6ba629 | 4489 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
cd633972 | 4490 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
4491 | } |
4492 | ||
68310a5e | 4493 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 4494 | { |
59641015 ID |
4495 | struct btrfs_balance_control *bctl; |
4496 | struct btrfs_balance_item *item; | |
4497 | struct btrfs_disk_balance_args disk_bargs; | |
4498 | struct btrfs_path *path; | |
4499 | struct extent_buffer *leaf; | |
4500 | struct btrfs_key key; | |
4501 | int ret; | |
4502 | ||
4503 | path = btrfs_alloc_path(); | |
4504 | if (!path) | |
4505 | return -ENOMEM; | |
4506 | ||
59641015 | 4507 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 4508 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
4509 | key.offset = 0; |
4510 | ||
68310a5e | 4511 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 4512 | if (ret < 0) |
68310a5e | 4513 | goto out; |
59641015 ID |
4514 | if (ret > 0) { /* ret = -ENOENT; */ |
4515 | ret = 0; | |
68310a5e ID |
4516 | goto out; |
4517 | } | |
4518 | ||
4519 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
4520 | if (!bctl) { | |
4521 | ret = -ENOMEM; | |
4522 | goto out; | |
59641015 ID |
4523 | } |
4524 | ||
4525 | leaf = path->nodes[0]; | |
4526 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4527 | ||
68310a5e ID |
4528 | bctl->flags = btrfs_balance_flags(leaf, item); |
4529 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4530 | |
4531 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4532 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4533 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4534 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4535 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4536 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4537 | ||
eee95e3f DS |
4538 | /* |
4539 | * This should never happen, as the paused balance state is recovered | |
4540 | * during mount without any chance of other exclusive ops to collide. | |
4541 | * | |
4542 | * This gives the exclusive op status to balance and keeps in paused | |
4543 | * state until user intervention (cancel or umount). If the ownership | |
4544 | * cannot be assigned, show a message but do not fail. The balance | |
4545 | * is in a paused state and must have fs_info::balance_ctl properly | |
4546 | * set up. | |
4547 | */ | |
efc0e69c | 4548 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) |
eee95e3f | 4549 | btrfs_warn(fs_info, |
6dac13f8 | 4550 | "balance: cannot set exclusive op status, resume manually"); |
ed0fb78f | 4551 | |
fb286100 JB |
4552 | btrfs_release_path(path); |
4553 | ||
68310a5e | 4554 | mutex_lock(&fs_info->balance_mutex); |
833aae18 DS |
4555 | BUG_ON(fs_info->balance_ctl); |
4556 | spin_lock(&fs_info->balance_lock); | |
4557 | fs_info->balance_ctl = bctl; | |
4558 | spin_unlock(&fs_info->balance_lock); | |
68310a5e | 4559 | mutex_unlock(&fs_info->balance_mutex); |
59641015 ID |
4560 | out: |
4561 | btrfs_free_path(path); | |
ec44a35c CM |
4562 | return ret; |
4563 | } | |
4564 | ||
837d5b6e ID |
4565 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4566 | { | |
4567 | int ret = 0; | |
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 | ||
3009a62f | 4575 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
837d5b6e ID |
4576 | atomic_inc(&fs_info->balance_pause_req); |
4577 | mutex_unlock(&fs_info->balance_mutex); | |
4578 | ||
4579 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4580 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4581 | |
4582 | mutex_lock(&fs_info->balance_mutex); | |
4583 | /* we are good with balance_ctl ripped off from under us */ | |
3009a62f | 4584 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
837d5b6e ID |
4585 | atomic_dec(&fs_info->balance_pause_req); |
4586 | } else { | |
4587 | ret = -ENOTCONN; | |
4588 | } | |
4589 | ||
4590 | mutex_unlock(&fs_info->balance_mutex); | |
4591 | return ret; | |
4592 | } | |
4593 | ||
a7e99c69 ID |
4594 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4595 | { | |
4596 | mutex_lock(&fs_info->balance_mutex); | |
4597 | if (!fs_info->balance_ctl) { | |
4598 | mutex_unlock(&fs_info->balance_mutex); | |
4599 | return -ENOTCONN; | |
4600 | } | |
4601 | ||
cf7d20f4 DS |
4602 | /* |
4603 | * A paused balance with the item stored on disk can be resumed at | |
4604 | * mount time if the mount is read-write. Otherwise it's still paused | |
4605 | * and we must not allow cancelling as it deletes the item. | |
4606 | */ | |
4607 | if (sb_rdonly(fs_info->sb)) { | |
4608 | mutex_unlock(&fs_info->balance_mutex); | |
4609 | return -EROFS; | |
4610 | } | |
4611 | ||
a7e99c69 ID |
4612 | atomic_inc(&fs_info->balance_cancel_req); |
4613 | /* | |
4614 | * if we are running just wait and return, balance item is | |
4615 | * deleted in btrfs_balance in this case | |
4616 | */ | |
3009a62f | 4617 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
a7e99c69 ID |
4618 | mutex_unlock(&fs_info->balance_mutex); |
4619 | wait_event(fs_info->balance_wait_q, | |
3009a62f | 4620 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
a7e99c69 ID |
4621 | mutex_lock(&fs_info->balance_mutex); |
4622 | } else { | |
a7e99c69 | 4623 | mutex_unlock(&fs_info->balance_mutex); |
dccdb07b DS |
4624 | /* |
4625 | * Lock released to allow other waiters to continue, we'll | |
4626 | * reexamine the status again. | |
4627 | */ | |
a7e99c69 ID |
4628 | mutex_lock(&fs_info->balance_mutex); |
4629 | ||
a17c95df | 4630 | if (fs_info->balance_ctl) { |
149196a2 | 4631 | reset_balance_state(fs_info); |
c3e1f96c | 4632 | btrfs_exclop_finish(fs_info); |
6dac13f8 | 4633 | btrfs_info(fs_info, "balance: canceled"); |
a17c95df | 4634 | } |
a7e99c69 ID |
4635 | } |
4636 | ||
3009a62f DS |
4637 | BUG_ON(fs_info->balance_ctl || |
4638 | test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); | |
a7e99c69 ID |
4639 | atomic_dec(&fs_info->balance_cancel_req); |
4640 | mutex_unlock(&fs_info->balance_mutex); | |
4641 | return 0; | |
4642 | } | |
4643 | ||
97f4dd09 | 4644 | int btrfs_uuid_scan_kthread(void *data) |
803b2f54 SB |
4645 | { |
4646 | struct btrfs_fs_info *fs_info = data; | |
4647 | struct btrfs_root *root = fs_info->tree_root; | |
4648 | struct btrfs_key key; | |
803b2f54 SB |
4649 | struct btrfs_path *path = NULL; |
4650 | int ret = 0; | |
4651 | struct extent_buffer *eb; | |
4652 | int slot; | |
4653 | struct btrfs_root_item root_item; | |
4654 | u32 item_size; | |
f45388f3 | 4655 | struct btrfs_trans_handle *trans = NULL; |
c94bec2c | 4656 | bool closing = false; |
803b2f54 SB |
4657 | |
4658 | path = btrfs_alloc_path(); | |
4659 | if (!path) { | |
4660 | ret = -ENOMEM; | |
4661 | goto out; | |
4662 | } | |
4663 | ||
4664 | key.objectid = 0; | |
4665 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4666 | key.offset = 0; | |
4667 | ||
803b2f54 | 4668 | while (1) { |
c94bec2c JB |
4669 | if (btrfs_fs_closing(fs_info)) { |
4670 | closing = true; | |
4671 | break; | |
4672 | } | |
7c829b72 AJ |
4673 | ret = btrfs_search_forward(root, &key, path, |
4674 | BTRFS_OLDEST_GENERATION); | |
803b2f54 SB |
4675 | if (ret) { |
4676 | if (ret > 0) | |
4677 | ret = 0; | |
4678 | break; | |
4679 | } | |
4680 | ||
4681 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4682 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4683 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4684 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4685 | goto skip; | |
4686 | ||
4687 | eb = path->nodes[0]; | |
4688 | slot = path->slots[0]; | |
3212fa14 | 4689 | item_size = btrfs_item_size(eb, slot); |
803b2f54 SB |
4690 | if (item_size < sizeof(root_item)) |
4691 | goto skip; | |
4692 | ||
803b2f54 SB |
4693 | read_extent_buffer(eb, &root_item, |
4694 | btrfs_item_ptr_offset(eb, slot), | |
4695 | (int)sizeof(root_item)); | |
4696 | if (btrfs_root_refs(&root_item) == 0) | |
4697 | goto skip; | |
f45388f3 FDBM |
4698 | |
4699 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4700 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4701 | if (trans) | |
4702 | goto update_tree; | |
4703 | ||
4704 | btrfs_release_path(path); | |
803b2f54 SB |
4705 | /* |
4706 | * 1 - subvol uuid item | |
4707 | * 1 - received_subvol uuid item | |
4708 | */ | |
4709 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4710 | if (IS_ERR(trans)) { | |
4711 | ret = PTR_ERR(trans); | |
4712 | break; | |
4713 | } | |
f45388f3 FDBM |
4714 | continue; |
4715 | } else { | |
4716 | goto skip; | |
4717 | } | |
4718 | update_tree: | |
9771a5cf | 4719 | btrfs_release_path(path); |
f45388f3 | 4720 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
cdb345a8 | 4721 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
803b2f54 SB |
4722 | BTRFS_UUID_KEY_SUBVOL, |
4723 | key.objectid); | |
4724 | if (ret < 0) { | |
efe120a0 | 4725 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4726 | ret); |
803b2f54 SB |
4727 | break; |
4728 | } | |
4729 | } | |
4730 | ||
4731 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
cdb345a8 | 4732 | ret = btrfs_uuid_tree_add(trans, |
803b2f54 SB |
4733 | root_item.received_uuid, |
4734 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4735 | key.objectid); | |
4736 | if (ret < 0) { | |
efe120a0 | 4737 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4738 | ret); |
803b2f54 SB |
4739 | break; |
4740 | } | |
4741 | } | |
4742 | ||
f45388f3 | 4743 | skip: |
9771a5cf | 4744 | btrfs_release_path(path); |
803b2f54 | 4745 | if (trans) { |
3a45bb20 | 4746 | ret = btrfs_end_transaction(trans); |
f45388f3 | 4747 | trans = NULL; |
803b2f54 SB |
4748 | if (ret) |
4749 | break; | |
4750 | } | |
4751 | ||
803b2f54 SB |
4752 | if (key.offset < (u64)-1) { |
4753 | key.offset++; | |
4754 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4755 | key.offset = 0; | |
4756 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4757 | } else if (key.objectid < (u64)-1) { | |
4758 | key.offset = 0; | |
4759 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4760 | key.objectid++; | |
4761 | } else { | |
4762 | break; | |
4763 | } | |
4764 | cond_resched(); | |
4765 | } | |
4766 | ||
4767 | out: | |
4768 | btrfs_free_path(path); | |
f45388f3 | 4769 | if (trans && !IS_ERR(trans)) |
3a45bb20 | 4770 | btrfs_end_transaction(trans); |
803b2f54 | 4771 | if (ret) |
efe120a0 | 4772 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
c94bec2c | 4773 | else if (!closing) |
afcdd129 | 4774 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4775 | up(&fs_info->uuid_tree_rescan_sem); |
4776 | return 0; | |
4777 | } | |
4778 | ||
f7a81ea4 SB |
4779 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4780 | { | |
4781 | struct btrfs_trans_handle *trans; | |
4782 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4783 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4784 | struct task_struct *task; |
4785 | int ret; | |
f7a81ea4 SB |
4786 | |
4787 | /* | |
4788 | * 1 - root node | |
4789 | * 1 - root item | |
4790 | */ | |
4791 | trans = btrfs_start_transaction(tree_root, 2); | |
4792 | if (IS_ERR(trans)) | |
4793 | return PTR_ERR(trans); | |
4794 | ||
9b7a2440 | 4795 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
f7a81ea4 | 4796 | if (IS_ERR(uuid_root)) { |
6d13f549 | 4797 | ret = PTR_ERR(uuid_root); |
66642832 | 4798 | btrfs_abort_transaction(trans, ret); |
3a45bb20 | 4799 | btrfs_end_transaction(trans); |
6d13f549 | 4800 | return ret; |
f7a81ea4 SB |
4801 | } |
4802 | ||
4803 | fs_info->uuid_root = uuid_root; | |
4804 | ||
3a45bb20 | 4805 | ret = btrfs_commit_transaction(trans); |
803b2f54 SB |
4806 | if (ret) |
4807 | return ret; | |
4808 | ||
4809 | down(&fs_info->uuid_tree_rescan_sem); | |
4810 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4811 | if (IS_ERR(task)) { | |
70f80175 | 4812 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4813 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4814 | up(&fs_info->uuid_tree_rescan_sem); |
4815 | return PTR_ERR(task); | |
4816 | } | |
4817 | ||
4818 | return 0; | |
f7a81ea4 | 4819 | } |
803b2f54 | 4820 | |
8f18cf13 CM |
4821 | /* |
4822 | * shrinking a device means finding all of the device extents past | |
4823 | * the new size, and then following the back refs to the chunks. | |
4824 | * The chunk relocation code actually frees the device extent | |
4825 | */ | |
4826 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4827 | { | |
0b246afa JM |
4828 | struct btrfs_fs_info *fs_info = device->fs_info; |
4829 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4830 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4831 | struct btrfs_dev_extent *dev_extent = NULL; |
4832 | struct btrfs_path *path; | |
4833 | u64 length; | |
8f18cf13 CM |
4834 | u64 chunk_offset; |
4835 | int ret; | |
4836 | int slot; | |
ba1bf481 JB |
4837 | int failed = 0; |
4838 | bool retried = false; | |
8f18cf13 CM |
4839 | struct extent_buffer *l; |
4840 | struct btrfs_key key; | |
0b246afa | 4841 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4842 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d | 4843 | u64 old_size = btrfs_device_get_total_bytes(device); |
7dfb8be1 | 4844 | u64 diff; |
61d0d0d2 | 4845 | u64 start; |
7dfb8be1 NB |
4846 | |
4847 | new_size = round_down(new_size, fs_info->sectorsize); | |
61d0d0d2 | 4848 | start = new_size; |
0e4324a4 | 4849 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
8f18cf13 | 4850 | |
401e29c1 | 4851 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
63a212ab SB |
4852 | return -EINVAL; |
4853 | ||
8f18cf13 CM |
4854 | path = btrfs_alloc_path(); |
4855 | if (!path) | |
4856 | return -ENOMEM; | |
4857 | ||
0338dff6 | 4858 | path->reada = READA_BACK; |
8f18cf13 | 4859 | |
61d0d0d2 NB |
4860 | trans = btrfs_start_transaction(root, 0); |
4861 | if (IS_ERR(trans)) { | |
4862 | btrfs_free_path(path); | |
4863 | return PTR_ERR(trans); | |
4864 | } | |
4865 | ||
34441361 | 4866 | mutex_lock(&fs_info->chunk_mutex); |
7d9eb12c | 4867 | |
7cc8e58d | 4868 | btrfs_device_set_total_bytes(device, new_size); |
ebbede42 | 4869 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
2b82032c | 4870 | device->fs_devices->total_rw_bytes -= diff; |
a5ed45f8 | 4871 | atomic64_sub(diff, &fs_info->free_chunk_space); |
2bf64758 | 4872 | } |
61d0d0d2 NB |
4873 | |
4874 | /* | |
4875 | * Once the device's size has been set to the new size, ensure all | |
4876 | * in-memory chunks are synced to disk so that the loop below sees them | |
4877 | * and relocates them accordingly. | |
4878 | */ | |
1c11b63e | 4879 | if (contains_pending_extent(device, &start, diff)) { |
61d0d0d2 NB |
4880 | mutex_unlock(&fs_info->chunk_mutex); |
4881 | ret = btrfs_commit_transaction(trans); | |
4882 | if (ret) | |
4883 | goto done; | |
4884 | } else { | |
4885 | mutex_unlock(&fs_info->chunk_mutex); | |
4886 | btrfs_end_transaction(trans); | |
4887 | } | |
8f18cf13 | 4888 | |
ba1bf481 | 4889 | again: |
8f18cf13 CM |
4890 | key.objectid = device->devid; |
4891 | key.offset = (u64)-1; | |
4892 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4893 | ||
213e64da | 4894 | do { |
f3372065 | 4895 | mutex_lock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4896 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4897 | if (ret < 0) { |
f3372065 | 4898 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
8f18cf13 | 4899 | goto done; |
67c5e7d4 | 4900 | } |
8f18cf13 CM |
4901 | |
4902 | ret = btrfs_previous_item(root, path, 0, key.type); | |
8f18cf13 | 4903 | if (ret) { |
f3372065 | 4904 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
7056bf69 NB |
4905 | if (ret < 0) |
4906 | goto done; | |
8f18cf13 | 4907 | ret = 0; |
b3b4aa74 | 4908 | btrfs_release_path(path); |
bf1fb512 | 4909 | break; |
8f18cf13 CM |
4910 | } |
4911 | ||
4912 | l = path->nodes[0]; | |
4913 | slot = path->slots[0]; | |
4914 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4915 | ||
ba1bf481 | 4916 | if (key.objectid != device->devid) { |
f3372065 | 4917 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4918 | btrfs_release_path(path); |
bf1fb512 | 4919 | break; |
ba1bf481 | 4920 | } |
8f18cf13 CM |
4921 | |
4922 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4923 | length = btrfs_dev_extent_length(l, dev_extent); | |
4924 | ||
ba1bf481 | 4925 | if (key.offset + length <= new_size) { |
f3372065 | 4926 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
b3b4aa74 | 4927 | btrfs_release_path(path); |
d6397bae | 4928 | break; |
ba1bf481 | 4929 | } |
8f18cf13 | 4930 | |
8f18cf13 | 4931 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4932 | btrfs_release_path(path); |
8f18cf13 | 4933 | |
a6f93c71 LB |
4934 | /* |
4935 | * We may be relocating the only data chunk we have, | |
4936 | * which could potentially end up with losing data's | |
4937 | * raid profile, so lets allocate an empty one in | |
4938 | * advance. | |
4939 | */ | |
4940 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); | |
4941 | if (ret < 0) { | |
f3372065 | 4942 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
a6f93c71 LB |
4943 | goto done; |
4944 | } | |
4945 | ||
0b246afa | 4946 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
f3372065 | 4947 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
eede2bf3 | 4948 | if (ret == -ENOSPC) { |
ba1bf481 | 4949 | failed++; |
eede2bf3 OS |
4950 | } else if (ret) { |
4951 | if (ret == -ETXTBSY) { | |
4952 | btrfs_warn(fs_info, | |
4953 | "could not shrink block group %llu due to active swapfile", | |
4954 | chunk_offset); | |
4955 | } | |
4956 | goto done; | |
4957 | } | |
213e64da | 4958 | } while (key.offset-- > 0); |
ba1bf481 JB |
4959 | |
4960 | if (failed && !retried) { | |
4961 | failed = 0; | |
4962 | retried = true; | |
4963 | goto again; | |
4964 | } else if (failed && retried) { | |
4965 | ret = -ENOSPC; | |
ba1bf481 | 4966 | goto done; |
8f18cf13 CM |
4967 | } |
4968 | ||
d6397bae | 4969 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4970 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4971 | if (IS_ERR(trans)) { |
4972 | ret = PTR_ERR(trans); | |
4973 | goto done; | |
4974 | } | |
4975 | ||
34441361 | 4976 | mutex_lock(&fs_info->chunk_mutex); |
c57dd1f2 QW |
4977 | /* Clear all state bits beyond the shrunk device size */ |
4978 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, | |
4979 | CHUNK_STATE_MASK); | |
4980 | ||
7cc8e58d | 4981 | btrfs_device_set_disk_total_bytes(device, new_size); |
bbbf7243 NB |
4982 | if (list_empty(&device->post_commit_list)) |
4983 | list_add_tail(&device->post_commit_list, | |
4984 | &trans->transaction->dev_update_list); | |
d6397bae | 4985 | |
d6397bae | 4986 | WARN_ON(diff > old_total); |
7dfb8be1 NB |
4987 | btrfs_set_super_total_bytes(super_copy, |
4988 | round_down(old_total - diff, fs_info->sectorsize)); | |
34441361 | 4989 | mutex_unlock(&fs_info->chunk_mutex); |
2196d6e8 | 4990 | |
2bb2e00e | 4991 | btrfs_reserve_chunk_metadata(trans, false); |
2196d6e8 MX |
4992 | /* Now btrfs_update_device() will change the on-disk size. */ |
4993 | ret = btrfs_update_device(trans, device); | |
2bb2e00e | 4994 | btrfs_trans_release_chunk_metadata(trans); |
801660b0 AJ |
4995 | if (ret < 0) { |
4996 | btrfs_abort_transaction(trans, ret); | |
4997 | btrfs_end_transaction(trans); | |
4998 | } else { | |
4999 | ret = btrfs_commit_transaction(trans); | |
5000 | } | |
8f18cf13 CM |
5001 | done: |
5002 | btrfs_free_path(path); | |
53e489bc | 5003 | if (ret) { |
34441361 | 5004 | mutex_lock(&fs_info->chunk_mutex); |
53e489bc | 5005 | btrfs_device_set_total_bytes(device, old_size); |
ebbede42 | 5006 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
53e489bc | 5007 | device->fs_devices->total_rw_bytes += diff; |
a5ed45f8 | 5008 | atomic64_add(diff, &fs_info->free_chunk_space); |
34441361 | 5009 | mutex_unlock(&fs_info->chunk_mutex); |
53e489bc | 5010 | } |
8f18cf13 CM |
5011 | return ret; |
5012 | } | |
5013 | ||
2ff7e61e | 5014 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
0b86a832 CM |
5015 | struct btrfs_key *key, |
5016 | struct btrfs_chunk *chunk, int item_size) | |
5017 | { | |
0b246afa | 5018 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
0b86a832 CM |
5019 | struct btrfs_disk_key disk_key; |
5020 | u32 array_size; | |
5021 | u8 *ptr; | |
5022 | ||
79bd3712 FM |
5023 | lockdep_assert_held(&fs_info->chunk_mutex); |
5024 | ||
0b86a832 | 5025 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 5026 | if (array_size + item_size + sizeof(disk_key) |
79bd3712 | 5027 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) |
0b86a832 CM |
5028 | return -EFBIG; |
5029 | ||
5030 | ptr = super_copy->sys_chunk_array + array_size; | |
5031 | btrfs_cpu_key_to_disk(&disk_key, key); | |
5032 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
5033 | ptr += sizeof(disk_key); | |
5034 | memcpy(ptr, chunk, item_size); | |
5035 | item_size += sizeof(disk_key); | |
5036 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 | 5037 | |
0b86a832 CM |
5038 | return 0; |
5039 | } | |
5040 | ||
73c5de00 AJ |
5041 | /* |
5042 | * sort the devices in descending order by max_avail, total_avail | |
5043 | */ | |
5044 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 5045 | { |
73c5de00 AJ |
5046 | const struct btrfs_device_info *di_a = a; |
5047 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 5048 | |
73c5de00 | 5049 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 5050 | return -1; |
73c5de00 | 5051 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 5052 | return 1; |
73c5de00 AJ |
5053 | if (di_a->total_avail > di_b->total_avail) |
5054 | return -1; | |
5055 | if (di_a->total_avail < di_b->total_avail) | |
5056 | return 1; | |
5057 | return 0; | |
b2117a39 | 5058 | } |
0b86a832 | 5059 | |
53b381b3 DW |
5060 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5061 | { | |
ffe2d203 | 5062 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
5063 | return; |
5064 | ||
ceda0864 | 5065 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
5066 | } |
5067 | ||
cfbb825c DS |
5068 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
5069 | { | |
5070 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) | |
5071 | return; | |
5072 | ||
5073 | btrfs_set_fs_incompat(info, RAID1C34); | |
5074 | } | |
5075 | ||
4f2bafe8 | 5076 | /* |
f6f39f7a | 5077 | * Structure used internally for btrfs_create_chunk() function. |
4f2bafe8 NA |
5078 | * Wraps needed parameters. |
5079 | */ | |
5080 | struct alloc_chunk_ctl { | |
5081 | u64 start; | |
5082 | u64 type; | |
5083 | /* Total number of stripes to allocate */ | |
5084 | int num_stripes; | |
5085 | /* sub_stripes info for map */ | |
5086 | int sub_stripes; | |
5087 | /* Stripes per device */ | |
5088 | int dev_stripes; | |
5089 | /* Maximum number of devices to use */ | |
5090 | int devs_max; | |
5091 | /* Minimum number of devices to use */ | |
5092 | int devs_min; | |
5093 | /* ndevs has to be a multiple of this */ | |
5094 | int devs_increment; | |
5095 | /* Number of copies */ | |
5096 | int ncopies; | |
5097 | /* Number of stripes worth of bytes to store parity information */ | |
5098 | int nparity; | |
5099 | u64 max_stripe_size; | |
5100 | u64 max_chunk_size; | |
6aafb303 | 5101 | u64 dev_extent_min; |
4f2bafe8 NA |
5102 | u64 stripe_size; |
5103 | u64 chunk_size; | |
5104 | int ndevs; | |
5105 | }; | |
5106 | ||
27c314d5 NA |
5107 | static void init_alloc_chunk_ctl_policy_regular( |
5108 | struct btrfs_fs_devices *fs_devices, | |
5109 | struct alloc_chunk_ctl *ctl) | |
5110 | { | |
f6fca391 | 5111 | struct btrfs_space_info *space_info; |
27c314d5 | 5112 | |
f6fca391 SR |
5113 | space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); |
5114 | ASSERT(space_info); | |
5115 | ||
5116 | ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); | |
5117 | ctl->max_stripe_size = ctl->max_chunk_size; | |
5118 | ||
5119 | if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) | |
5120 | ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); | |
27c314d5 NA |
5121 | |
5122 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5123 | ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), |
27c314d5 | 5124 | ctl->max_chunk_size); |
a97699d1 | 5125 | ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT; |
27c314d5 NA |
5126 | } |
5127 | ||
1cd6121f NA |
5128 | static void init_alloc_chunk_ctl_policy_zoned( |
5129 | struct btrfs_fs_devices *fs_devices, | |
5130 | struct alloc_chunk_ctl *ctl) | |
5131 | { | |
5132 | u64 zone_size = fs_devices->fs_info->zone_size; | |
5133 | u64 limit; | |
5134 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; | |
5135 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; | |
5136 | u64 min_chunk_size = min_data_stripes * zone_size; | |
5137 | u64 type = ctl->type; | |
5138 | ||
5139 | ctl->max_stripe_size = zone_size; | |
5140 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
5141 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, | |
5142 | zone_size); | |
5143 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
5144 | ctl->max_chunk_size = ctl->max_stripe_size; | |
5145 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
5146 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; | |
5147 | ctl->devs_max = min_t(int, ctl->devs_max, | |
5148 | BTRFS_MAX_DEVS_SYS_CHUNK); | |
bb05b298 AB |
5149 | } else { |
5150 | BUG(); | |
1cd6121f NA |
5151 | } |
5152 | ||
5153 | /* We don't want a chunk larger than 10% of writable space */ | |
428c8e03 | 5154 | limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), |
1cd6121f NA |
5155 | zone_size), |
5156 | min_chunk_size); | |
5157 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); | |
5158 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; | |
5159 | } | |
5160 | ||
27c314d5 NA |
5161 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
5162 | struct alloc_chunk_ctl *ctl) | |
5163 | { | |
5164 | int index = btrfs_bg_flags_to_raid_index(ctl->type); | |
5165 | ||
5166 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
5167 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; | |
5168 | ctl->devs_max = btrfs_raid_array[index].devs_max; | |
5169 | if (!ctl->devs_max) | |
5170 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); | |
5171 | ctl->devs_min = btrfs_raid_array[index].devs_min; | |
5172 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; | |
5173 | ctl->ncopies = btrfs_raid_array[index].ncopies; | |
5174 | ctl->nparity = btrfs_raid_array[index].nparity; | |
5175 | ctl->ndevs = 0; | |
5176 | ||
5177 | switch (fs_devices->chunk_alloc_policy) { | |
5178 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5179 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); | |
5180 | break; | |
1cd6121f NA |
5181 | case BTRFS_CHUNK_ALLOC_ZONED: |
5182 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); | |
5183 | break; | |
27c314d5 NA |
5184 | default: |
5185 | BUG(); | |
5186 | } | |
5187 | } | |
5188 | ||
560156cb NA |
5189 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
5190 | struct alloc_chunk_ctl *ctl, | |
5191 | struct btrfs_device_info *devices_info) | |
b2117a39 | 5192 | { |
560156cb | 5193 | struct btrfs_fs_info *info = fs_devices->fs_info; |
ebcc9301 | 5194 | struct btrfs_device *device; |
73c5de00 | 5195 | u64 total_avail; |
560156cb | 5196 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
73c5de00 | 5197 | int ret; |
560156cb NA |
5198 | int ndevs = 0; |
5199 | u64 max_avail; | |
5200 | u64 dev_offset; | |
0cad8a11 | 5201 | |
9f680ce0 | 5202 | /* |
73c5de00 AJ |
5203 | * in the first pass through the devices list, we gather information |
5204 | * about the available holes on each device. | |
9f680ce0 | 5205 | */ |
ebcc9301 | 5206 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
ebbede42 | 5207 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
31b1a2bd | 5208 | WARN(1, KERN_ERR |
efe120a0 | 5209 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
5210 | continue; |
5211 | } | |
b2117a39 | 5212 | |
e12c9621 AJ |
5213 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
5214 | &device->dev_state) || | |
401e29c1 | 5215 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
73c5de00 | 5216 | continue; |
b2117a39 | 5217 | |
73c5de00 AJ |
5218 | if (device->total_bytes > device->bytes_used) |
5219 | total_avail = device->total_bytes - device->bytes_used; | |
5220 | else | |
5221 | total_avail = 0; | |
38c01b96 | 5222 | |
5223 | /* If there is no space on this device, skip it. */ | |
6aafb303 | 5224 | if (total_avail < ctl->dev_extent_min) |
38c01b96 | 5225 | continue; |
b2117a39 | 5226 | |
560156cb NA |
5227 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
5228 | &max_avail); | |
73c5de00 | 5229 | if (ret && ret != -ENOSPC) |
560156cb | 5230 | return ret; |
b2117a39 | 5231 | |
73c5de00 | 5232 | if (ret == 0) |
560156cb | 5233 | max_avail = dev_extent_want; |
b2117a39 | 5234 | |
6aafb303 | 5235 | if (max_avail < ctl->dev_extent_min) { |
4117f207 QW |
5236 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
5237 | btrfs_debug(info, | |
560156cb | 5238 | "%s: devid %llu has no free space, have=%llu want=%llu", |
4117f207 | 5239 | __func__, device->devid, max_avail, |
6aafb303 | 5240 | ctl->dev_extent_min); |
73c5de00 | 5241 | continue; |
4117f207 | 5242 | } |
b2117a39 | 5243 | |
063d006f ES |
5244 | if (ndevs == fs_devices->rw_devices) { |
5245 | WARN(1, "%s: found more than %llu devices\n", | |
5246 | __func__, fs_devices->rw_devices); | |
5247 | break; | |
5248 | } | |
73c5de00 AJ |
5249 | devices_info[ndevs].dev_offset = dev_offset; |
5250 | devices_info[ndevs].max_avail = max_avail; | |
5251 | devices_info[ndevs].total_avail = total_avail; | |
5252 | devices_info[ndevs].dev = device; | |
5253 | ++ndevs; | |
5254 | } | |
560156cb | 5255 | ctl->ndevs = ndevs; |
b2117a39 | 5256 | |
73c5de00 AJ |
5257 | /* |
5258 | * now sort the devices by hole size / available space | |
5259 | */ | |
560156cb | 5260 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
73c5de00 | 5261 | btrfs_cmp_device_info, NULL); |
b2117a39 | 5262 | |
560156cb NA |
5263 | return 0; |
5264 | } | |
5265 | ||
5badf512 NA |
5266 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
5267 | struct btrfs_device_info *devices_info) | |
5268 | { | |
5269 | /* Number of stripes that count for block group size */ | |
5270 | int data_stripes; | |
5271 | ||
5272 | /* | |
5273 | * The primary goal is to maximize the number of stripes, so use as | |
5274 | * many devices as possible, even if the stripes are not maximum sized. | |
5275 | * | |
5276 | * The DUP profile stores more than one stripe per device, the | |
5277 | * max_avail is the total size so we have to adjust. | |
5278 | */ | |
5279 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, | |
5280 | ctl->dev_stripes); | |
5281 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5282 | ||
5283 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ | |
5284 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5285 | ||
5286 | /* | |
5287 | * Use the number of data stripes to figure out how big this chunk is | |
5288 | * really going to be in terms of logical address space, and compare | |
5289 | * that answer with the max chunk size. If it's higher, we try to | |
5290 | * reduce stripe_size. | |
5291 | */ | |
5292 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5293 | /* | |
5294 | * Reduce stripe_size, round it up to a 16MB boundary again and | |
5295 | * then use it, unless it ends up being even bigger than the | |
5296 | * previous value we had already. | |
5297 | */ | |
5298 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, | |
5299 | data_stripes), SZ_16M), | |
5300 | ctl->stripe_size); | |
5301 | } | |
5302 | ||
5da431b7 QW |
5303 | /* Stripe size should not go beyond 1G. */ |
5304 | ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); | |
5305 | ||
5badf512 NA |
5306 | /* Align to BTRFS_STRIPE_LEN */ |
5307 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); | |
5308 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5309 | ||
5310 | return 0; | |
5311 | } | |
5312 | ||
1cd6121f NA |
5313 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
5314 | struct btrfs_device_info *devices_info) | |
5315 | { | |
5316 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; | |
5317 | /* Number of stripes that count for block group size */ | |
5318 | int data_stripes; | |
5319 | ||
5320 | /* | |
5321 | * It should hold because: | |
5322 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes | |
5323 | */ | |
5324 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); | |
5325 | ||
5326 | ctl->stripe_size = zone_size; | |
5327 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5328 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5329 | ||
5330 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ | |
5331 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { | |
5332 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, | |
5333 | ctl->stripe_size) + ctl->nparity, | |
5334 | ctl->dev_stripes); | |
5335 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; | |
5336 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; | |
5337 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); | |
5338 | } | |
5339 | ||
5340 | ctl->chunk_size = ctl->stripe_size * data_stripes; | |
5341 | ||
5342 | return 0; | |
5343 | } | |
5344 | ||
5badf512 NA |
5345 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
5346 | struct alloc_chunk_ctl *ctl, | |
5347 | struct btrfs_device_info *devices_info) | |
5348 | { | |
5349 | struct btrfs_fs_info *info = fs_devices->fs_info; | |
5350 | ||
5351 | /* | |
5352 | * Round down to number of usable stripes, devs_increment can be any | |
5353 | * number so we can't use round_down() that requires power of 2, while | |
5354 | * rounddown is safe. | |
5355 | */ | |
5356 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); | |
5357 | ||
5358 | if (ctl->ndevs < ctl->devs_min) { | |
5359 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { | |
5360 | btrfs_debug(info, | |
5361 | "%s: not enough devices with free space: have=%d minimum required=%d", | |
5362 | __func__, ctl->ndevs, ctl->devs_min); | |
5363 | } | |
5364 | return -ENOSPC; | |
5365 | } | |
5366 | ||
5367 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); | |
5368 | ||
5369 | switch (fs_devices->chunk_alloc_policy) { | |
5370 | case BTRFS_CHUNK_ALLOC_REGULAR: | |
5371 | return decide_stripe_size_regular(ctl, devices_info); | |
1cd6121f NA |
5372 | case BTRFS_CHUNK_ALLOC_ZONED: |
5373 | return decide_stripe_size_zoned(ctl, devices_info); | |
5badf512 NA |
5374 | default: |
5375 | BUG(); | |
5376 | } | |
5377 | } | |
5378 | ||
79bd3712 | 5379 | static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, |
dce580ca NA |
5380 | struct alloc_chunk_ctl *ctl, |
5381 | struct btrfs_device_info *devices_info) | |
560156cb NA |
5382 | { |
5383 | struct btrfs_fs_info *info = trans->fs_info; | |
560156cb NA |
5384 | struct map_lookup *map = NULL; |
5385 | struct extent_map_tree *em_tree; | |
79bd3712 | 5386 | struct btrfs_block_group *block_group; |
560156cb | 5387 | struct extent_map *em; |
dce580ca NA |
5388 | u64 start = ctl->start; |
5389 | u64 type = ctl->type; | |
560156cb NA |
5390 | int ret; |
5391 | int i; | |
5392 | int j; | |
5393 | ||
dce580ca NA |
5394 | map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS); |
5395 | if (!map) | |
79bd3712 | 5396 | return ERR_PTR(-ENOMEM); |
dce580ca | 5397 | map->num_stripes = ctl->num_stripes; |
560156cb | 5398 | |
dce580ca NA |
5399 | for (i = 0; i < ctl->ndevs; ++i) { |
5400 | for (j = 0; j < ctl->dev_stripes; ++j) { | |
5401 | int s = i * ctl->dev_stripes + j; | |
73c5de00 AJ |
5402 | map->stripes[s].dev = devices_info[i].dev; |
5403 | map->stripes[s].physical = devices_info[i].dev_offset + | |
dce580ca | 5404 | j * ctl->stripe_size; |
6324fbf3 | 5405 | } |
6324fbf3 | 5406 | } |
500ceed8 NB |
5407 | map->io_align = BTRFS_STRIPE_LEN; |
5408 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 5409 | map->type = type; |
dce580ca | 5410 | map->sub_stripes = ctl->sub_stripes; |
0b86a832 | 5411 | |
dce580ca | 5412 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
1abe9b8a | 5413 | |
172ddd60 | 5414 | em = alloc_extent_map(); |
2b82032c | 5415 | if (!em) { |
298a8f9c | 5416 | kfree(map); |
79bd3712 | 5417 | return ERR_PTR(-ENOMEM); |
593060d7 | 5418 | } |
298a8f9c | 5419 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 5420 | em->map_lookup = map; |
2b82032c | 5421 | em->start = start; |
dce580ca | 5422 | em->len = ctl->chunk_size; |
2b82032c YZ |
5423 | em->block_start = 0; |
5424 | em->block_len = em->len; | |
dce580ca | 5425 | em->orig_block_len = ctl->stripe_size; |
593060d7 | 5426 | |
c8bf1b67 | 5427 | em_tree = &info->mapping_tree; |
890871be | 5428 | write_lock(&em_tree->lock); |
09a2a8f9 | 5429 | ret = add_extent_mapping(em_tree, em, 0); |
0f5d42b2 | 5430 | if (ret) { |
1efb72a3 | 5431 | write_unlock(&em_tree->lock); |
0f5d42b2 | 5432 | free_extent_map(em); |
79bd3712 | 5433 | return ERR_PTR(ret); |
0f5d42b2 | 5434 | } |
1efb72a3 NB |
5435 | write_unlock(&em_tree->lock); |
5436 | ||
5758d1bd | 5437 | block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); |
79bd3712 | 5438 | if (IS_ERR(block_group)) |
6df9a95e | 5439 | goto error_del_extent; |
2b82032c | 5440 | |
bbbf7243 NB |
5441 | for (i = 0; i < map->num_stripes; i++) { |
5442 | struct btrfs_device *dev = map->stripes[i].dev; | |
5443 | ||
4f2bafe8 | 5444 | btrfs_device_set_bytes_used(dev, |
dce580ca | 5445 | dev->bytes_used + ctl->stripe_size); |
bbbf7243 NB |
5446 | if (list_empty(&dev->post_commit_list)) |
5447 | list_add_tail(&dev->post_commit_list, | |
5448 | &trans->transaction->dev_update_list); | |
5449 | } | |
43530c46 | 5450 | |
dce580ca | 5451 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
4f2bafe8 | 5452 | &info->free_chunk_space); |
1c116187 | 5453 | |
0f5d42b2 | 5454 | free_extent_map(em); |
0b246afa | 5455 | check_raid56_incompat_flag(info, type); |
cfbb825c | 5456 | check_raid1c34_incompat_flag(info, type); |
53b381b3 | 5457 | |
79bd3712 | 5458 | return block_group; |
b2117a39 | 5459 | |
6df9a95e | 5460 | error_del_extent: |
0f5d42b2 JB |
5461 | write_lock(&em_tree->lock); |
5462 | remove_extent_mapping(em_tree, em); | |
5463 | write_unlock(&em_tree->lock); | |
5464 | ||
5465 | /* One for our allocation */ | |
5466 | free_extent_map(em); | |
5467 | /* One for the tree reference */ | |
5468 | free_extent_map(em); | |
dce580ca | 5469 | |
79bd3712 | 5470 | return block_group; |
dce580ca NA |
5471 | } |
5472 | ||
f6f39f7a | 5473 | struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, |
79bd3712 | 5474 | u64 type) |
dce580ca NA |
5475 | { |
5476 | struct btrfs_fs_info *info = trans->fs_info; | |
5477 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
5478 | struct btrfs_device_info *devices_info = NULL; | |
5479 | struct alloc_chunk_ctl ctl; | |
79bd3712 | 5480 | struct btrfs_block_group *block_group; |
dce580ca NA |
5481 | int ret; |
5482 | ||
11c67b1a NB |
5483 | lockdep_assert_held(&info->chunk_mutex); |
5484 | ||
dce580ca NA |
5485 | if (!alloc_profile_is_valid(type, 0)) { |
5486 | ASSERT(0); | |
79bd3712 | 5487 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5488 | } |
5489 | ||
5490 | if (list_empty(&fs_devices->alloc_list)) { | |
5491 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) | |
5492 | btrfs_debug(info, "%s: no writable device", __func__); | |
79bd3712 | 5493 | return ERR_PTR(-ENOSPC); |
dce580ca NA |
5494 | } |
5495 | ||
5496 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
5497 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); | |
5498 | ASSERT(0); | |
79bd3712 | 5499 | return ERR_PTR(-EINVAL); |
dce580ca NA |
5500 | } |
5501 | ||
11c67b1a | 5502 | ctl.start = find_next_chunk(info); |
dce580ca NA |
5503 | ctl.type = type; |
5504 | init_alloc_chunk_ctl(fs_devices, &ctl); | |
5505 | ||
5506 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
5507 | GFP_NOFS); | |
5508 | if (!devices_info) | |
79bd3712 | 5509 | return ERR_PTR(-ENOMEM); |
dce580ca NA |
5510 | |
5511 | ret = gather_device_info(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5512 | if (ret < 0) { |
5513 | block_group = ERR_PTR(ret); | |
dce580ca | 5514 | goto out; |
79bd3712 | 5515 | } |
dce580ca NA |
5516 | |
5517 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); | |
79bd3712 FM |
5518 | if (ret < 0) { |
5519 | block_group = ERR_PTR(ret); | |
dce580ca | 5520 | goto out; |
79bd3712 | 5521 | } |
dce580ca | 5522 | |
79bd3712 | 5523 | block_group = create_chunk(trans, &ctl, devices_info); |
dce580ca NA |
5524 | |
5525 | out: | |
b2117a39 | 5526 | kfree(devices_info); |
79bd3712 | 5527 | return block_group; |
2b82032c YZ |
5528 | } |
5529 | ||
79bd3712 FM |
5530 | /* |
5531 | * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the | |
5532 | * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system | |
5533 | * chunks. | |
5534 | * | |
5535 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation | |
5536 | * phases. | |
5537 | */ | |
5538 | int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, | |
5539 | struct btrfs_block_group *bg) | |
5540 | { | |
5541 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
79bd3712 FM |
5542 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
5543 | struct btrfs_key key; | |
5544 | struct btrfs_chunk *chunk; | |
5545 | struct btrfs_stripe *stripe; | |
5546 | struct extent_map *em; | |
5547 | struct map_lookup *map; | |
5548 | size_t item_size; | |
5549 | int i; | |
5550 | int ret; | |
5551 | ||
5552 | /* | |
5553 | * We take the chunk_mutex for 2 reasons: | |
5554 | * | |
5555 | * 1) Updates and insertions in the chunk btree must be done while holding | |
5556 | * the chunk_mutex, as well as updating the system chunk array in the | |
5557 | * superblock. See the comment on top of btrfs_chunk_alloc() for the | |
5558 | * details; | |
5559 | * | |
5560 | * 2) To prevent races with the final phase of a device replace operation | |
5561 | * that replaces the device object associated with the map's stripes, | |
5562 | * because the device object's id can change at any time during that | |
5563 | * final phase of the device replace operation | |
5564 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the | |
5565 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, | |
5566 | * which would cause a failure when updating the device item, which does | |
5567 | * not exists, or persisting a stripe of the chunk item with such ID. | |
5568 | * Here we can't use the device_list_mutex because our caller already | |
5569 | * has locked the chunk_mutex, and the final phase of device replace | |
5570 | * acquires both mutexes - first the device_list_mutex and then the | |
5571 | * chunk_mutex. Using any of those two mutexes protects us from a | |
5572 | * concurrent device replace. | |
5573 | */ | |
5574 | lockdep_assert_held(&fs_info->chunk_mutex); | |
5575 | ||
5576 | em = btrfs_get_chunk_map(fs_info, bg->start, bg->length); | |
5577 | if (IS_ERR(em)) { | |
5578 | ret = PTR_ERR(em); | |
5579 | btrfs_abort_transaction(trans, ret); | |
5580 | return ret; | |
5581 | } | |
5582 | ||
5583 | map = em->map_lookup; | |
5584 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
5585 | ||
5586 | chunk = kzalloc(item_size, GFP_NOFS); | |
5587 | if (!chunk) { | |
5588 | ret = -ENOMEM; | |
5589 | btrfs_abort_transaction(trans, ret); | |
50460e37 | 5590 | goto out; |
2b82032c YZ |
5591 | } |
5592 | ||
79bd3712 FM |
5593 | for (i = 0; i < map->num_stripes; i++) { |
5594 | struct btrfs_device *device = map->stripes[i].dev; | |
5595 | ||
5596 | ret = btrfs_update_device(trans, device); | |
5597 | if (ret) | |
5598 | goto out; | |
5599 | } | |
5600 | ||
2b82032c | 5601 | stripe = &chunk->stripe; |
6df9a95e | 5602 | for (i = 0; i < map->num_stripes; i++) { |
79bd3712 FM |
5603 | struct btrfs_device *device = map->stripes[i].dev; |
5604 | const u64 dev_offset = map->stripes[i].physical; | |
0b86a832 | 5605 | |
e17cade2 CM |
5606 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
5607 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
5608 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 5609 | stripe++; |
0b86a832 CM |
5610 | } |
5611 | ||
79bd3712 | 5612 | btrfs_set_stack_chunk_length(chunk, bg->length); |
fd51eb2f | 5613 | btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); |
a97699d1 | 5614 | btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); |
2b82032c YZ |
5615 | btrfs_set_stack_chunk_type(chunk, map->type); |
5616 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
a97699d1 QW |
5617 | btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); |
5618 | btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); | |
0b246afa | 5619 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 5620 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 5621 | |
2b82032c YZ |
5622 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
5623 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
79bd3712 | 5624 | key.offset = bg->start; |
0b86a832 | 5625 | |
2b82032c | 5626 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
79bd3712 FM |
5627 | if (ret) |
5628 | goto out; | |
5629 | ||
3349b57f | 5630 | set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); |
79bd3712 FM |
5631 | |
5632 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2ff7e61e | 5633 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
79bd3712 FM |
5634 | if (ret) |
5635 | goto out; | |
8f18cf13 | 5636 | } |
1abe9b8a | 5637 | |
6df9a95e | 5638 | out: |
0b86a832 | 5639 | kfree(chunk); |
6df9a95e | 5640 | free_extent_map(em); |
4ed1d16e | 5641 | return ret; |
2b82032c | 5642 | } |
0b86a832 | 5643 | |
6f8e0fc7 | 5644 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
2b82032c | 5645 | { |
6f8e0fc7 | 5646 | struct btrfs_fs_info *fs_info = trans->fs_info; |
2b82032c | 5647 | u64 alloc_profile; |
79bd3712 FM |
5648 | struct btrfs_block_group *meta_bg; |
5649 | struct btrfs_block_group *sys_bg; | |
5650 | ||
5651 | /* | |
5652 | * When adding a new device for sprouting, the seed device is read-only | |
5653 | * so we must first allocate a metadata and a system chunk. But before | |
5654 | * adding the block group items to the extent, device and chunk btrees, | |
5655 | * we must first: | |
5656 | * | |
5657 | * 1) Create both chunks without doing any changes to the btrees, as | |
5658 | * otherwise we would get -ENOSPC since the block groups from the | |
5659 | * seed device are read-only; | |
5660 | * | |
5661 | * 2) Add the device item for the new sprout device - finishing the setup | |
5662 | * of a new block group requires updating the device item in the chunk | |
5663 | * btree, so it must exist when we attempt to do it. The previous step | |
5664 | * ensures this does not fail with -ENOSPC. | |
5665 | * | |
5666 | * After that we can add the block group items to their btrees: | |
5667 | * update existing device item in the chunk btree, add a new block group | |
5668 | * item to the extent btree, add a new chunk item to the chunk btree and | |
5669 | * finally add the new device extent items to the devices btree. | |
5670 | */ | |
2b82032c | 5671 | |
1b86826d | 5672 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
f6f39f7a | 5673 | meta_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5674 | if (IS_ERR(meta_bg)) |
5675 | return PTR_ERR(meta_bg); | |
2b82032c | 5676 | |
1b86826d | 5677 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
f6f39f7a | 5678 | sys_bg = btrfs_create_chunk(trans, alloc_profile); |
79bd3712 FM |
5679 | if (IS_ERR(sys_bg)) |
5680 | return PTR_ERR(sys_bg); | |
5681 | ||
5682 | return 0; | |
2b82032c YZ |
5683 | } |
5684 | ||
d20983b4 MX |
5685 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5686 | { | |
fc9a2ac7 | 5687 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
2b82032c | 5688 | |
fc9a2ac7 | 5689 | return btrfs_raid_array[index].tolerated_failures; |
2b82032c YZ |
5690 | } |
5691 | ||
a09f23c3 | 5692 | bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
2b82032c YZ |
5693 | { |
5694 | struct extent_map *em; | |
5695 | struct map_lookup *map; | |
d20983b4 | 5696 | int miss_ndevs = 0; |
2b82032c | 5697 | int i; |
a09f23c3 | 5698 | bool ret = true; |
2b82032c | 5699 | |
60ca842e | 5700 | em = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
592d92ee | 5701 | if (IS_ERR(em)) |
a09f23c3 | 5702 | return false; |
2b82032c | 5703 | |
95617d69 | 5704 | map = em->map_lookup; |
2b82032c | 5705 | for (i = 0; i < map->num_stripes; i++) { |
e6e674bd AJ |
5706 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
5707 | &map->stripes[i].dev->dev_state)) { | |
d20983b4 MX |
5708 | miss_ndevs++; |
5709 | continue; | |
5710 | } | |
ebbede42 AJ |
5711 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
5712 | &map->stripes[i].dev->dev_state)) { | |
a09f23c3 | 5713 | ret = false; |
d20983b4 | 5714 | goto end; |
2b82032c YZ |
5715 | } |
5716 | } | |
d20983b4 MX |
5717 | |
5718 | /* | |
a09f23c3 AJ |
5719 | * If the number of missing devices is larger than max errors, we can |
5720 | * not write the data into that chunk successfully. | |
d20983b4 MX |
5721 | */ |
5722 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
a09f23c3 | 5723 | ret = false; |
d20983b4 | 5724 | end: |
0b86a832 | 5725 | free_extent_map(em); |
a09f23c3 | 5726 | return ret; |
0b86a832 CM |
5727 | } |
5728 | ||
c8bf1b67 | 5729 | void btrfs_mapping_tree_free(struct extent_map_tree *tree) |
0b86a832 CM |
5730 | { |
5731 | struct extent_map *em; | |
5732 | ||
d397712b | 5733 | while (1) { |
c8bf1b67 DS |
5734 | write_lock(&tree->lock); |
5735 | em = lookup_extent_mapping(tree, 0, (u64)-1); | |
0b86a832 | 5736 | if (em) |
c8bf1b67 DS |
5737 | remove_extent_mapping(tree, em); |
5738 | write_unlock(&tree->lock); | |
0b86a832 CM |
5739 | if (!em) |
5740 | break; | |
0b86a832 CM |
5741 | /* once for us */ |
5742 | free_extent_map(em); | |
5743 | /* once for the tree */ | |
5744 | free_extent_map(em); | |
5745 | } | |
5746 | } | |
5747 | ||
5d964051 | 5748 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e CM |
5749 | { |
5750 | struct extent_map *em; | |
5751 | struct map_lookup *map; | |
6d322b48 QW |
5752 | enum btrfs_raid_types index; |
5753 | int ret = 1; | |
f188591e | 5754 | |
60ca842e | 5755 | em = btrfs_get_chunk_map(fs_info, logical, len); |
592d92ee LB |
5756 | if (IS_ERR(em)) |
5757 | /* | |
5758 | * We could return errors for these cases, but that could get | |
5759 | * ugly and we'd probably do the same thing which is just not do | |
5760 | * anything else and exit, so return 1 so the callers don't try | |
5761 | * to use other copies. | |
5762 | */ | |
fb7669b5 | 5763 | return 1; |
fb7669b5 | 5764 | |
95617d69 | 5765 | map = em->map_lookup; |
6d322b48 QW |
5766 | index = btrfs_bg_flags_to_raid_index(map->type); |
5767 | ||
5768 | /* Non-RAID56, use their ncopies from btrfs_raid_array. */ | |
5769 | if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
5770 | ret = btrfs_raid_array[index].ncopies; | |
53b381b3 DW |
5771 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5772 | ret = 2; | |
5773 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
8810f751 LB |
5774 | /* |
5775 | * There could be two corrupted data stripes, we need | |
5776 | * to loop retry in order to rebuild the correct data. | |
e7e02096 | 5777 | * |
8810f751 LB |
5778 | * Fail a stripe at a time on every retry except the |
5779 | * stripe under reconstruction. | |
5780 | */ | |
5781 | ret = map->num_stripes; | |
f188591e | 5782 | free_extent_map(em); |
f188591e CM |
5783 | return ret; |
5784 | } | |
5785 | ||
2ff7e61e | 5786 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
53b381b3 DW |
5787 | u64 logical) |
5788 | { | |
5789 | struct extent_map *em; | |
5790 | struct map_lookup *map; | |
0b246afa | 5791 | unsigned long len = fs_info->sectorsize; |
53b381b3 | 5792 | |
b036f479 QW |
5793 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
5794 | return len; | |
5795 | ||
60ca842e | 5796 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5797 | |
69f03f13 NB |
5798 | if (!WARN_ON(IS_ERR(em))) { |
5799 | map = em->map_lookup; | |
5800 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
a97699d1 | 5801 | len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT; |
69f03f13 NB |
5802 | free_extent_map(em); |
5803 | } | |
53b381b3 DW |
5804 | return len; |
5805 | } | |
5806 | ||
e4ff5fb5 | 5807 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
53b381b3 DW |
5808 | { |
5809 | struct extent_map *em; | |
5810 | struct map_lookup *map; | |
53b381b3 DW |
5811 | int ret = 0; |
5812 | ||
b036f479 QW |
5813 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
5814 | return 0; | |
5815 | ||
60ca842e | 5816 | em = btrfs_get_chunk_map(fs_info, logical, len); |
53b381b3 | 5817 | |
69f03f13 NB |
5818 | if(!WARN_ON(IS_ERR(em))) { |
5819 | map = em->map_lookup; | |
5820 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5821 | ret = 1; | |
5822 | free_extent_map(em); | |
5823 | } | |
53b381b3 DW |
5824 | return ret; |
5825 | } | |
5826 | ||
30d9861f | 5827 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
99f92a7c | 5828 | struct map_lookup *map, int first, |
8ba0ae78 | 5829 | int dev_replace_is_ongoing) |
dfe25020 CM |
5830 | { |
5831 | int i; | |
99f92a7c | 5832 | int num_stripes; |
8ba0ae78 | 5833 | int preferred_mirror; |
30d9861f SB |
5834 | int tolerance; |
5835 | struct btrfs_device *srcdev; | |
5836 | ||
99f92a7c | 5837 | ASSERT((map->type & |
c7369b3f | 5838 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
99f92a7c AJ |
5839 | |
5840 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5841 | num_stripes = map->sub_stripes; | |
5842 | else | |
5843 | num_stripes = map->num_stripes; | |
5844 | ||
33fd2f71 AJ |
5845 | switch (fs_info->fs_devices->read_policy) { |
5846 | default: | |
5847 | /* Shouldn't happen, just warn and use pid instead of failing */ | |
5848 | btrfs_warn_rl(fs_info, | |
5849 | "unknown read_policy type %u, reset to pid", | |
5850 | fs_info->fs_devices->read_policy); | |
5851 | fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; | |
5852 | fallthrough; | |
5853 | case BTRFS_READ_POLICY_PID: | |
5854 | preferred_mirror = first + (current->pid % num_stripes); | |
5855 | break; | |
5856 | } | |
8ba0ae78 | 5857 | |
30d9861f SB |
5858 | if (dev_replace_is_ongoing && |
5859 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5860 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5861 | srcdev = fs_info->dev_replace.srcdev; | |
5862 | else | |
5863 | srcdev = NULL; | |
5864 | ||
5865 | /* | |
5866 | * try to avoid the drive that is the source drive for a | |
5867 | * dev-replace procedure, only choose it if no other non-missing | |
5868 | * mirror is available | |
5869 | */ | |
5870 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
8ba0ae78 AJ |
5871 | if (map->stripes[preferred_mirror].dev->bdev && |
5872 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) | |
5873 | return preferred_mirror; | |
99f92a7c | 5874 | for (i = first; i < first + num_stripes; i++) { |
30d9861f SB |
5875 | if (map->stripes[i].dev->bdev && |
5876 | (tolerance || map->stripes[i].dev != srcdev)) | |
5877 | return i; | |
5878 | } | |
dfe25020 | 5879 | } |
30d9861f | 5880 | |
dfe25020 CM |
5881 | /* we couldn't find one that doesn't fail. Just return something |
5882 | * and the io error handling code will clean up eventually | |
5883 | */ | |
8ba0ae78 | 5884 | return preferred_mirror; |
dfe25020 CM |
5885 | } |
5886 | ||
731ccf15 | 5887 | static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, |
1faf3885 | 5888 | u16 total_stripes) |
6e9606d2 | 5889 | { |
4ced85f8 QW |
5890 | struct btrfs_io_context *bioc; |
5891 | ||
5892 | bioc = kzalloc( | |
4c664611 QW |
5893 | /* The size of btrfs_io_context */ |
5894 | sizeof(struct btrfs_io_context) + | |
5895 | /* Plus the variable array for the stripes */ | |
18d758a2 | 5896 | sizeof(struct btrfs_io_stripe) * (total_stripes), |
9f0eac07 L |
5897 | GFP_NOFS); |
5898 | ||
5899 | if (!bioc) | |
5900 | return NULL; | |
6e9606d2 | 5901 | |
4c664611 | 5902 | refcount_set(&bioc->refs, 1); |
6e9606d2 | 5903 | |
731ccf15 | 5904 | bioc->fs_info = fs_info; |
1faf3885 | 5905 | bioc->replace_stripe_src = -1; |
18d758a2 | 5906 | bioc->full_stripe_logical = (u64)-1; |
608769a4 | 5907 | |
4c664611 | 5908 | return bioc; |
6e9606d2 ZL |
5909 | } |
5910 | ||
4c664611 | 5911 | void btrfs_get_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 5912 | { |
4c664611 QW |
5913 | WARN_ON(!refcount_read(&bioc->refs)); |
5914 | refcount_inc(&bioc->refs); | |
6e9606d2 ZL |
5915 | } |
5916 | ||
4c664611 | 5917 | void btrfs_put_bioc(struct btrfs_io_context *bioc) |
6e9606d2 | 5918 | { |
4c664611 | 5919 | if (!bioc) |
6e9606d2 | 5920 | return; |
4c664611 QW |
5921 | if (refcount_dec_and_test(&bioc->refs)) |
5922 | kfree(bioc); | |
6e9606d2 ZL |
5923 | } |
5924 | ||
0b3d4cd3 LB |
5925 | /* |
5926 | * Please note that, discard won't be sent to target device of device | |
5927 | * replace. | |
5928 | */ | |
a4012f06 CH |
5929 | struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, |
5930 | u64 logical, u64 *length_ret, | |
5931 | u32 *num_stripes) | |
0b3d4cd3 LB |
5932 | { |
5933 | struct extent_map *em; | |
5934 | struct map_lookup *map; | |
a4012f06 | 5935 | struct btrfs_discard_stripe *stripes; |
6b7faadd | 5936 | u64 length = *length_ret; |
0b3d4cd3 | 5937 | u64 offset; |
6ded22c1 QW |
5938 | u32 stripe_nr; |
5939 | u32 stripe_nr_end; | |
5940 | u32 stripe_cnt; | |
0b3d4cd3 | 5941 | u64 stripe_end_offset; |
0b3d4cd3 | 5942 | u64 stripe_offset; |
0b3d4cd3 LB |
5943 | u32 stripe_index; |
5944 | u32 factor = 0; | |
5945 | u32 sub_stripes = 0; | |
6ded22c1 | 5946 | u32 stripes_per_dev = 0; |
0b3d4cd3 LB |
5947 | u32 remaining_stripes = 0; |
5948 | u32 last_stripe = 0; | |
a4012f06 | 5949 | int ret; |
0b3d4cd3 LB |
5950 | int i; |
5951 | ||
60ca842e | 5952 | em = btrfs_get_chunk_map(fs_info, logical, length); |
0b3d4cd3 | 5953 | if (IS_ERR(em)) |
a4012f06 | 5954 | return ERR_CAST(em); |
0b3d4cd3 LB |
5955 | |
5956 | map = em->map_lookup; | |
a4012f06 | 5957 | |
0b3d4cd3 LB |
5958 | /* we don't discard raid56 yet */ |
5959 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5960 | ret = -EOPNOTSUPP; | |
a4012f06 | 5961 | goto out_free_map; |
a97699d1 | 5962 | } |
0b3d4cd3 LB |
5963 | |
5964 | offset = logical - em->start; | |
2d974619 | 5965 | length = min_t(u64, em->start + em->len - logical, length); |
6b7faadd | 5966 | *length_ret = length; |
0b3d4cd3 | 5967 | |
0b3d4cd3 LB |
5968 | /* |
5969 | * stripe_nr counts the total number of stripes we have to stride | |
5970 | * to get to this block | |
5971 | */ | |
a97699d1 | 5972 | stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; |
0b3d4cd3 LB |
5973 | |
5974 | /* stripe_offset is the offset of this block in its stripe */ | |
a97699d1 | 5975 | stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); |
0b3d4cd3 | 5976 | |
a97699d1 QW |
5977 | stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> |
5978 | BTRFS_STRIPE_LEN_SHIFT; | |
0b3d4cd3 | 5979 | stripe_cnt = stripe_nr_end - stripe_nr; |
a97699d1 | 5980 | stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) - |
0b3d4cd3 LB |
5981 | (offset + length); |
5982 | /* | |
5983 | * after this, stripe_nr is the number of stripes on this | |
5984 | * device we have to walk to find the data, and stripe_index is | |
5985 | * the number of our device in the stripe array | |
5986 | */ | |
a4012f06 | 5987 | *num_stripes = 1; |
0b3d4cd3 LB |
5988 | stripe_index = 0; |
5989 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5990 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5991 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5992 | sub_stripes = 1; | |
5993 | else | |
5994 | sub_stripes = map->sub_stripes; | |
5995 | ||
5996 | factor = map->num_stripes / sub_stripes; | |
a4012f06 | 5997 | *num_stripes = min_t(u64, map->num_stripes, |
0b3d4cd3 | 5998 | sub_stripes * stripe_cnt); |
6ded22c1 QW |
5999 | stripe_index = stripe_nr % factor; |
6000 | stripe_nr /= factor; | |
0b3d4cd3 | 6001 | stripe_index *= sub_stripes; |
6ded22c1 QW |
6002 | |
6003 | remaining_stripes = stripe_cnt % factor; | |
6004 | stripes_per_dev = stripe_cnt / factor; | |
6005 | last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; | |
c7369b3f | 6006 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
0b3d4cd3 | 6007 | BTRFS_BLOCK_GROUP_DUP)) { |
a4012f06 | 6008 | *num_stripes = map->num_stripes; |
0b3d4cd3 | 6009 | } else { |
6ded22c1 QW |
6010 | stripe_index = stripe_nr % map->num_stripes; |
6011 | stripe_nr /= map->num_stripes; | |
0b3d4cd3 LB |
6012 | } |
6013 | ||
a4012f06 CH |
6014 | stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); |
6015 | if (!stripes) { | |
0b3d4cd3 | 6016 | ret = -ENOMEM; |
a4012f06 | 6017 | goto out_free_map; |
0b3d4cd3 LB |
6018 | } |
6019 | ||
a4012f06 CH |
6020 | for (i = 0; i < *num_stripes; i++) { |
6021 | stripes[i].physical = | |
0b3d4cd3 | 6022 | map->stripes[stripe_index].physical + |
a97699d1 | 6023 | stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); |
a4012f06 | 6024 | stripes[i].dev = map->stripes[stripe_index].dev; |
0b3d4cd3 LB |
6025 | |
6026 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
6027 | BTRFS_BLOCK_GROUP_RAID10)) { | |
a97699d1 | 6028 | stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT; |
0b3d4cd3 LB |
6029 | |
6030 | if (i / sub_stripes < remaining_stripes) | |
a97699d1 | 6031 | stripes[i].length += BTRFS_STRIPE_LEN; |
0b3d4cd3 LB |
6032 | |
6033 | /* | |
6034 | * Special for the first stripe and | |
6035 | * the last stripe: | |
6036 | * | |
6037 | * |-------|...|-------| | |
6038 | * |----------| | |
6039 | * off end_off | |
6040 | */ | |
6041 | if (i < sub_stripes) | |
a4012f06 | 6042 | stripes[i].length -= stripe_offset; |
0b3d4cd3 LB |
6043 | |
6044 | if (stripe_index >= last_stripe && | |
6045 | stripe_index <= (last_stripe + | |
6046 | sub_stripes - 1)) | |
a4012f06 | 6047 | stripes[i].length -= stripe_end_offset; |
0b3d4cd3 LB |
6048 | |
6049 | if (i == sub_stripes - 1) | |
6050 | stripe_offset = 0; | |
6051 | } else { | |
a4012f06 | 6052 | stripes[i].length = length; |
0b3d4cd3 LB |
6053 | } |
6054 | ||
6055 | stripe_index++; | |
6056 | if (stripe_index == map->num_stripes) { | |
6057 | stripe_index = 0; | |
6058 | stripe_nr++; | |
6059 | } | |
6060 | } | |
6061 | ||
0b3d4cd3 | 6062 | free_extent_map(em); |
a4012f06 CH |
6063 | return stripes; |
6064 | out_free_map: | |
6065 | free_extent_map(em); | |
6066 | return ERR_PTR(ret); | |
0b3d4cd3 LB |
6067 | } |
6068 | ||
6143c23c NA |
6069 | static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) |
6070 | { | |
6071 | struct btrfs_block_group *cache; | |
6072 | bool ret; | |
6073 | ||
de17addc | 6074 | /* Non zoned filesystem does not use "to_copy" flag */ |
6143c23c NA |
6075 | if (!btrfs_is_zoned(fs_info)) |
6076 | return false; | |
6077 | ||
6078 | cache = btrfs_lookup_block_group(fs_info, logical); | |
6079 | ||
3349b57f | 6080 | ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); |
6143c23c NA |
6081 | |
6082 | btrfs_put_block_group(cache); | |
6083 | return ret; | |
6084 | } | |
6085 | ||
73c0f228 | 6086 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
be5c7edb | 6087 | struct btrfs_io_context *bioc, |
73c0f228 | 6088 | struct btrfs_dev_replace *dev_replace, |
6143c23c | 6089 | u64 logical, |
73c0f228 LB |
6090 | int *num_stripes_ret, int *max_errors_ret) |
6091 | { | |
73c0f228 | 6092 | u64 srcdev_devid = dev_replace->srcdev->devid; |
1faf3885 QW |
6093 | /* |
6094 | * At this stage, num_stripes is still the real number of stripes, | |
6095 | * excluding the duplicated stripes. | |
6096 | */ | |
73c0f228 | 6097 | int num_stripes = *num_stripes_ret; |
1faf3885 | 6098 | int nr_extra_stripes = 0; |
73c0f228 LB |
6099 | int max_errors = *max_errors_ret; |
6100 | int i; | |
6101 | ||
1faf3885 QW |
6102 | /* |
6103 | * A block group which has "to_copy" set will eventually be copied by | |
6104 | * the dev-replace process. We can avoid cloning IO here. | |
6105 | */ | |
6106 | if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) | |
6107 | return; | |
73c0f228 | 6108 | |
1faf3885 QW |
6109 | /* |
6110 | * Duplicate the write operations while the dev-replace procedure is | |
6111 | * running. Since the copying of the old disk to the new disk takes | |
6112 | * place at run time while the filesystem is mounted writable, the | |
6113 | * regular write operations to the old disk have to be duplicated to go | |
6114 | * to the new disk as well. | |
6115 | * | |
6116 | * Note that device->missing is handled by the caller, and that the | |
6117 | * write to the old disk is already set up in the stripes array. | |
6118 | */ | |
6119 | for (i = 0; i < num_stripes; i++) { | |
6120 | struct btrfs_io_stripe *old = &bioc->stripes[i]; | |
6121 | struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; | |
6143c23c | 6122 | |
1faf3885 QW |
6123 | if (old->dev->devid != srcdev_devid) |
6124 | continue; | |
73c0f228 | 6125 | |
1faf3885 QW |
6126 | new->physical = old->physical; |
6127 | new->dev = dev_replace->tgtdev; | |
6128 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
6129 | bioc->replace_stripe_src = i; | |
6130 | nr_extra_stripes++; | |
6131 | } | |
73c0f228 | 6132 | |
1faf3885 QW |
6133 | /* We can only have at most 2 extra nr_stripes (for DUP). */ |
6134 | ASSERT(nr_extra_stripes <= 2); | |
6135 | /* | |
6136 | * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for | |
6137 | * replace. | |
6138 | * If we have 2 extra stripes, only choose the one with smaller physical. | |
6139 | */ | |
6140 | if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { | |
6141 | struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; | |
6142 | struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; | |
73c0f228 | 6143 | |
1faf3885 QW |
6144 | /* Only DUP can have two extra stripes. */ |
6145 | ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); | |
6146 | ||
6147 | /* | |
6148 | * Swap the last stripe stripes and reduce @nr_extra_stripes. | |
6149 | * The extra stripe would still be there, but won't be accessed. | |
6150 | */ | |
6151 | if (first->physical > second->physical) { | |
6152 | swap(second->physical, first->physical); | |
6153 | swap(second->dev, first->dev); | |
6154 | nr_extra_stripes--; | |
73c0f228 LB |
6155 | } |
6156 | } | |
6157 | ||
1faf3885 QW |
6158 | *num_stripes_ret = num_stripes + nr_extra_stripes; |
6159 | *max_errors_ret = max_errors + nr_extra_stripes; | |
6160 | bioc->replace_nr_stripes = nr_extra_stripes; | |
73c0f228 LB |
6161 | } |
6162 | ||
2b19a1fe LB |
6163 | static bool need_full_stripe(enum btrfs_map_op op) |
6164 | { | |
6165 | return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS); | |
6166 | } | |
6167 | ||
f8a02dc6 | 6168 | static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op, |
6ded22c1 | 6169 | u64 offset, u32 *stripe_nr, u64 *stripe_offset, |
f8a02dc6 | 6170 | u64 *full_stripe_start) |
5f141126 | 6171 | { |
5f141126 NB |
6172 | ASSERT(op != BTRFS_MAP_DISCARD); |
6173 | ||
cc353a8b | 6174 | /* |
f8a02dc6 CH |
6175 | * Stripe_nr is the stripe where this block falls. stripe_offset is |
6176 | * the offset of this block in its stripe. | |
cc353a8b | 6177 | */ |
a97699d1 QW |
6178 | *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; |
6179 | *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; | |
f8a02dc6 | 6180 | ASSERT(*stripe_offset < U32_MAX); |
5f141126 | 6181 | |
f8a02dc6 | 6182 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
a97699d1 QW |
6183 | unsigned long full_stripe_len = nr_data_stripes(map) << |
6184 | BTRFS_STRIPE_LEN_SHIFT; | |
5f141126 | 6185 | |
a97699d1 QW |
6186 | /* |
6187 | * For full stripe start, we use previously calculated | |
6188 | * @stripe_nr. Align it to nr_data_stripes, then multiply with | |
6189 | * STRIPE_LEN. | |
6190 | * | |
6191 | * By this we can avoid u64 division completely. And we have | |
6192 | * to go rounddown(), not round_down(), as nr_data_stripes is | |
6193 | * not ensured to be power of 2. | |
6194 | */ | |
f8a02dc6 | 6195 | *full_stripe_start = |
a97699d1 QW |
6196 | rounddown(*stripe_nr, nr_data_stripes(map)) << |
6197 | BTRFS_STRIPE_LEN_SHIFT; | |
5f141126 NB |
6198 | |
6199 | /* | |
f8a02dc6 CH |
6200 | * For writes to RAID56, allow to write a full stripe set, but |
6201 | * no straddling of stripe sets. | |
5f141126 | 6202 | */ |
f8a02dc6 CH |
6203 | if (op == BTRFS_MAP_WRITE) |
6204 | return full_stripe_len - (offset - *full_stripe_start); | |
5f141126 NB |
6205 | } |
6206 | ||
f8a02dc6 CH |
6207 | /* |
6208 | * For other RAID types and for RAID56 reads, allow a single stripe (on | |
6209 | * a single disk). | |
6210 | */ | |
6211 | if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) | |
a97699d1 | 6212 | return BTRFS_STRIPE_LEN - *stripe_offset; |
f8a02dc6 | 6213 | return U64_MAX; |
5f141126 NB |
6214 | } |
6215 | ||
03793cbb | 6216 | static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map, |
6ded22c1 | 6217 | u32 stripe_index, u64 stripe_offset, u32 stripe_nr) |
03793cbb CH |
6218 | { |
6219 | dst->dev = map->stripes[stripe_index].dev; | |
6220 | dst->physical = map->stripes[stripe_index].physical + | |
a97699d1 | 6221 | stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT); |
03793cbb CH |
6222 | } |
6223 | ||
103c1972 CH |
6224 | int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
6225 | u64 logical, u64 *length, | |
6226 | struct btrfs_io_context **bioc_ret, | |
6227 | struct btrfs_io_stripe *smap, int *mirror_num_ret, | |
6228 | int need_raid_map) | |
0b86a832 CM |
6229 | { |
6230 | struct extent_map *em; | |
6231 | struct map_lookup *map; | |
f8a02dc6 | 6232 | u64 map_offset; |
593060d7 | 6233 | u64 stripe_offset; |
6ded22c1 | 6234 | u32 stripe_nr; |
9d644a62 | 6235 | u32 stripe_index; |
cff82672 | 6236 | int data_stripes; |
cea9e445 | 6237 | int i; |
de11cc12 | 6238 | int ret = 0; |
03793cbb | 6239 | int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); |
f2d8d74d | 6240 | int num_stripes; |
5f50fa91 | 6241 | int num_copies; |
a236aed1 | 6242 | int max_errors = 0; |
4c664611 | 6243 | struct btrfs_io_context *bioc = NULL; |
472262f3 SB |
6244 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
6245 | int dev_replace_is_ongoing = 0; | |
4ced85f8 | 6246 | u16 num_alloc_stripes; |
53b381b3 | 6247 | u64 raid56_full_stripe_start = (u64)-1; |
f8a02dc6 | 6248 | u64 max_len; |
89b798ad | 6249 | |
4c664611 | 6250 | ASSERT(bioc_ret); |
75fb2e9e | 6251 | ASSERT(op != BTRFS_MAP_DISCARD); |
0b3d4cd3 | 6252 | |
5f50fa91 QW |
6253 | num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); |
6254 | if (mirror_num > num_copies) | |
6255 | return -EINVAL; | |
6256 | ||
42034313 | 6257 | em = btrfs_get_chunk_map(fs_info, logical, *length); |
1c3ab6df QW |
6258 | if (IS_ERR(em)) |
6259 | return PTR_ERR(em); | |
42034313 | 6260 | |
95617d69 | 6261 | map = em->map_lookup; |
cff82672 | 6262 | data_stripes = nr_data_stripes(map); |
f8a02dc6 CH |
6263 | |
6264 | map_offset = logical - em->start; | |
6265 | max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr, | |
6266 | &stripe_offset, &raid56_full_stripe_start); | |
6267 | *length = min_t(u64, em->len - map_offset, max_len); | |
593060d7 | 6268 | |
cb5583dd | 6269 | down_read(&dev_replace->rwsem); |
472262f3 | 6270 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
53176dde DS |
6271 | /* |
6272 | * Hold the semaphore for read during the whole operation, write is | |
6273 | * requested at commit time but must wait. | |
6274 | */ | |
472262f3 | 6275 | if (!dev_replace_is_ongoing) |
cb5583dd | 6276 | up_read(&dev_replace->rwsem); |
472262f3 | 6277 | |
f2d8d74d | 6278 | num_stripes = 1; |
cea9e445 | 6279 | stripe_index = 0; |
fce3bb9a | 6280 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
6ded22c1 QW |
6281 | stripe_index = stripe_nr % map->num_stripes; |
6282 | stripe_nr /= map->num_stripes; | |
de483734 | 6283 | if (!need_full_stripe(op)) |
28e1cc7d | 6284 | mirror_num = 1; |
c7369b3f | 6285 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) { |
de483734 | 6286 | if (need_full_stripe(op)) |
f2d8d74d | 6287 | num_stripes = map->num_stripes; |
2fff734f | 6288 | else if (mirror_num) |
f188591e | 6289 | stripe_index = mirror_num - 1; |
dfe25020 | 6290 | else { |
30d9861f | 6291 | stripe_index = find_live_mirror(fs_info, map, 0, |
30d9861f | 6292 | dev_replace_is_ongoing); |
a1d3c478 | 6293 | mirror_num = stripe_index + 1; |
dfe25020 | 6294 | } |
2fff734f | 6295 | |
611f0e00 | 6296 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
de483734 | 6297 | if (need_full_stripe(op)) { |
f2d8d74d | 6298 | num_stripes = map->num_stripes; |
a1d3c478 | 6299 | } else if (mirror_num) { |
f188591e | 6300 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
6301 | } else { |
6302 | mirror_num = 1; | |
6303 | } | |
2fff734f | 6304 | |
321aecc6 | 6305 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 6306 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 6307 | |
6ded22c1 QW |
6308 | stripe_index = (stripe_nr % factor) * map->sub_stripes; |
6309 | stripe_nr /= factor; | |
321aecc6 | 6310 | |
de483734 | 6311 | if (need_full_stripe(op)) |
f2d8d74d | 6312 | num_stripes = map->sub_stripes; |
321aecc6 CM |
6313 | else if (mirror_num) |
6314 | stripe_index += mirror_num - 1; | |
dfe25020 | 6315 | else { |
3e74317a | 6316 | int old_stripe_index = stripe_index; |
30d9861f SB |
6317 | stripe_index = find_live_mirror(fs_info, map, |
6318 | stripe_index, | |
30d9861f | 6319 | dev_replace_is_ongoing); |
3e74317a | 6320 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 6321 | } |
53b381b3 | 6322 | |
ffe2d203 | 6323 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
de483734 | 6324 | if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) { |
6ded22c1 QW |
6325 | /* |
6326 | * Push stripe_nr back to the start of the full stripe | |
6327 | * For those cases needing a full stripe, @stripe_nr | |
6328 | * is the full stripe number. | |
6329 | * | |
6330 | * Originally we go raid56_full_stripe_start / full_stripe_len, | |
6331 | * but that can be expensive. Here we just divide | |
6332 | * @stripe_nr with @data_stripes. | |
6333 | */ | |
6334 | stripe_nr /= data_stripes; | |
53b381b3 DW |
6335 | |
6336 | /* RAID[56] write or recovery. Return all stripes */ | |
6337 | num_stripes = map->num_stripes; | |
6dead96c | 6338 | max_errors = btrfs_chunk_max_errors(map); |
53b381b3 | 6339 | |
462b0b2a QW |
6340 | /* Return the length to the full stripe end */ |
6341 | *length = min(logical + *length, | |
6342 | raid56_full_stripe_start + em->start + | |
a97699d1 | 6343 | (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical; |
53b381b3 DW |
6344 | stripe_index = 0; |
6345 | stripe_offset = 0; | |
6346 | } else { | |
6347 | /* | |
6348 | * Mirror #0 or #1 means the original data block. | |
6349 | * Mirror #2 is RAID5 parity block. | |
6350 | * Mirror #3 is RAID6 Q block. | |
6351 | */ | |
6ded22c1 QW |
6352 | stripe_index = stripe_nr % data_stripes; |
6353 | stripe_nr /= data_stripes; | |
53b381b3 | 6354 | if (mirror_num > 1) |
cff82672 | 6355 | stripe_index = data_stripes + mirror_num - 2; |
53b381b3 DW |
6356 | |
6357 | /* We distribute the parity blocks across stripes */ | |
6ded22c1 | 6358 | stripe_index = (stripe_nr + stripe_index) % map->num_stripes; |
de483734 | 6359 | if (!need_full_stripe(op) && mirror_num <= 1) |
28e1cc7d | 6360 | mirror_num = 1; |
53b381b3 | 6361 | } |
8790d502 CM |
6362 | } else { |
6363 | /* | |
6ded22c1 | 6364 | * After this, stripe_nr is the number of stripes on this |
47c5713f DS |
6365 | * device we have to walk to find the data, and stripe_index is |
6366 | * the number of our device in the stripe array | |
8790d502 | 6367 | */ |
6ded22c1 QW |
6368 | stripe_index = stripe_nr % map->num_stripes; |
6369 | stripe_nr /= map->num_stripes; | |
a1d3c478 | 6370 | mirror_num = stripe_index + 1; |
8790d502 | 6371 | } |
e042d1ec | 6372 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
6373 | btrfs_crit(fs_info, |
6374 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
6375 | stripe_index, map->num_stripes); |
6376 | ret = -EINVAL; | |
6377 | goto out; | |
6378 | } | |
cea9e445 | 6379 | |
472262f3 | 6380 | num_alloc_stripes = num_stripes; |
1faf3885 QW |
6381 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
6382 | op != BTRFS_MAP_READ) | |
6383 | /* | |
6384 | * For replace case, we need to add extra stripes for extra | |
6385 | * duplicated stripes. | |
6386 | * | |
6387 | * For both WRITE and GET_READ_MIRRORS, we may have at most | |
6388 | * 2 more stripes (DUP types, otherwise 1). | |
6389 | */ | |
6390 | num_alloc_stripes += 2; | |
2c8cdd6e | 6391 | |
03793cbb CH |
6392 | /* |
6393 | * If this I/O maps to a single device, try to return the device and | |
6394 | * physical block information on the stack instead of allocating an | |
6395 | * I/O context structure. | |
6396 | */ | |
6397 | if (smap && num_alloc_stripes == 1 && | |
6398 | !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) && | |
6399 | (!need_full_stripe(op) || !dev_replace_is_ongoing || | |
6400 | !dev_replace->tgtdev)) { | |
5f50fa91 QW |
6401 | set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr); |
6402 | *mirror_num_ret = mirror_num; | |
03793cbb CH |
6403 | *bioc_ret = NULL; |
6404 | ret = 0; | |
6405 | goto out; | |
6406 | } | |
6407 | ||
1faf3885 | 6408 | bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes); |
4c664611 | 6409 | if (!bioc) { |
de11cc12 LZ |
6410 | ret = -ENOMEM; |
6411 | goto out; | |
6412 | } | |
1faf3885 | 6413 | bioc->map_type = map->type; |
608769a4 | 6414 | |
18d758a2 QW |
6415 | /* |
6416 | * For RAID56 full map, we need to make sure the stripes[] follows the | |
6417 | * rule that data stripes are all ordered, then followed with P and Q | |
6418 | * (if we have). | |
6419 | * | |
6420 | * It's still mostly the same as other profiles, just with extra rotation. | |
6421 | */ | |
2b19a1fe LB |
6422 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map && |
6423 | (need_full_stripe(op) || mirror_num > 1)) { | |
18d758a2 QW |
6424 | /* |
6425 | * For RAID56 @stripe_nr is already the number of full stripes | |
6426 | * before us, which is also the rotation value (needs to modulo | |
6427 | * with num_stripes). | |
6428 | * | |
6429 | * In this case, we just add @stripe_nr with @i, then do the | |
6430 | * modulo, to reduce one modulo call. | |
6431 | */ | |
6432 | bioc->full_stripe_logical = em->start + | |
6433 | ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT); | |
6434 | for (i = 0; i < num_stripes; i++) | |
6435 | set_io_stripe(&bioc->stripes[i], map, | |
6436 | (i + stripe_nr) % num_stripes, | |
6437 | stripe_offset, stripe_nr); | |
6438 | } else { | |
6439 | /* | |
6440 | * For all other non-RAID56 profiles, just copy the target | |
6441 | * stripe into the bioc. | |
6442 | */ | |
6443 | for (i = 0; i < num_stripes; i++) { | |
6444 | set_io_stripe(&bioc->stripes[i], map, stripe_index, | |
6445 | stripe_offset, stripe_nr); | |
6446 | stripe_index++; | |
6447 | } | |
593060d7 | 6448 | } |
de11cc12 | 6449 | |
2b19a1fe | 6450 | if (need_full_stripe(op)) |
d20983b4 | 6451 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 6452 | |
73c0f228 | 6453 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
2b19a1fe | 6454 | need_full_stripe(op)) { |
be5c7edb | 6455 | handle_ops_on_dev_replace(op, bioc, dev_replace, logical, |
6143c23c | 6456 | &num_stripes, &max_errors); |
472262f3 SB |
6457 | } |
6458 | ||
4c664611 | 6459 | *bioc_ret = bioc; |
4c664611 QW |
6460 | bioc->num_stripes = num_stripes; |
6461 | bioc->max_errors = max_errors; | |
6462 | bioc->mirror_num = mirror_num; | |
ad6d620e | 6463 | |
cea9e445 | 6464 | out: |
73beece9 | 6465 | if (dev_replace_is_ongoing) { |
53176dde DS |
6466 | lockdep_assert_held(&dev_replace->rwsem); |
6467 | /* Unlock and let waiting writers proceed */ | |
cb5583dd | 6468 | up_read(&dev_replace->rwsem); |
73beece9 | 6469 | } |
0b86a832 | 6470 | free_extent_map(em); |
de11cc12 | 6471 | return ret; |
0b86a832 CM |
6472 | } |
6473 | ||
cf8cddd3 | 6474 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 6475 | u64 logical, u64 *length, |
4c664611 | 6476 | struct btrfs_io_context **bioc_ret, int mirror_num) |
f2d8d74d | 6477 | { |
4c664611 | 6478 | return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, |
03793cbb | 6479 | NULL, &mirror_num, 0); |
f2d8d74d CM |
6480 | } |
6481 | ||
af8e2d1d | 6482 | /* For Scrub/replace */ |
cf8cddd3 | 6483 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d | 6484 | u64 logical, u64 *length, |
4c664611 | 6485 | struct btrfs_io_context **bioc_ret) |
af8e2d1d | 6486 | { |
03793cbb CH |
6487 | return __btrfs_map_block(fs_info, op, logical, length, bioc_ret, |
6488 | NULL, NULL, 1); | |
af8e2d1d MX |
6489 | } |
6490 | ||
562d7b15 JB |
6491 | static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, |
6492 | const struct btrfs_fs_devices *fs_devices) | |
6493 | { | |
6494 | if (args->fsid == NULL) | |
6495 | return true; | |
6496 | if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) | |
6497 | return true; | |
6498 | return false; | |
6499 | } | |
6500 | ||
6501 | static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, | |
6502 | const struct btrfs_device *device) | |
6503 | { | |
0fca385d LS |
6504 | if (args->missing) { |
6505 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && | |
6506 | !device->bdev) | |
6507 | return true; | |
6508 | return false; | |
6509 | } | |
562d7b15 | 6510 | |
0fca385d | 6511 | if (device->devid != args->devid) |
562d7b15 JB |
6512 | return false; |
6513 | if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) | |
6514 | return false; | |
0fca385d | 6515 | return true; |
562d7b15 JB |
6516 | } |
6517 | ||
09ba3bc9 AJ |
6518 | /* |
6519 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or | |
6520 | * return NULL. | |
6521 | * | |
6522 | * If devid and uuid are both specified, the match must be exact, otherwise | |
6523 | * only devid is used. | |
09ba3bc9 | 6524 | */ |
562d7b15 JB |
6525 | struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, |
6526 | const struct btrfs_dev_lookup_args *args) | |
0b86a832 | 6527 | { |
2b82032c | 6528 | struct btrfs_device *device; |
944d3f9f NB |
6529 | struct btrfs_fs_devices *seed_devs; |
6530 | ||
562d7b15 | 6531 | if (dev_args_match_fs_devices(args, fs_devices)) { |
944d3f9f | 6532 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
562d7b15 | 6533 | if (dev_args_match_device(args, device)) |
944d3f9f NB |
6534 | return device; |
6535 | } | |
6536 | } | |
2b82032c | 6537 | |
944d3f9f | 6538 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
562d7b15 JB |
6539 | if (!dev_args_match_fs_devices(args, seed_devs)) |
6540 | continue; | |
6541 | list_for_each_entry(device, &seed_devs->devices, dev_list) { | |
6542 | if (dev_args_match_device(args, device)) | |
6543 | return device; | |
2b82032c | 6544 | } |
2b82032c | 6545 | } |
944d3f9f | 6546 | |
2b82032c | 6547 | return NULL; |
0b86a832 CM |
6548 | } |
6549 | ||
2ff7e61e | 6550 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6551 | u64 devid, u8 *dev_uuid) |
6552 | { | |
6553 | struct btrfs_device *device; | |
fccc0007 | 6554 | unsigned int nofs_flag; |
dfe25020 | 6555 | |
fccc0007 JB |
6556 | /* |
6557 | * We call this under the chunk_mutex, so we want to use NOFS for this | |
6558 | * allocation, however we don't want to change btrfs_alloc_device() to | |
6559 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe | |
6560 | * places. | |
6561 | */ | |
bb21e302 | 6562 | |
fccc0007 | 6563 | nofs_flag = memalloc_nofs_save(); |
bb21e302 | 6564 | device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); |
fccc0007 | 6565 | memalloc_nofs_restore(nofs_flag); |
12bd2fc0 | 6566 | if (IS_ERR(device)) |
adfb69af | 6567 | return device; |
12bd2fc0 ID |
6568 | |
6569 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6570 | device->fs_devices = fs_devices; |
dfe25020 | 6571 | fs_devices->num_devices++; |
12bd2fc0 | 6572 | |
e6e674bd | 6573 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
cd02dca5 | 6574 | fs_devices->missing_devices++; |
12bd2fc0 | 6575 | |
dfe25020 CM |
6576 | return device; |
6577 | } | |
6578 | ||
43dd529a DS |
6579 | /* |
6580 | * Allocate new device struct, set up devid and UUID. | |
6581 | * | |
12bd2fc0 ID |
6582 | * @fs_info: used only for generating a new devid, can be NULL if |
6583 | * devid is provided (i.e. @devid != NULL). | |
6584 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6585 | * is generated. | |
6586 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6587 | * is generated. | |
bb21e302 | 6588 | * @path: a pointer to device path if available, NULL otherwise. |
12bd2fc0 ID |
6589 | * |
6590 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
48dae9cf | 6591 | * on error. Returned struct is not linked onto any lists and must be |
a425f9d4 | 6592 | * destroyed with btrfs_free_device. |
12bd2fc0 ID |
6593 | */ |
6594 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
bb21e302 AJ |
6595 | const u64 *devid, const u8 *uuid, |
6596 | const char *path) | |
12bd2fc0 ID |
6597 | { |
6598 | struct btrfs_device *dev; | |
6599 | u64 tmp; | |
6600 | ||
fae7f21c | 6601 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6602 | return ERR_PTR(-EINVAL); |
12bd2fc0 | 6603 | |
fe4f46d4 DS |
6604 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
6605 | if (!dev) | |
6606 | return ERR_PTR(-ENOMEM); | |
6607 | ||
fe4f46d4 DS |
6608 | INIT_LIST_HEAD(&dev->dev_list); |
6609 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
6610 | INIT_LIST_HEAD(&dev->post_commit_list); | |
6611 | ||
fe4f46d4 DS |
6612 | atomic_set(&dev->dev_stats_ccnt, 0); |
6613 | btrfs_device_data_ordered_init(dev); | |
35da5a7e | 6614 | extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); |
12bd2fc0 ID |
6615 | |
6616 | if (devid) | |
6617 | tmp = *devid; | |
6618 | else { | |
6619 | int ret; | |
6620 | ||
6621 | ret = find_next_devid(fs_info, &tmp); | |
6622 | if (ret) { | |
a425f9d4 | 6623 | btrfs_free_device(dev); |
12bd2fc0 ID |
6624 | return ERR_PTR(ret); |
6625 | } | |
6626 | } | |
6627 | dev->devid = tmp; | |
6628 | ||
6629 | if (uuid) | |
6630 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6631 | else | |
6632 | generate_random_uuid(dev->uuid); | |
6633 | ||
bb21e302 AJ |
6634 | if (path) { |
6635 | struct rcu_string *name; | |
6636 | ||
6637 | name = rcu_string_strdup(path, GFP_KERNEL); | |
6638 | if (!name) { | |
6639 | btrfs_free_device(dev); | |
6640 | return ERR_PTR(-ENOMEM); | |
6641 | } | |
6642 | rcu_assign_pointer(dev->name, name); | |
6643 | } | |
6644 | ||
12bd2fc0 ID |
6645 | return dev; |
6646 | } | |
6647 | ||
5a2b8e60 | 6648 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
2b902dfc | 6649 | u64 devid, u8 *uuid, bool error) |
5a2b8e60 | 6650 | { |
2b902dfc AJ |
6651 | if (error) |
6652 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", | |
6653 | devid, uuid); | |
6654 | else | |
6655 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", | |
6656 | devid, uuid); | |
5a2b8e60 AJ |
6657 | } |
6658 | ||
bc88b486 | 6659 | u64 btrfs_calc_stripe_length(const struct extent_map *em) |
39e264a4 | 6660 | { |
bc88b486 QW |
6661 | const struct map_lookup *map = em->map_lookup; |
6662 | const int data_stripes = calc_data_stripes(map->type, map->num_stripes); | |
e4f6c6be | 6663 | |
bc88b486 | 6664 | return div_u64(em->len, data_stripes); |
39e264a4 NB |
6665 | } |
6666 | ||
e9306ad4 QW |
6667 | #if BITS_PER_LONG == 32 |
6668 | /* | |
6669 | * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE | |
6670 | * can't be accessed on 32bit systems. | |
6671 | * | |
6672 | * This function do mount time check to reject the fs if it already has | |
6673 | * metadata chunk beyond that limit. | |
6674 | */ | |
6675 | static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6676 | u64 logical, u64 length, u64 type) | |
6677 | { | |
6678 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6679 | return 0; | |
6680 | ||
6681 | if (logical + length < MAX_LFS_FILESIZE) | |
6682 | return 0; | |
6683 | ||
6684 | btrfs_err_32bit_limit(fs_info); | |
6685 | return -EOVERFLOW; | |
6686 | } | |
6687 | ||
6688 | /* | |
6689 | * This is to give early warning for any metadata chunk reaching | |
6690 | * BTRFS_32BIT_EARLY_WARN_THRESHOLD. | |
6691 | * Although we can still access the metadata, it's not going to be possible | |
6692 | * once the limit is reached. | |
6693 | */ | |
6694 | static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, | |
6695 | u64 logical, u64 length, u64 type) | |
6696 | { | |
6697 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) | |
6698 | return; | |
6699 | ||
6700 | if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6701 | return; | |
6702 | ||
6703 | btrfs_warn_32bit_limit(fs_info); | |
6704 | } | |
6705 | #endif | |
6706 | ||
ff37c89f NB |
6707 | static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, |
6708 | u64 devid, u8 *uuid) | |
6709 | { | |
6710 | struct btrfs_device *dev; | |
6711 | ||
6712 | if (!btrfs_test_opt(fs_info, DEGRADED)) { | |
6713 | btrfs_report_missing_device(fs_info, devid, uuid, true); | |
6714 | return ERR_PTR(-ENOENT); | |
6715 | } | |
6716 | ||
6717 | dev = add_missing_dev(fs_info->fs_devices, devid, uuid); | |
6718 | if (IS_ERR(dev)) { | |
6719 | btrfs_err(fs_info, "failed to init missing device %llu: %ld", | |
6720 | devid, PTR_ERR(dev)); | |
6721 | return dev; | |
6722 | } | |
6723 | btrfs_report_missing_device(fs_info, devid, uuid, false); | |
6724 | ||
6725 | return dev; | |
6726 | } | |
6727 | ||
9690ac09 | 6728 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
e06cd3dd LB |
6729 | struct btrfs_chunk *chunk) |
6730 | { | |
562d7b15 | 6731 | BTRFS_DEV_LOOKUP_ARGS(args); |
9690ac09 | 6732 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
c8bf1b67 | 6733 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
e06cd3dd LB |
6734 | struct map_lookup *map; |
6735 | struct extent_map *em; | |
6736 | u64 logical; | |
6737 | u64 length; | |
e06cd3dd | 6738 | u64 devid; |
e9306ad4 | 6739 | u64 type; |
e06cd3dd | 6740 | u8 uuid[BTRFS_UUID_SIZE]; |
76a66ba1 | 6741 | int index; |
e06cd3dd LB |
6742 | int num_stripes; |
6743 | int ret; | |
6744 | int i; | |
6745 | ||
6746 | logical = key->offset; | |
6747 | length = btrfs_chunk_length(leaf, chunk); | |
e9306ad4 | 6748 | type = btrfs_chunk_type(leaf, chunk); |
76a66ba1 | 6749 | index = btrfs_bg_flags_to_raid_index(type); |
e06cd3dd LB |
6750 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6751 | ||
e9306ad4 QW |
6752 | #if BITS_PER_LONG == 32 |
6753 | ret = check_32bit_meta_chunk(fs_info, logical, length, type); | |
6754 | if (ret < 0) | |
6755 | return ret; | |
6756 | warn_32bit_meta_chunk(fs_info, logical, length, type); | |
6757 | #endif | |
6758 | ||
075cb3c7 QW |
6759 | /* |
6760 | * Only need to verify chunk item if we're reading from sys chunk array, | |
6761 | * as chunk item in tree block is already verified by tree-checker. | |
6762 | */ | |
6763 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { | |
ddaf1d5a | 6764 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
075cb3c7 QW |
6765 | if (ret) |
6766 | return ret; | |
6767 | } | |
a061fc8d | 6768 | |
c8bf1b67 DS |
6769 | read_lock(&map_tree->lock); |
6770 | em = lookup_extent_mapping(map_tree, logical, 1); | |
6771 | read_unlock(&map_tree->lock); | |
0b86a832 CM |
6772 | |
6773 | /* already mapped? */ | |
6774 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6775 | free_extent_map(em); | |
0b86a832 CM |
6776 | return 0; |
6777 | } else if (em) { | |
6778 | free_extent_map(em); | |
6779 | } | |
0b86a832 | 6780 | |
172ddd60 | 6781 | em = alloc_extent_map(); |
0b86a832 CM |
6782 | if (!em) |
6783 | return -ENOMEM; | |
593060d7 | 6784 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6785 | if (!map) { |
6786 | free_extent_map(em); | |
6787 | return -ENOMEM; | |
6788 | } | |
6789 | ||
298a8f9c | 6790 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6791 | em->map_lookup = map; |
0b86a832 CM |
6792 | em->start = logical; |
6793 | em->len = length; | |
70c8a91c | 6794 | em->orig_start = 0; |
0b86a832 | 6795 | em->block_start = 0; |
c8b97818 | 6796 | em->block_len = em->len; |
0b86a832 | 6797 | |
593060d7 CM |
6798 | map->num_stripes = num_stripes; |
6799 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6800 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
e9306ad4 | 6801 | map->type = type; |
76a66ba1 QW |
6802 | /* |
6803 | * We can't use the sub_stripes value, as for profiles other than | |
6804 | * RAID10, they may have 0 as sub_stripes for filesystems created by | |
6805 | * older mkfs (<v5.4). | |
6806 | * In that case, it can cause divide-by-zero errors later. | |
6807 | * Since currently sub_stripes is fixed for each profile, let's | |
6808 | * use the trusted value instead. | |
6809 | */ | |
6810 | map->sub_stripes = btrfs_raid_array[index].sub_stripes; | |
cf90d884 | 6811 | map->verified_stripes = 0; |
bc88b486 | 6812 | em->orig_block_len = btrfs_calc_stripe_length(em); |
593060d7 CM |
6813 | for (i = 0; i < num_stripes; i++) { |
6814 | map->stripes[i].physical = | |
6815 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6816 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
562d7b15 | 6817 | args.devid = devid; |
a443755f CM |
6818 | read_extent_buffer(leaf, uuid, (unsigned long) |
6819 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6820 | BTRFS_UUID_SIZE); | |
562d7b15 JB |
6821 | args.uuid = uuid; |
6822 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); | |
dfe25020 | 6823 | if (!map->stripes[i].dev) { |
ff37c89f NB |
6824 | map->stripes[i].dev = handle_missing_device(fs_info, |
6825 | devid, uuid); | |
adfb69af | 6826 | if (IS_ERR(map->stripes[i].dev)) { |
1742e1c9 | 6827 | ret = PTR_ERR(map->stripes[i].dev); |
dfe25020 | 6828 | free_extent_map(em); |
1742e1c9 | 6829 | return ret; |
dfe25020 CM |
6830 | } |
6831 | } | |
ff37c89f | 6832 | |
e12c9621 AJ |
6833 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
6834 | &(map->stripes[i].dev->dev_state)); | |
0b86a832 CM |
6835 | } |
6836 | ||
c8bf1b67 DS |
6837 | write_lock(&map_tree->lock); |
6838 | ret = add_extent_mapping(map_tree, em, 0); | |
6839 | write_unlock(&map_tree->lock); | |
64f64f43 QW |
6840 | if (ret < 0) { |
6841 | btrfs_err(fs_info, | |
6842 | "failed to add chunk map, start=%llu len=%llu: %d", | |
6843 | em->start, em->len, ret); | |
6844 | } | |
0b86a832 CM |
6845 | free_extent_map(em); |
6846 | ||
64f64f43 | 6847 | return ret; |
0b86a832 CM |
6848 | } |
6849 | ||
143bede5 | 6850 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6851 | struct btrfs_dev_item *dev_item, |
6852 | struct btrfs_device *device) | |
6853 | { | |
6854 | unsigned long ptr; | |
0b86a832 CM |
6855 | |
6856 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6857 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6858 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6859 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6860 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6861 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6862 | device->type = btrfs_device_type(leaf, dev_item); |
6863 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6864 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6865 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6866 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
401e29c1 | 6867 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
0b86a832 | 6868 | |
410ba3a2 | 6869 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6870 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6871 | } |
6872 | ||
2ff7e61e | 6873 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
5f375835 | 6874 | u8 *fsid) |
2b82032c YZ |
6875 | { |
6876 | struct btrfs_fs_devices *fs_devices; | |
6877 | int ret; | |
6878 | ||
a32bf9a3 | 6879 | lockdep_assert_held(&uuid_mutex); |
2dfeca9b | 6880 | ASSERT(fsid); |
2b82032c | 6881 | |
427c8fdd | 6882 | /* This will match only for multi-device seed fs */ |
944d3f9f | 6883 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
44880fdc | 6884 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
5f375835 MX |
6885 | return fs_devices; |
6886 | ||
2b82032c | 6887 | |
7239ff4b | 6888 | fs_devices = find_fsid(fsid, NULL); |
2b82032c | 6889 | if (!fs_devices) { |
0b246afa | 6890 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6891 | return ERR_PTR(-ENOENT); |
6892 | ||
7239ff4b | 6893 | fs_devices = alloc_fs_devices(fsid, NULL); |
5f375835 MX |
6894 | if (IS_ERR(fs_devices)) |
6895 | return fs_devices; | |
6896 | ||
0395d84f | 6897 | fs_devices->seeding = true; |
5f375835 MX |
6898 | fs_devices->opened = 1; |
6899 | return fs_devices; | |
2b82032c | 6900 | } |
e4404d6e | 6901 | |
427c8fdd NB |
6902 | /* |
6903 | * Upon first call for a seed fs fsid, just create a private copy of the | |
6904 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list | |
6905 | */ | |
e4404d6e | 6906 | fs_devices = clone_fs_devices(fs_devices); |
5f375835 MX |
6907 | if (IS_ERR(fs_devices)) |
6908 | return fs_devices; | |
2b82032c | 6909 | |
05bdb996 | 6910 | ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder); |
48d28232 JL |
6911 | if (ret) { |
6912 | free_fs_devices(fs_devices); | |
c83b60c0 | 6913 | return ERR_PTR(ret); |
48d28232 | 6914 | } |
2b82032c YZ |
6915 | |
6916 | if (!fs_devices->seeding) { | |
0226e0eb | 6917 | close_fs_devices(fs_devices); |
e4404d6e | 6918 | free_fs_devices(fs_devices); |
c83b60c0 | 6919 | return ERR_PTR(-EINVAL); |
2b82032c YZ |
6920 | } |
6921 | ||
944d3f9f | 6922 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
c83b60c0 | 6923 | |
5f375835 | 6924 | return fs_devices; |
2b82032c YZ |
6925 | } |
6926 | ||
17850759 | 6927 | static int read_one_dev(struct extent_buffer *leaf, |
0b86a832 CM |
6928 | struct btrfs_dev_item *dev_item) |
6929 | { | |
562d7b15 | 6930 | BTRFS_DEV_LOOKUP_ARGS(args); |
17850759 | 6931 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
0b246afa | 6932 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
0b86a832 CM |
6933 | struct btrfs_device *device; |
6934 | u64 devid; | |
6935 | int ret; | |
44880fdc | 6936 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
a443755f CM |
6937 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6938 | ||
c1867eb3 DS |
6939 | devid = btrfs_device_id(leaf, dev_item); |
6940 | args.devid = devid; | |
410ba3a2 | 6941 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6942 | BTRFS_UUID_SIZE); |
1473b24e | 6943 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
44880fdc | 6944 | BTRFS_FSID_SIZE); |
562d7b15 JB |
6945 | args.uuid = dev_uuid; |
6946 | args.fsid = fs_uuid; | |
2b82032c | 6947 | |
de37aa51 | 6948 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
2ff7e61e | 6949 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
5f375835 MX |
6950 | if (IS_ERR(fs_devices)) |
6951 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6952 | } |
6953 | ||
562d7b15 | 6954 | device = btrfs_find_device(fs_info->fs_devices, &args); |
5f375835 | 6955 | if (!device) { |
c5502451 | 6956 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
2b902dfc AJ |
6957 | btrfs_report_missing_device(fs_info, devid, |
6958 | dev_uuid, true); | |
45dbdbc9 | 6959 | return -ENOENT; |
c5502451 | 6960 | } |
2b82032c | 6961 | |
2ff7e61e | 6962 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
adfb69af AJ |
6963 | if (IS_ERR(device)) { |
6964 | btrfs_err(fs_info, | |
6965 | "failed to add missing dev %llu: %ld", | |
6966 | devid, PTR_ERR(device)); | |
6967 | return PTR_ERR(device); | |
6968 | } | |
2b902dfc | 6969 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
5f375835 | 6970 | } else { |
c5502451 | 6971 | if (!device->bdev) { |
2b902dfc AJ |
6972 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
6973 | btrfs_report_missing_device(fs_info, | |
6974 | devid, dev_uuid, true); | |
45dbdbc9 | 6975 | return -ENOENT; |
2b902dfc AJ |
6976 | } |
6977 | btrfs_report_missing_device(fs_info, devid, | |
6978 | dev_uuid, false); | |
c5502451 | 6979 | } |
5f375835 | 6980 | |
e6e674bd AJ |
6981 | if (!device->bdev && |
6982 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { | |
cd02dca5 CM |
6983 | /* |
6984 | * this happens when a device that was properly setup | |
6985 | * in the device info lists suddenly goes bad. | |
6986 | * device->bdev is NULL, and so we have to set | |
6987 | * device->missing to one here | |
6988 | */ | |
5f375835 | 6989 | device->fs_devices->missing_devices++; |
e6e674bd | 6990 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
2b82032c | 6991 | } |
5f375835 MX |
6992 | |
6993 | /* Move the device to its own fs_devices */ | |
6994 | if (device->fs_devices != fs_devices) { | |
e6e674bd AJ |
6995 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
6996 | &device->dev_state)); | |
5f375835 MX |
6997 | |
6998 | list_move(&device->dev_list, &fs_devices->devices); | |
6999 | device->fs_devices->num_devices--; | |
7000 | fs_devices->num_devices++; | |
7001 | ||
7002 | device->fs_devices->missing_devices--; | |
7003 | fs_devices->missing_devices++; | |
7004 | ||
7005 | device->fs_devices = fs_devices; | |
7006 | } | |
2b82032c YZ |
7007 | } |
7008 | ||
0b246afa | 7009 | if (device->fs_devices != fs_info->fs_devices) { |
ebbede42 | 7010 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
2b82032c YZ |
7011 | if (device->generation != |
7012 | btrfs_device_generation(leaf, dev_item)) | |
7013 | return -EINVAL; | |
6324fbf3 | 7014 | } |
0b86a832 CM |
7015 | |
7016 | fill_device_from_item(leaf, dev_item, device); | |
3a160a93 | 7017 | if (device->bdev) { |
cda00eba | 7018 | u64 max_total_bytes = bdev_nr_bytes(device->bdev); |
3a160a93 AJ |
7019 | |
7020 | if (device->total_bytes > max_total_bytes) { | |
7021 | btrfs_err(fs_info, | |
7022 | "device total_bytes should be at most %llu but found %llu", | |
7023 | max_total_bytes, device->total_bytes); | |
7024 | return -EINVAL; | |
7025 | } | |
7026 | } | |
e12c9621 | 7027 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
ebbede42 | 7028 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
401e29c1 | 7029 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
2b82032c | 7030 | device->fs_devices->total_rw_bytes += device->total_bytes; |
a5ed45f8 NB |
7031 | atomic64_add(device->total_bytes - device->bytes_used, |
7032 | &fs_info->free_chunk_space); | |
2bf64758 | 7033 | } |
0b86a832 | 7034 | ret = 0; |
0b86a832 CM |
7035 | return ret; |
7036 | } | |
7037 | ||
6bccf3ab | 7038 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 7039 | { |
ab8d0fc4 | 7040 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 7041 | struct extent_buffer *sb; |
0b86a832 | 7042 | struct btrfs_disk_key *disk_key; |
0b86a832 | 7043 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
7044 | u8 *array_ptr; |
7045 | unsigned long sb_array_offset; | |
84eed90f | 7046 | int ret = 0; |
0b86a832 CM |
7047 | u32 num_stripes; |
7048 | u32 array_size; | |
7049 | u32 len = 0; | |
1ffb22cf | 7050 | u32 cur_offset; |
e06cd3dd | 7051 | u64 type; |
84eed90f | 7052 | struct btrfs_key key; |
0b86a832 | 7053 | |
0b246afa | 7054 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
e959d3c1 | 7055 | |
a83fffb7 | 7056 | /* |
e959d3c1 QW |
7057 | * We allocated a dummy extent, just to use extent buffer accessors. |
7058 | * There will be unused space after BTRFS_SUPER_INFO_SIZE, but | |
7059 | * that's fine, we will not go beyond system chunk array anyway. | |
a83fffb7 | 7060 | */ |
e959d3c1 QW |
7061 | sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); |
7062 | if (!sb) | |
7063 | return -ENOMEM; | |
4db8c528 | 7064 | set_extent_buffer_uptodate(sb); |
4008c04a | 7065 | |
a061fc8d | 7066 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
7067 | array_size = btrfs_super_sys_array_size(super_copy); |
7068 | ||
1ffb22cf DS |
7069 | array_ptr = super_copy->sys_chunk_array; |
7070 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
7071 | cur_offset = 0; | |
0b86a832 | 7072 | |
1ffb22cf DS |
7073 | while (cur_offset < array_size) { |
7074 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
7075 | len = sizeof(*disk_key); |
7076 | if (cur_offset + len > array_size) | |
7077 | goto out_short_read; | |
7078 | ||
0b86a832 CM |
7079 | btrfs_disk_key_to_cpu(&key, disk_key); |
7080 | ||
1ffb22cf DS |
7081 | array_ptr += len; |
7082 | sb_array_offset += len; | |
7083 | cur_offset += len; | |
0b86a832 | 7084 | |
32ab3d1b JT |
7085 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
7086 | btrfs_err(fs_info, | |
7087 | "unexpected item type %u in sys_array at offset %u", | |
7088 | (u32)key.type, cur_offset); | |
7089 | ret = -EIO; | |
7090 | break; | |
7091 | } | |
f5cdedd7 | 7092 | |
32ab3d1b JT |
7093 | chunk = (struct btrfs_chunk *)sb_array_offset; |
7094 | /* | |
7095 | * At least one btrfs_chunk with one stripe must be present, | |
7096 | * exact stripe count check comes afterwards | |
7097 | */ | |
7098 | len = btrfs_chunk_item_size(1); | |
7099 | if (cur_offset + len > array_size) | |
7100 | goto out_short_read; | |
e06cd3dd | 7101 | |
32ab3d1b JT |
7102 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
7103 | if (!num_stripes) { | |
7104 | btrfs_err(fs_info, | |
7105 | "invalid number of stripes %u in sys_array at offset %u", | |
7106 | num_stripes, cur_offset); | |
7107 | ret = -EIO; | |
7108 | break; | |
7109 | } | |
e3540eab | 7110 | |
32ab3d1b JT |
7111 | type = btrfs_chunk_type(sb, chunk); |
7112 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 7113 | btrfs_err(fs_info, |
32ab3d1b JT |
7114 | "invalid chunk type %llu in sys_array at offset %u", |
7115 | type, cur_offset); | |
84eed90f CM |
7116 | ret = -EIO; |
7117 | break; | |
0b86a832 | 7118 | } |
32ab3d1b JT |
7119 | |
7120 | len = btrfs_chunk_item_size(num_stripes); | |
7121 | if (cur_offset + len > array_size) | |
7122 | goto out_short_read; | |
7123 | ||
7124 | ret = read_one_chunk(&key, sb, chunk); | |
7125 | if (ret) | |
7126 | break; | |
7127 | ||
1ffb22cf DS |
7128 | array_ptr += len; |
7129 | sb_array_offset += len; | |
7130 | cur_offset += len; | |
0b86a832 | 7131 | } |
d865177a | 7132 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7133 | free_extent_buffer_stale(sb); |
84eed90f | 7134 | return ret; |
e3540eab DS |
7135 | |
7136 | out_short_read: | |
ab8d0fc4 | 7137 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 7138 | len, cur_offset); |
d865177a | 7139 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 7140 | free_extent_buffer_stale(sb); |
e3540eab | 7141 | return -EIO; |
0b86a832 CM |
7142 | } |
7143 | ||
21634a19 QW |
7144 | /* |
7145 | * Check if all chunks in the fs are OK for read-write degraded mount | |
7146 | * | |
6528b99d AJ |
7147 | * If the @failing_dev is specified, it's accounted as missing. |
7148 | * | |
21634a19 QW |
7149 | * Return true if all chunks meet the minimal RW mount requirements. |
7150 | * Return false if any chunk doesn't meet the minimal RW mount requirements. | |
7151 | */ | |
6528b99d AJ |
7152 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
7153 | struct btrfs_device *failing_dev) | |
21634a19 | 7154 | { |
c8bf1b67 | 7155 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
21634a19 QW |
7156 | struct extent_map *em; |
7157 | u64 next_start = 0; | |
7158 | bool ret = true; | |
7159 | ||
c8bf1b67 DS |
7160 | read_lock(&map_tree->lock); |
7161 | em = lookup_extent_mapping(map_tree, 0, (u64)-1); | |
7162 | read_unlock(&map_tree->lock); | |
21634a19 QW |
7163 | /* No chunk at all? Return false anyway */ |
7164 | if (!em) { | |
7165 | ret = false; | |
7166 | goto out; | |
7167 | } | |
7168 | while (em) { | |
7169 | struct map_lookup *map; | |
7170 | int missing = 0; | |
7171 | int max_tolerated; | |
7172 | int i; | |
7173 | ||
7174 | map = em->map_lookup; | |
7175 | max_tolerated = | |
7176 | btrfs_get_num_tolerated_disk_barrier_failures( | |
7177 | map->type); | |
7178 | for (i = 0; i < map->num_stripes; i++) { | |
7179 | struct btrfs_device *dev = map->stripes[i].dev; | |
7180 | ||
e6e674bd AJ |
7181 | if (!dev || !dev->bdev || |
7182 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || | |
21634a19 QW |
7183 | dev->last_flush_error) |
7184 | missing++; | |
6528b99d AJ |
7185 | else if (failing_dev && failing_dev == dev) |
7186 | missing++; | |
21634a19 QW |
7187 | } |
7188 | if (missing > max_tolerated) { | |
6528b99d AJ |
7189 | if (!failing_dev) |
7190 | btrfs_warn(fs_info, | |
52042d8e | 7191 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
21634a19 QW |
7192 | em->start, missing, max_tolerated); |
7193 | free_extent_map(em); | |
7194 | ret = false; | |
7195 | goto out; | |
7196 | } | |
7197 | next_start = extent_map_end(em); | |
7198 | free_extent_map(em); | |
7199 | ||
c8bf1b67 DS |
7200 | read_lock(&map_tree->lock); |
7201 | em = lookup_extent_mapping(map_tree, next_start, | |
21634a19 | 7202 | (u64)(-1) - next_start); |
c8bf1b67 | 7203 | read_unlock(&map_tree->lock); |
21634a19 QW |
7204 | } |
7205 | out: | |
7206 | return ret; | |
7207 | } | |
7208 | ||
d85327b1 DS |
7209 | static void readahead_tree_node_children(struct extent_buffer *node) |
7210 | { | |
7211 | int i; | |
7212 | const int nr_items = btrfs_header_nritems(node); | |
7213 | ||
bfb484d9 JB |
7214 | for (i = 0; i < nr_items; i++) |
7215 | btrfs_readahead_node_child(node, i); | |
d85327b1 DS |
7216 | } |
7217 | ||
5b4aacef | 7218 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 7219 | { |
5b4aacef | 7220 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
7221 | struct btrfs_path *path; |
7222 | struct extent_buffer *leaf; | |
7223 | struct btrfs_key key; | |
7224 | struct btrfs_key found_key; | |
7225 | int ret; | |
7226 | int slot; | |
43cb1478 | 7227 | int iter_ret = 0; |
99e3ecfc | 7228 | u64 total_dev = 0; |
d85327b1 | 7229 | u64 last_ra_node = 0; |
0b86a832 | 7230 | |
0b86a832 CM |
7231 | path = btrfs_alloc_path(); |
7232 | if (!path) | |
7233 | return -ENOMEM; | |
7234 | ||
3dd0f7a3 AJ |
7235 | /* |
7236 | * uuid_mutex is needed only if we are mounting a sprout FS | |
7237 | * otherwise we don't need it. | |
7238 | */ | |
b367e47f | 7239 | mutex_lock(&uuid_mutex); |
b367e47f | 7240 | |
48cfa61b BB |
7241 | /* |
7242 | * It is possible for mount and umount to race in such a way that | |
7243 | * we execute this code path, but open_fs_devices failed to clear | |
7244 | * total_rw_bytes. We certainly want it cleared before reading the | |
7245 | * device items, so clear it here. | |
7246 | */ | |
7247 | fs_info->fs_devices->total_rw_bytes = 0; | |
7248 | ||
4d9380e0 FM |
7249 | /* |
7250 | * Lockdep complains about possible circular locking dependency between | |
7251 | * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores | |
7252 | * used for freeze procection of a fs (struct super_block.s_writers), | |
7253 | * which we take when starting a transaction, and extent buffers of the | |
7254 | * chunk tree if we call read_one_dev() while holding a lock on an | |
7255 | * extent buffer of the chunk tree. Since we are mounting the filesystem | |
7256 | * and at this point there can't be any concurrent task modifying the | |
7257 | * chunk tree, to keep it simple, just skip locking on the chunk tree. | |
7258 | */ | |
7259 | ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); | |
7260 | path->skip_locking = 1; | |
7261 | ||
395927a9 FDBM |
7262 | /* |
7263 | * Read all device items, and then all the chunk items. All | |
7264 | * device items are found before any chunk item (their object id | |
7265 | * is smaller than the lowest possible object id for a chunk | |
7266 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
7267 | */ |
7268 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
7269 | key.offset = 0; | |
7270 | key.type = 0; | |
43cb1478 GN |
7271 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
7272 | struct extent_buffer *node = path->nodes[1]; | |
d85327b1 | 7273 | |
0b86a832 CM |
7274 | leaf = path->nodes[0]; |
7275 | slot = path->slots[0]; | |
43cb1478 | 7276 | |
d85327b1 DS |
7277 | if (node) { |
7278 | if (last_ra_node != node->start) { | |
7279 | readahead_tree_node_children(node); | |
7280 | last_ra_node = node->start; | |
7281 | } | |
7282 | } | |
395927a9 FDBM |
7283 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
7284 | struct btrfs_dev_item *dev_item; | |
7285 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 7286 | struct btrfs_dev_item); |
17850759 | 7287 | ret = read_one_dev(leaf, dev_item); |
395927a9 FDBM |
7288 | if (ret) |
7289 | goto error; | |
99e3ecfc | 7290 | total_dev++; |
0b86a832 CM |
7291 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
7292 | struct btrfs_chunk *chunk; | |
79bd3712 FM |
7293 | |
7294 | /* | |
7295 | * We are only called at mount time, so no need to take | |
7296 | * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, | |
7297 | * we always lock first fs_info->chunk_mutex before | |
7298 | * acquiring any locks on the chunk tree. This is a | |
7299 | * requirement for chunk allocation, see the comment on | |
7300 | * top of btrfs_chunk_alloc() for details. | |
7301 | */ | |
0b86a832 | 7302 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
9690ac09 | 7303 | ret = read_one_chunk(&found_key, leaf, chunk); |
2b82032c YZ |
7304 | if (ret) |
7305 | goto error; | |
0b86a832 | 7306 | } |
43cb1478 GN |
7307 | } |
7308 | /* Catch error found during iteration */ | |
7309 | if (iter_ret < 0) { | |
7310 | ret = iter_ret; | |
7311 | goto error; | |
0b86a832 | 7312 | } |
99e3ecfc LB |
7313 | |
7314 | /* | |
7315 | * After loading chunk tree, we've got all device information, | |
7316 | * do another round of validation checks. | |
7317 | */ | |
0b246afa | 7318 | if (total_dev != fs_info->fs_devices->total_devices) { |
d201238c QW |
7319 | btrfs_warn(fs_info, |
7320 | "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", | |
0b246afa | 7321 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc | 7322 | total_dev); |
d201238c QW |
7323 | fs_info->fs_devices->total_devices = total_dev; |
7324 | btrfs_set_super_num_devices(fs_info->super_copy, total_dev); | |
99e3ecfc | 7325 | } |
0b246afa JM |
7326 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
7327 | fs_info->fs_devices->total_rw_bytes) { | |
7328 | btrfs_err(fs_info, | |
99e3ecfc | 7329 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
7330 | btrfs_super_total_bytes(fs_info->super_copy), |
7331 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
7332 | ret = -EINVAL; |
7333 | goto error; | |
7334 | } | |
0b86a832 CM |
7335 | ret = 0; |
7336 | error: | |
b367e47f LZ |
7337 | mutex_unlock(&uuid_mutex); |
7338 | ||
2b82032c | 7339 | btrfs_free_path(path); |
0b86a832 CM |
7340 | return ret; |
7341 | } | |
442a4f63 | 7342 | |
a8d1b164 | 7343 | int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
cb517eab | 7344 | { |
944d3f9f | 7345 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
cb517eab | 7346 | struct btrfs_device *device; |
a8d1b164 | 7347 | int ret = 0; |
cb517eab | 7348 | |
944d3f9f NB |
7349 | fs_devices->fs_info = fs_info; |
7350 | ||
7351 | mutex_lock(&fs_devices->device_list_mutex); | |
7352 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
7353 | device->fs_info = fs_info; | |
944d3f9f NB |
7354 | |
7355 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { | |
a8d1b164 | 7356 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
fb456252 | 7357 | device->fs_info = fs_info; |
a8d1b164 JT |
7358 | ret = btrfs_get_dev_zone_info(device, false); |
7359 | if (ret) | |
7360 | break; | |
7361 | } | |
29cc83f6 | 7362 | |
944d3f9f | 7363 | seed_devs->fs_info = fs_info; |
29cc83f6 | 7364 | } |
e17125b5 | 7365 | mutex_unlock(&fs_devices->device_list_mutex); |
a8d1b164 JT |
7366 | |
7367 | return ret; | |
cb517eab MX |
7368 | } |
7369 | ||
1dc990df DS |
7370 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
7371 | const struct btrfs_dev_stats_item *ptr, | |
7372 | int index) | |
7373 | { | |
7374 | u64 val; | |
7375 | ||
7376 | read_extent_buffer(eb, &val, | |
7377 | offsetof(struct btrfs_dev_stats_item, values) + | |
7378 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7379 | sizeof(val)); | |
7380 | return val; | |
7381 | } | |
7382 | ||
7383 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, | |
7384 | struct btrfs_dev_stats_item *ptr, | |
7385 | int index, u64 val) | |
7386 | { | |
7387 | write_extent_buffer(eb, &val, | |
7388 | offsetof(struct btrfs_dev_stats_item, values) + | |
7389 | ((unsigned long)ptr) + (index * sizeof(u64)), | |
7390 | sizeof(val)); | |
7391 | } | |
7392 | ||
92e26df4 JB |
7393 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
7394 | struct btrfs_path *path) | |
733f4fbb | 7395 | { |
124604eb | 7396 | struct btrfs_dev_stats_item *ptr; |
733f4fbb | 7397 | struct extent_buffer *eb; |
124604eb JB |
7398 | struct btrfs_key key; |
7399 | int item_size; | |
7400 | int i, ret, slot; | |
7401 | ||
82d62d06 JB |
7402 | if (!device->fs_info->dev_root) |
7403 | return 0; | |
7404 | ||
124604eb JB |
7405 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7406 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
7407 | key.offset = device->devid; | |
7408 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); | |
7409 | if (ret) { | |
7410 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7411 | btrfs_dev_stat_set(device, i, 0); | |
7412 | device->dev_stats_valid = 1; | |
7413 | btrfs_release_path(path); | |
92e26df4 | 7414 | return ret < 0 ? ret : 0; |
124604eb JB |
7415 | } |
7416 | slot = path->slots[0]; | |
7417 | eb = path->nodes[0]; | |
3212fa14 | 7418 | item_size = btrfs_item_size(eb, slot); |
124604eb JB |
7419 | |
7420 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); | |
7421 | ||
7422 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
7423 | if (item_size >= (1 + i) * sizeof(__le64)) | |
7424 | btrfs_dev_stat_set(device, i, | |
7425 | btrfs_dev_stats_value(eb, ptr, i)); | |
7426 | else | |
7427 | btrfs_dev_stat_set(device, i, 0); | |
7428 | } | |
7429 | ||
7430 | device->dev_stats_valid = 1; | |
7431 | btrfs_dev_stat_print_on_load(device); | |
7432 | btrfs_release_path(path); | |
92e26df4 JB |
7433 | |
7434 | return 0; | |
124604eb JB |
7435 | } |
7436 | ||
7437 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
7438 | { | |
7439 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; | |
733f4fbb SB |
7440 | struct btrfs_device *device; |
7441 | struct btrfs_path *path = NULL; | |
92e26df4 | 7442 | int ret = 0; |
733f4fbb SB |
7443 | |
7444 | path = btrfs_alloc_path(); | |
3b80a984 AJ |
7445 | if (!path) |
7446 | return -ENOMEM; | |
733f4fbb SB |
7447 | |
7448 | mutex_lock(&fs_devices->device_list_mutex); | |
92e26df4 JB |
7449 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
7450 | ret = btrfs_device_init_dev_stats(device, path); | |
7451 | if (ret) | |
7452 | goto out; | |
7453 | } | |
124604eb | 7454 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
92e26df4 JB |
7455 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
7456 | ret = btrfs_device_init_dev_stats(device, path); | |
7457 | if (ret) | |
7458 | goto out; | |
7459 | } | |
733f4fbb | 7460 | } |
92e26df4 | 7461 | out: |
733f4fbb SB |
7462 | mutex_unlock(&fs_devices->device_list_mutex); |
7463 | ||
733f4fbb | 7464 | btrfs_free_path(path); |
92e26df4 | 7465 | return ret; |
733f4fbb SB |
7466 | } |
7467 | ||
7468 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
733f4fbb SB |
7469 | struct btrfs_device *device) |
7470 | { | |
5495f195 | 7471 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 7472 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
7473 | struct btrfs_path *path; |
7474 | struct btrfs_key key; | |
7475 | struct extent_buffer *eb; | |
7476 | struct btrfs_dev_stats_item *ptr; | |
7477 | int ret; | |
7478 | int i; | |
7479 | ||
242e2956 DS |
7480 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
7481 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
7482 | key.offset = device->devid; |
7483 | ||
7484 | path = btrfs_alloc_path(); | |
fa252992 DS |
7485 | if (!path) |
7486 | return -ENOMEM; | |
733f4fbb SB |
7487 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
7488 | if (ret < 0) { | |
0b246afa | 7489 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7490 | "error %d while searching for dev_stats item for device %s", |
cb3e217b | 7491 | ret, btrfs_dev_name(device)); |
733f4fbb SB |
7492 | goto out; |
7493 | } | |
7494 | ||
7495 | if (ret == 0 && | |
3212fa14 | 7496 | btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { |
733f4fbb SB |
7497 | /* need to delete old one and insert a new one */ |
7498 | ret = btrfs_del_item(trans, dev_root, path); | |
7499 | if (ret != 0) { | |
0b246afa | 7500 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7501 | "delete too small dev_stats item for device %s failed %d", |
cb3e217b | 7502 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7503 | goto out; |
7504 | } | |
7505 | ret = 1; | |
7506 | } | |
7507 | ||
7508 | if (ret == 1) { | |
7509 | /* need to insert a new item */ | |
7510 | btrfs_release_path(path); | |
7511 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7512 | &key, sizeof(*ptr)); | |
7513 | if (ret < 0) { | |
0b246afa | 7514 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7515 | "insert dev_stats item for device %s failed %d", |
cb3e217b | 7516 | btrfs_dev_name(device), ret); |
733f4fbb SB |
7517 | goto out; |
7518 | } | |
7519 | } | |
7520 | ||
7521 | eb = path->nodes[0]; | |
7522 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7523 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7524 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7525 | btrfs_dev_stat_read(device, i)); | |
7526 | btrfs_mark_buffer_dirty(eb); | |
7527 | ||
7528 | out: | |
7529 | btrfs_free_path(path); | |
7530 | return ret; | |
7531 | } | |
7532 | ||
7533 | /* | |
7534 | * called from commit_transaction. Writes all changed device stats to disk. | |
7535 | */ | |
196c9d8d | 7536 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
733f4fbb | 7537 | { |
196c9d8d | 7538 | struct btrfs_fs_info *fs_info = trans->fs_info; |
733f4fbb SB |
7539 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7540 | struct btrfs_device *device; | |
addc3fa7 | 7541 | int stats_cnt; |
733f4fbb SB |
7542 | int ret = 0; |
7543 | ||
7544 | mutex_lock(&fs_devices->device_list_mutex); | |
7545 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
9deae968 NB |
7546 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
7547 | if (!device->dev_stats_valid || stats_cnt == 0) | |
733f4fbb SB |
7548 | continue; |
7549 | ||
9deae968 NB |
7550 | |
7551 | /* | |
7552 | * There is a LOAD-LOAD control dependency between the value of | |
7553 | * dev_stats_ccnt and updating the on-disk values which requires | |
7554 | * reading the in-memory counters. Such control dependencies | |
7555 | * require explicit read memory barriers. | |
7556 | * | |
7557 | * This memory barriers pairs with smp_mb__before_atomic in | |
7558 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full | |
7559 | * barrier implied by atomic_xchg in | |
7560 | * btrfs_dev_stats_read_and_reset | |
7561 | */ | |
7562 | smp_rmb(); | |
7563 | ||
5495f195 | 7564 | ret = update_dev_stat_item(trans, device); |
733f4fbb | 7565 | if (!ret) |
addc3fa7 | 7566 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7567 | } |
7568 | mutex_unlock(&fs_devices->device_list_mutex); | |
7569 | ||
7570 | return ret; | |
7571 | } | |
7572 | ||
442a4f63 SB |
7573 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7574 | { | |
7575 | btrfs_dev_stat_inc(dev, index); | |
442a4f63 | 7576 | |
733f4fbb SB |
7577 | if (!dev->dev_stats_valid) |
7578 | return; | |
fb456252 | 7579 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7580 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7581 | btrfs_dev_name(dev), |
442a4f63 SB |
7582 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7583 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7584 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7585 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7586 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7587 | } |
c11d2c23 | 7588 | |
733f4fbb SB |
7589 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7590 | { | |
a98cdb85 SB |
7591 | int i; |
7592 | ||
7593 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7594 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7595 | break; | |
7596 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7597 | return; /* all values == 0, suppress message */ | |
7598 | ||
fb456252 | 7599 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7600 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
cb3e217b | 7601 | btrfs_dev_name(dev), |
733f4fbb SB |
7602 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7603 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7604 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7605 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7606 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7607 | } | |
7608 | ||
2ff7e61e | 7609 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
b27f7c0c | 7610 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 | 7611 | { |
562d7b15 | 7612 | BTRFS_DEV_LOOKUP_ARGS(args); |
c11d2c23 | 7613 | struct btrfs_device *dev; |
0b246afa | 7614 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7615 | int i; |
7616 | ||
7617 | mutex_lock(&fs_devices->device_list_mutex); | |
562d7b15 JB |
7618 | args.devid = stats->devid; |
7619 | dev = btrfs_find_device(fs_info->fs_devices, &args); | |
c11d2c23 SB |
7620 | mutex_unlock(&fs_devices->device_list_mutex); |
7621 | ||
7622 | if (!dev) { | |
0b246afa | 7623 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7624 | return -ENODEV; |
733f4fbb | 7625 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7626 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7627 | return -ENODEV; |
b27f7c0c | 7628 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7629 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7630 | if (stats->nr_items > i) | |
7631 | stats->values[i] = | |
7632 | btrfs_dev_stat_read_and_reset(dev, i); | |
7633 | else | |
4e411a7d | 7634 | btrfs_dev_stat_set(dev, i, 0); |
c11d2c23 | 7635 | } |
a69976bc AJ |
7636 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
7637 | current->comm, task_pid_nr(current)); | |
c11d2c23 SB |
7638 | } else { |
7639 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7640 | if (stats->nr_items > i) | |
7641 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7642 | } | |
7643 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7644 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7645 | return 0; | |
7646 | } | |
a8a6dab7 | 7647 | |
935e5cc9 | 7648 | /* |
bbbf7243 NB |
7649 | * Update the size and bytes used for each device where it changed. This is |
7650 | * delayed since we would otherwise get errors while writing out the | |
7651 | * superblocks. | |
7652 | * | |
7653 | * Must be invoked during transaction commit. | |
935e5cc9 | 7654 | */ |
bbbf7243 | 7655 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
935e5cc9 | 7656 | { |
935e5cc9 MX |
7657 | struct btrfs_device *curr, *next; |
7658 | ||
bbbf7243 | 7659 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
ce7213c7 | 7660 | |
bbbf7243 | 7661 | if (list_empty(&trans->dev_update_list)) |
ce7213c7 MX |
7662 | return; |
7663 | ||
bbbf7243 NB |
7664 | /* |
7665 | * We don't need the device_list_mutex here. This list is owned by the | |
7666 | * transaction and the transaction must complete before the device is | |
7667 | * released. | |
7668 | */ | |
7669 | mutex_lock(&trans->fs_info->chunk_mutex); | |
7670 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, | |
7671 | post_commit_list) { | |
7672 | list_del_init(&curr->post_commit_list); | |
7673 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7674 | curr->commit_bytes_used = curr->bytes_used; | |
ce7213c7 | 7675 | } |
bbbf7243 | 7676 | mutex_unlock(&trans->fs_info->chunk_mutex); |
ce7213c7 | 7677 | } |
5a13f430 | 7678 | |
46df06b8 DS |
7679 | /* |
7680 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. | |
7681 | */ | |
7682 | int btrfs_bg_type_to_factor(u64 flags) | |
7683 | { | |
44b28ada DS |
7684 | const int index = btrfs_bg_flags_to_raid_index(flags); |
7685 | ||
7686 | return btrfs_raid_array[index].ncopies; | |
46df06b8 | 7687 | } |
cf90d884 QW |
7688 | |
7689 | ||
cf90d884 QW |
7690 | |
7691 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, | |
7692 | u64 chunk_offset, u64 devid, | |
7693 | u64 physical_offset, u64 physical_len) | |
7694 | { | |
562d7b15 | 7695 | struct btrfs_dev_lookup_args args = { .devid = devid }; |
c8bf1b67 | 7696 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7697 | struct extent_map *em; |
7698 | struct map_lookup *map; | |
05a37c48 | 7699 | struct btrfs_device *dev; |
cf90d884 QW |
7700 | u64 stripe_len; |
7701 | bool found = false; | |
7702 | int ret = 0; | |
7703 | int i; | |
7704 | ||
7705 | read_lock(&em_tree->lock); | |
7706 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
7707 | read_unlock(&em_tree->lock); | |
7708 | ||
7709 | if (!em) { | |
7710 | btrfs_err(fs_info, | |
7711 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", | |
7712 | physical_offset, devid); | |
7713 | ret = -EUCLEAN; | |
7714 | goto out; | |
7715 | } | |
7716 | ||
7717 | map = em->map_lookup; | |
bc88b486 | 7718 | stripe_len = btrfs_calc_stripe_length(em); |
cf90d884 QW |
7719 | if (physical_len != stripe_len) { |
7720 | btrfs_err(fs_info, | |
7721 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", | |
7722 | physical_offset, devid, em->start, physical_len, | |
7723 | stripe_len); | |
7724 | ret = -EUCLEAN; | |
7725 | goto out; | |
7726 | } | |
7727 | ||
3613249a QW |
7728 | /* |
7729 | * Very old mkfs.btrfs (before v4.1) will not respect the reserved | |
7730 | * space. Although kernel can handle it without problem, better to warn | |
7731 | * the users. | |
7732 | */ | |
7733 | if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) | |
7734 | btrfs_warn(fs_info, | |
7735 | "devid %llu physical %llu len %llu inside the reserved space", | |
7736 | devid, physical_offset, physical_len); | |
7737 | ||
cf90d884 QW |
7738 | for (i = 0; i < map->num_stripes; i++) { |
7739 | if (map->stripes[i].dev->devid == devid && | |
7740 | map->stripes[i].physical == physical_offset) { | |
7741 | found = true; | |
7742 | if (map->verified_stripes >= map->num_stripes) { | |
7743 | btrfs_err(fs_info, | |
7744 | "too many dev extents for chunk %llu found", | |
7745 | em->start); | |
7746 | ret = -EUCLEAN; | |
7747 | goto out; | |
7748 | } | |
7749 | map->verified_stripes++; | |
7750 | break; | |
7751 | } | |
7752 | } | |
7753 | if (!found) { | |
7754 | btrfs_err(fs_info, | |
7755 | "dev extent physical offset %llu devid %llu has no corresponding chunk", | |
7756 | physical_offset, devid); | |
7757 | ret = -EUCLEAN; | |
7758 | } | |
05a37c48 | 7759 | |
1a9fd417 | 7760 | /* Make sure no dev extent is beyond device boundary */ |
562d7b15 | 7761 | dev = btrfs_find_device(fs_info->fs_devices, &args); |
05a37c48 QW |
7762 | if (!dev) { |
7763 | btrfs_err(fs_info, "failed to find devid %llu", devid); | |
7764 | ret = -EUCLEAN; | |
7765 | goto out; | |
7766 | } | |
1b3922a8 | 7767 | |
05a37c48 QW |
7768 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
7769 | btrfs_err(fs_info, | |
7770 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", | |
7771 | devid, physical_offset, physical_len, | |
7772 | dev->disk_total_bytes); | |
7773 | ret = -EUCLEAN; | |
7774 | goto out; | |
7775 | } | |
381a696e NA |
7776 | |
7777 | if (dev->zone_info) { | |
7778 | u64 zone_size = dev->zone_info->zone_size; | |
7779 | ||
7780 | if (!IS_ALIGNED(physical_offset, zone_size) || | |
7781 | !IS_ALIGNED(physical_len, zone_size)) { | |
7782 | btrfs_err(fs_info, | |
7783 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", | |
7784 | devid, physical_offset, physical_len); | |
7785 | ret = -EUCLEAN; | |
7786 | goto out; | |
7787 | } | |
7788 | } | |
7789 | ||
cf90d884 QW |
7790 | out: |
7791 | free_extent_map(em); | |
7792 | return ret; | |
7793 | } | |
7794 | ||
7795 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) | |
7796 | { | |
c8bf1b67 | 7797 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
cf90d884 QW |
7798 | struct extent_map *em; |
7799 | struct rb_node *node; | |
7800 | int ret = 0; | |
7801 | ||
7802 | read_lock(&em_tree->lock); | |
07e1ce09 | 7803 | for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) { |
cf90d884 QW |
7804 | em = rb_entry(node, struct extent_map, rb_node); |
7805 | if (em->map_lookup->num_stripes != | |
7806 | em->map_lookup->verified_stripes) { | |
7807 | btrfs_err(fs_info, | |
7808 | "chunk %llu has missing dev extent, have %d expect %d", | |
7809 | em->start, em->map_lookup->verified_stripes, | |
7810 | em->map_lookup->num_stripes); | |
7811 | ret = -EUCLEAN; | |
7812 | goto out; | |
7813 | } | |
7814 | } | |
7815 | out: | |
7816 | read_unlock(&em_tree->lock); | |
7817 | return ret; | |
7818 | } | |
7819 | ||
7820 | /* | |
7821 | * Ensure that all dev extents are mapped to correct chunk, otherwise | |
7822 | * later chunk allocation/free would cause unexpected behavior. | |
7823 | * | |
7824 | * NOTE: This will iterate through the whole device tree, which should be of | |
7825 | * the same size level as the chunk tree. This slightly increases mount time. | |
7826 | */ | |
7827 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) | |
7828 | { | |
7829 | struct btrfs_path *path; | |
7830 | struct btrfs_root *root = fs_info->dev_root; | |
7831 | struct btrfs_key key; | |
5eb19381 QW |
7832 | u64 prev_devid = 0; |
7833 | u64 prev_dev_ext_end = 0; | |
cf90d884 QW |
7834 | int ret = 0; |
7835 | ||
42437a63 JB |
7836 | /* |
7837 | * We don't have a dev_root because we mounted with ignorebadroots and | |
7838 | * failed to load the root, so we want to skip the verification in this | |
7839 | * case for sure. | |
7840 | * | |
7841 | * However if the dev root is fine, but the tree itself is corrupted | |
7842 | * we'd still fail to mount. This verification is only to make sure | |
7843 | * writes can happen safely, so instead just bypass this check | |
7844 | * completely in the case of IGNOREBADROOTS. | |
7845 | */ | |
7846 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) | |
7847 | return 0; | |
7848 | ||
cf90d884 QW |
7849 | key.objectid = 1; |
7850 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7851 | key.offset = 0; | |
7852 | ||
7853 | path = btrfs_alloc_path(); | |
7854 | if (!path) | |
7855 | return -ENOMEM; | |
7856 | ||
7857 | path->reada = READA_FORWARD; | |
7858 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
7859 | if (ret < 0) | |
7860 | goto out; | |
7861 | ||
7862 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
ad9a9378 | 7863 | ret = btrfs_next_leaf(root, path); |
cf90d884 QW |
7864 | if (ret < 0) |
7865 | goto out; | |
7866 | /* No dev extents at all? Not good */ | |
7867 | if (ret > 0) { | |
7868 | ret = -EUCLEAN; | |
7869 | goto out; | |
7870 | } | |
7871 | } | |
7872 | while (1) { | |
7873 | struct extent_buffer *leaf = path->nodes[0]; | |
7874 | struct btrfs_dev_extent *dext; | |
7875 | int slot = path->slots[0]; | |
7876 | u64 chunk_offset; | |
7877 | u64 physical_offset; | |
7878 | u64 physical_len; | |
7879 | u64 devid; | |
7880 | ||
7881 | btrfs_item_key_to_cpu(leaf, &key, slot); | |
7882 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
7883 | break; | |
7884 | devid = key.objectid; | |
7885 | physical_offset = key.offset; | |
7886 | ||
7887 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); | |
7888 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); | |
7889 | physical_len = btrfs_dev_extent_length(leaf, dext); | |
7890 | ||
5eb19381 QW |
7891 | /* Check if this dev extent overlaps with the previous one */ |
7892 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { | |
7893 | btrfs_err(fs_info, | |
7894 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", | |
7895 | devid, physical_offset, prev_dev_ext_end); | |
7896 | ret = -EUCLEAN; | |
7897 | goto out; | |
7898 | } | |
7899 | ||
cf90d884 QW |
7900 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
7901 | physical_offset, physical_len); | |
7902 | if (ret < 0) | |
7903 | goto out; | |
5eb19381 QW |
7904 | prev_devid = devid; |
7905 | prev_dev_ext_end = physical_offset + physical_len; | |
7906 | ||
cf90d884 QW |
7907 | ret = btrfs_next_item(root, path); |
7908 | if (ret < 0) | |
7909 | goto out; | |
7910 | if (ret > 0) { | |
7911 | ret = 0; | |
7912 | break; | |
7913 | } | |
7914 | } | |
7915 | ||
7916 | /* Ensure all chunks have corresponding dev extents */ | |
7917 | ret = verify_chunk_dev_extent_mapping(fs_info); | |
7918 | out: | |
7919 | btrfs_free_path(path); | |
7920 | return ret; | |
7921 | } | |
eede2bf3 OS |
7922 | |
7923 | /* | |
7924 | * Check whether the given block group or device is pinned by any inode being | |
7925 | * used as a swapfile. | |
7926 | */ | |
7927 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) | |
7928 | { | |
7929 | struct btrfs_swapfile_pin *sp; | |
7930 | struct rb_node *node; | |
7931 | ||
7932 | spin_lock(&fs_info->swapfile_pins_lock); | |
7933 | node = fs_info->swapfile_pins.rb_node; | |
7934 | while (node) { | |
7935 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); | |
7936 | if (ptr < sp->ptr) | |
7937 | node = node->rb_left; | |
7938 | else if (ptr > sp->ptr) | |
7939 | node = node->rb_right; | |
7940 | else | |
7941 | break; | |
7942 | } | |
7943 | spin_unlock(&fs_info->swapfile_pins_lock); | |
7944 | return node != NULL; | |
7945 | } | |
f7ef5287 NA |
7946 | |
7947 | static int relocating_repair_kthread(void *data) | |
7948 | { | |
0d031dc4 | 7949 | struct btrfs_block_group *cache = data; |
f7ef5287 NA |
7950 | struct btrfs_fs_info *fs_info = cache->fs_info; |
7951 | u64 target; | |
7952 | int ret = 0; | |
7953 | ||
7954 | target = cache->start; | |
7955 | btrfs_put_block_group(cache); | |
7956 | ||
ca5e4ea0 | 7957 | sb_start_write(fs_info->sb); |
f7ef5287 NA |
7958 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { |
7959 | btrfs_info(fs_info, | |
7960 | "zoned: skip relocating block group %llu to repair: EBUSY", | |
7961 | target); | |
ca5e4ea0 | 7962 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
7963 | return -EBUSY; |
7964 | } | |
7965 | ||
f3372065 | 7966 | mutex_lock(&fs_info->reclaim_bgs_lock); |
f7ef5287 NA |
7967 | |
7968 | /* Ensure block group still exists */ | |
7969 | cache = btrfs_lookup_block_group(fs_info, target); | |
7970 | if (!cache) | |
7971 | goto out; | |
7972 | ||
3349b57f | 7973 | if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) |
f7ef5287 NA |
7974 | goto out; |
7975 | ||
7976 | ret = btrfs_may_alloc_data_chunk(fs_info, target); | |
7977 | if (ret < 0) | |
7978 | goto out; | |
7979 | ||
7980 | btrfs_info(fs_info, | |
7981 | "zoned: relocating block group %llu to repair IO failure", | |
7982 | target); | |
7983 | ret = btrfs_relocate_chunk(fs_info, target); | |
7984 | ||
7985 | out: | |
7986 | if (cache) | |
7987 | btrfs_put_block_group(cache); | |
f3372065 | 7988 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
f7ef5287 | 7989 | btrfs_exclop_finish(fs_info); |
ca5e4ea0 | 7990 | sb_end_write(fs_info->sb); |
f7ef5287 NA |
7991 | |
7992 | return ret; | |
7993 | } | |
7994 | ||
554aed7d | 7995 | bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) |
f7ef5287 NA |
7996 | { |
7997 | struct btrfs_block_group *cache; | |
7998 | ||
554aed7d JT |
7999 | if (!btrfs_is_zoned(fs_info)) |
8000 | return false; | |
8001 | ||
f7ef5287 NA |
8002 | /* Do not attempt to repair in degraded state */ |
8003 | if (btrfs_test_opt(fs_info, DEGRADED)) | |
554aed7d | 8004 | return true; |
f7ef5287 NA |
8005 | |
8006 | cache = btrfs_lookup_block_group(fs_info, logical); | |
8007 | if (!cache) | |
554aed7d | 8008 | return true; |
f7ef5287 | 8009 | |
3349b57f | 8010 | if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { |
f7ef5287 | 8011 | btrfs_put_block_group(cache); |
554aed7d | 8012 | return true; |
f7ef5287 | 8013 | } |
f7ef5287 NA |
8014 | |
8015 | kthread_run(relocating_repair_kthread, cache, | |
8016 | "btrfs-relocating-repair"); | |
8017 | ||
554aed7d | 8018 | return true; |
f7ef5287 | 8019 | } |
4886ff7b QW |
8020 | |
8021 | static void map_raid56_repair_block(struct btrfs_io_context *bioc, | |
8022 | struct btrfs_io_stripe *smap, | |
8023 | u64 logical) | |
8024 | { | |
8025 | int data_stripes = nr_bioc_data_stripes(bioc); | |
8026 | int i; | |
8027 | ||
8028 | for (i = 0; i < data_stripes; i++) { | |
8029 | u64 stripe_start = bioc->full_stripe_logical + | |
8030 | (i << BTRFS_STRIPE_LEN_SHIFT); | |
8031 | ||
8032 | if (logical >= stripe_start && | |
8033 | logical < stripe_start + BTRFS_STRIPE_LEN) | |
8034 | break; | |
8035 | } | |
8036 | ASSERT(i < data_stripes); | |
8037 | smap->dev = bioc->stripes[i].dev; | |
8038 | smap->physical = bioc->stripes[i].physical + | |
8039 | ((logical - bioc->full_stripe_logical) & | |
8040 | BTRFS_STRIPE_LEN_MASK); | |
8041 | } | |
8042 | ||
8043 | /* | |
8044 | * Map a repair write into a single device. | |
8045 | * | |
8046 | * A repair write is triggered by read time repair or scrub, which would only | |
8047 | * update the contents of a single device. | |
8048 | * Not update any other mirrors nor go through RMW path. | |
8049 | * | |
8050 | * Callers should ensure: | |
8051 | * | |
8052 | * - Call btrfs_bio_counter_inc_blocked() first | |
8053 | * - The range does not cross stripe boundary | |
8054 | * - Has a valid @mirror_num passed in. | |
8055 | */ | |
8056 | int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, | |
8057 | struct btrfs_io_stripe *smap, u64 logical, | |
8058 | u32 length, int mirror_num) | |
8059 | { | |
8060 | struct btrfs_io_context *bioc = NULL; | |
8061 | u64 map_length = length; | |
8062 | int mirror_ret = mirror_num; | |
8063 | int ret; | |
8064 | ||
8065 | ASSERT(mirror_num > 0); | |
8066 | ||
8067 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, | |
8068 | &bioc, smap, &mirror_ret, true); | |
8069 | if (ret < 0) | |
8070 | return ret; | |
8071 | ||
8072 | /* The map range should not cross stripe boundary. */ | |
8073 | ASSERT(map_length >= length); | |
8074 | ||
8075 | /* Already mapped to single stripe. */ | |
8076 | if (!bioc) | |
8077 | goto out; | |
8078 | ||
8079 | /* Map the RAID56 multi-stripe writes to a single one. */ | |
8080 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
8081 | map_raid56_repair_block(bioc, smap, logical); | |
8082 | goto out; | |
8083 | } | |
8084 | ||
8085 | ASSERT(mirror_num <= bioc->num_stripes); | |
8086 | smap->dev = bioc->stripes[mirror_num - 1].dev; | |
8087 | smap->physical = bioc->stripes[mirror_num - 1].physical; | |
8088 | out: | |
8089 | btrfs_put_bioc(bioc); | |
8090 | ASSERT(smap->dev); | |
8091 | return 0; | |
8092 | } |