Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * raid10.c : Multiple Devices driver for Linux | |
4 | * | |
5 | * Copyright (C) 2000-2004 Neil Brown | |
6 | * | |
7 | * RAID-10 support for md. | |
8 | * | |
25985edc | 9 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
10 | */ |
11 | ||
5a0e3ad6 | 12 | #include <linux/slab.h> |
25570727 | 13 | #include <linux/delay.h> |
bff61975 | 14 | #include <linux/blkdev.h> |
056075c7 | 15 | #include <linux/module.h> |
bff61975 | 16 | #include <linux/seq_file.h> |
8bda470e | 17 | #include <linux/ratelimit.h> |
3ea7daa5 | 18 | #include <linux/kthread.h> |
afd75628 | 19 | #include <linux/raid/md_p.h> |
109e3765 | 20 | #include <trace/events/block.h> |
43b2e5d8 | 21 | #include "md.h" |
1979dbbe LN |
22 | |
23 | #define RAID_1_10_NAME "raid10" | |
ef740c37 | 24 | #include "raid10.h" |
dab8b292 | 25 | #include "raid0.h" |
935fe098 | 26 | #include "md-bitmap.h" |
1da177e4 LT |
27 | |
28 | /* | |
29 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
30 | * The layout of data is defined by | |
31 | * chunk_size | |
32 | * raid_disks | |
33 | * near_copies (stored in low byte of layout) | |
34 | * far_copies (stored in second byte of layout) | |
c93983bf | 35 | * far_offset (stored in bit 16 of layout ) |
475901af | 36 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 37 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 38 | * |
475901af JB |
39 | * The data to be stored is divided into chunks using chunksize. Each device |
40 | * is divided into far_copies sections. In each section, chunks are laid out | |
41 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
42 | * (each on a different drive). The starting device for each section is offset | |
43 | * near_copies from the starting device of the previous section. Thus there | |
44 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
45 | * drive. near_copies and far_copies must be at least one, and their product | |
46 | * is at most raid_disks. | |
c93983bf N |
47 | * |
48 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
49 | * The copies are still in different stripes, but instead of being very far |
50 | * apart on disk, there are adjacent stripes. | |
51 | * | |
52 | * The far and offset algorithms are handled slightly differently if | |
53 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
54 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
55 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
56 | * to the set rather than the entire array. This is done to improve the number | |
57 | * of device combinations that can fail without causing the array to fail. | |
58 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
59 | * on a device): | |
60 | * A B C D A B C D E | |
61 | * ... ... | |
62 | * D A B C E A B C D | |
63 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
64 | * [A B] [C D] [A B] [C D E] | |
65 | * |...| |...| |...| | ... | | |
66 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
67 | */ |
68 | ||
e879a879 N |
69 | static void allow_barrier(struct r10conf *conf); |
70 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 71 | static int _enough(struct r10conf *conf, int previous, int ignore); |
1919cbb2 | 72 | static int enough(struct r10conf *conf, int ignore); |
3ea7daa5 N |
73 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
74 | int *skipped); | |
75 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 76 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 77 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 78 | |
fb0eb5df ML |
79 | #include "raid1-10.c" |
80 | ||
b9b083f9 YK |
81 | #define NULL_CMD |
82 | #define cmd_before(conf, cmd) \ | |
83 | do { \ | |
84 | write_sequnlock_irq(&(conf)->resync_lock); \ | |
85 | cmd; \ | |
86 | } while (0) | |
87 | #define cmd_after(conf) write_seqlock_irq(&(conf)->resync_lock) | |
88 | ||
89 | #define wait_event_barrier_cmd(conf, cond, cmd) \ | |
90 | wait_event_cmd((conf)->wait_barrier, cond, cmd_before(conf, cmd), \ | |
91 | cmd_after(conf)) | |
92 | ||
93 | #define wait_event_barrier(conf, cond) \ | |
94 | wait_event_barrier_cmd(conf, cond, NULL_CMD) | |
95 | ||
f0250618 ML |
96 | /* |
97 | * for resync bio, r10bio pointer can be retrieved from the per-bio | |
98 | * 'struct resync_pages'. | |
99 | */ | |
100 | static inline struct r10bio *get_resync_r10bio(struct bio *bio) | |
101 | { | |
102 | return get_resync_pages(bio)->raid_bio; | |
103 | } | |
104 | ||
dd0fc66f | 105 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 106 | { |
e879a879 | 107 | struct r10conf *conf = data; |
c2968285 | 108 | int size = offsetof(struct r10bio, devs[conf->geo.raid_disks]); |
1da177e4 | 109 | |
69335ef3 N |
110 | /* allocate a r10bio with room for raid_disks entries in the |
111 | * bios array */ | |
7eaceacc | 112 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
113 | } |
114 | ||
8db87912 | 115 | #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) |
0310fa21 N |
116 | /* amount of memory to reserve for resync requests */ |
117 | #define RESYNC_WINDOW (1024*1024) | |
118 | /* maximum number of concurrent requests, memory permitting */ | |
119 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
4b242e97 | 120 | #define CLUSTER_RESYNC_WINDOW (32 * RESYNC_WINDOW) |
8db87912 | 121 | #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9) |
1da177e4 LT |
122 | |
123 | /* | |
124 | * When performing a resync, we need to read and compare, so | |
125 | * we need as many pages are there are copies. | |
126 | * When performing a recovery, we need 2 bios, one for read, | |
127 | * one for write (we recover only one drive per r10buf) | |
128 | * | |
129 | */ | |
dd0fc66f | 130 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 131 | { |
e879a879 | 132 | struct r10conf *conf = data; |
9f2c9d12 | 133 | struct r10bio *r10_bio; |
1da177e4 | 134 | struct bio *bio; |
f0250618 ML |
135 | int j; |
136 | int nalloc, nalloc_rp; | |
137 | struct resync_pages *rps; | |
1da177e4 LT |
138 | |
139 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 140 | if (!r10_bio) |
1da177e4 | 141 | return NULL; |
1da177e4 | 142 | |
3ea7daa5 N |
143 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
144 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
145 | nalloc = conf->copies; /* resync */ |
146 | else | |
147 | nalloc = 2; /* recovery */ | |
148 | ||
f0250618 ML |
149 | /* allocate once for all bios */ |
150 | if (!conf->have_replacement) | |
151 | nalloc_rp = nalloc; | |
152 | else | |
153 | nalloc_rp = nalloc * 2; | |
6da2ec56 | 154 | rps = kmalloc_array(nalloc_rp, sizeof(struct resync_pages), gfp_flags); |
f0250618 ML |
155 | if (!rps) |
156 | goto out_free_r10bio; | |
157 | ||
1da177e4 LT |
158 | /* |
159 | * Allocate bios. | |
160 | */ | |
161 | for (j = nalloc ; j-- ; ) { | |
066ff571 | 162 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags); |
1da177e4 LT |
163 | if (!bio) |
164 | goto out_free_bio; | |
066ff571 | 165 | bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); |
1da177e4 | 166 | r10_bio->devs[j].bio = bio; |
69335ef3 N |
167 | if (!conf->have_replacement) |
168 | continue; | |
066ff571 | 169 | bio = bio_kmalloc(RESYNC_PAGES, gfp_flags); |
69335ef3 N |
170 | if (!bio) |
171 | goto out_free_bio; | |
066ff571 | 172 | bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); |
69335ef3 | 173 | r10_bio->devs[j].repl_bio = bio; |
1da177e4 LT |
174 | } |
175 | /* | |
176 | * Allocate RESYNC_PAGES data pages and attach them | |
177 | * where needed. | |
178 | */ | |
f0250618 | 179 | for (j = 0; j < nalloc; j++) { |
69335ef3 | 180 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
f0250618 ML |
181 | struct resync_pages *rp, *rp_repl; |
182 | ||
183 | rp = &rps[j]; | |
184 | if (rbio) | |
185 | rp_repl = &rps[nalloc + j]; | |
186 | ||
1da177e4 | 187 | bio = r10_bio->devs[j].bio; |
f0250618 ML |
188 | |
189 | if (!j || test_bit(MD_RECOVERY_SYNC, | |
190 | &conf->mddev->recovery)) { | |
191 | if (resync_alloc_pages(rp, gfp_flags)) | |
1da177e4 | 192 | goto out_free_pages; |
f0250618 ML |
193 | } else { |
194 | memcpy(rp, &rps[0], sizeof(*rp)); | |
195 | resync_get_all_pages(rp); | |
196 | } | |
1da177e4 | 197 | |
f0250618 ML |
198 | rp->raid_bio = r10_bio; |
199 | bio->bi_private = rp; | |
200 | if (rbio) { | |
201 | memcpy(rp_repl, rp, sizeof(*rp)); | |
202 | rbio->bi_private = rp_repl; | |
1da177e4 LT |
203 | } |
204 | } | |
205 | ||
206 | return r10_bio; | |
207 | ||
208 | out_free_pages: | |
f0250618 | 209 | while (--j >= 0) |
45422b70 | 210 | resync_free_pages(&rps[j]); |
f0250618 | 211 | |
5fdd2cf8 | 212 | j = 0; |
1da177e4 | 213 | out_free_bio: |
5fdd2cf8 | 214 | for ( ; j < nalloc; j++) { |
215 | if (r10_bio->devs[j].bio) | |
066ff571 CH |
216 | bio_uninit(r10_bio->devs[j].bio); |
217 | kfree(r10_bio->devs[j].bio); | |
69335ef3 | 218 | if (r10_bio->devs[j].repl_bio) |
066ff571 CH |
219 | bio_uninit(r10_bio->devs[j].repl_bio); |
220 | kfree(r10_bio->devs[j].repl_bio); | |
69335ef3 | 221 | } |
f0250618 ML |
222 | kfree(rps); |
223 | out_free_r10bio: | |
c7afa803 | 224 | rbio_pool_free(r10_bio, conf); |
1da177e4 LT |
225 | return NULL; |
226 | } | |
227 | ||
228 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
229 | { | |
e879a879 | 230 | struct r10conf *conf = data; |
9f2c9d12 | 231 | struct r10bio *r10bio = __r10_bio; |
1da177e4 | 232 | int j; |
f0250618 | 233 | struct resync_pages *rp = NULL; |
1da177e4 | 234 | |
f0250618 | 235 | for (j = conf->copies; j--; ) { |
1da177e4 | 236 | struct bio *bio = r10bio->devs[j].bio; |
f0250618 | 237 | |
eb81b328 GJ |
238 | if (bio) { |
239 | rp = get_resync_pages(bio); | |
240 | resync_free_pages(rp); | |
066ff571 CH |
241 | bio_uninit(bio); |
242 | kfree(bio); | |
eb81b328 | 243 | } |
f0250618 | 244 | |
69335ef3 | 245 | bio = r10bio->devs[j].repl_bio; |
066ff571 CH |
246 | if (bio) { |
247 | bio_uninit(bio); | |
248 | kfree(bio); | |
249 | } | |
1da177e4 | 250 | } |
f0250618 ML |
251 | |
252 | /* resync pages array stored in the 1st bio's .bi_private */ | |
253 | kfree(rp); | |
254 | ||
c7afa803 | 255 | rbio_pool_free(r10bio, conf); |
1da177e4 LT |
256 | } |
257 | ||
e879a879 | 258 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
259 | { |
260 | int i; | |
261 | ||
c2968285 | 262 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 263 | struct bio **bio = & r10_bio->devs[i].bio; |
749c55e9 | 264 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
265 | bio_put(*bio); |
266 | *bio = NULL; | |
69335ef3 N |
267 | bio = &r10_bio->devs[i].repl_bio; |
268 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
269 | bio_put(*bio); | |
270 | *bio = NULL; | |
1da177e4 LT |
271 | } |
272 | } | |
273 | ||
9f2c9d12 | 274 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 275 | { |
e879a879 | 276 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 277 | |
1da177e4 | 278 | put_all_bios(conf, r10_bio); |
afeee514 | 279 | mempool_free(r10_bio, &conf->r10bio_pool); |
1da177e4 LT |
280 | } |
281 | ||
9f2c9d12 | 282 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 283 | { |
e879a879 | 284 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 285 | |
afeee514 | 286 | mempool_free(r10_bio, &conf->r10buf_pool); |
1da177e4 | 287 | |
0a27ec96 | 288 | lower_barrier(conf); |
1da177e4 LT |
289 | } |
290 | ||
0c0be98b YK |
291 | static void wake_up_barrier(struct r10conf *conf) |
292 | { | |
293 | if (wq_has_sleeper(&conf->wait_barrier)) | |
294 | wake_up(&conf->wait_barrier); | |
295 | } | |
296 | ||
9f2c9d12 | 297 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
298 | { |
299 | unsigned long flags; | |
fd01b88c | 300 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 301 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
302 | |
303 | spin_lock_irqsave(&conf->device_lock, flags); | |
304 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 305 | conf->nr_queued ++; |
1da177e4 LT |
306 | spin_unlock_irqrestore(&conf->device_lock, flags); |
307 | ||
388667be AJ |
308 | /* wake up frozen array... */ |
309 | wake_up(&conf->wait_barrier); | |
310 | ||
1da177e4 LT |
311 | md_wakeup_thread(mddev->thread); |
312 | } | |
313 | ||
314 | /* | |
315 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
316 | * operation and are ready to return a success/failure code to the buffer | |
317 | * cache layer. | |
318 | */ | |
9f2c9d12 | 319 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
320 | { |
321 | struct bio *bio = r10_bio->master_bio; | |
e879a879 | 322 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 323 | |
856e08e2 | 324 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) |
4e4cbee9 | 325 | bio->bi_status = BLK_STS_IOERR; |
fd16f2e8 N |
326 | |
327 | bio_endio(bio); | |
328 | /* | |
329 | * Wake up any possible resync thread that waits for the device | |
330 | * to go idle. | |
331 | */ | |
332 | allow_barrier(conf); | |
333 | ||
1da177e4 LT |
334 | free_r10bio(r10_bio); |
335 | } | |
336 | ||
337 | /* | |
338 | * Update disk head position estimator based on IRQ completion info. | |
339 | */ | |
9f2c9d12 | 340 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 341 | { |
e879a879 | 342 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
343 | |
344 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
345 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
346 | } | |
347 | ||
778ca018 NK |
348 | /* |
349 | * Find the disk number which triggered given bio | |
350 | */ | |
e879a879 | 351 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 352 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
353 | { |
354 | int slot; | |
69335ef3 | 355 | int repl = 0; |
778ca018 | 356 | |
c2968285 | 357 | for (slot = 0; slot < conf->geo.raid_disks; slot++) { |
778ca018 NK |
358 | if (r10_bio->devs[slot].bio == bio) |
359 | break; | |
69335ef3 N |
360 | if (r10_bio->devs[slot].repl_bio == bio) { |
361 | repl = 1; | |
362 | break; | |
363 | } | |
364 | } | |
778ca018 | 365 | |
778ca018 NK |
366 | update_head_pos(slot, r10_bio); |
367 | ||
749c55e9 N |
368 | if (slotp) |
369 | *slotp = slot; | |
69335ef3 N |
370 | if (replp) |
371 | *replp = repl; | |
778ca018 NK |
372 | return r10_bio->devs[slot].devnum; |
373 | } | |
374 | ||
4246a0b6 | 375 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 376 | { |
4e4cbee9 | 377 | int uptodate = !bio->bi_status; |
9f2c9d12 | 378 | struct r10bio *r10_bio = bio->bi_private; |
a0e764c5 | 379 | int slot; |
abbf098e | 380 | struct md_rdev *rdev; |
e879a879 | 381 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 382 | |
1da177e4 | 383 | slot = r10_bio->read_slot; |
abbf098e | 384 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
385 | /* |
386 | * this branch is our 'one mirror IO has finished' event handler: | |
387 | */ | |
4443ae10 N |
388 | update_head_pos(slot, r10_bio); |
389 | ||
390 | if (uptodate) { | |
1da177e4 LT |
391 | /* |
392 | * Set R10BIO_Uptodate in our master bio, so that | |
393 | * we will return a good error code to the higher | |
394 | * levels even if IO on some other mirrored buffer fails. | |
395 | * | |
396 | * The 'master' represents the composite IO operation to | |
397 | * user-side. So if something waits for IO, then it will | |
398 | * wait for the 'master' bio. | |
399 | */ | |
400 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
401 | } else { |
402 | /* If all other devices that store this block have | |
403 | * failed, we want to return the error upwards rather | |
404 | * than fail the last device. Here we redefine | |
405 | * "uptodate" to mean "Don't want to retry" | |
406 | */ | |
635f6416 N |
407 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
408 | rdev->raid_disk)) | |
fae8cc5e | 409 | uptodate = 1; |
fae8cc5e N |
410 | } |
411 | if (uptodate) { | |
1da177e4 | 412 | raid_end_bio_io(r10_bio); |
abbf098e | 413 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 414 | } else { |
1da177e4 | 415 | /* |
7c4e06ff | 416 | * oops, read error - keep the refcount on the rdev |
1da177e4 | 417 | */ |
913cce5a | 418 | pr_err_ratelimited("md/raid10:%s: %pg: rescheduling sector %llu\n", |
8bda470e | 419 | mdname(conf->mddev), |
913cce5a | 420 | rdev->bdev, |
8bda470e | 421 | (unsigned long long)r10_bio->sector); |
856e08e2 | 422 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
423 | reschedule_retry(r10_bio); |
424 | } | |
1da177e4 LT |
425 | } |
426 | ||
9f2c9d12 | 427 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
428 | { |
429 | /* clear the bitmap if all writes complete successfully */ | |
e64e4018 AS |
430 | md_bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, |
431 | r10_bio->sectors, | |
432 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
433 | 0); | |
bd870a16 N |
434 | md_write_end(r10_bio->mddev); |
435 | } | |
436 | ||
9f2c9d12 | 437 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
438 | { |
439 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
440 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
441 | reschedule_retry(r10_bio); | |
442 | else { | |
443 | close_write(r10_bio); | |
444 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
445 | reschedule_retry(r10_bio); | |
446 | else | |
447 | raid_end_bio_io(r10_bio); | |
448 | } | |
449 | } | |
450 | } | |
451 | ||
4246a0b6 | 452 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 453 | { |
9f2c9d12 | 454 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 455 | int dev; |
749c55e9 | 456 | int dec_rdev = 1; |
e879a879 | 457 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 458 | int slot, repl; |
4ca40c2c | 459 | struct md_rdev *rdev = NULL; |
1919cbb2 | 460 | struct bio *to_put = NULL; |
579ed34f SL |
461 | bool discard_error; |
462 | ||
4e4cbee9 | 463 | discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; |
1da177e4 | 464 | |
475b0321 | 465 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 466 | |
475b0321 N |
467 | if (repl) |
468 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
469 | if (!rdev) { |
470 | smp_rmb(); | |
471 | repl = 0; | |
475b0321 | 472 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 473 | } |
1da177e4 LT |
474 | /* |
475 | * this branch is our 'one mirror IO has finished' event handler: | |
476 | */ | |
4e4cbee9 | 477 | if (bio->bi_status && !discard_error) { |
475b0321 N |
478 | if (repl) |
479 | /* Never record new bad blocks to replacement, | |
480 | * just fail it. | |
481 | */ | |
482 | md_error(rdev->mddev, rdev); | |
483 | else { | |
484 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
485 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
486 | set_bit(MD_RECOVERY_NEEDED, | |
487 | &rdev->mddev->recovery); | |
1919cbb2 | 488 | |
475b0321 | 489 | dec_rdev = 0; |
1919cbb2 N |
490 | if (test_bit(FailFast, &rdev->flags) && |
491 | (bio->bi_opf & MD_FAILFAST)) { | |
492 | md_error(rdev->mddev, rdev); | |
7cee6d4e YY |
493 | } |
494 | ||
495 | /* | |
496 | * When the device is faulty, it is not necessary to | |
497 | * handle write error. | |
7cee6d4e YY |
498 | */ |
499 | if (!test_bit(Faulty, &rdev->flags)) | |
1919cbb2 | 500 | set_bit(R10BIO_WriteError, &r10_bio->state); |
7cee6d4e | 501 | else { |
5ba03936 WS |
502 | /* Fail the request */ |
503 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
7cee6d4e YY |
504 | r10_bio->devs[slot].bio = NULL; |
505 | to_put = bio; | |
506 | dec_rdev = 1; | |
507 | } | |
475b0321 | 508 | } |
749c55e9 | 509 | } else { |
1da177e4 LT |
510 | /* |
511 | * Set R10BIO_Uptodate in our master bio, so that | |
512 | * we will return a good error code for to the higher | |
513 | * levels even if IO on some other mirrored buffer fails. | |
514 | * | |
515 | * The 'master' represents the composite IO operation to | |
516 | * user-side. So if something waits for IO, then it will | |
517 | * wait for the 'master' bio. | |
3a0f007b | 518 | * |
3056e3ae AL |
519 | * Do not set R10BIO_Uptodate if the current device is |
520 | * rebuilding or Faulty. This is because we cannot use | |
521 | * such device for properly reading the data back (we could | |
522 | * potentially use it, if the current write would have felt | |
523 | * before rdev->recovery_offset, but for simplicity we don't | |
524 | * check this here. | |
525 | */ | |
526 | if (test_bit(In_sync, &rdev->flags) && | |
527 | !test_bit(Faulty, &rdev->flags)) | |
528 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 529 | |
749c55e9 | 530 | /* Maybe we can clear some bad blocks. */ |
3a0f007b YK |
531 | if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr, |
532 | r10_bio->sectors) && | |
533 | !discard_error) { | |
749c55e9 | 534 | bio_put(bio); |
475b0321 N |
535 | if (repl) |
536 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
537 | else | |
538 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
539 | dec_rdev = 0; |
540 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
541 | } | |
542 | } | |
543 | ||
1da177e4 LT |
544 | /* |
545 | * | |
546 | * Let's see if all mirrored write operations have finished | |
547 | * already. | |
548 | */ | |
19d5f834 | 549 | one_write_done(r10_bio); |
749c55e9 | 550 | if (dec_rdev) |
884162df | 551 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
552 | if (to_put) |
553 | bio_put(to_put); | |
1da177e4 LT |
554 | } |
555 | ||
1da177e4 LT |
556 | /* |
557 | * RAID10 layout manager | |
25985edc | 558 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
559 | * parameters: near_copies and far_copies. |
560 | * near_copies * far_copies must be <= raid_disks. | |
561 | * Normally one of these will be 1. | |
562 | * If both are 1, we get raid0. | |
563 | * If near_copies == raid_disks, we get raid1. | |
564 | * | |
25985edc | 565 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
566 | * first chunk, followed by near_copies copies of the next chunk and |
567 | * so on. | |
568 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
569 | * as described above, we start again with a device offset of near_copies. | |
570 | * So we effectively have another copy of the whole array further down all | |
571 | * the drives, but with blocks on different drives. | |
572 | * With this layout, and block is never stored twice on the one device. | |
573 | * | |
574 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 575 | * on each device that it is on. |
1da177e4 LT |
576 | * |
577 | * raid10_find_virt does the reverse mapping, from a device and a | |
578 | * sector offset to a virtual address | |
579 | */ | |
580 | ||
f8c9e74f | 581 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
582 | { |
583 | int n,f; | |
584 | sector_t sector; | |
585 | sector_t chunk; | |
586 | sector_t stripe; | |
587 | int dev; | |
1da177e4 | 588 | int slot = 0; |
9a3152ab JB |
589 | int last_far_set_start, last_far_set_size; |
590 | ||
591 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
592 | last_far_set_start *= geo->far_set_size; | |
593 | ||
594 | last_far_set_size = geo->far_set_size; | |
595 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
596 | |
597 | /* now calculate first sector/dev */ | |
5cf00fcd N |
598 | chunk = r10bio->sector >> geo->chunk_shift; |
599 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 600 | |
5cf00fcd | 601 | chunk *= geo->near_copies; |
1da177e4 | 602 | stripe = chunk; |
5cf00fcd N |
603 | dev = sector_div(stripe, geo->raid_disks); |
604 | if (geo->far_offset) | |
605 | stripe *= geo->far_copies; | |
1da177e4 | 606 | |
5cf00fcd | 607 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
608 | |
609 | /* and calculate all the others */ | |
5cf00fcd | 610 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 611 | int d = dev; |
475901af | 612 | int set; |
1da177e4 | 613 | sector_t s = sector; |
1da177e4 | 614 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 615 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
616 | slot++; |
617 | ||
5cf00fcd | 618 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 619 | set = d / geo->far_set_size; |
5cf00fcd | 620 | d += geo->near_copies; |
475901af | 621 | |
9a3152ab JB |
622 | if ((geo->raid_disks % geo->far_set_size) && |
623 | (d > last_far_set_start)) { | |
624 | d -= last_far_set_start; | |
625 | d %= last_far_set_size; | |
626 | d += last_far_set_start; | |
627 | } else { | |
628 | d %= geo->far_set_size; | |
629 | d += geo->far_set_size * set; | |
630 | } | |
5cf00fcd | 631 | s += geo->stride; |
1da177e4 LT |
632 | r10bio->devs[slot].devnum = d; |
633 | r10bio->devs[slot].addr = s; | |
634 | slot++; | |
635 | } | |
636 | dev++; | |
5cf00fcd | 637 | if (dev >= geo->raid_disks) { |
1da177e4 | 638 | dev = 0; |
5cf00fcd | 639 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
640 | } |
641 | } | |
f8c9e74f N |
642 | } |
643 | ||
644 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
645 | { | |
646 | struct geom *geo = &conf->geo; | |
647 | ||
648 | if (conf->reshape_progress != MaxSector && | |
649 | ((r10bio->sector >= conf->reshape_progress) != | |
650 | conf->mddev->reshape_backwards)) { | |
651 | set_bit(R10BIO_Previous, &r10bio->state); | |
652 | geo = &conf->prev; | |
653 | } else | |
654 | clear_bit(R10BIO_Previous, &r10bio->state); | |
655 | ||
656 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
657 | } |
658 | ||
e879a879 | 659 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
660 | { |
661 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
662 | /* Never use conf->prev as this is only called during resync |
663 | * or recovery, so reshape isn't happening | |
664 | */ | |
5cf00fcd | 665 | struct geom *geo = &conf->geo; |
475901af JB |
666 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
667 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
668 | int last_far_set_start; |
669 | ||
670 | if (geo->raid_disks % geo->far_set_size) { | |
671 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
672 | last_far_set_start *= geo->far_set_size; | |
673 | ||
674 | if (dev >= last_far_set_start) { | |
675 | far_set_size = geo->far_set_size; | |
676 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
677 | far_set_start = last_far_set_start; | |
678 | } | |
679 | } | |
1da177e4 | 680 | |
5cf00fcd N |
681 | offset = sector & geo->chunk_mask; |
682 | if (geo->far_offset) { | |
c93983bf | 683 | int fc; |
5cf00fcd N |
684 | chunk = sector >> geo->chunk_shift; |
685 | fc = sector_div(chunk, geo->far_copies); | |
686 | dev -= fc * geo->near_copies; | |
475901af JB |
687 | if (dev < far_set_start) |
688 | dev += far_set_size; | |
c93983bf | 689 | } else { |
5cf00fcd N |
690 | while (sector >= geo->stride) { |
691 | sector -= geo->stride; | |
475901af JB |
692 | if (dev < (geo->near_copies + far_set_start)) |
693 | dev += far_set_size - geo->near_copies; | |
c93983bf | 694 | else |
5cf00fcd | 695 | dev -= geo->near_copies; |
c93983bf | 696 | } |
5cf00fcd | 697 | chunk = sector >> geo->chunk_shift; |
c93983bf | 698 | } |
5cf00fcd N |
699 | vchunk = chunk * geo->raid_disks + dev; |
700 | sector_div(vchunk, geo->near_copies); | |
701 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
702 | } |
703 | ||
1da177e4 LT |
704 | /* |
705 | * This routine returns the disk from which the requested read should | |
706 | * be done. There is a per-array 'next expected sequential IO' sector | |
707 | * number - if this matches on the next IO then we use the last disk. | |
708 | * There is also a per-disk 'last know head position' sector that is | |
709 | * maintained from IRQ contexts, both the normal and the resync IO | |
710 | * completion handlers update this position correctly. If there is no | |
711 | * perfect sequential match then we pick the disk whose head is closest. | |
712 | * | |
713 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
714 | * because position is mirror, not device based. | |
715 | * | |
716 | * The rdev for the device selected will have nr_pending incremented. | |
717 | */ | |
718 | ||
719 | /* | |
720 | * FIXME: possibly should rethink readbalancing and do it differently | |
721 | * depending on near_copies / far_copies geometry. | |
722 | */ | |
96c3fd1f N |
723 | static struct md_rdev *read_balance(struct r10conf *conf, |
724 | struct r10bio *r10_bio, | |
725 | int *max_sectors) | |
1da177e4 | 726 | { |
af3a2cd6 | 727 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 728 | int disk, slot; |
856e08e2 N |
729 | int sectors = r10_bio->sectors; |
730 | int best_good_sectors; | |
56d99121 | 731 | sector_t new_distance, best_dist; |
e9eeba28 | 732 | struct md_rdev *best_dist_rdev, *best_pending_rdev, *rdev = NULL; |
56d99121 | 733 | int do_balance; |
e9eeba28 GJ |
734 | int best_dist_slot, best_pending_slot; |
735 | bool has_nonrot_disk = false; | |
736 | unsigned int min_pending; | |
5cf00fcd | 737 | struct geom *geo = &conf->geo; |
1da177e4 LT |
738 | |
739 | raid10_find_phys(conf, r10_bio); | |
e9eeba28 GJ |
740 | best_dist_slot = -1; |
741 | min_pending = UINT_MAX; | |
742 | best_dist_rdev = NULL; | |
743 | best_pending_rdev = NULL; | |
56d99121 | 744 | best_dist = MaxSector; |
856e08e2 | 745 | best_good_sectors = 0; |
56d99121 | 746 | do_balance = 1; |
8d3ca83d | 747 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
f1092076 YK |
748 | |
749 | if (raid1_should_read_first(conf->mddev, this_sector, sectors)) | |
56d99121 | 750 | do_balance = 0; |
1da177e4 | 751 | |
56d99121 | 752 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
753 | sector_t first_bad; |
754 | int bad_sectors; | |
755 | sector_t dev_sector; | |
e9eeba28 GJ |
756 | unsigned int pending; |
757 | bool nonrot; | |
856e08e2 | 758 | |
56d99121 N |
759 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
760 | continue; | |
1da177e4 | 761 | disk = r10_bio->devs[slot].devnum; |
a448af25 | 762 | rdev = conf->mirrors[disk].replacement; |
abbf098e | 763 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || |
2ae6aaf7 | 764 | r10_bio->devs[slot].addr + sectors > |
a448af25 YK |
765 | rdev->recovery_offset) |
766 | rdev = conf->mirrors[disk].rdev; | |
050b6615 | 767 | if (rdev == NULL || |
8ae12666 | 768 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
769 | continue; |
770 | if (!test_bit(In_sync, &rdev->flags) && | |
771 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
772 | continue; |
773 | ||
856e08e2 N |
774 | dev_sector = r10_bio->devs[slot].addr; |
775 | if (is_badblock(rdev, dev_sector, sectors, | |
776 | &first_bad, &bad_sectors)) { | |
777 | if (best_dist < MaxSector) | |
778 | /* Already have a better slot */ | |
779 | continue; | |
780 | if (first_bad <= dev_sector) { | |
781 | /* Cannot read here. If this is the | |
782 | * 'primary' device, then we must not read | |
783 | * beyond 'bad_sectors' from another device. | |
784 | */ | |
785 | bad_sectors -= (dev_sector - first_bad); | |
786 | if (!do_balance && sectors > bad_sectors) | |
787 | sectors = bad_sectors; | |
788 | if (best_good_sectors > sectors) | |
789 | best_good_sectors = sectors; | |
790 | } else { | |
791 | sector_t good_sectors = | |
792 | first_bad - dev_sector; | |
793 | if (good_sectors > best_good_sectors) { | |
794 | best_good_sectors = good_sectors; | |
e9eeba28 GJ |
795 | best_dist_slot = slot; |
796 | best_dist_rdev = rdev; | |
856e08e2 N |
797 | } |
798 | if (!do_balance) | |
799 | /* Must read from here */ | |
800 | break; | |
801 | } | |
802 | continue; | |
803 | } else | |
804 | best_good_sectors = sectors; | |
805 | ||
56d99121 N |
806 | if (!do_balance) |
807 | break; | |
1da177e4 | 808 | |
10f0d2a5 | 809 | nonrot = bdev_nonrot(rdev->bdev); |
e9eeba28 GJ |
810 | has_nonrot_disk |= nonrot; |
811 | pending = atomic_read(&rdev->nr_pending); | |
812 | if (min_pending > pending && nonrot) { | |
813 | min_pending = pending; | |
814 | best_pending_slot = slot; | |
815 | best_pending_rdev = rdev; | |
816 | } | |
817 | ||
818 | if (best_dist_slot >= 0) | |
8d3ca83d N |
819 | /* At least 2 disks to choose from so failfast is OK */ |
820 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
821 | /* This optimisation is debatable, and completely destroys |
822 | * sequential read speed for 'far copies' arrays. So only | |
823 | * keep it for 'near' arrays, and review those later. | |
824 | */ | |
e9eeba28 | 825 | if (geo->near_copies > 1 && !pending) |
8d3ca83d | 826 | new_distance = 0; |
8ed3a195 KS |
827 | |
828 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 829 | else if (geo->far_copies > 1) |
56d99121 | 830 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 831 | else |
56d99121 N |
832 | new_distance = abs(r10_bio->devs[slot].addr - |
833 | conf->mirrors[disk].head_position); | |
e9eeba28 | 834 | |
56d99121 N |
835 | if (new_distance < best_dist) { |
836 | best_dist = new_distance; | |
e9eeba28 GJ |
837 | best_dist_slot = slot; |
838 | best_dist_rdev = rdev; | |
1da177e4 LT |
839 | } |
840 | } | |
abbf098e | 841 | if (slot >= conf->copies) { |
e9eeba28 GJ |
842 | if (has_nonrot_disk) { |
843 | slot = best_pending_slot; | |
844 | rdev = best_pending_rdev; | |
845 | } else { | |
846 | slot = best_dist_slot; | |
847 | rdev = best_dist_rdev; | |
848 | } | |
abbf098e | 849 | } |
1da177e4 | 850 | |
56d99121 | 851 | if (slot >= 0) { |
56d99121 | 852 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
853 | r10_bio->read_slot = slot; |
854 | } else | |
96c3fd1f | 855 | rdev = NULL; |
856e08e2 | 856 | *max_sectors = best_good_sectors; |
1da177e4 | 857 | |
96c3fd1f | 858 | return rdev; |
1da177e4 LT |
859 | } |
860 | ||
e879a879 | 861 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
862 | { |
863 | /* Any writes that have been queued but are awaiting | |
864 | * bitmap updates get flushed here. | |
a35e63ef | 865 | */ |
a35e63ef N |
866 | spin_lock_irq(&conf->device_lock); |
867 | ||
868 | if (conf->pending_bio_list.head) { | |
18022a1b | 869 | struct blk_plug plug; |
a35e63ef | 870 | struct bio *bio; |
18022a1b | 871 | |
a35e63ef | 872 | bio = bio_list_get(&conf->pending_bio_list); |
a35e63ef | 873 | spin_unlock_irq(&conf->device_lock); |
474beb57 N |
874 | |
875 | /* | |
876 | * As this is called in a wait_event() loop (see freeze_array), | |
877 | * current->state might be TASK_UNINTERRUPTIBLE which will | |
878 | * cause a warning when we prepare to wait again. As it is | |
879 | * rare that this path is taken, it is perfectly safe to force | |
880 | * us to go around the wait_event() loop again, so the warning | |
881 | * is a false-positive. Silence the warning by resetting | |
882 | * thread state | |
883 | */ | |
884 | __set_current_state(TASK_RUNNING); | |
885 | ||
18022a1b | 886 | blk_start_plug(&plug); |
9efcc2c3 | 887 | raid1_prepare_flush_writes(conf->mddev->bitmap); |
34db0cd6 | 888 | wake_up(&conf->wait_barrier); |
a35e63ef N |
889 | |
890 | while (bio) { /* submit pending writes */ | |
891 | struct bio *next = bio->bi_next; | |
8295efbe YK |
892 | |
893 | raid1_submit_write(bio); | |
a35e63ef | 894 | bio = next; |
01044462 | 895 | cond_resched(); |
a35e63ef | 896 | } |
18022a1b | 897 | blk_finish_plug(&plug); |
a35e63ef N |
898 | } else |
899 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 900 | } |
7eaceacc | 901 | |
0a27ec96 N |
902 | /* Barriers.... |
903 | * Sometimes we need to suspend IO while we do something else, | |
904 | * either some resync/recovery, or reconfigure the array. | |
905 | * To do this we raise a 'barrier'. | |
906 | * The 'barrier' is a counter that can be raised multiple times | |
907 | * to count how many activities are happening which preclude | |
908 | * normal IO. | |
909 | * We can only raise the barrier if there is no pending IO. | |
910 | * i.e. if nr_pending == 0. | |
911 | * We choose only to raise the barrier if no-one is waiting for the | |
912 | * barrier to go down. This means that as soon as an IO request | |
913 | * is ready, no other operations which require a barrier will start | |
914 | * until the IO request has had a chance. | |
915 | * | |
916 | * So: regular IO calls 'wait_barrier'. When that returns there | |
917 | * is no backgroup IO happening, It must arrange to call | |
918 | * allow_barrier when it has finished its IO. | |
919 | * backgroup IO calls must call raise_barrier. Once that returns | |
920 | * there is no normal IO happeing. It must arrange to call | |
921 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 922 | */ |
1da177e4 | 923 | |
e879a879 | 924 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 925 | { |
b9b083f9 | 926 | write_seqlock_irq(&conf->resync_lock); |
9fdfe6d4 YK |
927 | |
928 | if (WARN_ON_ONCE(force && !conf->barrier)) | |
929 | force = false; | |
0a27ec96 | 930 | |
6cce3b23 | 931 | /* Wait until no block IO is waiting (unless 'force') */ |
b9b083f9 | 932 | wait_event_barrier(conf, force || !conf->nr_waiting); |
0a27ec96 N |
933 | |
934 | /* block any new IO from starting */ | |
b9b083f9 | 935 | WRITE_ONCE(conf->barrier, conf->barrier + 1); |
0a27ec96 | 936 | |
c3b328ac | 937 | /* Now wait for all pending IO to complete */ |
b9b083f9 YK |
938 | wait_event_barrier(conf, !atomic_read(&conf->nr_pending) && |
939 | conf->barrier < RESYNC_DEPTH); | |
0a27ec96 | 940 | |
b9b083f9 | 941 | write_sequnlock_irq(&conf->resync_lock); |
0a27ec96 N |
942 | } |
943 | ||
e879a879 | 944 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
945 | { |
946 | unsigned long flags; | |
b9b083f9 YK |
947 | |
948 | write_seqlock_irqsave(&conf->resync_lock, flags); | |
949 | WRITE_ONCE(conf->barrier, conf->barrier - 1); | |
950 | write_sequnlock_irqrestore(&conf->resync_lock, flags); | |
0a27ec96 N |
951 | wake_up(&conf->wait_barrier); |
952 | } | |
953 | ||
ed2e063f YK |
954 | static bool stop_waiting_barrier(struct r10conf *conf) |
955 | { | |
956 | struct bio_list *bio_list = current->bio_list; | |
44693154 | 957 | struct md_thread *thread; |
ed2e063f YK |
958 | |
959 | /* barrier is dropped */ | |
960 | if (!conf->barrier) | |
961 | return true; | |
962 | ||
963 | /* | |
964 | * If there are already pending requests (preventing the barrier from | |
965 | * rising completely), and the pre-process bio queue isn't empty, then | |
966 | * don't wait, as we need to empty that queue to get the nr_pending | |
967 | * count down. | |
968 | */ | |
969 | if (atomic_read(&conf->nr_pending) && bio_list && | |
970 | (!bio_list_empty(&bio_list[0]) || !bio_list_empty(&bio_list[1]))) | |
971 | return true; | |
972 | ||
44693154 YK |
973 | /* daemon thread must exist while handling io */ |
974 | thread = rcu_dereference_protected(conf->mddev->thread, true); | |
72c215ed LN |
975 | /* |
976 | * move on if io is issued from raid10d(), nr_pending is not released | |
977 | * from original io(see handle_read_error()). All raise barrier is | |
978 | * blocked until this io is done. | |
979 | */ | |
44693154 | 980 | if (thread->tsk == current) { |
72c215ed | 981 | WARN_ON_ONCE(atomic_read(&conf->nr_pending) == 0); |
ed2e063f | 982 | return true; |
72c215ed | 983 | } |
ed2e063f YK |
984 | |
985 | return false; | |
986 | } | |
987 | ||
b9b083f9 YK |
988 | static bool wait_barrier_nolock(struct r10conf *conf) |
989 | { | |
990 | unsigned int seq = read_seqbegin(&conf->resync_lock); | |
991 | ||
992 | if (READ_ONCE(conf->barrier)) | |
993 | return false; | |
994 | ||
995 | atomic_inc(&conf->nr_pending); | |
996 | if (!read_seqretry(&conf->resync_lock, seq)) | |
997 | return true; | |
998 | ||
999 | if (atomic_dec_and_test(&conf->nr_pending)) | |
1000 | wake_up_barrier(conf); | |
1001 | ||
1002 | return false; | |
1003 | } | |
1004 | ||
c9aa889b | 1005 | static bool wait_barrier(struct r10conf *conf, bool nowait) |
0a27ec96 | 1006 | { |
c9aa889b VV |
1007 | bool ret = true; |
1008 | ||
b9b083f9 YK |
1009 | if (wait_barrier_nolock(conf)) |
1010 | return true; | |
1011 | ||
1012 | write_seqlock_irq(&conf->resync_lock); | |
0a27ec96 | 1013 | if (conf->barrier) { |
c9aa889b VV |
1014 | /* Return false when nowait flag is set */ |
1015 | if (nowait) { | |
1016 | ret = false; | |
1017 | } else { | |
0de57e54 | 1018 | conf->nr_waiting++; |
28be4fd3 | 1019 | mddev_add_trace_msg(conf->mddev, "raid10 wait barrier"); |
b9b083f9 | 1020 | wait_event_barrier(conf, stop_waiting_barrier(conf)); |
0de57e54 | 1021 | conf->nr_waiting--; |
c9aa889b | 1022 | } |
0e5313e2 TM |
1023 | if (!conf->nr_waiting) |
1024 | wake_up(&conf->wait_barrier); | |
1da177e4 | 1025 | } |
c9aa889b VV |
1026 | /* Only increment nr_pending when we wait */ |
1027 | if (ret) | |
1028 | atomic_inc(&conf->nr_pending); | |
b9b083f9 | 1029 | write_sequnlock_irq(&conf->resync_lock); |
c9aa889b | 1030 | return ret; |
1da177e4 LT |
1031 | } |
1032 | ||
e879a879 | 1033 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 1034 | { |
0e5313e2 TM |
1035 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
1036 | (conf->array_freeze_pending)) | |
0c0be98b | 1037 | wake_up_barrier(conf); |
0a27ec96 N |
1038 | } |
1039 | ||
e2d59925 | 1040 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
1041 | { |
1042 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 1043 | * go quiet. |
4443ae10 | 1044 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1045 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1046 | * This is called in the context of one normal IO request |
1047 | * that has failed. Thus any sync request that might be pending | |
1048 | * will be blocked by nr_pending, and we need to wait for | |
1049 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1050 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1051 | * must match the number of pending IOs (nr_pending) before |
1052 | * we continue. | |
4443ae10 | 1053 | */ |
b9b083f9 | 1054 | write_seqlock_irq(&conf->resync_lock); |
0e5313e2 | 1055 | conf->array_freeze_pending++; |
b9b083f9 | 1056 | WRITE_ONCE(conf->barrier, conf->barrier + 1); |
4443ae10 | 1057 | conf->nr_waiting++; |
b9b083f9 YK |
1058 | wait_event_barrier_cmd(conf, atomic_read(&conf->nr_pending) == |
1059 | conf->nr_queued + extra, flush_pending_writes(conf)); | |
0e5313e2 | 1060 | conf->array_freeze_pending--; |
b9b083f9 | 1061 | write_sequnlock_irq(&conf->resync_lock); |
4443ae10 N |
1062 | } |
1063 | ||
e879a879 | 1064 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1065 | { |
1066 | /* reverse the effect of the freeze */ | |
b9b083f9 YK |
1067 | write_seqlock_irq(&conf->resync_lock); |
1068 | WRITE_ONCE(conf->barrier, conf->barrier - 1); | |
4443ae10 N |
1069 | conf->nr_waiting--; |
1070 | wake_up(&conf->wait_barrier); | |
b9b083f9 | 1071 | write_sequnlock_irq(&conf->resync_lock); |
4443ae10 N |
1072 | } |
1073 | ||
f8c9e74f N |
1074 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1075 | struct md_rdev *rdev) | |
1076 | { | |
1077 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1078 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1079 | return rdev->data_offset; | |
1080 | else | |
1081 | return rdev->new_data_offset; | |
1082 | } | |
1083 | ||
57c67df4 N |
1084 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) |
1085 | { | |
daae161f | 1086 | struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb, cb); |
57c67df4 N |
1087 | struct mddev *mddev = plug->cb.data; |
1088 | struct r10conf *conf = mddev->private; | |
1089 | struct bio *bio; | |
1090 | ||
9efcc2c3 | 1091 | if (from_schedule) { |
57c67df4 N |
1092 | spin_lock_irq(&conf->device_lock); |
1093 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
57c67df4 | 1094 | spin_unlock_irq(&conf->device_lock); |
a8d5fdd4 | 1095 | wake_up_barrier(conf); |
57c67df4 N |
1096 | md_wakeup_thread(mddev->thread); |
1097 | kfree(plug); | |
1098 | return; | |
1099 | } | |
1100 | ||
1101 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1102 | bio = bio_list_get(&plug->pending); | |
9efcc2c3 | 1103 | raid1_prepare_flush_writes(mddev->bitmap); |
a8d5fdd4 | 1104 | wake_up_barrier(conf); |
57c67df4 N |
1105 | |
1106 | while (bio) { /* submit pending writes */ | |
1107 | struct bio *next = bio->bi_next; | |
8295efbe YK |
1108 | |
1109 | raid1_submit_write(bio); | |
57c67df4 | 1110 | bio = next; |
01044462 | 1111 | cond_resched(); |
57c67df4 N |
1112 | } |
1113 | kfree(plug); | |
1114 | } | |
1115 | ||
caea3c47 GJ |
1116 | /* |
1117 | * 1. Register the new request and wait if the reconstruction thread has put | |
1118 | * up a bar for new requests. Continue immediately if no resync is active | |
1119 | * currently. | |
1120 | * 2. If IO spans the reshape position. Need to wait for reshape to pass. | |
1121 | */ | |
c9aa889b | 1122 | static bool regular_request_wait(struct mddev *mddev, struct r10conf *conf, |
caea3c47 GJ |
1123 | struct bio *bio, sector_t sectors) |
1124 | { | |
c9aa889b VV |
1125 | /* Bail out if REQ_NOWAIT is set for the bio */ |
1126 | if (!wait_barrier(conf, bio->bi_opf & REQ_NOWAIT)) { | |
1127 | bio_wouldblock_error(bio); | |
1128 | return false; | |
1129 | } | |
caea3c47 GJ |
1130 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
1131 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1132 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
caea3c47 | 1133 | allow_barrier(conf); |
c9aa889b VV |
1134 | if (bio->bi_opf & REQ_NOWAIT) { |
1135 | bio_wouldblock_error(bio); | |
1136 | return false; | |
1137 | } | |
28be4fd3 | 1138 | mddev_add_trace_msg(conf->mddev, "raid10 wait reshape"); |
caea3c47 GJ |
1139 | wait_event(conf->wait_barrier, |
1140 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1141 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1142 | sectors); | |
c9aa889b | 1143 | wait_barrier(conf, false); |
caea3c47 | 1144 | } |
c9aa889b | 1145 | return true; |
caea3c47 GJ |
1146 | } |
1147 | ||
bb5f1ed7 | 1148 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
82045523 | 1149 | struct r10bio *r10_bio, bool io_accounting) |
1da177e4 | 1150 | { |
e879a879 | 1151 | struct r10conf *conf = mddev->private; |
1da177e4 | 1152 | struct bio *read_bio; |
cb1802ff BVA |
1153 | const enum req_op op = bio_op(bio); |
1154 | const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; | |
bb5f1ed7 | 1155 | int max_sectors; |
bb5f1ed7 | 1156 | struct md_rdev *rdev; |
545250f2 N |
1157 | char b[BDEVNAME_SIZE]; |
1158 | int slot = r10_bio->read_slot; | |
1159 | struct md_rdev *err_rdev = NULL; | |
1160 | gfp_t gfp = GFP_NOIO; | |
bb5f1ed7 | 1161 | |
93decc56 | 1162 | if (slot >= 0 && r10_bio->devs[slot].rdev) { |
545250f2 N |
1163 | /* |
1164 | * This is an error retry, but we cannot | |
1165 | * safely dereference the rdev in the r10_bio, | |
1166 | * we must use the one in conf. | |
1167 | * If it has already been disconnected (unlikely) | |
1168 | * we lose the device name in error messages. | |
1169 | */ | |
1170 | int disk; | |
1171 | /* | |
1172 | * As we are blocking raid10, it is a little safer to | |
1173 | * use __GFP_HIGH. | |
1174 | */ | |
1175 | gfp = GFP_NOIO | __GFP_HIGH; | |
1176 | ||
545250f2 | 1177 | disk = r10_bio->devs[slot].devnum; |
a448af25 | 1178 | err_rdev = conf->mirrors[disk].rdev; |
545250f2 | 1179 | if (err_rdev) |
900d156b | 1180 | snprintf(b, sizeof(b), "%pg", err_rdev->bdev); |
545250f2 N |
1181 | else { |
1182 | strcpy(b, "???"); | |
1183 | /* This never gets dereferenced */ | |
1184 | err_rdev = r10_bio->devs[slot].rdev; | |
1185 | } | |
545250f2 | 1186 | } |
bb5f1ed7 | 1187 | |
c9aa889b VV |
1188 | if (!regular_request_wait(mddev, conf, bio, r10_bio->sectors)) |
1189 | return; | |
bb5f1ed7 RL |
1190 | rdev = read_balance(conf, r10_bio, &max_sectors); |
1191 | if (!rdev) { | |
545250f2 N |
1192 | if (err_rdev) { |
1193 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", | |
1194 | mdname(mddev), b, | |
1195 | (unsigned long long)r10_bio->sector); | |
1196 | } | |
bb5f1ed7 RL |
1197 | raid_end_bio_io(r10_bio); |
1198 | return; | |
1199 | } | |
545250f2 | 1200 | if (err_rdev) |
913cce5a | 1201 | pr_err_ratelimited("md/raid10:%s: %pg: redirecting sector %llu to another mirror\n", |
545250f2 | 1202 | mdname(mddev), |
913cce5a | 1203 | rdev->bdev, |
545250f2 | 1204 | (unsigned long long)r10_bio->sector); |
fc9977dd N |
1205 | if (max_sectors < bio_sectors(bio)) { |
1206 | struct bio *split = bio_split(bio, max_sectors, | |
afeee514 | 1207 | gfp, &conf->bio_split); |
fc9977dd | 1208 | bio_chain(split, bio); |
e820d55c | 1209 | allow_barrier(conf); |
ed00aabd | 1210 | submit_bio_noacct(bio); |
c9aa889b | 1211 | wait_barrier(conf, false); |
fc9977dd N |
1212 | bio = split; |
1213 | r10_bio->master_bio = bio; | |
1214 | r10_bio->sectors = max_sectors; | |
1215 | } | |
bb5f1ed7 RL |
1216 | slot = r10_bio->read_slot; |
1217 | ||
82045523 YK |
1218 | if (io_accounting) { |
1219 | md_account_bio(mddev, &bio); | |
1220 | r10_bio->master_bio = bio; | |
1221 | } | |
abfc426d | 1222 | read_bio = bio_alloc_clone(rdev->bdev, bio, gfp, &mddev->bio_set); |
bb5f1ed7 RL |
1223 | |
1224 | r10_bio->devs[slot].bio = read_bio; | |
1225 | r10_bio->devs[slot].rdev = rdev; | |
1226 | ||
1227 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1228 | choose_data_offset(r10_bio, rdev); | |
bb5f1ed7 | 1229 | read_bio->bi_end_io = raid10_end_read_request; |
c34b7ac6 | 1230 | read_bio->bi_opf = op | do_sync; |
bb5f1ed7 RL |
1231 | if (test_bit(FailFast, &rdev->flags) && |
1232 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1233 | read_bio->bi_opf |= MD_FAILFAST; | |
1234 | read_bio->bi_private = r10_bio; | |
c396b90e | 1235 | mddev_trace_remap(mddev, read_bio, r10_bio->sector); |
ed00aabd | 1236 | submit_bio_noacct(read_bio); |
bb5f1ed7 RL |
1237 | return; |
1238 | } | |
1239 | ||
27f26a0f GJ |
1240 | static void raid10_write_one_disk(struct mddev *mddev, struct r10bio *r10_bio, |
1241 | struct bio *bio, bool replacement, | |
fc9977dd | 1242 | int n_copy) |
bb5f1ed7 | 1243 | { |
cb1802ff BVA |
1244 | const enum req_op op = bio_op(bio); |
1245 | const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; | |
1246 | const blk_opf_t do_fua = bio->bi_opf & REQ_FUA; | |
6cce3b23 | 1247 | unsigned long flags; |
27f26a0f GJ |
1248 | struct r10conf *conf = mddev->private; |
1249 | struct md_rdev *rdev; | |
1250 | int devnum = r10_bio->devs[n_copy].devnum; | |
1251 | struct bio *mbio; | |
1252 | ||
a448af25 YK |
1253 | rdev = replacement ? conf->mirrors[devnum].replacement : |
1254 | conf->mirrors[devnum].rdev; | |
27f26a0f | 1255 | |
abfc426d | 1256 | mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, &mddev->bio_set); |
27f26a0f GJ |
1257 | if (replacement) |
1258 | r10_bio->devs[n_copy].repl_bio = mbio; | |
1259 | else | |
1260 | r10_bio->devs[n_copy].bio = mbio; | |
1261 | ||
1262 | mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + | |
1263 | choose_data_offset(r10_bio, rdev)); | |
27f26a0f | 1264 | mbio->bi_end_io = raid10_end_write_request; |
c34b7ac6 | 1265 | mbio->bi_opf = op | do_sync | do_fua; |
27f26a0f GJ |
1266 | if (!replacement && test_bit(FailFast, |
1267 | &conf->mirrors[devnum].rdev->flags) | |
1268 | && enough(conf, devnum)) | |
1269 | mbio->bi_opf |= MD_FAILFAST; | |
1270 | mbio->bi_private = r10_bio; | |
c396b90e | 1271 | mddev_trace_remap(mddev, mbio, r10_bio->sector); |
27f26a0f | 1272 | /* flush_pending_writes() needs access to the rdev so...*/ |
309dca30 | 1273 | mbio->bi_bdev = (void *)rdev; |
27f26a0f GJ |
1274 | |
1275 | atomic_inc(&r10_bio->remaining); | |
1276 | ||
460af1f9 | 1277 | if (!raid1_add_bio_to_plug(mddev, mbio, raid10_unplug, conf->copies)) { |
23b245c0 | 1278 | spin_lock_irqsave(&conf->device_lock, flags); |
27f26a0f | 1279 | bio_list_add(&conf->pending_bio_list, mbio); |
23b245c0 | 1280 | spin_unlock_irqrestore(&conf->device_lock, flags); |
27f26a0f | 1281 | md_wakeup_thread(mddev->thread); |
23b245c0 | 1282 | } |
27f26a0f GJ |
1283 | } |
1284 | ||
f2e7e269 XN |
1285 | static void wait_blocked_dev(struct mddev *mddev, struct r10bio *r10_bio) |
1286 | { | |
1287 | int i; | |
1288 | struct r10conf *conf = mddev->private; | |
1289 | struct md_rdev *blocked_rdev; | |
1290 | ||
1291 | retry_wait: | |
1292 | blocked_rdev = NULL; | |
f2e7e269 | 1293 | for (i = 0; i < conf->copies; i++) { |
67364349 LN |
1294 | struct md_rdev *rdev, *rrdev; |
1295 | ||
a448af25 YK |
1296 | rdev = conf->mirrors[i].rdev; |
1297 | rrdev = conf->mirrors[i].replacement; | |
f2e7e269 XN |
1298 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1299 | atomic_inc(&rdev->nr_pending); | |
1300 | blocked_rdev = rdev; | |
1301 | break; | |
1302 | } | |
1303 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { | |
1304 | atomic_inc(&rrdev->nr_pending); | |
1305 | blocked_rdev = rrdev; | |
1306 | break; | |
1307 | } | |
1308 | ||
1309 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { | |
f2e7e269 | 1310 | sector_t dev_sector = r10_bio->devs[i].addr; |
f2e7e269 XN |
1311 | |
1312 | /* | |
1313 | * Discard request doesn't care the write result | |
1314 | * so it doesn't need to wait blocked disk here. | |
1315 | */ | |
1316 | if (!r10_bio->sectors) | |
1317 | continue; | |
1318 | ||
3a0f007b YK |
1319 | if (rdev_has_badblock(rdev, dev_sector, |
1320 | r10_bio->sectors) < 0) { | |
f2e7e269 XN |
1321 | /* |
1322 | * Mustn't write here until the bad block | |
1323 | * is acknowledged | |
1324 | */ | |
1325 | atomic_inc(&rdev->nr_pending); | |
1326 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1327 | blocked_rdev = rdev; | |
1328 | break; | |
1329 | } | |
1330 | } | |
1331 | } | |
f2e7e269 XN |
1332 | |
1333 | if (unlikely(blocked_rdev)) { | |
1334 | /* Have to wait for this device to get unblocked, then retry */ | |
1335 | allow_barrier(conf); | |
28be4fd3 CH |
1336 | mddev_add_trace_msg(conf->mddev, |
1337 | "raid10 %s wait rdev %d blocked", | |
1338 | __func__, blocked_rdev->raid_disk); | |
f2e7e269 | 1339 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
c9aa889b | 1340 | wait_barrier(conf, false); |
f2e7e269 XN |
1341 | goto retry_wait; |
1342 | } | |
1343 | } | |
1344 | ||
27f26a0f GJ |
1345 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, |
1346 | struct r10bio *r10_bio) | |
1347 | { | |
1348 | struct r10conf *conf = mddev->private; | |
1349 | int i; | |
bb5f1ed7 | 1350 | sector_t sectors; |
d4432c23 | 1351 | int max_sectors; |
1da177e4 | 1352 | |
cb8a7a7e GJ |
1353 | if ((mddev_is_clustered(mddev) && |
1354 | md_cluster_ops->area_resyncing(mddev, WRITE, | |
1355 | bio->bi_iter.bi_sector, | |
1356 | bio_end_sector(bio)))) { | |
1357 | DEFINE_WAIT(w); | |
c9aa889b VV |
1358 | /* Bail out if REQ_NOWAIT is set for the bio */ |
1359 | if (bio->bi_opf & REQ_NOWAIT) { | |
1360 | bio_wouldblock_error(bio); | |
1361 | return; | |
1362 | } | |
cb8a7a7e GJ |
1363 | for (;;) { |
1364 | prepare_to_wait(&conf->wait_barrier, | |
1365 | &w, TASK_IDLE); | |
1366 | if (!md_cluster_ops->area_resyncing(mddev, WRITE, | |
1367 | bio->bi_iter.bi_sector, bio_end_sector(bio))) | |
1368 | break; | |
1369 | schedule(); | |
1370 | } | |
1371 | finish_wait(&conf->wait_barrier, &w); | |
1372 | } | |
1373 | ||
fc9977dd | 1374 | sectors = r10_bio->sectors; |
c9aa889b VV |
1375 | if (!regular_request_wait(mddev, conf, bio, sectors)) |
1376 | return; | |
3ea7daa5 | 1377 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1378 | (mddev->reshape_backwards |
4f024f37 KO |
1379 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1380 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1381 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1382 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1383 | /* Need to update reshape_position in metadata */ |
1384 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1385 | set_mask_bits(&mddev->sb_flags, 0, |
1386 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1387 | md_wakeup_thread(mddev->thread); |
c9aa889b VV |
1388 | if (bio->bi_opf & REQ_NOWAIT) { |
1389 | allow_barrier(conf); | |
1390 | bio_wouldblock_error(bio); | |
1391 | return; | |
1392 | } | |
28be4fd3 CH |
1393 | mddev_add_trace_msg(conf->mddev, |
1394 | "raid10 wait reshape metadata"); | |
3ea7daa5 | 1395 | wait_event(mddev->sb_wait, |
2953079c | 1396 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1397 | |
1398 | conf->reshape_safe = mddev->reshape_position; | |
1399 | } | |
1400 | ||
6bfe0b49 | 1401 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1402 | * inc refcount on their rdev. Record them by setting |
1403 | * bios[x] to bio | |
d4432c23 N |
1404 | * If there are known/acknowledged bad blocks on any device |
1405 | * on which we have seen a write error, we want to avoid | |
1406 | * writing to those blocks. This potentially requires several | |
1407 | * writes to write around the bad blocks. Each set of writes | |
fd16f2e8 | 1408 | * gets its own r10_bio with a set of bios attached. |
1da177e4 | 1409 | */ |
c3b328ac | 1410 | |
69335ef3 | 1411 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1412 | raid10_find_phys(conf, r10_bio); |
f2e7e269 XN |
1413 | |
1414 | wait_blocked_dev(mddev, r10_bio); | |
1415 | ||
d4432c23 N |
1416 | max_sectors = r10_bio->sectors; |
1417 | ||
1da177e4 LT |
1418 | for (i = 0; i < conf->copies; i++) { |
1419 | int d = r10_bio->devs[i].devnum; | |
2ae6aaf7 LN |
1420 | struct md_rdev *rdev, *rrdev; |
1421 | ||
a448af25 YK |
1422 | rdev = conf->mirrors[d].rdev; |
1423 | rrdev = conf->mirrors[d].replacement; | |
8ae12666 | 1424 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1425 | rdev = NULL; |
8ae12666 | 1426 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1427 | rrdev = NULL; |
1428 | ||
d4432c23 | 1429 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1430 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1431 | |
1432 | if (!rdev && !rrdev) { | |
6cce3b23 | 1433 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1434 | continue; |
1435 | } | |
e7c0c3fa | 1436 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1437 | sector_t first_bad; |
1438 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1439 | int bad_sectors; | |
1440 | int is_bad; | |
1441 | ||
bb5f1ed7 | 1442 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 | 1443 | &first_bad, &bad_sectors); |
d4432c23 N |
1444 | if (is_bad && first_bad <= dev_sector) { |
1445 | /* Cannot write here at all */ | |
1446 | bad_sectors -= (dev_sector - first_bad); | |
1447 | if (bad_sectors < max_sectors) | |
1448 | /* Mustn't write more than bad_sectors | |
1449 | * to other devices yet | |
1450 | */ | |
1451 | max_sectors = bad_sectors; | |
1452 | /* We don't set R10BIO_Degraded as that | |
1453 | * only applies if the disk is missing, | |
1454 | * so it might be re-added, and we want to | |
1455 | * know to recover this chunk. | |
1456 | * In this case the device is here, and the | |
1457 | * fact that this chunk is not in-sync is | |
1458 | * recorded in the bad block log. | |
1459 | */ | |
1460 | continue; | |
1461 | } | |
1462 | if (is_bad) { | |
1463 | int good_sectors = first_bad - dev_sector; | |
1464 | if (good_sectors < max_sectors) | |
1465 | max_sectors = good_sectors; | |
1466 | } | |
6cce3b23 | 1467 | } |
e7c0c3fa N |
1468 | if (rdev) { |
1469 | r10_bio->devs[i].bio = bio; | |
1470 | atomic_inc(&rdev->nr_pending); | |
1471 | } | |
475b0321 N |
1472 | if (rrdev) { |
1473 | r10_bio->devs[i].repl_bio = bio; | |
1474 | atomic_inc(&rrdev->nr_pending); | |
1475 | } | |
1da177e4 | 1476 | } |
1da177e4 | 1477 | |
6b6c8110 | 1478 | if (max_sectors < r10_bio->sectors) |
d4432c23 | 1479 | r10_bio->sectors = max_sectors; |
fc9977dd N |
1480 | |
1481 | if (r10_bio->sectors < bio_sectors(bio)) { | |
1482 | struct bio *split = bio_split(bio, r10_bio->sectors, | |
afeee514 | 1483 | GFP_NOIO, &conf->bio_split); |
fc9977dd | 1484 | bio_chain(split, bio); |
e820d55c | 1485 | allow_barrier(conf); |
ed00aabd | 1486 | submit_bio_noacct(bio); |
c9aa889b | 1487 | wait_barrier(conf, false); |
fc9977dd N |
1488 | bio = split; |
1489 | r10_bio->master_bio = bio; | |
d4432c23 | 1490 | } |
d4432c23 | 1491 | |
82045523 YK |
1492 | md_account_bio(mddev, &bio); |
1493 | r10_bio->master_bio = bio; | |
4e78064f | 1494 | atomic_set(&r10_bio->remaining, 1); |
e64e4018 | 1495 | md_bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1496 | |
1da177e4 | 1497 | for (i = 0; i < conf->copies; i++) { |
27f26a0f | 1498 | if (r10_bio->devs[i].bio) |
fc9977dd | 1499 | raid10_write_one_disk(mddev, r10_bio, bio, false, i); |
27f26a0f | 1500 | if (r10_bio->devs[i].repl_bio) |
fc9977dd | 1501 | raid10_write_one_disk(mddev, r10_bio, bio, true, i); |
d4432c23 | 1502 | } |
079fa166 | 1503 | one_write_done(r10_bio); |
20d0189b KO |
1504 | } |
1505 | ||
fc9977dd | 1506 | static void __make_request(struct mddev *mddev, struct bio *bio, int sectors) |
bb5f1ed7 RL |
1507 | { |
1508 | struct r10conf *conf = mddev->private; | |
1509 | struct r10bio *r10_bio; | |
1510 | ||
afeee514 | 1511 | r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); |
bb5f1ed7 RL |
1512 | |
1513 | r10_bio->master_bio = bio; | |
fc9977dd | 1514 | r10_bio->sectors = sectors; |
bb5f1ed7 RL |
1515 | |
1516 | r10_bio->mddev = mddev; | |
1517 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1518 | r10_bio->state = 0; | |
93decc56 | 1519 | r10_bio->read_slot = -1; |
c2968285 XN |
1520 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * |
1521 | conf->geo.raid_disks); | |
bb5f1ed7 RL |
1522 | |
1523 | if (bio_data_dir(bio) == READ) | |
82045523 | 1524 | raid10_read_request(mddev, bio, r10_bio, true); |
bb5f1ed7 RL |
1525 | else |
1526 | raid10_write_request(mddev, bio, r10_bio); | |
1527 | } | |
1528 | ||
254c271d XN |
1529 | static void raid_end_discard_bio(struct r10bio *r10bio) |
1530 | { | |
1531 | struct r10conf *conf = r10bio->mddev->private; | |
1532 | struct r10bio *first_r10bio; | |
1533 | ||
1534 | while (atomic_dec_and_test(&r10bio->remaining)) { | |
1535 | ||
1536 | allow_barrier(conf); | |
1537 | ||
1538 | if (!test_bit(R10BIO_Discard, &r10bio->state)) { | |
1539 | first_r10bio = (struct r10bio *)r10bio->master_bio; | |
1540 | free_r10bio(r10bio); | |
1541 | r10bio = first_r10bio; | |
1542 | } else { | |
1543 | md_write_end(r10bio->mddev); | |
1544 | bio_endio(r10bio->master_bio); | |
1545 | free_r10bio(r10bio); | |
1546 | break; | |
1547 | } | |
1548 | } | |
1549 | } | |
1550 | ||
d30588b2 XN |
1551 | static void raid10_end_discard_request(struct bio *bio) |
1552 | { | |
1553 | struct r10bio *r10_bio = bio->bi_private; | |
1554 | struct r10conf *conf = r10_bio->mddev->private; | |
1555 | struct md_rdev *rdev = NULL; | |
1556 | int dev; | |
1557 | int slot, repl; | |
1558 | ||
1559 | /* | |
1560 | * We don't care the return value of discard bio | |
1561 | */ | |
1562 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
1563 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1564 | ||
1565 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
a448af25 YK |
1566 | rdev = repl ? conf->mirrors[dev].replacement : |
1567 | conf->mirrors[dev].rdev; | |
d30588b2 | 1568 | |
254c271d | 1569 | raid_end_discard_bio(r10_bio); |
d30588b2 XN |
1570 | rdev_dec_pending(rdev, conf->mddev); |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * There are some limitations to handle discard bio | |
1575 | * 1st, the discard size is bigger than stripe_size*2. | |
1576 | * 2st, if the discard bio spans reshape progress, we use the old way to | |
1577 | * handle discard bio | |
1578 | */ | |
1579 | static int raid10_handle_discard(struct mddev *mddev, struct bio *bio) | |
1580 | { | |
1581 | struct r10conf *conf = mddev->private; | |
1582 | struct geom *geo = &conf->geo; | |
254c271d XN |
1583 | int far_copies = geo->far_copies; |
1584 | bool first_copy = true; | |
1585 | struct r10bio *r10_bio, *first_r10bio; | |
d30588b2 XN |
1586 | struct bio *split; |
1587 | int disk; | |
1588 | sector_t chunk; | |
1589 | unsigned int stripe_size; | |
1590 | unsigned int stripe_data_disks; | |
1591 | sector_t split_size; | |
1592 | sector_t bio_start, bio_end; | |
1593 | sector_t first_stripe_index, last_stripe_index; | |
1594 | sector_t start_disk_offset; | |
1595 | unsigned int start_disk_index; | |
1596 | sector_t end_disk_offset; | |
1597 | unsigned int end_disk_index; | |
1598 | unsigned int remainder; | |
1599 | ||
1600 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1601 | return -EAGAIN; | |
1602 | ||
c9aa889b VV |
1603 | if (WARN_ON_ONCE(bio->bi_opf & REQ_NOWAIT)) { |
1604 | bio_wouldblock_error(bio); | |
1605 | return 0; | |
1606 | } | |
1607 | wait_barrier(conf, false); | |
d30588b2 XN |
1608 | |
1609 | /* | |
1610 | * Check reshape again to avoid reshape happens after checking | |
1611 | * MD_RECOVERY_RESHAPE and before wait_barrier | |
1612 | */ | |
1613 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1614 | goto out; | |
1615 | ||
1616 | if (geo->near_copies) | |
1617 | stripe_data_disks = geo->raid_disks / geo->near_copies + | |
1618 | geo->raid_disks % geo->near_copies; | |
1619 | else | |
1620 | stripe_data_disks = geo->raid_disks; | |
1621 | ||
1622 | stripe_size = stripe_data_disks << geo->chunk_shift; | |
1623 | ||
1624 | bio_start = bio->bi_iter.bi_sector; | |
1625 | bio_end = bio_end_sector(bio); | |
1626 | ||
1627 | /* | |
1628 | * Maybe one discard bio is smaller than strip size or across one | |
1629 | * stripe and discard region is larger than one stripe size. For far | |
1630 | * offset layout, if the discard region is not aligned with stripe | |
1631 | * size, there is hole when we submit discard bio to member disk. | |
1632 | * For simplicity, we only handle discard bio which discard region | |
1633 | * is bigger than stripe_size * 2 | |
1634 | */ | |
1635 | if (bio_sectors(bio) < stripe_size*2) | |
1636 | goto out; | |
1637 | ||
1638 | /* | |
1639 | * Keep bio aligned with strip size. | |
1640 | */ | |
1641 | div_u64_rem(bio_start, stripe_size, &remainder); | |
1642 | if (remainder) { | |
1643 | split_size = stripe_size - remainder; | |
1644 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); | |
1645 | bio_chain(split, bio); | |
1646 | allow_barrier(conf); | |
1647 | /* Resend the fist split part */ | |
1648 | submit_bio_noacct(split); | |
c9aa889b | 1649 | wait_barrier(conf, false); |
d30588b2 XN |
1650 | } |
1651 | div_u64_rem(bio_end, stripe_size, &remainder); | |
1652 | if (remainder) { | |
1653 | split_size = bio_sectors(bio) - remainder; | |
1654 | split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); | |
1655 | bio_chain(split, bio); | |
1656 | allow_barrier(conf); | |
1657 | /* Resend the second split part */ | |
1658 | submit_bio_noacct(bio); | |
1659 | bio = split; | |
c9aa889b | 1660 | wait_barrier(conf, false); |
d30588b2 XN |
1661 | } |
1662 | ||
d30588b2 XN |
1663 | bio_start = bio->bi_iter.bi_sector; |
1664 | bio_end = bio_end_sector(bio); | |
1665 | ||
1666 | /* | |
1667 | * Raid10 uses chunk as the unit to store data. It's similar like raid0. | |
1668 | * One stripe contains the chunks from all member disk (one chunk from | |
1669 | * one disk at the same HBA address). For layout detail, see 'man md 4' | |
1670 | */ | |
1671 | chunk = bio_start >> geo->chunk_shift; | |
1672 | chunk *= geo->near_copies; | |
1673 | first_stripe_index = chunk; | |
1674 | start_disk_index = sector_div(first_stripe_index, geo->raid_disks); | |
1675 | if (geo->far_offset) | |
1676 | first_stripe_index *= geo->far_copies; | |
1677 | start_disk_offset = (bio_start & geo->chunk_mask) + | |
1678 | (first_stripe_index << geo->chunk_shift); | |
1679 | ||
1680 | chunk = bio_end >> geo->chunk_shift; | |
1681 | chunk *= geo->near_copies; | |
1682 | last_stripe_index = chunk; | |
1683 | end_disk_index = sector_div(last_stripe_index, geo->raid_disks); | |
1684 | if (geo->far_offset) | |
1685 | last_stripe_index *= geo->far_copies; | |
1686 | end_disk_offset = (bio_end & geo->chunk_mask) + | |
1687 | (last_stripe_index << geo->chunk_shift); | |
1688 | ||
254c271d XN |
1689 | retry_discard: |
1690 | r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); | |
1691 | r10_bio->mddev = mddev; | |
1692 | r10_bio->state = 0; | |
1693 | r10_bio->sectors = 0; | |
1694 | memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks); | |
1695 | wait_blocked_dev(mddev, r10_bio); | |
1696 | ||
1697 | /* | |
1698 | * For far layout it needs more than one r10bio to cover all regions. | |
1699 | * Inspired by raid10_sync_request, we can use the first r10bio->master_bio | |
1700 | * to record the discard bio. Other r10bio->master_bio record the first | |
1701 | * r10bio. The first r10bio only release after all other r10bios finish. | |
1702 | * The discard bio returns only first r10bio finishes | |
1703 | */ | |
1704 | if (first_copy) { | |
1705 | r10_bio->master_bio = bio; | |
1706 | set_bit(R10BIO_Discard, &r10_bio->state); | |
1707 | first_copy = false; | |
1708 | first_r10bio = r10_bio; | |
1709 | } else | |
1710 | r10_bio->master_bio = (struct bio *)first_r10bio; | |
1711 | ||
46d4703b XN |
1712 | /* |
1713 | * first select target devices under rcu_lock and | |
1714 | * inc refcount on their rdev. Record them by setting | |
1715 | * bios[x] to bio | |
1716 | */ | |
d30588b2 | 1717 | for (disk = 0; disk < geo->raid_disks; disk++) { |
67364349 | 1718 | struct md_rdev *rdev, *rrdev; |
d30588b2 | 1719 | |
a448af25 YK |
1720 | rdev = conf->mirrors[disk].rdev; |
1721 | rrdev = conf->mirrors[disk].replacement; | |
d30588b2 XN |
1722 | r10_bio->devs[disk].bio = NULL; |
1723 | r10_bio->devs[disk].repl_bio = NULL; | |
1724 | ||
1725 | if (rdev && (test_bit(Faulty, &rdev->flags))) | |
1726 | rdev = NULL; | |
1727 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) | |
1728 | rrdev = NULL; | |
1729 | if (!rdev && !rrdev) | |
1730 | continue; | |
1731 | ||
1732 | if (rdev) { | |
1733 | r10_bio->devs[disk].bio = bio; | |
1734 | atomic_inc(&rdev->nr_pending); | |
1735 | } | |
1736 | if (rrdev) { | |
1737 | r10_bio->devs[disk].repl_bio = bio; | |
1738 | atomic_inc(&rrdev->nr_pending); | |
1739 | } | |
1740 | } | |
d30588b2 XN |
1741 | |
1742 | atomic_set(&r10_bio->remaining, 1); | |
1743 | for (disk = 0; disk < geo->raid_disks; disk++) { | |
1744 | sector_t dev_start, dev_end; | |
1745 | struct bio *mbio, *rbio = NULL; | |
d30588b2 XN |
1746 | |
1747 | /* | |
1748 | * Now start to calculate the start and end address for each disk. | |
1749 | * The space between dev_start and dev_end is the discard region. | |
1750 | * | |
1751 | * For dev_start, it needs to consider three conditions: | |
1752 | * 1st, the disk is before start_disk, you can imagine the disk in | |
1753 | * the next stripe. So the dev_start is the start address of next | |
1754 | * stripe. | |
1755 | * 2st, the disk is after start_disk, it means the disk is at the | |
1756 | * same stripe of first disk | |
1757 | * 3st, the first disk itself, we can use start_disk_offset directly | |
1758 | */ | |
1759 | if (disk < start_disk_index) | |
1760 | dev_start = (first_stripe_index + 1) * mddev->chunk_sectors; | |
1761 | else if (disk > start_disk_index) | |
1762 | dev_start = first_stripe_index * mddev->chunk_sectors; | |
1763 | else | |
1764 | dev_start = start_disk_offset; | |
1765 | ||
1766 | if (disk < end_disk_index) | |
1767 | dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; | |
1768 | else if (disk > end_disk_index) | |
1769 | dev_end = last_stripe_index * mddev->chunk_sectors; | |
1770 | else | |
1771 | dev_end = end_disk_offset; | |
1772 | ||
1773 | /* | |
1774 | * It only handles discard bio which size is >= stripe size, so | |
46d4703b XN |
1775 | * dev_end > dev_start all the time. |
1776 | * It doesn't need to use rcu lock to get rdev here. We already | |
1777 | * add rdev->nr_pending in the first loop. | |
d30588b2 XN |
1778 | */ |
1779 | if (r10_bio->devs[disk].bio) { | |
46d4703b | 1780 | struct md_rdev *rdev = conf->mirrors[disk].rdev; |
abfc426d CH |
1781 | mbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, |
1782 | &mddev->bio_set); | |
d30588b2 XN |
1783 | mbio->bi_end_io = raid10_end_discard_request; |
1784 | mbio->bi_private = r10_bio; | |
1785 | r10_bio->devs[disk].bio = mbio; | |
1786 | r10_bio->devs[disk].devnum = disk; | |
1787 | atomic_inc(&r10_bio->remaining); | |
1788 | md_submit_discard_bio(mddev, rdev, mbio, | |
1789 | dev_start + choose_data_offset(r10_bio, rdev), | |
1790 | dev_end - dev_start); | |
1791 | bio_endio(mbio); | |
1792 | } | |
1793 | if (r10_bio->devs[disk].repl_bio) { | |
46d4703b | 1794 | struct md_rdev *rrdev = conf->mirrors[disk].replacement; |
abfc426d CH |
1795 | rbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, |
1796 | &mddev->bio_set); | |
d30588b2 XN |
1797 | rbio->bi_end_io = raid10_end_discard_request; |
1798 | rbio->bi_private = r10_bio; | |
1799 | r10_bio->devs[disk].repl_bio = rbio; | |
1800 | r10_bio->devs[disk].devnum = disk; | |
1801 | atomic_inc(&r10_bio->remaining); | |
1802 | md_submit_discard_bio(mddev, rrdev, rbio, | |
1803 | dev_start + choose_data_offset(r10_bio, rrdev), | |
1804 | dev_end - dev_start); | |
1805 | bio_endio(rbio); | |
1806 | } | |
1807 | } | |
1808 | ||
254c271d XN |
1809 | if (!geo->far_offset && --far_copies) { |
1810 | first_stripe_index += geo->stride >> geo->chunk_shift; | |
1811 | start_disk_offset += geo->stride; | |
1812 | last_stripe_index += geo->stride >> geo->chunk_shift; | |
1813 | end_disk_offset += geo->stride; | |
1814 | atomic_inc(&first_r10bio->remaining); | |
1815 | raid_end_discard_bio(r10_bio); | |
c9aa889b | 1816 | wait_barrier(conf, false); |
254c271d | 1817 | goto retry_discard; |
d30588b2 XN |
1818 | } |
1819 | ||
254c271d XN |
1820 | raid_end_discard_bio(r10_bio); |
1821 | ||
d30588b2 XN |
1822 | return 0; |
1823 | out: | |
1824 | allow_barrier(conf); | |
1825 | return -EAGAIN; | |
1826 | } | |
1827 | ||
cc27b0c7 | 1828 | static bool raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1829 | { |
1830 | struct r10conf *conf = mddev->private; | |
1831 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1832 | int chunk_sects = chunk_mask + 1; | |
fc9977dd | 1833 | int sectors = bio_sectors(bio); |
20d0189b | 1834 | |
775d7831 DJ |
1835 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
1836 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 1837 | return true; |
20d0189b | 1838 | |
cc27b0c7 N |
1839 | if (!md_write_start(mddev, bio)) |
1840 | return false; | |
1841 | ||
d30588b2 XN |
1842 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) |
1843 | if (!raid10_handle_discard(mddev, bio)) | |
1844 | return true; | |
1845 | ||
fc9977dd N |
1846 | /* |
1847 | * If this request crosses a chunk boundary, we need to split | |
1848 | * it. | |
1849 | */ | |
1850 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1851 | sectors > chunk_sects | |
1852 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1853 | || conf->prev.near_copies < | |
1854 | conf->prev.raid_disks))) | |
1855 | sectors = chunk_sects - | |
1856 | (bio->bi_iter.bi_sector & | |
1857 | (chunk_sects - 1)); | |
1858 | __make_request(mddev, bio, sectors); | |
079fa166 N |
1859 | |
1860 | /* In case raid10d snuck in to freeze_array */ | |
0c0be98b | 1861 | wake_up_barrier(conf); |
cc27b0c7 | 1862 | return true; |
1da177e4 LT |
1863 | } |
1864 | ||
849674e4 | 1865 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1866 | { |
e879a879 | 1867 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1868 | int i; |
1869 | ||
a448af25 YK |
1870 | lockdep_assert_held(&mddev->lock); |
1871 | ||
5cf00fcd | 1872 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1873 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1874 | if (conf->geo.near_copies > 1) |
1875 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1876 | if (conf->geo.far_copies > 1) { | |
1877 | if (conf->geo.far_offset) | |
1878 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1879 | else |
5cf00fcd | 1880 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1881 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1882 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1883 | } |
5cf00fcd N |
1884 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1885 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 | 1886 | for (i = 0; i < conf->geo.raid_disks; i++) { |
a448af25 YK |
1887 | struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev); |
1888 | ||
d44b0a92 N |
1889 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); |
1890 | } | |
1da177e4 LT |
1891 | seq_printf(seq, "]"); |
1892 | } | |
1893 | ||
700c7213 N |
1894 | /* check if there are enough drives for |
1895 | * every block to appear on atleast one. | |
1896 | * Don't consider the device numbered 'ignore' | |
1897 | * as we might be about to remove it. | |
1898 | */ | |
635f6416 | 1899 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1900 | { |
1901 | int first = 0; | |
725d6e57 | 1902 | int has_enough = 0; |
635f6416 N |
1903 | int disks, ncopies; |
1904 | if (previous) { | |
1905 | disks = conf->prev.raid_disks; | |
1906 | ncopies = conf->prev.near_copies; | |
1907 | } else { | |
1908 | disks = conf->geo.raid_disks; | |
1909 | ncopies = conf->geo.near_copies; | |
1910 | } | |
700c7213 N |
1911 | |
1912 | do { | |
1913 | int n = conf->copies; | |
1914 | int cnt = 0; | |
80b48124 | 1915 | int this = first; |
700c7213 | 1916 | while (n--) { |
725d6e57 N |
1917 | struct md_rdev *rdev; |
1918 | if (this != ignore && | |
a448af25 | 1919 | (rdev = conf->mirrors[this].rdev) && |
725d6e57 | 1920 | test_bit(In_sync, &rdev->flags)) |
700c7213 | 1921 | cnt++; |
635f6416 | 1922 | this = (this+1) % disks; |
700c7213 N |
1923 | } |
1924 | if (cnt == 0) | |
725d6e57 | 1925 | goto out; |
635f6416 | 1926 | first = (first + ncopies) % disks; |
700c7213 | 1927 | } while (first != 0); |
725d6e57 N |
1928 | has_enough = 1; |
1929 | out: | |
725d6e57 | 1930 | return has_enough; |
700c7213 N |
1931 | } |
1932 | ||
f8c9e74f N |
1933 | static int enough(struct r10conf *conf, int ignore) |
1934 | { | |
635f6416 N |
1935 | /* when calling 'enough', both 'prev' and 'geo' must |
1936 | * be stable. | |
1937 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1938 | * is held. | |
1939 | */ | |
1940 | return _enough(conf, 0, ignore) && | |
1941 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1942 | } |
1943 | ||
9631abdb MT |
1944 | /** |
1945 | * raid10_error() - RAID10 error handler. | |
1946 | * @mddev: affected md device. | |
1947 | * @rdev: member device to fail. | |
1948 | * | |
1949 | * The routine acknowledges &rdev failure and determines new @mddev state. | |
1950 | * If it failed, then: | |
1951 | * - &MD_BROKEN flag is set in &mddev->flags. | |
1952 | * Otherwise, it must be degraded: | |
1953 | * - recovery is interrupted. | |
1954 | * - &mddev->degraded is bumped. | |
62bca04b | 1955 | * |
9631abdb MT |
1956 | * @rdev is marked as &Faulty excluding case when array is failed and |
1957 | * &mddev->fail_last_dev is off. | |
1958 | */ | |
849674e4 | 1959 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1960 | { |
e879a879 | 1961 | struct r10conf *conf = mddev->private; |
635f6416 | 1962 | unsigned long flags; |
1da177e4 | 1963 | |
635f6416 | 1964 | spin_lock_irqsave(&conf->device_lock, flags); |
9631abdb MT |
1965 | |
1966 | if (test_bit(In_sync, &rdev->flags) && !enough(conf, rdev->raid_disk)) { | |
1967 | set_bit(MD_BROKEN, &mddev->flags); | |
1968 | ||
1969 | if (!mddev->fail_last_dev) { | |
1970 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1971 | return; | |
1972 | } | |
635f6416 | 1973 | } |
2446dba0 | 1974 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1975 | mddev->degraded++; |
9631abdb | 1976 | |
2446dba0 | 1977 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
de393cde | 1978 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1979 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1980 | set_mask_bits(&mddev->sb_flags, 0, |
1981 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1982 | spin_unlock_irqrestore(&conf->device_lock, flags); |
913cce5a | 1983 | pr_crit("md/raid10:%s: Disk failure on %pg, disabling device.\n" |
08464e09 | 1984 | "md/raid10:%s: Operation continuing on %d devices.\n", |
913cce5a | 1985 | mdname(mddev), rdev->bdev, |
08464e09 | 1986 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); |
1da177e4 LT |
1987 | } |
1988 | ||
e879a879 | 1989 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1990 | { |
1991 | int i; | |
4056ca51 | 1992 | struct md_rdev *rdev; |
1da177e4 | 1993 | |
08464e09 | 1994 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1995 | if (!conf) { |
08464e09 | 1996 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1997 | return; |
1998 | } | |
08464e09 N |
1999 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
2000 | conf->geo.raid_disks); | |
1da177e4 | 2001 | |
a448af25 | 2002 | lockdep_assert_held(&conf->mddev->reconfig_mutex); |
5cf00fcd | 2003 | for (i = 0; i < conf->geo.raid_disks; i++) { |
4056ca51 N |
2004 | rdev = conf->mirrors[i].rdev; |
2005 | if (rdev) | |
913cce5a | 2006 | pr_debug(" disk %d, wo:%d, o:%d, dev:%pg\n", |
08464e09 N |
2007 | i, !test_bit(In_sync, &rdev->flags), |
2008 | !test_bit(Faulty, &rdev->flags), | |
913cce5a | 2009 | rdev->bdev); |
1da177e4 LT |
2010 | } |
2011 | } | |
2012 | ||
e879a879 | 2013 | static void close_sync(struct r10conf *conf) |
1da177e4 | 2014 | { |
c9aa889b | 2015 | wait_barrier(conf, false); |
0a27ec96 | 2016 | allow_barrier(conf); |
1da177e4 | 2017 | |
afeee514 | 2018 | mempool_exit(&conf->r10buf_pool); |
1da177e4 LT |
2019 | } |
2020 | ||
fd01b88c | 2021 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
2022 | { |
2023 | int i; | |
e879a879 | 2024 | struct r10conf *conf = mddev->private; |
dc280d98 | 2025 | struct raid10_info *tmp; |
6b965620 N |
2026 | int count = 0; |
2027 | unsigned long flags; | |
1da177e4 LT |
2028 | |
2029 | /* | |
2030 | * Find all non-in_sync disks within the RAID10 configuration | |
2031 | * and mark them in_sync | |
2032 | */ | |
5cf00fcd | 2033 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 2034 | tmp = conf->mirrors + i; |
4ca40c2c N |
2035 | if (tmp->replacement |
2036 | && tmp->replacement->recovery_offset == MaxSector | |
2037 | && !test_bit(Faulty, &tmp->replacement->flags) | |
2038 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
2039 | /* Replacement has just become active */ | |
2040 | if (!tmp->rdev | |
2041 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
2042 | count++; | |
2043 | if (tmp->rdev) { | |
2044 | /* Replaced device not technically faulty, | |
2045 | * but we need to be sure it gets removed | |
2046 | * and never re-added. | |
2047 | */ | |
2048 | set_bit(Faulty, &tmp->rdev->flags); | |
2049 | sysfs_notify_dirent_safe( | |
2050 | tmp->rdev->sysfs_state); | |
2051 | } | |
2052 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
2053 | } else if (tmp->rdev | |
61e4947c | 2054 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
2055 | && !test_bit(Faulty, &tmp->rdev->flags) |
2056 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 2057 | count++; |
2863b9eb | 2058 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
2059 | } |
2060 | } | |
6b965620 N |
2061 | spin_lock_irqsave(&conf->device_lock, flags); |
2062 | mddev->degraded -= count; | |
2063 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
2064 | |
2065 | print_conf(conf); | |
6b965620 | 2066 | return count; |
1da177e4 LT |
2067 | } |
2068 | ||
fd01b88c | 2069 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 2070 | { |
e879a879 | 2071 | struct r10conf *conf = mddev->private; |
199050ea | 2072 | int err = -EEXIST; |
6090368a | 2073 | int mirror, repl_slot = -1; |
6c2fce2e | 2074 | int first = 0; |
5cf00fcd | 2075 | int last = conf->geo.raid_disks - 1; |
6090368a | 2076 | struct raid10_info *p; |
1da177e4 LT |
2077 | |
2078 | if (mddev->recovery_cp < MaxSector) | |
2079 | /* only hot-add to in-sync arrays, as recovery is | |
2080 | * very different from resync | |
2081 | */ | |
199050ea | 2082 | return -EBUSY; |
635f6416 | 2083 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 2084 | return -EINVAL; |
1da177e4 | 2085 | |
1501efad DW |
2086 | if (md_integrity_add_rdev(rdev, mddev)) |
2087 | return -ENXIO; | |
2088 | ||
a53a6c85 | 2089 | if (rdev->raid_disk >= 0) |
6c2fce2e | 2090 | first = last = rdev->raid_disk; |
1da177e4 | 2091 | |
2c4193df | 2092 | if (rdev->saved_raid_disk >= first && |
9e753ba9 | 2093 | rdev->saved_raid_disk < conf->geo.raid_disks && |
6cce3b23 N |
2094 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
2095 | mirror = rdev->saved_raid_disk; | |
2096 | else | |
6c2fce2e | 2097 | mirror = first; |
2bb77736 | 2098 | for ( ; mirror <= last ; mirror++) { |
6090368a | 2099 | p = &conf->mirrors[mirror]; |
2bb77736 N |
2100 | if (p->recovery_disabled == mddev->recovery_disabled) |
2101 | continue; | |
b7044d41 | 2102 | if (p->rdev) { |
6090368a LN |
2103 | if (test_bit(WantReplacement, &p->rdev->flags) && |
2104 | p->replacement == NULL && repl_slot < 0) | |
2105 | repl_slot = mirror; | |
2106 | continue; | |
b7044d41 | 2107 | } |
1da177e4 | 2108 | |
3d8466ba CH |
2109 | err = mddev_stack_new_rdev(mddev, rdev); |
2110 | if (err) | |
2111 | return err; | |
2bb77736 | 2112 | p->head_position = 0; |
d890fa2b | 2113 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
2114 | rdev->raid_disk = mirror; |
2115 | err = 0; | |
2116 | if (rdev->saved_raid_disk != mirror) | |
2117 | conf->fullsync = 1; | |
a448af25 | 2118 | WRITE_ONCE(p->rdev, rdev); |
2bb77736 N |
2119 | break; |
2120 | } | |
532a2a3f | 2121 | |
6090368a LN |
2122 | if (err && repl_slot >= 0) { |
2123 | p = &conf->mirrors[repl_slot]; | |
2124 | clear_bit(In_sync, &rdev->flags); | |
2125 | set_bit(Replacement, &rdev->flags); | |
2126 | rdev->raid_disk = repl_slot; | |
3d8466ba CH |
2127 | err = mddev_stack_new_rdev(mddev, rdev); |
2128 | if (err) | |
2129 | return err; | |
6090368a | 2130 | conf->fullsync = 1; |
a448af25 | 2131 | WRITE_ONCE(p->replacement, rdev); |
6090368a LN |
2132 | } |
2133 | ||
1da177e4 | 2134 | print_conf(conf); |
199050ea | 2135 | return err; |
1da177e4 LT |
2136 | } |
2137 | ||
b8321b68 | 2138 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 2139 | { |
e879a879 | 2140 | struct r10conf *conf = mddev->private; |
1da177e4 | 2141 | int err = 0; |
b8321b68 | 2142 | int number = rdev->raid_disk; |
c8ab903e | 2143 | struct md_rdev **rdevp; |
d17f744e | 2144 | struct raid10_info *p; |
1da177e4 LT |
2145 | |
2146 | print_conf(conf); | |
d17f744e MP |
2147 | if (unlikely(number >= mddev->raid_disks)) |
2148 | return 0; | |
2149 | p = conf->mirrors + number; | |
c8ab903e N |
2150 | if (rdev == p->rdev) |
2151 | rdevp = &p->rdev; | |
2152 | else if (rdev == p->replacement) | |
2153 | rdevp = &p->replacement; | |
2154 | else | |
2155 | return 0; | |
2156 | ||
2157 | if (test_bit(In_sync, &rdev->flags) || | |
2158 | atomic_read(&rdev->nr_pending)) { | |
2159 | err = -EBUSY; | |
2160 | goto abort; | |
2161 | } | |
d787be40 | 2162 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
2163 | * is not possible. |
2164 | */ | |
2165 | if (!test_bit(Faulty, &rdev->flags) && | |
2166 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 2167 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 2168 | number < conf->geo.raid_disks && |
c8ab903e N |
2169 | enough(conf, -1)) { |
2170 | err = -EBUSY; | |
2171 | goto abort; | |
1da177e4 | 2172 | } |
a448af25 | 2173 | WRITE_ONCE(*rdevp, NULL); |
d787be40 | 2174 | if (p->replacement) { |
4ca40c2c | 2175 | /* We must have just cleared 'rdev' */ |
a448af25 | 2176 | WRITE_ONCE(p->rdev, p->replacement); |
4ca40c2c | 2177 | clear_bit(Replacement, &p->replacement->flags); |
a448af25 | 2178 | WRITE_ONCE(p->replacement, NULL); |
e5bc9c3c | 2179 | } |
4ca40c2c | 2180 | |
e5bc9c3c | 2181 | clear_bit(WantReplacement, &rdev->flags); |
c8ab903e N |
2182 | err = md_integrity_register(mddev); |
2183 | ||
1da177e4 LT |
2184 | abort: |
2185 | ||
2186 | print_conf(conf); | |
2187 | return err; | |
2188 | } | |
2189 | ||
81fa1520 | 2190 | static void __end_sync_read(struct r10bio *r10_bio, struct bio *bio, int d) |
1da177e4 | 2191 | { |
e879a879 | 2192 | struct r10conf *conf = r10_bio->mddev->private; |
0eb3ff12 | 2193 | |
4e4cbee9 | 2194 | if (!bio->bi_status) |
0eb3ff12 | 2195 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
2196 | else |
2197 | /* The write handler will notice the lack of | |
2198 | * R10BIO_Uptodate and record any errors etc | |
2199 | */ | |
4dbcdc75 N |
2200 | atomic_add(r10_bio->sectors, |
2201 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
2202 | |
2203 | /* for reconstruct, we always reschedule after a read. | |
2204 | * for resync, only after all reads | |
2205 | */ | |
73d5c38a | 2206 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
2207 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
2208 | atomic_dec_and_test(&r10_bio->remaining)) { | |
2209 | /* we have read all the blocks, | |
2210 | * do the comparison in process context in raid10d | |
2211 | */ | |
2212 | reschedule_retry(r10_bio); | |
2213 | } | |
1da177e4 LT |
2214 | } |
2215 | ||
81fa1520 ML |
2216 | static void end_sync_read(struct bio *bio) |
2217 | { | |
f0250618 | 2218 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
81fa1520 ML |
2219 | struct r10conf *conf = r10_bio->mddev->private; |
2220 | int d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
2221 | ||
2222 | __end_sync_read(r10_bio, bio, d); | |
2223 | } | |
2224 | ||
2225 | static void end_reshape_read(struct bio *bio) | |
2226 | { | |
f0250618 | 2227 | /* reshape read bio isn't allocated from r10buf_pool */ |
81fa1520 ML |
2228 | struct r10bio *r10_bio = bio->bi_private; |
2229 | ||
2230 | __end_sync_read(r10_bio, bio, r10_bio->read_slot); | |
2231 | } | |
2232 | ||
9f2c9d12 | 2233 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 2234 | { |
fd01b88c | 2235 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 2236 | |
1da177e4 LT |
2237 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
2238 | if (r10_bio->master_bio == NULL) { | |
2239 | /* the primary of several recovery bios */ | |
73d5c38a | 2240 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
2241 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2242 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2243 | reschedule_retry(r10_bio); |
2244 | else | |
2245 | put_buf(r10_bio); | |
73d5c38a | 2246 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
2247 | break; |
2248 | } else { | |
9f2c9d12 | 2249 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
2250 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2251 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
2252 | reschedule_retry(r10_bio); |
2253 | else | |
2254 | put_buf(r10_bio); | |
1da177e4 LT |
2255 | r10_bio = r10_bio2; |
2256 | } | |
2257 | } | |
1da177e4 LT |
2258 | } |
2259 | ||
4246a0b6 | 2260 | static void end_sync_write(struct bio *bio) |
5e570289 | 2261 | { |
f0250618 | 2262 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
fd01b88c | 2263 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2264 | struct r10conf *conf = mddev->private; |
5e570289 | 2265 | int d; |
5e570289 | 2266 | int slot; |
9ad1aefc | 2267 | int repl; |
4ca40c2c | 2268 | struct md_rdev *rdev = NULL; |
5e570289 | 2269 | |
9ad1aefc N |
2270 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
2271 | if (repl) | |
2272 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 2273 | else |
9ad1aefc | 2274 | rdev = conf->mirrors[d].rdev; |
5e570289 | 2275 | |
4e4cbee9 | 2276 | if (bio->bi_status) { |
9ad1aefc N |
2277 | if (repl) |
2278 | md_error(mddev, rdev); | |
2279 | else { | |
2280 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
2281 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2282 | set_bit(MD_RECOVERY_NEEDED, | |
2283 | &rdev->mddev->recovery); | |
9ad1aefc N |
2284 | set_bit(R10BIO_WriteError, &r10_bio->state); |
2285 | } | |
3a0f007b YK |
2286 | } else if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr, |
2287 | r10_bio->sectors)) { | |
5e570289 | 2288 | set_bit(R10BIO_MadeGood, &r10_bio->state); |
3a0f007b | 2289 | } |
5e570289 | 2290 | |
9ad1aefc | 2291 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
2292 | |
2293 | end_sync_request(r10_bio); | |
2294 | } | |
2295 | ||
1da177e4 LT |
2296 | /* |
2297 | * Note: sync and recover and handled very differently for raid10 | |
2298 | * This code is for resync. | |
2299 | * For resync, we read through virtual addresses and read all blocks. | |
2300 | * If there is any error, we schedule a write. The lowest numbered | |
2301 | * drive is authoritative. | |
2302 | * However requests come for physical address, so we need to map. | |
2303 | * For every physical address there are raid_disks/copies virtual addresses, | |
2304 | * which is always are least one, but is not necessarly an integer. | |
2305 | * This means that a physical address can span multiple chunks, so we may | |
2306 | * have to submit multiple io requests for a single sync request. | |
2307 | */ | |
2308 | /* | |
2309 | * We check if all blocks are in-sync and only write to blocks that | |
2310 | * aren't in sync | |
2311 | */ | |
9f2c9d12 | 2312 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2313 | { |
e879a879 | 2314 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
2315 | int i, first; |
2316 | struct bio *tbio, *fbio; | |
f4380a91 | 2317 | int vcnt; |
cdb76be3 | 2318 | struct page **tpages, **fpages; |
1da177e4 LT |
2319 | |
2320 | atomic_set(&r10_bio->remaining, 1); | |
2321 | ||
2322 | /* find the first device with a block */ | |
2323 | for (i=0; i<conf->copies; i++) | |
4e4cbee9 | 2324 | if (!r10_bio->devs[i].bio->bi_status) |
1da177e4 LT |
2325 | break; |
2326 | ||
2327 | if (i == conf->copies) | |
2328 | goto done; | |
2329 | ||
2330 | first = i; | |
2331 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2332 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2333 | fbio->bi_iter.bi_idx = 0; | |
cdb76be3 | 2334 | fpages = get_resync_pages(fbio)->pages; |
1da177e4 | 2335 | |
f4380a91 | 2336 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2337 | /* now find blocks with errors */ |
0eb3ff12 N |
2338 | for (i=0 ; i < conf->copies ; i++) { |
2339 | int j, d; | |
8d3ca83d | 2340 | struct md_rdev *rdev; |
f0250618 | 2341 | struct resync_pages *rp; |
1da177e4 | 2342 | |
1da177e4 | 2343 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2344 | |
2345 | if (tbio->bi_end_io != end_sync_read) | |
2346 | continue; | |
2347 | if (i == first) | |
1da177e4 | 2348 | continue; |
cdb76be3 ML |
2349 | |
2350 | tpages = get_resync_pages(tbio)->pages; | |
8d3ca83d N |
2351 | d = r10_bio->devs[i].devnum; |
2352 | rdev = conf->mirrors[d].rdev; | |
4e4cbee9 | 2353 | if (!r10_bio->devs[i].bio->bi_status) { |
0eb3ff12 N |
2354 | /* We know that the bi_io_vec layout is the same for |
2355 | * both 'first' and 'i', so we just compare them. | |
2356 | * All vec entries are PAGE_SIZE; | |
2357 | */ | |
7bb23c49 N |
2358 | int sectors = r10_bio->sectors; |
2359 | for (j = 0; j < vcnt; j++) { | |
2360 | int len = PAGE_SIZE; | |
2361 | if (sectors < (len / 512)) | |
2362 | len = sectors * 512; | |
cdb76be3 ML |
2363 | if (memcmp(page_address(fpages[j]), |
2364 | page_address(tpages[j]), | |
7bb23c49 | 2365 | len)) |
0eb3ff12 | 2366 | break; |
7bb23c49 N |
2367 | sectors -= len/512; |
2368 | } | |
0eb3ff12 N |
2369 | if (j == vcnt) |
2370 | continue; | |
7f7583d4 | 2371 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2372 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2373 | /* Don't fix anything. */ | |
2374 | continue; | |
8d3ca83d N |
2375 | } else if (test_bit(FailFast, &rdev->flags)) { |
2376 | /* Just give up on this device */ | |
2377 | md_error(rdev->mddev, rdev); | |
2378 | continue; | |
0eb3ff12 | 2379 | } |
f84ee364 N |
2380 | /* Ok, we need to write this bio, either to correct an |
2381 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2382 | * First we need to fixup bv_offset, bv_len and |
2383 | * bi_vecs, as the read request might have corrupted these | |
2384 | */ | |
f0250618 | 2385 | rp = get_resync_pages(tbio); |
a7c50c94 | 2386 | bio_reset(tbio, conf->mirrors[d].rdev->bdev, REQ_OP_WRITE); |
8be185f2 | 2387 | |
fb0eb5df ML |
2388 | md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); |
2389 | ||
f0250618 ML |
2390 | rp->raid_bio = r10_bio; |
2391 | tbio->bi_private = rp; | |
4f024f37 | 2392 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 LT |
2393 | tbio->bi_end_io = end_sync_write; |
2394 | ||
c31df25f KO |
2395 | bio_copy_data(tbio, fbio); |
2396 | ||
1da177e4 LT |
2397 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2398 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2399 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2400 | |
1919cbb2 N |
2401 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2402 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2403 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
ed00aabd | 2404 | submit_bio_noacct(tbio); |
1da177e4 LT |
2405 | } |
2406 | ||
9ad1aefc N |
2407 | /* Now write out to any replacement devices |
2408 | * that are active | |
2409 | */ | |
2410 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2411 | int d; |
9ad1aefc N |
2412 | |
2413 | tbio = r10_bio->devs[i].repl_bio; | |
2414 | if (!tbio || !tbio->bi_end_io) | |
2415 | continue; | |
2416 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2417 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2418 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2419 | d = r10_bio->devs[i].devnum; |
2420 | atomic_inc(&r10_bio->remaining); | |
2421 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2422 | bio_sectors(tbio)); |
ed00aabd | 2423 | submit_bio_noacct(tbio); |
9ad1aefc N |
2424 | } |
2425 | ||
1da177e4 LT |
2426 | done: |
2427 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2428 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2429 | put_buf(r10_bio); | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | /* | |
2434 | * Now for the recovery code. | |
2435 | * Recovery happens across physical sectors. | |
2436 | * We recover all non-is_sync drives by finding the virtual address of | |
2437 | * each, and then choose a working drive that also has that virt address. | |
2438 | * There is a separate r10_bio for each non-in_sync drive. | |
2439 | * Only the first two slots are in use. The first for reading, | |
2440 | * The second for writing. | |
2441 | * | |
2442 | */ | |
9f2c9d12 | 2443 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2444 | { |
2445 | /* We got a read error during recovery. | |
2446 | * We repeat the read in smaller page-sized sections. | |
2447 | * If a read succeeds, write it to the new device or record | |
2448 | * a bad block if we cannot. | |
2449 | * If a read fails, record a bad block on both old and | |
2450 | * new devices. | |
2451 | */ | |
fd01b88c | 2452 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2453 | struct r10conf *conf = mddev->private; |
5e570289 N |
2454 | struct bio *bio = r10_bio->devs[0].bio; |
2455 | sector_t sect = 0; | |
2456 | int sectors = r10_bio->sectors; | |
2457 | int idx = 0; | |
2458 | int dr = r10_bio->devs[0].devnum; | |
2459 | int dw = r10_bio->devs[1].devnum; | |
cdb76be3 | 2460 | struct page **pages = get_resync_pages(bio)->pages; |
5e570289 N |
2461 | |
2462 | while (sectors) { | |
2463 | int s = sectors; | |
3cb03002 | 2464 | struct md_rdev *rdev; |
5e570289 N |
2465 | sector_t addr; |
2466 | int ok; | |
2467 | ||
2468 | if (s > (PAGE_SIZE>>9)) | |
2469 | s = PAGE_SIZE >> 9; | |
2470 | ||
2471 | rdev = conf->mirrors[dr].rdev; | |
2472 | addr = r10_bio->devs[0].addr + sect, | |
2473 | ok = sync_page_io(rdev, | |
2474 | addr, | |
2475 | s << 9, | |
cdb76be3 | 2476 | pages[idx], |
4ce4c73f | 2477 | REQ_OP_READ, false); |
5e570289 N |
2478 | if (ok) { |
2479 | rdev = conf->mirrors[dw].rdev; | |
2480 | addr = r10_bio->devs[1].addr + sect; | |
2481 | ok = sync_page_io(rdev, | |
2482 | addr, | |
2483 | s << 9, | |
cdb76be3 | 2484 | pages[idx], |
4ce4c73f | 2485 | REQ_OP_WRITE, false); |
b7044d41 | 2486 | if (!ok) { |
5e570289 | 2487 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2488 | if (!test_and_set_bit(WantReplacement, |
2489 | &rdev->flags)) | |
2490 | set_bit(MD_RECOVERY_NEEDED, | |
2491 | &rdev->mddev->recovery); | |
2492 | } | |
5e570289 N |
2493 | } |
2494 | if (!ok) { | |
2495 | /* We don't worry if we cannot set a bad block - | |
2496 | * it really is bad so there is no loss in not | |
2497 | * recording it yet | |
2498 | */ | |
2499 | rdev_set_badblocks(rdev, addr, s, 0); | |
2500 | ||
2501 | if (rdev != conf->mirrors[dw].rdev) { | |
2502 | /* need bad block on destination too */ | |
3cb03002 | 2503 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2504 | addr = r10_bio->devs[1].addr + sect; |
2505 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2506 | if (!ok) { | |
2507 | /* just abort the recovery */ | |
08464e09 N |
2508 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2509 | mdname(mddev)); | |
5e570289 N |
2510 | |
2511 | conf->mirrors[dw].recovery_disabled | |
2512 | = mddev->recovery_disabled; | |
2513 | set_bit(MD_RECOVERY_INTR, | |
2514 | &mddev->recovery); | |
2515 | break; | |
2516 | } | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | sectors -= s; | |
2521 | sect += s; | |
2522 | idx++; | |
2523 | } | |
2524 | } | |
1da177e4 | 2525 | |
9f2c9d12 | 2526 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2527 | { |
e879a879 | 2528 | struct r10conf *conf = mddev->private; |
c65060ad | 2529 | int d; |
26208a7c YK |
2530 | struct bio *wbio = r10_bio->devs[1].bio; |
2531 | struct bio *wbio2 = r10_bio->devs[1].repl_bio; | |
2532 | ||
2533 | /* Need to test wbio2->bi_end_io before we call | |
2534 | * submit_bio_noacct as if the former is NULL, | |
2535 | * the latter is free to free wbio2. | |
2536 | */ | |
2537 | if (wbio2 && !wbio2->bi_end_io) | |
2538 | wbio2 = NULL; | |
1da177e4 | 2539 | |
5e570289 N |
2540 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2541 | fix_recovery_read_error(r10_bio); | |
26208a7c YK |
2542 | if (wbio->bi_end_io) |
2543 | end_sync_request(r10_bio); | |
2544 | if (wbio2) | |
2545 | end_sync_request(r10_bio); | |
5e570289 N |
2546 | return; |
2547 | } | |
2548 | ||
c65060ad NK |
2549 | /* |
2550 | * share the pages with the first bio | |
1da177e4 LT |
2551 | * and submit the write request |
2552 | */ | |
1da177e4 | 2553 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2554 | if (wbio->bi_end_io) { |
2555 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2556 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
ed00aabd | 2557 | submit_bio_noacct(wbio); |
24afd80d | 2558 | } |
0eb25bb0 | 2559 | if (wbio2) { |
24afd80d N |
2560 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2561 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2562 | bio_sectors(wbio2)); |
ed00aabd | 2563 | submit_bio_noacct(wbio2); |
24afd80d | 2564 | } |
1da177e4 LT |
2565 | } |
2566 | ||
3cb03002 | 2567 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
265ad47a | 2568 | int sectors, struct page *page, enum req_op op) |
58c54fcc | 2569 | { |
3a0f007b YK |
2570 | if (rdev_has_badblock(rdev, sector, sectors) && |
2571 | (op == REQ_OP_READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
58c54fcc | 2572 | return -1; |
265ad47a | 2573 | if (sync_page_io(rdev, sector, sectors << 9, page, op, false)) |
58c54fcc N |
2574 | /* success */ |
2575 | return 1; | |
265ad47a | 2576 | if (op == REQ_OP_WRITE) { |
58c54fcc | 2577 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2578 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2579 | set_bit(MD_RECOVERY_NEEDED, | |
2580 | &rdev->mddev->recovery); | |
2581 | } | |
58c54fcc N |
2582 | /* need to record an error - either for the block or the device */ |
2583 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2584 | md_error(rdev->mddev, rdev); | |
2585 | return 0; | |
2586 | } | |
2587 | ||
1da177e4 LT |
2588 | /* |
2589 | * This is a kernel thread which: | |
2590 | * | |
2591 | * 1. Retries failed read operations on working mirrors. | |
2592 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2593 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2594 | */ |
2595 | ||
e879a879 | 2596 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2597 | { |
2598 | int sect = 0; /* Offset from r10_bio->sector */ | |
605eeda6 | 2599 | int sectors = r10_bio->sectors, slot = r10_bio->read_slot; |
13db16d7 | 2600 | struct md_rdev *rdev; |
605eeda6 | 2601 | int d = r10_bio->devs[slot].devnum; |
1e50915f | 2602 | |
7c4e06ff N |
2603 | /* still own a reference to this rdev, so it cannot |
2604 | * have been cleared recently. | |
2605 | */ | |
2606 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2607 | |
7c4e06ff N |
2608 | if (test_bit(Faulty, &rdev->flags)) |
2609 | /* drive has already been failed, just ignore any | |
2610 | more fix_read_error() attempts */ | |
2611 | return; | |
1e50915f | 2612 | |
1979dbbe | 2613 | if (exceed_read_errors(mddev, rdev)) { |
605eeda6 | 2614 | r10_bio->devs[slot].bio = IO_BLOCKED; |
7c4e06ff | 2615 | return; |
1e50915f | 2616 | } |
1e50915f | 2617 | |
6814d536 N |
2618 | while(sectors) { |
2619 | int s = sectors; | |
605eeda6 | 2620 | int sl = slot; |
6814d536 N |
2621 | int success = 0; |
2622 | int start; | |
2623 | ||
2624 | if (s > (PAGE_SIZE>>9)) | |
2625 | s = PAGE_SIZE >> 9; | |
2626 | ||
6814d536 | 2627 | do { |
0544a21d | 2628 | d = r10_bio->devs[sl].devnum; |
a448af25 | 2629 | rdev = conf->mirrors[d].rdev; |
6814d536 | 2630 | if (rdev && |
8dbed5ce | 2631 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2632 | !test_bit(Faulty, &rdev->flags) && |
3a0f007b YK |
2633 | rdev_has_badblock(rdev, |
2634 | r10_bio->devs[sl].addr + sect, | |
2635 | s) == 0) { | |
6814d536 | 2636 | atomic_inc(&rdev->nr_pending); |
2b193363 | 2637 | success = sync_page_io(rdev, |
6814d536 | 2638 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2639 | sect, |
6814d536 | 2640 | s<<9, |
796a5cf0 | 2641 | conf->tmppage, |
4ce4c73f | 2642 | REQ_OP_READ, false); |
6814d536 | 2643 | rdev_dec_pending(rdev, mddev); |
6814d536 N |
2644 | if (success) |
2645 | break; | |
2646 | } | |
2647 | sl++; | |
2648 | if (sl == conf->copies) | |
2649 | sl = 0; | |
02c67a3b | 2650 | } while (sl != slot); |
6814d536 N |
2651 | |
2652 | if (!success) { | |
58c54fcc N |
2653 | /* Cannot read from anywhere, just mark the block |
2654 | * as bad on the first device to discourage future | |
2655 | * reads. | |
2656 | */ | |
605eeda6 | 2657 | int dn = r10_bio->devs[slot].devnum; |
58c54fcc N |
2658 | rdev = conf->mirrors[dn].rdev; |
2659 | ||
2660 | if (!rdev_set_badblocks( | |
2661 | rdev, | |
605eeda6 | 2662 | r10_bio->devs[slot].addr |
58c54fcc | 2663 | + sect, |
fae8cc5e | 2664 | s, 0)) { |
58c54fcc | 2665 | md_error(mddev, rdev); |
605eeda6 | 2666 | r10_bio->devs[slot].bio |
fae8cc5e N |
2667 | = IO_BLOCKED; |
2668 | } | |
6814d536 N |
2669 | break; |
2670 | } | |
2671 | ||
2672 | start = sl; | |
2673 | /* write it back and re-read */ | |
605eeda6 | 2674 | while (sl != slot) { |
6814d536 N |
2675 | if (sl==0) |
2676 | sl = conf->copies; | |
2677 | sl--; | |
2678 | d = r10_bio->devs[sl].devnum; | |
a448af25 | 2679 | rdev = conf->mirrors[d].rdev; |
1294b9c9 | 2680 | if (!rdev || |
f5b67ae8 | 2681 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2682 | !test_bit(In_sync, &rdev->flags)) |
2683 | continue; | |
2684 | ||
2685 | atomic_inc(&rdev->nr_pending); | |
58c54fcc N |
2686 | if (r10_sync_page_io(rdev, |
2687 | r10_bio->devs[sl].addr + | |
2688 | sect, | |
265ad47a | 2689 | s, conf->tmppage, REQ_OP_WRITE) |
1294b9c9 N |
2690 | == 0) { |
2691 | /* Well, this device is dead */ | |
913cce5a | 2692 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %pg)\n", |
08464e09 N |
2693 | mdname(mddev), s, |
2694 | (unsigned long long)( | |
2695 | sect + | |
2696 | choose_data_offset(r10_bio, | |
2697 | rdev)), | |
913cce5a CH |
2698 | rdev->bdev); |
2699 | pr_notice("md/raid10:%s: %pg: failing drive\n", | |
08464e09 | 2700 | mdname(mddev), |
913cce5a | 2701 | rdev->bdev); |
6814d536 | 2702 | } |
1294b9c9 | 2703 | rdev_dec_pending(rdev, mddev); |
6814d536 N |
2704 | } |
2705 | sl = start; | |
605eeda6 | 2706 | while (sl != slot) { |
6814d536 N |
2707 | if (sl==0) |
2708 | sl = conf->copies; | |
2709 | sl--; | |
2710 | d = r10_bio->devs[sl].devnum; | |
a448af25 | 2711 | rdev = conf->mirrors[d].rdev; |
1294b9c9 | 2712 | if (!rdev || |
f5b67ae8 | 2713 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2714 | !test_bit(In_sync, &rdev->flags)) |
2715 | continue; | |
6814d536 | 2716 | |
1294b9c9 | 2717 | atomic_inc(&rdev->nr_pending); |
58c54fcc N |
2718 | switch (r10_sync_page_io(rdev, |
2719 | r10_bio->devs[sl].addr + | |
2720 | sect, | |
265ad47a | 2721 | s, conf->tmppage, REQ_OP_READ)) { |
58c54fcc | 2722 | case 0: |
1294b9c9 | 2723 | /* Well, this device is dead */ |
913cce5a | 2724 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %pg)\n", |
1294b9c9 N |
2725 | mdname(mddev), s, |
2726 | (unsigned long long)( | |
f8c9e74f N |
2727 | sect + |
2728 | choose_data_offset(r10_bio, rdev)), | |
913cce5a CH |
2729 | rdev->bdev); |
2730 | pr_notice("md/raid10:%s: %pg: failing drive\n", | |
1294b9c9 | 2731 | mdname(mddev), |
913cce5a | 2732 | rdev->bdev); |
58c54fcc N |
2733 | break; |
2734 | case 1: | |
913cce5a | 2735 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %pg)\n", |
1294b9c9 N |
2736 | mdname(mddev), s, |
2737 | (unsigned long long)( | |
f8c9e74f N |
2738 | sect + |
2739 | choose_data_offset(r10_bio, rdev)), | |
913cce5a | 2740 | rdev->bdev); |
1294b9c9 | 2741 | atomic_add(s, &rdev->corrected_errors); |
6814d536 | 2742 | } |
1294b9c9 N |
2743 | |
2744 | rdev_dec_pending(rdev, mddev); | |
6814d536 | 2745 | } |
6814d536 N |
2746 | |
2747 | sectors -= s; | |
2748 | sect += s; | |
2749 | } | |
2750 | } | |
2751 | ||
9f2c9d12 | 2752 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2753 | { |
2754 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2755 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2756 | struct r10conf *conf = mddev->private; |
3cb03002 | 2757 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2758 | /* bio has the data to be written to slot 'i' where |
2759 | * we just recently had a write error. | |
2760 | * We repeatedly clone the bio and trim down to one block, | |
2761 | * then try the write. Where the write fails we record | |
2762 | * a bad block. | |
2763 | * It is conceivable that the bio doesn't exactly align with | |
2764 | * blocks. We must handle this. | |
2765 | * | |
2766 | * We currently own a reference to the rdev. | |
2767 | */ | |
2768 | ||
2769 | int block_sectors; | |
2770 | sector_t sector; | |
2771 | int sectors; | |
2772 | int sect_to_write = r10_bio->sectors; | |
2773 | int ok = 1; | |
2774 | ||
2775 | if (rdev->badblocks.shift < 0) | |
2776 | return 0; | |
2777 | ||
f04ebb0b N |
2778 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2779 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2780 | sector = r10_bio->sector; |
2781 | sectors = ((r10_bio->sector + block_sectors) | |
2782 | & ~(sector_t)(block_sectors - 1)) | |
2783 | - sector; | |
2784 | ||
2785 | while (sect_to_write) { | |
2786 | struct bio *wbio; | |
27028626 | 2787 | sector_t wsector; |
bd870a16 N |
2788 | if (sectors > sect_to_write) |
2789 | sectors = sect_to_write; | |
2790 | /* Write at 'sector' for 'sectors' */ | |
abfc426d CH |
2791 | wbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, |
2792 | &mddev->bio_set); | |
4f024f37 | 2793 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2794 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2795 | wbio->bi_iter.bi_sector = wsector + | |
2796 | choose_data_offset(r10_bio, rdev); | |
c34b7ac6 | 2797 | wbio->bi_opf = REQ_OP_WRITE; |
4e49ea4a MC |
2798 | |
2799 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2800 | /* Failure! */ |
27028626 | 2801 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2802 | sectors, 0) |
2803 | && ok; | |
2804 | ||
2805 | bio_put(wbio); | |
2806 | sect_to_write -= sectors; | |
2807 | sector += sectors; | |
2808 | sectors = block_sectors; | |
2809 | } | |
2810 | return ok; | |
2811 | } | |
2812 | ||
9f2c9d12 | 2813 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2814 | { |
2815 | int slot = r10_bio->read_slot; | |
560f8e55 | 2816 | struct bio *bio; |
e879a879 | 2817 | struct r10conf *conf = mddev->private; |
abbf098e | 2818 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2819 | |
2820 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2821 | * the block and we can fix it. | |
2822 | * We freeze all other IO, and try reading the block from | |
2823 | * other devices. When we find one, we re-write | |
2824 | * and check it that fixes the read error. | |
2825 | * This is all done synchronously while the array is | |
2826 | * frozen. | |
2827 | */ | |
fae8cc5e | 2828 | bio = r10_bio->devs[slot].bio; |
fae8cc5e N |
2829 | bio_put(bio); |
2830 | r10_bio->devs[slot].bio = NULL; | |
2831 | ||
8d3ca83d N |
2832 | if (mddev->ro) |
2833 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2834 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2835 | freeze_array(conf, 1); |
560f8e55 N |
2836 | fix_read_error(conf, mddev, r10_bio); |
2837 | unfreeze_array(conf); | |
fae8cc5e | 2838 | } else |
8d3ca83d | 2839 | md_error(mddev, rdev); |
fae8cc5e | 2840 | |
abbf098e | 2841 | rdev_dec_pending(rdev, mddev); |
545250f2 | 2842 | r10_bio->state = 0; |
82045523 | 2843 | raid10_read_request(mddev, r10_bio->master_bio, r10_bio, false); |
72c215ed LN |
2844 | /* |
2845 | * allow_barrier after re-submit to ensure no sync io | |
2846 | * can be issued while regular io pending. | |
2847 | */ | |
2848 | allow_barrier(conf); | |
560f8e55 N |
2849 | } |
2850 | ||
e879a879 | 2851 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2852 | { |
2853 | /* Some sort of write request has finished and it | |
2854 | * succeeded in writing where we thought there was a | |
2855 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2856 | * Or possibly if failed and we need to record |
2857 | * a bad block. | |
749c55e9 N |
2858 | */ |
2859 | int m; | |
3cb03002 | 2860 | struct md_rdev *rdev; |
749c55e9 N |
2861 | |
2862 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2863 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2864 | for (m = 0; m < conf->copies; m++) { |
2865 | int dev = r10_bio->devs[m].devnum; | |
2866 | rdev = conf->mirrors[dev].rdev; | |
01a69cab YY |
2867 | if (r10_bio->devs[m].bio == NULL || |
2868 | r10_bio->devs[m].bio->bi_end_io == NULL) | |
1a0b7cd8 | 2869 | continue; |
4e4cbee9 | 2870 | if (!r10_bio->devs[m].bio->bi_status) { |
749c55e9 N |
2871 | rdev_clear_badblocks( |
2872 | rdev, | |
2873 | r10_bio->devs[m].addr, | |
c6563a8c | 2874 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2875 | } else { |
2876 | if (!rdev_set_badblocks( | |
2877 | rdev, | |
2878 | r10_bio->devs[m].addr, | |
2879 | r10_bio->sectors, 0)) | |
2880 | md_error(conf->mddev, rdev); | |
749c55e9 | 2881 | } |
9ad1aefc | 2882 | rdev = conf->mirrors[dev].replacement; |
01a69cab YY |
2883 | if (r10_bio->devs[m].repl_bio == NULL || |
2884 | r10_bio->devs[m].repl_bio->bi_end_io == NULL) | |
9ad1aefc | 2885 | continue; |
4246a0b6 | 2886 | |
4e4cbee9 | 2887 | if (!r10_bio->devs[m].repl_bio->bi_status) { |
9ad1aefc N |
2888 | rdev_clear_badblocks( |
2889 | rdev, | |
2890 | r10_bio->devs[m].addr, | |
c6563a8c | 2891 | r10_bio->sectors, 0); |
9ad1aefc N |
2892 | } else { |
2893 | if (!rdev_set_badblocks( | |
2894 | rdev, | |
2895 | r10_bio->devs[m].addr, | |
2896 | r10_bio->sectors, 0)) | |
2897 | md_error(conf->mddev, rdev); | |
2898 | } | |
1a0b7cd8 | 2899 | } |
749c55e9 N |
2900 | put_buf(r10_bio); |
2901 | } else { | |
95af587e | 2902 | bool fail = false; |
bd870a16 N |
2903 | for (m = 0; m < conf->copies; m++) { |
2904 | int dev = r10_bio->devs[m].devnum; | |
2905 | struct bio *bio = r10_bio->devs[m].bio; | |
2906 | rdev = conf->mirrors[dev].rdev; | |
2907 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2908 | rdev_clear_badblocks( |
2909 | rdev, | |
2910 | r10_bio->devs[m].addr, | |
c6563a8c | 2911 | r10_bio->sectors, 0); |
749c55e9 | 2912 | rdev_dec_pending(rdev, conf->mddev); |
4e4cbee9 | 2913 | } else if (bio != NULL && bio->bi_status) { |
95af587e | 2914 | fail = true; |
bd870a16 N |
2915 | if (!narrow_write_error(r10_bio, m)) { |
2916 | md_error(conf->mddev, rdev); | |
2917 | set_bit(R10BIO_Degraded, | |
2918 | &r10_bio->state); | |
2919 | } | |
2920 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2921 | } |
475b0321 N |
2922 | bio = r10_bio->devs[m].repl_bio; |
2923 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2924 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2925 | rdev_clear_badblocks( |
2926 | rdev, | |
2927 | r10_bio->devs[m].addr, | |
c6563a8c | 2928 | r10_bio->sectors, 0); |
475b0321 N |
2929 | rdev_dec_pending(rdev, conf->mddev); |
2930 | } | |
bd870a16 | 2931 | } |
95af587e N |
2932 | if (fail) { |
2933 | spin_lock_irq(&conf->device_lock); | |
2934 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2935 | conf->nr_queued++; |
95af587e | 2936 | spin_unlock_irq(&conf->device_lock); |
cf25ae78 GJ |
2937 | /* |
2938 | * In case freeze_array() is waiting for condition | |
2939 | * nr_pending == nr_queued + extra to be true. | |
2940 | */ | |
2941 | wake_up(&conf->wait_barrier); | |
95af587e | 2942 | md_wakeup_thread(conf->mddev->thread); |
c340702c N |
2943 | } else { |
2944 | if (test_bit(R10BIO_WriteError, | |
2945 | &r10_bio->state)) | |
2946 | close_write(r10_bio); | |
95af587e | 2947 | raid_end_bio_io(r10_bio); |
c340702c | 2948 | } |
749c55e9 N |
2949 | } |
2950 | } | |
2951 | ||
4ed8731d | 2952 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2953 | { |
4ed8731d | 2954 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2955 | struct r10bio *r10_bio; |
1da177e4 | 2956 | unsigned long flags; |
e879a879 | 2957 | struct r10conf *conf = mddev->private; |
1da177e4 | 2958 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2959 | struct blk_plug plug; |
1da177e4 LT |
2960 | |
2961 | md_check_recovery(mddev); | |
1da177e4 | 2962 | |
95af587e | 2963 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2964 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
2965 | LIST_HEAD(tmp); |
2966 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2967 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
2968 | while (!list_empty(&conf->bio_end_io_list)) { |
2969 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2970 | conf->nr_queued--; | |
2971 | } | |
95af587e N |
2972 | } |
2973 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2974 | while (!list_empty(&tmp)) { | |
a452744b MP |
2975 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2976 | retry_list); | |
95af587e | 2977 | list_del(&r10_bio->retry_list); |
c340702c N |
2978 | if (mddev->degraded) |
2979 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2980 | ||
2981 | if (test_bit(R10BIO_WriteError, | |
2982 | &r10_bio->state)) | |
2983 | close_write(r10_bio); | |
95af587e N |
2984 | raid_end_bio_io(r10_bio); |
2985 | } | |
2986 | } | |
2987 | ||
e1dfa0a2 | 2988 | blk_start_plug(&plug); |
1da177e4 | 2989 | for (;;) { |
6cce3b23 | 2990 | |
0021b7bc | 2991 | flush_pending_writes(conf); |
6cce3b23 | 2992 | |
a35e63ef N |
2993 | spin_lock_irqsave(&conf->device_lock, flags); |
2994 | if (list_empty(head)) { | |
2995 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2996 | break; |
a35e63ef | 2997 | } |
9f2c9d12 | 2998 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2999 | list_del(head->prev); |
4443ae10 | 3000 | conf->nr_queued--; |
1da177e4 LT |
3001 | spin_unlock_irqrestore(&conf->device_lock, flags); |
3002 | ||
3003 | mddev = r10_bio->mddev; | |
070ec55d | 3004 | conf = mddev->private; |
bd870a16 N |
3005 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
3006 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 3007 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
3008 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
3009 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 3010 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 3011 | sync_request_write(mddev, r10_bio); |
7eaceacc | 3012 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 3013 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 3014 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 3015 | handle_read_error(mddev, r10_bio); |
fc9977dd N |
3016 | else |
3017 | WARN_ON_ONCE(1); | |
560f8e55 | 3018 | |
1d9d5241 | 3019 | cond_resched(); |
2953079c | 3020 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 3021 | md_check_recovery(mddev); |
1da177e4 | 3022 | } |
e1dfa0a2 | 3023 | blk_finish_plug(&plug); |
1da177e4 LT |
3024 | } |
3025 | ||
e879a879 | 3026 | static int init_resync(struct r10conf *conf) |
1da177e4 | 3027 | { |
afeee514 | 3028 | int ret, buffs, i; |
1da177e4 LT |
3029 | |
3030 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
afeee514 | 3031 | BUG_ON(mempool_initialized(&conf->r10buf_pool)); |
69335ef3 | 3032 | conf->have_replacement = 0; |
5cf00fcd | 3033 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
3034 | if (conf->mirrors[i].replacement) |
3035 | conf->have_replacement = 1; | |
afeee514 KO |
3036 | ret = mempool_init(&conf->r10buf_pool, buffs, |
3037 | r10buf_pool_alloc, r10buf_pool_free, conf); | |
3038 | if (ret) | |
3039 | return ret; | |
1da177e4 LT |
3040 | conf->next_resync = 0; |
3041 | return 0; | |
3042 | } | |
3043 | ||
208410b5 SL |
3044 | static struct r10bio *raid10_alloc_init_r10buf(struct r10conf *conf) |
3045 | { | |
afeee514 | 3046 | struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); |
208410b5 SL |
3047 | struct rsync_pages *rp; |
3048 | struct bio *bio; | |
3049 | int nalloc; | |
3050 | int i; | |
3051 | ||
3052 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || | |
3053 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
3054 | nalloc = conf->copies; /* resync */ | |
3055 | else | |
3056 | nalloc = 2; /* recovery */ | |
3057 | ||
3058 | for (i = 0; i < nalloc; i++) { | |
3059 | bio = r10bio->devs[i].bio; | |
3060 | rp = bio->bi_private; | |
a7c50c94 | 3061 | bio_reset(bio, NULL, 0); |
208410b5 SL |
3062 | bio->bi_private = rp; |
3063 | bio = r10bio->devs[i].repl_bio; | |
3064 | if (bio) { | |
3065 | rp = bio->bi_private; | |
a7c50c94 | 3066 | bio_reset(bio, NULL, 0); |
208410b5 SL |
3067 | bio->bi_private = rp; |
3068 | } | |
3069 | } | |
3070 | return r10bio; | |
3071 | } | |
3072 | ||
8db87912 GJ |
3073 | /* |
3074 | * Set cluster_sync_high since we need other nodes to add the | |
3075 | * range [cluster_sync_low, cluster_sync_high] to suspend list. | |
3076 | */ | |
3077 | static void raid10_set_cluster_sync_high(struct r10conf *conf) | |
3078 | { | |
3079 | sector_t window_size; | |
3080 | int extra_chunk, chunks; | |
3081 | ||
3082 | /* | |
3083 | * First, here we define "stripe" as a unit which across | |
3084 | * all member devices one time, so we get chunks by use | |
3085 | * raid_disks / near_copies. Otherwise, if near_copies is | |
3086 | * close to raid_disks, then resync window could increases | |
3087 | * linearly with the increase of raid_disks, which means | |
3088 | * we will suspend a really large IO window while it is not | |
3089 | * necessary. If raid_disks is not divisible by near_copies, | |
3090 | * an extra chunk is needed to ensure the whole "stripe" is | |
3091 | * covered. | |
3092 | */ | |
3093 | ||
3094 | chunks = conf->geo.raid_disks / conf->geo.near_copies; | |
3095 | if (conf->geo.raid_disks % conf->geo.near_copies == 0) | |
3096 | extra_chunk = 0; | |
3097 | else | |
3098 | extra_chunk = 1; | |
3099 | window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; | |
3100 | ||
3101 | /* | |
3102 | * At least use a 32M window to align with raid1's resync window | |
3103 | */ | |
3104 | window_size = (CLUSTER_RESYNC_WINDOW_SECTORS > window_size) ? | |
3105 | CLUSTER_RESYNC_WINDOW_SECTORS : window_size; | |
3106 | ||
3107 | conf->cluster_sync_high = conf->cluster_sync_low + window_size; | |
3108 | } | |
3109 | ||
1da177e4 LT |
3110 | /* |
3111 | * perform a "sync" on one "block" | |
3112 | * | |
3113 | * We need to make sure that no normal I/O request - particularly write | |
3114 | * requests - conflict with active sync requests. | |
3115 | * | |
3116 | * This is achieved by tracking pending requests and a 'barrier' concept | |
3117 | * that can be installed to exclude normal IO requests. | |
3118 | * | |
3119 | * Resync and recovery are handled very differently. | |
3120 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
3121 | * | |
3122 | * For resync, we iterate over virtual addresses, read all copies, | |
3123 | * and update if there are differences. If only one copy is live, | |
3124 | * skip it. | |
3125 | * For recovery, we iterate over physical addresses, read a good | |
3126 | * value for each non-in_sync drive, and over-write. | |
3127 | * | |
3128 | * So, for recovery we may have several outstanding complex requests for a | |
3129 | * given address, one for each out-of-sync device. We model this by allocating | |
3130 | * a number of r10_bio structures, one for each out-of-sync device. | |
3131 | * As we setup these structures, we collect all bio's together into a list | |
3132 | * which we then process collectively to add pages, and then process again | |
ed00aabd | 3133 | * to pass to submit_bio_noacct. |
1da177e4 LT |
3134 | * |
3135 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
3136 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
3137 | * has its remaining count decremented to 0, the whole complex operation | |
3138 | * is complete. | |
3139 | * | |
3140 | */ | |
3141 | ||
849674e4 | 3142 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 3143 | int *skipped) |
1da177e4 | 3144 | { |
e879a879 | 3145 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 3146 | struct r10bio *r10_bio; |
1da177e4 LT |
3147 | struct bio *biolist = NULL, *bio; |
3148 | sector_t max_sector, nr_sectors; | |
1da177e4 | 3149 | int i; |
6cce3b23 | 3150 | int max_sync; |
57dab0bd | 3151 | sector_t sync_blocks; |
1da177e4 LT |
3152 | sector_t sectors_skipped = 0; |
3153 | int chunks_skipped = 0; | |
5cf00fcd | 3154 | sector_t chunk_mask = conf->geo.chunk_mask; |
022e510f | 3155 | int page_idx = 0; |
8d355a46 | 3156 | int error_disk = -1; |
1da177e4 | 3157 | |
7e83ccbe MW |
3158 | /* |
3159 | * Allow skipping a full rebuild for incremental assembly | |
3160 | * of a clean array, like RAID1 does. | |
3161 | */ | |
3162 | if (mddev->bitmap == NULL && | |
3163 | mddev->recovery_cp == MaxSector && | |
13765120 N |
3164 | mddev->reshape_position == MaxSector && |
3165 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 3166 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 3167 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
3168 | conf->fullsync == 0) { |
3169 | *skipped = 1; | |
13765120 | 3170 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
3171 | } |
3172 | ||
a405c6f0 LN |
3173 | if (!mempool_initialized(&conf->r10buf_pool)) |
3174 | if (init_resync(conf)) | |
3175 | return 0; | |
3176 | ||
1da177e4 | 3177 | skipped: |
58c0fed4 | 3178 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
3179 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
3180 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
3181 | max_sector = mddev->resync_max_sectors; |
3182 | if (sector_nr >= max_sector) { | |
8db87912 GJ |
3183 | conf->cluster_sync_low = 0; |
3184 | conf->cluster_sync_high = 0; | |
3185 | ||
6cce3b23 N |
3186 | /* If we aborted, we need to abort the |
3187 | * sync on the 'current' bitmap chucks (there can | |
3188 | * be several when recovering multiple devices). | |
3189 | * as we may have started syncing it but not finished. | |
3190 | * We can find the current address in | |
3191 | * mddev->curr_resync, but for recovery, | |
3192 | * we need to convert that to several | |
3193 | * virtual addresses. | |
3194 | */ | |
3ea7daa5 N |
3195 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
3196 | end_reshape(conf); | |
b3968552 | 3197 | close_sync(conf); |
3ea7daa5 N |
3198 | return 0; |
3199 | } | |
3200 | ||
6cce3b23 N |
3201 | if (mddev->curr_resync < max_sector) { /* aborted */ |
3202 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
e64e4018 AS |
3203 | md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, |
3204 | &sync_blocks, 1); | |
5cf00fcd | 3205 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
3206 | sector_t sect = |
3207 | raid10_find_virt(conf, mddev->curr_resync, i); | |
e64e4018 AS |
3208 | md_bitmap_end_sync(mddev->bitmap, sect, |
3209 | &sync_blocks, 1); | |
6cce3b23 | 3210 | } |
9ad1aefc N |
3211 | } else { |
3212 | /* completed sync */ | |
3213 | if ((!mddev->bitmap || conf->fullsync) | |
3214 | && conf->have_replacement | |
3215 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3216 | /* Completed a full sync so the replacements | |
3217 | * are now fully recovered. | |
3218 | */ | |
f90145f3 N |
3219 | for (i = 0; i < conf->geo.raid_disks; i++) { |
3220 | struct md_rdev *rdev = | |
a448af25 YK |
3221 | conf->mirrors[i].replacement; |
3222 | ||
f90145f3 N |
3223 | if (rdev) |
3224 | rdev->recovery_offset = MaxSector; | |
3225 | } | |
9ad1aefc | 3226 | } |
6cce3b23 | 3227 | conf->fullsync = 0; |
9ad1aefc | 3228 | } |
e64e4018 | 3229 | md_bitmap_close_sync(mddev->bitmap); |
1da177e4 | 3230 | close_sync(conf); |
57afd89f | 3231 | *skipped = 1; |
1da177e4 LT |
3232 | return sectors_skipped; |
3233 | } | |
3ea7daa5 N |
3234 | |
3235 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
3236 | return reshape_request(mddev, sector_nr, skipped); | |
3237 | ||
5cf00fcd | 3238 | if (chunks_skipped >= conf->geo.raid_disks) { |
8d355a46 LN |
3239 | pr_err("md/raid10:%s: %s fails\n", mdname(mddev), |
3240 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? "resync" : "recovery"); | |
3241 | if (error_disk >= 0 && | |
3242 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3243 | /* | |
3244 | * recovery fails, set mirrors.recovery_disabled, | |
3245 | * device shouldn't be added to there. | |
3246 | */ | |
3247 | conf->mirrors[error_disk].recovery_disabled = | |
3248 | mddev->recovery_disabled; | |
3249 | return 0; | |
3250 | } | |
3251 | /* | |
3252 | * if there has been nothing to do on any drive, | |
3253 | * then there is nothing to do at all. | |
1da177e4 | 3254 | */ |
57afd89f N |
3255 | *skipped = 1; |
3256 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
3257 | } |
3258 | ||
c6207277 N |
3259 | if (max_sector > mddev->resync_max) |
3260 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
3261 | ||
1da177e4 LT |
3262 | /* make sure whole request will fit in a chunk - if chunks |
3263 | * are meaningful | |
3264 | */ | |
5cf00fcd N |
3265 | if (conf->geo.near_copies < conf->geo.raid_disks && |
3266 | max_sector > (sector_nr | chunk_mask)) | |
3267 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 3268 | |
7ac50447 TM |
3269 | /* |
3270 | * If there is non-resync activity waiting for a turn, then let it | |
3271 | * though before starting on this new sync request. | |
3272 | */ | |
3273 | if (conf->nr_waiting) | |
3274 | schedule_timeout_uninterruptible(1); | |
3275 | ||
1da177e4 LT |
3276 | /* Again, very different code for resync and recovery. |
3277 | * Both must result in an r10bio with a list of bios that | |
309dca30 | 3278 | * have bi_end_io, bi_sector, bi_bdev set, |
1da177e4 LT |
3279 | * and bi_private set to the r10bio. |
3280 | * For recovery, we may actually create several r10bios | |
3281 | * with 2 bios in each, that correspond to the bios in the main one. | |
3282 | * In this case, the subordinate r10bios link back through a | |
3283 | * borrowed master_bio pointer, and the counter in the master | |
3284 | * includes a ref from each subordinate. | |
3285 | */ | |
3286 | /* First, we decide what to do and set ->bi_end_io | |
3287 | * To end_sync_read if we want to read, and | |
3288 | * end_sync_write if we will want to write. | |
3289 | */ | |
3290 | ||
6cce3b23 | 3291 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
3292 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
3293 | /* recovery... the complicated one */ | |
e875ecea | 3294 | int j; |
1da177e4 LT |
3295 | r10_bio = NULL; |
3296 | ||
5cf00fcd | 3297 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 3298 | int still_degraded; |
9f2c9d12 | 3299 | struct r10bio *rb2; |
ab9d47e9 N |
3300 | sector_t sect; |
3301 | int must_sync; | |
e875ecea | 3302 | int any_working; |
dc280d98 | 3303 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3304 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3305 | |
a448af25 YK |
3306 | mrdev = mirror->rdev; |
3307 | mreplace = mirror->replacement; | |
f90145f3 | 3308 | |
59f8f0b5 LN |
3309 | if (mrdev && (test_bit(Faulty, &mrdev->flags) || |
3310 | test_bit(In_sync, &mrdev->flags))) | |
3311 | mrdev = NULL; | |
34817a24 LN |
3312 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3313 | mreplace = NULL; | |
ee37d731 | 3314 | |
a448af25 | 3315 | if (!mrdev && !mreplace) |
ab9d47e9 | 3316 | continue; |
1da177e4 | 3317 | |
ab9d47e9 N |
3318 | still_degraded = 0; |
3319 | /* want to reconstruct this device */ | |
3320 | rb2 = r10_bio; | |
3321 | sect = raid10_find_virt(conf, sector_nr, i); | |
a448af25 | 3322 | if (sect >= mddev->resync_max_sectors) |
fc448a18 N |
3323 | /* last stripe is not complete - don't |
3324 | * try to recover this sector. | |
3325 | */ | |
3326 | continue; | |
24afd80d N |
3327 | /* Unless we are doing a full sync, or a replacement |
3328 | * we only need to recover the block if it is set in | |
3329 | * the bitmap | |
ab9d47e9 | 3330 | */ |
e64e4018 AS |
3331 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3332 | &sync_blocks, 1); | |
ab9d47e9 N |
3333 | if (sync_blocks < max_sync) |
3334 | max_sync = sync_blocks; | |
3335 | if (!must_sync && | |
f90145f3 | 3336 | mreplace == NULL && |
ab9d47e9 N |
3337 | !conf->fullsync) { |
3338 | /* yep, skip the sync_blocks here, but don't assume | |
3339 | * that there will never be anything to do here | |
3340 | */ | |
3341 | chunks_skipped = -1; | |
3342 | continue; | |
3343 | } | |
59f8f0b5 LN |
3344 | if (mrdev) |
3345 | atomic_inc(&mrdev->nr_pending); | |
f90145f3 N |
3346 | if (mreplace) |
3347 | atomic_inc(&mreplace->nr_pending); | |
6cce3b23 | 3348 | |
208410b5 | 3349 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3350 | r10_bio->state = 0; |
ab9d47e9 N |
3351 | raise_barrier(conf, rb2 != NULL); |
3352 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3353 | |
ab9d47e9 N |
3354 | r10_bio->master_bio = (struct bio*)rb2; |
3355 | if (rb2) | |
3356 | atomic_inc(&rb2->remaining); | |
3357 | r10_bio->mddev = mddev; | |
3358 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3359 | r10_bio->sector = sect; | |
1da177e4 | 3360 | |
ab9d47e9 N |
3361 | raid10_find_phys(conf, r10_bio); |
3362 | ||
3363 | /* Need to check if the array will still be | |
3364 | * degraded | |
3365 | */ | |
f90145f3 | 3366 | for (j = 0; j < conf->geo.raid_disks; j++) { |
a448af25 YK |
3367 | struct md_rdev *rdev = conf->mirrors[j].rdev; |
3368 | ||
f90145f3 | 3369 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { |
ab9d47e9 | 3370 | still_degraded = 1; |
87fc767b | 3371 | break; |
1da177e4 | 3372 | } |
f90145f3 | 3373 | } |
ab9d47e9 | 3374 | |
e64e4018 AS |
3375 | must_sync = md_bitmap_start_sync(mddev->bitmap, sect, |
3376 | &sync_blocks, still_degraded); | |
ab9d47e9 | 3377 | |
e875ecea | 3378 | any_working = 0; |
ab9d47e9 | 3379 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3380 | int k; |
ab9d47e9 | 3381 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3382 | sector_t from_addr, to_addr; |
a448af25 | 3383 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
40c356ce N |
3384 | sector_t sector, first_bad; |
3385 | int bad_sectors; | |
f90145f3 N |
3386 | if (!rdev || |
3387 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3388 | continue; |
3389 | /* This is where we read from */ | |
e875ecea | 3390 | any_working = 1; |
40c356ce N |
3391 | sector = r10_bio->devs[j].addr; |
3392 | ||
3393 | if (is_badblock(rdev, sector, max_sync, | |
3394 | &first_bad, &bad_sectors)) { | |
3395 | if (first_bad > sector) | |
3396 | max_sync = first_bad - sector; | |
3397 | else { | |
3398 | bad_sectors -= (sector | |
3399 | - first_bad); | |
3400 | if (max_sync > bad_sectors) | |
3401 | max_sync = bad_sectors; | |
3402 | continue; | |
3403 | } | |
3404 | } | |
ab9d47e9 N |
3405 | bio = r10_bio->devs[0].bio; |
3406 | bio->bi_next = biolist; | |
3407 | biolist = bio; | |
ab9d47e9 | 3408 | bio->bi_end_io = end_sync_read; |
c34b7ac6 | 3409 | bio->bi_opf = REQ_OP_READ; |
8d3ca83d N |
3410 | if (test_bit(FailFast, &rdev->flags)) |
3411 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3412 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3413 | bio->bi_iter.bi_sector = from_addr + |
3414 | rdev->data_offset; | |
74d46992 | 3415 | bio_set_dev(bio, rdev->bdev); |
24afd80d N |
3416 | atomic_inc(&rdev->nr_pending); |
3417 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3418 | |
3419 | for (k=0; k<conf->copies; k++) | |
3420 | if (r10_bio->devs[k].devnum == i) | |
3421 | break; | |
3422 | BUG_ON(k == conf->copies); | |
5e570289 | 3423 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3424 | r10_bio->devs[0].devnum = d; |
5e570289 | 3425 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3426 | r10_bio->devs[1].devnum = i; |
5e570289 | 3427 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3428 | |
59f8f0b5 | 3429 | if (mrdev) { |
24afd80d N |
3430 | bio = r10_bio->devs[1].bio; |
3431 | bio->bi_next = biolist; | |
3432 | biolist = bio; | |
24afd80d | 3433 | bio->bi_end_io = end_sync_write; |
c34b7ac6 | 3434 | bio->bi_opf = REQ_OP_WRITE; |
4f024f37 | 3435 | bio->bi_iter.bi_sector = to_addr |
f90145f3 | 3436 | + mrdev->data_offset; |
74d46992 | 3437 | bio_set_dev(bio, mrdev->bdev); |
24afd80d N |
3438 | atomic_inc(&r10_bio->remaining); |
3439 | } else | |
3440 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3441 | ||
3442 | /* and maybe write to replacement */ | |
3443 | bio = r10_bio->devs[1].repl_bio; | |
3444 | if (bio) | |
3445 | bio->bi_end_io = NULL; | |
34817a24 | 3446 | /* Note: if replace is not NULL, then bio |
24afd80d N |
3447 | * cannot be NULL as r10buf_pool_alloc will |
3448 | * have allocated it. | |
24afd80d | 3449 | */ |
34817a24 | 3450 | if (!mreplace) |
24afd80d N |
3451 | break; |
3452 | bio->bi_next = biolist; | |
3453 | biolist = bio; | |
24afd80d | 3454 | bio->bi_end_io = end_sync_write; |
c34b7ac6 | 3455 | bio->bi_opf = REQ_OP_WRITE; |
4f024f37 | 3456 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 | 3457 | mreplace->data_offset; |
74d46992 | 3458 | bio_set_dev(bio, mreplace->bdev); |
24afd80d | 3459 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3460 | break; |
3461 | } | |
3462 | if (j == conf->copies) { | |
e875ecea N |
3463 | /* Cannot recover, so abort the recovery or |
3464 | * record a bad block */ | |
e875ecea N |
3465 | if (any_working) { |
3466 | /* problem is that there are bad blocks | |
3467 | * on other device(s) | |
3468 | */ | |
3469 | int k; | |
3470 | for (k = 0; k < conf->copies; k++) | |
3471 | if (r10_bio->devs[k].devnum == i) | |
3472 | break; | |
59f8f0b5 | 3473 | if (mrdev && !test_bit(In_sync, |
f90145f3 | 3474 | &mrdev->flags) |
24afd80d | 3475 | && !rdev_set_badblocks( |
f90145f3 | 3476 | mrdev, |
24afd80d N |
3477 | r10_bio->devs[k].addr, |
3478 | max_sync, 0)) | |
3479 | any_working = 0; | |
f90145f3 | 3480 | if (mreplace && |
24afd80d | 3481 | !rdev_set_badblocks( |
f90145f3 | 3482 | mreplace, |
e875ecea N |
3483 | r10_bio->devs[k].addr, |
3484 | max_sync, 0)) | |
3485 | any_working = 0; | |
3486 | } | |
3487 | if (!any_working) { | |
3488 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3489 | &mddev->recovery)) | |
08464e09 | 3490 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3491 | mdname(mddev)); |
24afd80d | 3492 | mirror->recovery_disabled |
e875ecea | 3493 | = mddev->recovery_disabled; |
8d355a46 LN |
3494 | } else { |
3495 | error_disk = i; | |
e875ecea | 3496 | } |
e8b84915 N |
3497 | put_buf(r10_bio); |
3498 | if (rb2) | |
3499 | atomic_dec(&rb2->remaining); | |
3500 | r10_bio = rb2; | |
59f8f0b5 LN |
3501 | if (mrdev) |
3502 | rdev_dec_pending(mrdev, mddev); | |
f90145f3 N |
3503 | if (mreplace) |
3504 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3505 | break; |
1da177e4 | 3506 | } |
59f8f0b5 LN |
3507 | if (mrdev) |
3508 | rdev_dec_pending(mrdev, mddev); | |
f90145f3 N |
3509 | if (mreplace) |
3510 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3511 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3512 | /* Only want this if there is elsewhere to | |
3513 | * read from. 'j' is currently the first | |
3514 | * readable copy. | |
3515 | */ | |
3516 | int targets = 1; | |
3517 | for (; j < conf->copies; j++) { | |
3518 | int d = r10_bio->devs[j].devnum; | |
3519 | if (conf->mirrors[d].rdev && | |
3520 | test_bit(In_sync, | |
3521 | &conf->mirrors[d].rdev->flags)) | |
3522 | targets++; | |
3523 | } | |
3524 | if (targets == 1) | |
3525 | r10_bio->devs[0].bio->bi_opf | |
3526 | &= ~MD_FAILFAST; | |
3527 | } | |
ab9d47e9 | 3528 | } |
1da177e4 LT |
3529 | if (biolist == NULL) { |
3530 | while (r10_bio) { | |
9f2c9d12 N |
3531 | struct r10bio *rb2 = r10_bio; |
3532 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3533 | rb2->master_bio = NULL; |
3534 | put_buf(rb2); | |
3535 | } | |
3536 | goto giveup; | |
3537 | } | |
3538 | } else { | |
3539 | /* resync. Schedule a read for every block at this virt offset */ | |
3540 | int count = 0; | |
6cce3b23 | 3541 | |
8db87912 GJ |
3542 | /* |
3543 | * Since curr_resync_completed could probably not update in | |
3544 | * time, and we will set cluster_sync_low based on it. | |
3545 | * Let's check against "sector_nr + 2 * RESYNC_SECTORS" for | |
3546 | * safety reason, which ensures curr_resync_completed is | |
3547 | * updated in bitmap_cond_end_sync. | |
3548 | */ | |
e64e4018 AS |
3549 | md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, |
3550 | mddev_is_clustered(mddev) && | |
3551 | (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); | |
78200d45 | 3552 | |
e64e4018 AS |
3553 | if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, |
3554 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3555 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3556 | &mddev->recovery)) { | |
6cce3b23 N |
3557 | /* We can skip this block */ |
3558 | *skipped = 1; | |
3559 | return sync_blocks + sectors_skipped; | |
3560 | } | |
3561 | if (sync_blocks < max_sync) | |
3562 | max_sync = sync_blocks; | |
208410b5 | 3563 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 3564 | r10_bio->state = 0; |
1da177e4 | 3565 | |
1da177e4 LT |
3566 | r10_bio->mddev = mddev; |
3567 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3568 | raise_barrier(conf, 0); |
3569 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3570 | |
3571 | r10_bio->master_bio = NULL; | |
3572 | r10_bio->sector = sector_nr; | |
3573 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3574 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3575 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3576 | |
5cf00fcd | 3577 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3578 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3579 | sector_t first_bad, sector; |
3580 | int bad_sectors; | |
f90145f3 | 3581 | struct md_rdev *rdev; |
40c356ce | 3582 | |
9ad1aefc N |
3583 | if (r10_bio->devs[i].repl_bio) |
3584 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3585 | ||
1da177e4 | 3586 | bio = r10_bio->devs[i].bio; |
4e4cbee9 | 3587 | bio->bi_status = BLK_STS_IOERR; |
a448af25 YK |
3588 | rdev = conf->mirrors[d].rdev; |
3589 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
1da177e4 | 3590 | continue; |
a448af25 | 3591 | |
40c356ce | 3592 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3593 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3594 | &first_bad, &bad_sectors)) { |
3595 | if (first_bad > sector) | |
3596 | max_sync = first_bad - sector; | |
3597 | else { | |
3598 | bad_sectors -= (sector - first_bad); | |
3599 | if (max_sync > bad_sectors) | |
91502f09 | 3600 | max_sync = bad_sectors; |
40c356ce N |
3601 | continue; |
3602 | } | |
3603 | } | |
f90145f3 | 3604 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3605 | atomic_inc(&r10_bio->remaining); |
3606 | bio->bi_next = biolist; | |
3607 | biolist = bio; | |
1da177e4 | 3608 | bio->bi_end_io = end_sync_read; |
c34b7ac6 | 3609 | bio->bi_opf = REQ_OP_READ; |
1cdd1257 | 3610 | if (test_bit(FailFast, &rdev->flags)) |
8d3ca83d | 3611 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3612 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3613 | bio_set_dev(bio, rdev->bdev); |
1da177e4 | 3614 | count++; |
9ad1aefc | 3615 | |
a448af25 YK |
3616 | rdev = conf->mirrors[d].replacement; |
3617 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) | |
9ad1aefc | 3618 | continue; |
a448af25 | 3619 | |
f90145f3 | 3620 | atomic_inc(&rdev->nr_pending); |
9ad1aefc N |
3621 | |
3622 | /* Need to set up for writing to the replacement */ | |
3623 | bio = r10_bio->devs[i].repl_bio; | |
4e4cbee9 | 3624 | bio->bi_status = BLK_STS_IOERR; |
9ad1aefc N |
3625 | |
3626 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3627 | bio->bi_next = biolist; |
3628 | biolist = bio; | |
9ad1aefc | 3629 | bio->bi_end_io = end_sync_write; |
c34b7ac6 | 3630 | bio->bi_opf = REQ_OP_WRITE; |
1cdd1257 | 3631 | if (test_bit(FailFast, &rdev->flags)) |
1919cbb2 | 3632 | bio->bi_opf |= MD_FAILFAST; |
f90145f3 | 3633 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
74d46992 | 3634 | bio_set_dev(bio, rdev->bdev); |
9ad1aefc | 3635 | count++; |
1da177e4 LT |
3636 | } |
3637 | ||
3638 | if (count < 2) { | |
3639 | for (i=0; i<conf->copies; i++) { | |
3640 | int d = r10_bio->devs[i].devnum; | |
3641 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3642 | rdev_dec_pending(conf->mirrors[d].rdev, |
3643 | mddev); | |
9ad1aefc N |
3644 | if (r10_bio->devs[i].repl_bio && |
3645 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3646 | rdev_dec_pending( | |
3647 | conf->mirrors[d].replacement, | |
3648 | mddev); | |
1da177e4 LT |
3649 | } |
3650 | put_buf(r10_bio); | |
3651 | biolist = NULL; | |
3652 | goto giveup; | |
3653 | } | |
3654 | } | |
3655 | ||
1da177e4 | 3656 | nr_sectors = 0; |
6cce3b23 N |
3657 | if (sector_nr + max_sync < max_sector) |
3658 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3659 | do { |
3660 | struct page *page; | |
3661 | int len = PAGE_SIZE; | |
1da177e4 LT |
3662 | if (sector_nr + (len>>9) > max_sector) |
3663 | len = (max_sector - sector_nr) << 9; | |
3664 | if (len == 0) | |
3665 | break; | |
3666 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
f0250618 | 3667 | struct resync_pages *rp = get_resync_pages(bio); |
022e510f | 3668 | page = resync_fetch_page(rp, page_idx); |
0c67dd64 JT |
3669 | if (WARN_ON(!bio_add_page(bio, page, len, 0))) { |
3670 | bio->bi_status = BLK_STS_RESOURCE; | |
3671 | bio_endio(bio); | |
3672 | goto giveup; | |
3673 | } | |
1da177e4 LT |
3674 | } |
3675 | nr_sectors += len>>9; | |
3676 | sector_nr += len>>9; | |
022e510f | 3677 | } while (++page_idx < RESYNC_PAGES); |
1da177e4 LT |
3678 | r10_bio->sectors = nr_sectors; |
3679 | ||
8db87912 GJ |
3680 | if (mddev_is_clustered(mddev) && |
3681 | test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3682 | /* It is resync not recovery */ | |
3683 | if (conf->cluster_sync_high < sector_nr + nr_sectors) { | |
3684 | conf->cluster_sync_low = mddev->curr_resync_completed; | |
3685 | raid10_set_cluster_sync_high(conf); | |
3686 | /* Send resync message */ | |
3687 | md_cluster_ops->resync_info_update(mddev, | |
3688 | conf->cluster_sync_low, | |
3689 | conf->cluster_sync_high); | |
3690 | } | |
3691 | } else if (mddev_is_clustered(mddev)) { | |
3692 | /* This is recovery not resync */ | |
3693 | sector_t sect_va1, sect_va2; | |
3694 | bool broadcast_msg = false; | |
3695 | ||
3696 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3697 | /* | |
3698 | * sector_nr is a device address for recovery, so we | |
3699 | * need translate it to array address before compare | |
3700 | * with cluster_sync_high. | |
3701 | */ | |
3702 | sect_va1 = raid10_find_virt(conf, sector_nr, i); | |
3703 | ||
3704 | if (conf->cluster_sync_high < sect_va1 + nr_sectors) { | |
3705 | broadcast_msg = true; | |
3706 | /* | |
3707 | * curr_resync_completed is similar as | |
3708 | * sector_nr, so make the translation too. | |
3709 | */ | |
3710 | sect_va2 = raid10_find_virt(conf, | |
3711 | mddev->curr_resync_completed, i); | |
3712 | ||
3713 | if (conf->cluster_sync_low == 0 || | |
3714 | conf->cluster_sync_low > sect_va2) | |
3715 | conf->cluster_sync_low = sect_va2; | |
3716 | } | |
3717 | } | |
3718 | if (broadcast_msg) { | |
3719 | raid10_set_cluster_sync_high(conf); | |
3720 | md_cluster_ops->resync_info_update(mddev, | |
3721 | conf->cluster_sync_low, | |
3722 | conf->cluster_sync_high); | |
3723 | } | |
3724 | } | |
3725 | ||
1da177e4 LT |
3726 | while (biolist) { |
3727 | bio = biolist; | |
3728 | biolist = biolist->bi_next; | |
3729 | ||
3730 | bio->bi_next = NULL; | |
f0250618 | 3731 | r10_bio = get_resync_r10bio(bio); |
1da177e4 LT |
3732 | r10_bio->sectors = nr_sectors; |
3733 | ||
3734 | if (bio->bi_end_io == end_sync_read) { | |
74d46992 | 3735 | md_sync_acct_bio(bio, nr_sectors); |
4e4cbee9 | 3736 | bio->bi_status = 0; |
ed00aabd | 3737 | submit_bio_noacct(bio); |
1da177e4 LT |
3738 | } |
3739 | } | |
3740 | ||
57afd89f N |
3741 | if (sectors_skipped) |
3742 | /* pretend they weren't skipped, it makes | |
3743 | * no important difference in this case | |
3744 | */ | |
3745 | md_done_sync(mddev, sectors_skipped, 1); | |
3746 | ||
1da177e4 LT |
3747 | return sectors_skipped + nr_sectors; |
3748 | giveup: | |
3749 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3750 | * drives must be failed or in resync, all drives |
3751 | * have a bad block, so try the next chunk... | |
1da177e4 | 3752 | */ |
09b4068a N |
3753 | if (sector_nr + max_sync < max_sector) |
3754 | max_sector = sector_nr + max_sync; | |
3755 | ||
3756 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3757 | chunks_skipped ++; |
3758 | sector_nr = max_sector; | |
1da177e4 | 3759 | goto skipped; |
1da177e4 LT |
3760 | } |
3761 | ||
80c3a6ce | 3762 | static sector_t |
fd01b88c | 3763 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3764 | { |
3765 | sector_t size; | |
e879a879 | 3766 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3767 | |
3768 | if (!raid_disks) | |
3ea7daa5 N |
3769 | raid_disks = min(conf->geo.raid_disks, |
3770 | conf->prev.raid_disks); | |
80c3a6ce | 3771 | if (!sectors) |
dab8b292 | 3772 | sectors = conf->dev_sectors; |
80c3a6ce | 3773 | |
5cf00fcd N |
3774 | size = sectors >> conf->geo.chunk_shift; |
3775 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3776 | size = size * raid_disks; |
5cf00fcd | 3777 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3778 | |
5cf00fcd | 3779 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3780 | } |
3781 | ||
6508fdbf N |
3782 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3783 | { | |
3784 | /* Calculate the number of sectors-per-device that will | |
3785 | * actually be used, and set conf->dev_sectors and | |
3786 | * conf->stride | |
3787 | */ | |
3788 | ||
5cf00fcd N |
3789 | size = size >> conf->geo.chunk_shift; |
3790 | sector_div(size, conf->geo.far_copies); | |
3791 | size = size * conf->geo.raid_disks; | |
3792 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3793 | /* 'size' is now the number of chunks in the array */ |
3794 | /* calculate "used chunks per device" */ | |
3795 | size = size * conf->copies; | |
3796 | ||
3797 | /* We need to round up when dividing by raid_disks to | |
3798 | * get the stride size. | |
3799 | */ | |
5cf00fcd | 3800 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3801 | |
5cf00fcd | 3802 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3803 | |
5cf00fcd N |
3804 | if (conf->geo.far_offset) |
3805 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3806 | else { |
5cf00fcd N |
3807 | sector_div(size, conf->geo.far_copies); |
3808 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3809 | } |
3810 | } | |
dab8b292 | 3811 | |
deb200d0 N |
3812 | enum geo_type {geo_new, geo_old, geo_start}; |
3813 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3814 | { | |
3815 | int nc, fc, fo; | |
3816 | int layout, chunk, disks; | |
3817 | switch (new) { | |
3818 | case geo_old: | |
3819 | layout = mddev->layout; | |
3820 | chunk = mddev->chunk_sectors; | |
3821 | disks = mddev->raid_disks - mddev->delta_disks; | |
3822 | break; | |
3823 | case geo_new: | |
3824 | layout = mddev->new_layout; | |
3825 | chunk = mddev->new_chunk_sectors; | |
3826 | disks = mddev->raid_disks; | |
3827 | break; | |
3828 | default: /* avoid 'may be unused' warnings */ | |
3829 | case geo_start: /* new when starting reshape - raid_disks not | |
3830 | * updated yet. */ | |
3831 | layout = mddev->new_layout; | |
3832 | chunk = mddev->new_chunk_sectors; | |
3833 | disks = mddev->raid_disks + mddev->delta_disks; | |
3834 | break; | |
3835 | } | |
8bce6d35 | 3836 | if (layout >> 19) |
deb200d0 N |
3837 | return -1; |
3838 | if (chunk < (PAGE_SIZE >> 9) || | |
3839 | !is_power_of_2(chunk)) | |
3840 | return -2; | |
3841 | nc = layout & 255; | |
3842 | fc = (layout >> 8) & 255; | |
3843 | fo = layout & (1<<16); | |
3844 | geo->raid_disks = disks; | |
3845 | geo->near_copies = nc; | |
3846 | geo->far_copies = fc; | |
3847 | geo->far_offset = fo; | |
8bce6d35 N |
3848 | switch (layout >> 17) { |
3849 | case 0: /* original layout. simple but not always optimal */ | |
3850 | geo->far_set_size = disks; | |
3851 | break; | |
3852 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3853 | * actually using this, but leave code here just in case.*/ | |
3854 | geo->far_set_size = disks/fc; | |
3855 | WARN(geo->far_set_size < fc, | |
3856 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3857 | break; | |
3858 | case 2: /* "improved" layout fixed to match documentation */ | |
3859 | geo->far_set_size = fc * nc; | |
3860 | break; | |
3861 | default: /* Not a valid layout */ | |
3862 | return -1; | |
3863 | } | |
deb200d0 N |
3864 | geo->chunk_mask = chunk - 1; |
3865 | geo->chunk_shift = ffz(~chunk); | |
3866 | return nc*fc; | |
3867 | } | |
3868 | ||
c9ac2acd YK |
3869 | static void raid10_free_conf(struct r10conf *conf) |
3870 | { | |
3871 | if (!conf) | |
3872 | return; | |
3873 | ||
3874 | mempool_exit(&conf->r10bio_pool); | |
3875 | kfree(conf->mirrors); | |
3876 | kfree(conf->mirrors_old); | |
3877 | kfree(conf->mirrors_new); | |
3878 | safe_put_page(conf->tmppage); | |
3879 | bioset_exit(&conf->bio_split); | |
3880 | kfree(conf); | |
3881 | } | |
3882 | ||
e879a879 | 3883 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3884 | { |
e879a879 | 3885 | struct r10conf *conf = NULL; |
dab8b292 | 3886 | int err = -EINVAL; |
deb200d0 N |
3887 | struct geom geo; |
3888 | int copies; | |
3889 | ||
3890 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3891 | |
deb200d0 | 3892 | if (copies == -2) { |
08464e09 N |
3893 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3894 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3895 | goto out; |
1da177e4 | 3896 | } |
2604b703 | 3897 | |
deb200d0 | 3898 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3899 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3900 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3901 | goto out; |
3902 | } | |
dab8b292 TM |
3903 | |
3904 | err = -ENOMEM; | |
e879a879 | 3905 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3906 | if (!conf) |
1da177e4 | 3907 | goto out; |
dab8b292 | 3908 | |
3ea7daa5 | 3909 | /* FIXME calc properly */ |
6396bb22 KC |
3910 | conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), |
3911 | sizeof(struct raid10_info), | |
dab8b292 TM |
3912 | GFP_KERNEL); |
3913 | if (!conf->mirrors) | |
3914 | goto out; | |
4443ae10 N |
3915 | |
3916 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3917 | if (!conf->tmppage) | |
dab8b292 TM |
3918 | goto out; |
3919 | ||
deb200d0 N |
3920 | conf->geo = geo; |
3921 | conf->copies = copies; | |
3f677f9c | 3922 | err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, |
c7afa803 | 3923 | rbio_pool_free, conf); |
afeee514 | 3924 | if (err) |
dab8b292 TM |
3925 | goto out; |
3926 | ||
afeee514 KO |
3927 | err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); |
3928 | if (err) | |
fc9977dd N |
3929 | goto out; |
3930 | ||
6508fdbf | 3931 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3932 | if (mddev->reshape_position == MaxSector) { |
3933 | conf->prev = conf->geo; | |
3934 | conf->reshape_progress = MaxSector; | |
3935 | } else { | |
3936 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3937 | err = -EINVAL; | |
3938 | goto out; | |
3939 | } | |
3940 | conf->reshape_progress = mddev->reshape_position; | |
3941 | if (conf->prev.far_offset) | |
3942 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3943 | else | |
3944 | /* far_copies must be 1 */ | |
3945 | conf->prev.stride = conf->dev_sectors; | |
3946 | } | |
299b0685 | 3947 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3948 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3949 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3950 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 | 3951 | |
b9b083f9 | 3952 | seqlock_init(&conf->resync_lock); |
dab8b292 | 3953 | init_waitqueue_head(&conf->wait_barrier); |
0e5313e2 | 3954 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3955 | |
afeee514 | 3956 | err = -ENOMEM; |
44693154 YK |
3957 | rcu_assign_pointer(conf->thread, |
3958 | md_register_thread(raid10d, mddev, "raid10")); | |
dab8b292 TM |
3959 | if (!conf->thread) |
3960 | goto out; | |
3961 | ||
dab8b292 TM |
3962 | conf->mddev = mddev; |
3963 | return conf; | |
3964 | ||
3965 | out: | |
c9ac2acd | 3966 | raid10_free_conf(conf); |
dab8b292 TM |
3967 | return ERR_PTR(err); |
3968 | } | |
3969 | ||
3d8466ba | 3970 | static unsigned int raid10_nr_stripes(struct r10conf *conf) |
16ef5101 | 3971 | { |
3d8466ba | 3972 | unsigned int raid_disks = conf->geo.raid_disks; |
16ef5101 | 3973 | |
3d8466ba CH |
3974 | if (conf->geo.raid_disks % conf->geo.near_copies) |
3975 | return raid_disks; | |
3976 | return raid_disks / conf->geo.near_copies; | |
3977 | } | |
16ef5101 | 3978 | |
3d8466ba CH |
3979 | static int raid10_set_queue_limits(struct mddev *mddev) |
3980 | { | |
3981 | struct r10conf *conf = mddev->private; | |
3982 | struct queue_limits lim; | |
3983 | ||
3984 | blk_set_stacking_limits(&lim); | |
3985 | lim.max_write_zeroes_sectors = 0; | |
3986 | lim.io_min = mddev->chunk_sectors << 9; | |
3987 | lim.io_opt = lim.io_min * raid10_nr_stripes(conf); | |
3988 | mddev_stack_rdev_limits(mddev, &lim); | |
396799eb | 3989 | return queue_limits_set(mddev->gendisk->queue, &lim); |
16ef5101 CH |
3990 | } |
3991 | ||
849674e4 | 3992 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3993 | { |
e879a879 | 3994 | struct r10conf *conf; |
16ef5101 | 3995 | int i, disk_idx; |
dc280d98 | 3996 | struct raid10_info *disk; |
3cb03002 | 3997 | struct md_rdev *rdev; |
dab8b292 | 3998 | sector_t size; |
3ea7daa5 N |
3999 | sector_t min_offset_diff = 0; |
4000 | int first = 1; | |
3d8466ba | 4001 | int ret = -EIO; |
dab8b292 TM |
4002 | |
4003 | if (mddev->private == NULL) { | |
4004 | conf = setup_conf(mddev); | |
4005 | if (IS_ERR(conf)) | |
4006 | return PTR_ERR(conf); | |
4007 | mddev->private = conf; | |
4008 | } | |
4009 | conf = mddev->private; | |
4010 | if (!conf) | |
4011 | goto out; | |
4012 | ||
44693154 YK |
4013 | rcu_assign_pointer(mddev->thread, conf->thread); |
4014 | rcu_assign_pointer(conf->thread, NULL); | |
f0ddb83d | 4015 | |
8db87912 GJ |
4016 | if (mddev_is_clustered(conf->mddev)) { |
4017 | int fc, fo; | |
4018 | ||
4019 | fc = (mddev->layout >> 8) & 255; | |
4020 | fo = mddev->layout & (1<<16); | |
4021 | if (fc > 1 || fo > 0) { | |
4022 | pr_err("only near layout is supported by clustered" | |
4023 | " raid10\n"); | |
43a52123 | 4024 | goto out_free_conf; |
8db87912 GJ |
4025 | } |
4026 | } | |
4027 | ||
dafb20fa | 4028 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 4029 | long long diff; |
34b343cf | 4030 | |
1da177e4 | 4031 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
4032 | if (disk_idx < 0) |
4033 | continue; | |
4034 | if (disk_idx >= conf->geo.raid_disks && | |
4035 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
4036 | continue; |
4037 | disk = conf->mirrors + disk_idx; | |
4038 | ||
56a2559b N |
4039 | if (test_bit(Replacement, &rdev->flags)) { |
4040 | if (disk->replacement) | |
4041 | goto out_free_conf; | |
4042 | disk->replacement = rdev; | |
4043 | } else { | |
4044 | if (disk->rdev) | |
4045 | goto out_free_conf; | |
4046 | disk->rdev = rdev; | |
4047 | } | |
3ea7daa5 N |
4048 | diff = (rdev->new_data_offset - rdev->data_offset); |
4049 | if (!mddev->reshape_backwards) | |
4050 | diff = -diff; | |
4051 | if (diff < 0) | |
4052 | diff = 0; | |
4053 | if (first || diff < min_offset_diff) | |
4054 | min_offset_diff = diff; | |
56a2559b | 4055 | |
1da177e4 | 4056 | disk->head_position = 0; |
6f287ca6 | 4057 | first = 0; |
1da177e4 | 4058 | } |
3ea7daa5 | 4059 | |
3d8466ba CH |
4060 | if (!mddev_is_dm(conf->mddev)) { |
4061 | ret = raid10_set_queue_limits(mddev); | |
4062 | if (ret) | |
4063 | goto out_free_conf; | |
4064 | } | |
4065 | ||
6d508242 | 4066 | /* need to check that every block has at least one working mirror */ |
700c7213 | 4067 | if (!enough(conf, -1)) { |
08464e09 | 4068 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 4069 | mdname(mddev)); |
1da177e4 LT |
4070 | goto out_free_conf; |
4071 | } | |
4072 | ||
3ea7daa5 N |
4073 | if (conf->reshape_progress != MaxSector) { |
4074 | /* must ensure that shape change is supported */ | |
4075 | if (conf->geo.far_copies != 1 && | |
4076 | conf->geo.far_offset == 0) | |
4077 | goto out_free_conf; | |
4078 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 4079 | conf->prev.far_offset == 0) |
3ea7daa5 N |
4080 | goto out_free_conf; |
4081 | } | |
4082 | ||
1da177e4 | 4083 | mddev->degraded = 0; |
f8c9e74f N |
4084 | for (i = 0; |
4085 | i < conf->geo.raid_disks | |
4086 | || i < conf->prev.raid_disks; | |
4087 | i++) { | |
1da177e4 LT |
4088 | |
4089 | disk = conf->mirrors + i; | |
4090 | ||
56a2559b N |
4091 | if (!disk->rdev && disk->replacement) { |
4092 | /* The replacement is all we have - use it */ | |
4093 | disk->rdev = disk->replacement; | |
4094 | disk->replacement = NULL; | |
4095 | clear_bit(Replacement, &disk->rdev->flags); | |
4096 | } | |
4097 | ||
5fd6c1dc | 4098 | if (!disk->rdev || |
2e333e89 | 4099 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
4100 | disk->head_position = 0; |
4101 | mddev->degraded++; | |
0b59bb64 N |
4102 | if (disk->rdev && |
4103 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 4104 | conf->fullsync = 1; |
1da177e4 | 4105 | } |
bda31539 BC |
4106 | |
4107 | if (disk->replacement && | |
4108 | !test_bit(In_sync, &disk->replacement->flags) && | |
4109 | disk->replacement->saved_raid_disk < 0) { | |
4110 | conf->fullsync = 1; | |
4111 | } | |
4112 | ||
d890fa2b | 4113 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
4114 | } |
4115 | ||
8c6ac868 | 4116 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
4117 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
4118 | mdname(mddev)); | |
4119 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
4120 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
4121 | conf->geo.raid_disks); | |
1da177e4 LT |
4122 | /* |
4123 | * Ok, everything is just fine now | |
4124 | */ | |
dab8b292 TM |
4125 | mddev->dev_sectors = conf->dev_sectors; |
4126 | size = raid10_size(mddev, 0, 0); | |
4127 | md_set_array_sectors(mddev, size); | |
4128 | mddev->resync_max_sectors = size; | |
46533ff7 | 4129 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 4130 | |
a91a2785 MP |
4131 | if (md_integrity_register(mddev)) |
4132 | goto out_free_conf; | |
4133 | ||
3ea7daa5 N |
4134 | if (conf->reshape_progress != MaxSector) { |
4135 | unsigned long before_length, after_length; | |
4136 | ||
4137 | before_length = ((1 << conf->prev.chunk_shift) * | |
4138 | conf->prev.far_copies); | |
4139 | after_length = ((1 << conf->geo.chunk_shift) * | |
4140 | conf->geo.far_copies); | |
4141 | ||
4142 | if (max(before_length, after_length) > min_offset_diff) { | |
4143 | /* This cannot work */ | |
08464e09 | 4144 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
4145 | goto out_free_conf; |
4146 | } | |
4147 | conf->offset_diff = min_offset_diff; | |
4148 | ||
3ea7daa5 N |
4149 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
4150 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
4151 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
ad39c081 | 4152 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3ea7daa5 N |
4153 | } |
4154 | ||
1da177e4 LT |
4155 | return 0; |
4156 | ||
4157 | out_free_conf: | |
7eb8ff02 | 4158 | md_unregister_thread(mddev, &mddev->thread); |
c9ac2acd | 4159 | raid10_free_conf(conf); |
1da177e4 LT |
4160 | mddev->private = NULL; |
4161 | out: | |
3d8466ba | 4162 | return ret; |
1da177e4 LT |
4163 | } |
4164 | ||
afa0f557 | 4165 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 4166 | { |
c9ac2acd | 4167 | raid10_free_conf(priv); |
1da177e4 LT |
4168 | } |
4169 | ||
b03e0ccb | 4170 | static void raid10_quiesce(struct mddev *mddev, int quiesce) |
6cce3b23 | 4171 | { |
e879a879 | 4172 | struct r10conf *conf = mddev->private; |
6cce3b23 | 4173 | |
b03e0ccb | 4174 | if (quiesce) |
6cce3b23 | 4175 | raise_barrier(conf, 0); |
b03e0ccb | 4176 | else |
6cce3b23 | 4177 | lower_barrier(conf); |
6cce3b23 | 4178 | } |
1da177e4 | 4179 | |
006a09a0 N |
4180 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
4181 | { | |
4182 | /* Resize of 'far' arrays is not supported. | |
4183 | * For 'near' and 'offset' arrays we can set the | |
4184 | * number of sectors used to be an appropriate multiple | |
4185 | * of the chunk size. | |
4186 | * For 'offset', this is far_copies*chunksize. | |
4187 | * For 'near' the multiplier is the LCM of | |
4188 | * near_copies and raid_disks. | |
4189 | * So if far_copies > 1 && !far_offset, fail. | |
4190 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
4191 | * multiply by chunk_size. Then round to this number. | |
4192 | * This is mostly done by raid10_size() | |
4193 | */ | |
4194 | struct r10conf *conf = mddev->private; | |
4195 | sector_t oldsize, size; | |
4196 | ||
f8c9e74f N |
4197 | if (mddev->reshape_position != MaxSector) |
4198 | return -EBUSY; | |
4199 | ||
5cf00fcd | 4200 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
4201 | return -EINVAL; |
4202 | ||
4203 | oldsize = raid10_size(mddev, 0, 0); | |
4204 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
4205 | if (mddev->external_size && |
4206 | mddev->array_sectors > size) | |
006a09a0 | 4207 | return -EINVAL; |
a4a6125a | 4208 | if (mddev->bitmap) { |
e64e4018 | 4209 | int ret = md_bitmap_resize(mddev->bitmap, size, 0, 0); |
a4a6125a N |
4210 | if (ret) |
4211 | return ret; | |
4212 | } | |
4213 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
4214 | if (sectors > mddev->dev_sectors && |
4215 | mddev->recovery_cp > oldsize) { | |
4216 | mddev->recovery_cp = oldsize; | |
4217 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4218 | } | |
6508fdbf N |
4219 | calc_sectors(conf, sectors); |
4220 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
4221 | mddev->resync_max_sectors = size; |
4222 | return 0; | |
4223 | } | |
4224 | ||
53a6ab4d | 4225 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 4226 | { |
3cb03002 | 4227 | struct md_rdev *rdev; |
e879a879 | 4228 | struct r10conf *conf; |
dab8b292 TM |
4229 | |
4230 | if (mddev->degraded > 0) { | |
08464e09 N |
4231 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
4232 | mdname(mddev)); | |
dab8b292 TM |
4233 | return ERR_PTR(-EINVAL); |
4234 | } | |
53a6ab4d | 4235 | sector_div(size, devs); |
dab8b292 | 4236 | |
dab8b292 TM |
4237 | /* Set new parameters */ |
4238 | mddev->new_level = 10; | |
4239 | /* new layout: far_copies = 1, near_copies = 2 */ | |
4240 | mddev->new_layout = (1<<8) + 2; | |
4241 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
4242 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
4243 | mddev->raid_disks *= 2; |
4244 | /* make sure it will be not marked as dirty */ | |
4245 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 4246 | mddev->dev_sectors = size; |
dab8b292 TM |
4247 | |
4248 | conf = setup_conf(mddev); | |
02214dc5 | 4249 | if (!IS_ERR(conf)) { |
dafb20fa | 4250 | rdev_for_each(rdev, mddev) |
53a6ab4d | 4251 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 4252 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
4253 | rdev->sectors = size; |
4254 | } | |
02214dc5 KW |
4255 | } |
4256 | ||
dab8b292 TM |
4257 | return conf; |
4258 | } | |
4259 | ||
fd01b88c | 4260 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 4261 | { |
e373ab10 | 4262 | struct r0conf *raid0_conf; |
dab8b292 TM |
4263 | |
4264 | /* raid10 can take over: | |
4265 | * raid0 - providing it has only two drives | |
4266 | */ | |
4267 | if (mddev->level == 0) { | |
4268 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
4269 | raid0_conf = mddev->private; |
4270 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
4271 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
4272 | mdname(mddev)); | |
dab8b292 TM |
4273 | return ERR_PTR(-EINVAL); |
4274 | } | |
53a6ab4d N |
4275 | return raid10_takeover_raid0(mddev, |
4276 | raid0_conf->strip_zone->zone_end, | |
4277 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
4278 | } |
4279 | return ERR_PTR(-EINVAL); | |
4280 | } | |
4281 | ||
3ea7daa5 N |
4282 | static int raid10_check_reshape(struct mddev *mddev) |
4283 | { | |
4284 | /* Called when there is a request to change | |
4285 | * - layout (to ->new_layout) | |
4286 | * - chunk size (to ->new_chunk_sectors) | |
4287 | * - raid_disks (by delta_disks) | |
4288 | * or when trying to restart a reshape that was ongoing. | |
4289 | * | |
4290 | * We need to validate the request and possibly allocate | |
4291 | * space if that might be an issue later. | |
4292 | * | |
4293 | * Currently we reject any reshape of a 'far' mode array, | |
4294 | * allow chunk size to change if new is generally acceptable, | |
4295 | * allow raid_disks to increase, and allow | |
4296 | * a switch between 'near' mode and 'offset' mode. | |
4297 | */ | |
4298 | struct r10conf *conf = mddev->private; | |
4299 | struct geom geo; | |
4300 | ||
4301 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
4302 | return -EINVAL; | |
4303 | ||
4304 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
4305 | /* mustn't change number of copies */ | |
4306 | return -EINVAL; | |
4307 | if (geo.far_copies > 1 && !geo.far_offset) | |
4308 | /* Cannot switch to 'far' mode */ | |
4309 | return -EINVAL; | |
4310 | ||
4311 | if (mddev->array_sectors & geo.chunk_mask) | |
4312 | /* not factor of array size */ | |
4313 | return -EINVAL; | |
4314 | ||
3ea7daa5 N |
4315 | if (!enough(conf, -1)) |
4316 | return -EINVAL; | |
4317 | ||
4318 | kfree(conf->mirrors_new); | |
4319 | conf->mirrors_new = NULL; | |
4320 | if (mddev->delta_disks > 0) { | |
4321 | /* allocate new 'mirrors' list */ | |
6396bb22 KC |
4322 | conf->mirrors_new = |
4323 | kcalloc(mddev->raid_disks + mddev->delta_disks, | |
4324 | sizeof(struct raid10_info), | |
4325 | GFP_KERNEL); | |
3ea7daa5 N |
4326 | if (!conf->mirrors_new) |
4327 | return -ENOMEM; | |
4328 | } | |
4329 | return 0; | |
4330 | } | |
4331 | ||
4332 | /* | |
4333 | * Need to check if array has failed when deciding whether to: | |
4334 | * - start an array | |
4335 | * - remove non-faulty devices | |
4336 | * - add a spare | |
4337 | * - allow a reshape | |
4338 | * This determination is simple when no reshape is happening. | |
4339 | * However if there is a reshape, we need to carefully check | |
4340 | * both the before and after sections. | |
4341 | * This is because some failed devices may only affect one | |
4342 | * of the two sections, and some non-in_sync devices may | |
4343 | * be insync in the section most affected by failed devices. | |
4344 | */ | |
4345 | static int calc_degraded(struct r10conf *conf) | |
4346 | { | |
4347 | int degraded, degraded2; | |
4348 | int i; | |
4349 | ||
3ea7daa5 N |
4350 | degraded = 0; |
4351 | /* 'prev' section first */ | |
4352 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
a448af25 YK |
4353 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
4354 | ||
3ea7daa5 N |
4355 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
4356 | degraded++; | |
4357 | else if (!test_bit(In_sync, &rdev->flags)) | |
4358 | /* When we can reduce the number of devices in | |
4359 | * an array, this might not contribute to | |
4360 | * 'degraded'. It does now. | |
4361 | */ | |
4362 | degraded++; | |
4363 | } | |
3ea7daa5 N |
4364 | if (conf->geo.raid_disks == conf->prev.raid_disks) |
4365 | return degraded; | |
3ea7daa5 N |
4366 | degraded2 = 0; |
4367 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
a448af25 YK |
4368 | struct md_rdev *rdev = conf->mirrors[i].rdev; |
4369 | ||
3ea7daa5 N |
4370 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
4371 | degraded2++; | |
4372 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4373 | /* If reshape is increasing the number of devices, | |
4374 | * this section has already been recovered, so | |
4375 | * it doesn't contribute to degraded. | |
4376 | * else it does. | |
4377 | */ | |
4378 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4379 | degraded2++; | |
4380 | } | |
4381 | } | |
3ea7daa5 N |
4382 | if (degraded2 > degraded) |
4383 | return degraded2; | |
4384 | return degraded; | |
4385 | } | |
4386 | ||
4387 | static int raid10_start_reshape(struct mddev *mddev) | |
4388 | { | |
4389 | /* A 'reshape' has been requested. This commits | |
4390 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4391 | * This also checks if there are enough spares and adds them | |
4392 | * to the array. | |
4393 | * We currently require enough spares to make the final | |
4394 | * array non-degraded. We also require that the difference | |
4395 | * between old and new data_offset - on each device - is | |
4396 | * enough that we never risk over-writing. | |
4397 | */ | |
4398 | ||
4399 | unsigned long before_length, after_length; | |
4400 | sector_t min_offset_diff = 0; | |
4401 | int first = 1; | |
4402 | struct geom new; | |
4403 | struct r10conf *conf = mddev->private; | |
4404 | struct md_rdev *rdev; | |
4405 | int spares = 0; | |
bb63a701 | 4406 | int ret; |
3ea7daa5 N |
4407 | |
4408 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4409 | return -EBUSY; | |
4410 | ||
4411 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4412 | return -EINVAL; | |
4413 | ||
4414 | before_length = ((1 << conf->prev.chunk_shift) * | |
4415 | conf->prev.far_copies); | |
4416 | after_length = ((1 << conf->geo.chunk_shift) * | |
4417 | conf->geo.far_copies); | |
4418 | ||
4419 | rdev_for_each(rdev, mddev) { | |
4420 | if (!test_bit(In_sync, &rdev->flags) | |
4421 | && !test_bit(Faulty, &rdev->flags)) | |
4422 | spares++; | |
4423 | if (rdev->raid_disk >= 0) { | |
4424 | long long diff = (rdev->new_data_offset | |
4425 | - rdev->data_offset); | |
4426 | if (!mddev->reshape_backwards) | |
4427 | diff = -diff; | |
4428 | if (diff < 0) | |
4429 | diff = 0; | |
4430 | if (first || diff < min_offset_diff) | |
4431 | min_offset_diff = diff; | |
b506335e | 4432 | first = 0; |
3ea7daa5 N |
4433 | } |
4434 | } | |
4435 | ||
4436 | if (max(before_length, after_length) > min_offset_diff) | |
4437 | return -EINVAL; | |
4438 | ||
4439 | if (spares < mddev->delta_disks) | |
4440 | return -EINVAL; | |
4441 | ||
4442 | conf->offset_diff = min_offset_diff; | |
4443 | spin_lock_irq(&conf->device_lock); | |
4444 | if (conf->mirrors_new) { | |
4445 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4446 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4447 | smp_mb(); |
c4796e21 | 4448 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4449 | conf->mirrors_old = conf->mirrors; |
4450 | conf->mirrors = conf->mirrors_new; | |
4451 | conf->mirrors_new = NULL; | |
4452 | } | |
4453 | setup_geo(&conf->geo, mddev, geo_start); | |
4454 | smp_mb(); | |
4455 | if (mddev->reshape_backwards) { | |
4456 | sector_t size = raid10_size(mddev, 0, 0); | |
4457 | if (size < mddev->array_sectors) { | |
4458 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4459 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4460 | mdname(mddev)); | |
3ea7daa5 N |
4461 | return -EINVAL; |
4462 | } | |
4463 | mddev->resync_max_sectors = size; | |
4464 | conf->reshape_progress = size; | |
4465 | } else | |
4466 | conf->reshape_progress = 0; | |
299b0685 | 4467 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4468 | spin_unlock_irq(&conf->device_lock); |
4469 | ||
bb63a701 | 4470 | if (mddev->delta_disks && mddev->bitmap) { |
afd75628 GJ |
4471 | struct mdp_superblock_1 *sb = NULL; |
4472 | sector_t oldsize, newsize; | |
4473 | ||
4474 | oldsize = raid10_size(mddev, 0, 0); | |
4475 | newsize = raid10_size(mddev, 0, conf->geo.raid_disks); | |
4476 | ||
4477 | if (!mddev_is_clustered(mddev)) { | |
4478 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
4479 | if (ret) | |
4480 | goto abort; | |
4481 | else | |
4482 | goto out; | |
4483 | } | |
4484 | ||
4485 | rdev_for_each(rdev, mddev) { | |
4486 | if (rdev->raid_disk > -1 && | |
4487 | !test_bit(Faulty, &rdev->flags)) | |
4488 | sb = page_address(rdev->sb_page); | |
4489 | } | |
4490 | ||
4491 | /* | |
4492 | * some node is already performing reshape, and no need to | |
4493 | * call md_bitmap_resize again since it should be called when | |
4494 | * receiving BITMAP_RESIZE msg | |
4495 | */ | |
4496 | if ((sb && (le32_to_cpu(sb->feature_map) & | |
4497 | MD_FEATURE_RESHAPE_ACTIVE)) || (oldsize == newsize)) | |
4498 | goto out; | |
4499 | ||
4500 | ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); | |
bb63a701 N |
4501 | if (ret) |
4502 | goto abort; | |
afd75628 GJ |
4503 | |
4504 | ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); | |
4505 | if (ret) { | |
4506 | md_bitmap_resize(mddev->bitmap, oldsize, 0, 0); | |
4507 | goto abort; | |
4508 | } | |
bb63a701 | 4509 | } |
afd75628 | 4510 | out: |
3ea7daa5 N |
4511 | if (mddev->delta_disks > 0) { |
4512 | rdev_for_each(rdev, mddev) | |
4513 | if (rdev->raid_disk < 0 && | |
4514 | !test_bit(Faulty, &rdev->flags)) { | |
4515 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4516 | if (rdev->raid_disk >= | |
4517 | conf->prev.raid_disks) | |
4518 | set_bit(In_sync, &rdev->flags); | |
4519 | else | |
4520 | rdev->recovery_offset = 0; | |
4521 | ||
38ffc01f DLM |
4522 | /* Failure here is OK */ |
4523 | sysfs_link_rdev(mddev, rdev); | |
3ea7daa5 N |
4524 | } |
4525 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4526 | && !test_bit(Faulty, &rdev->flags)) { | |
4527 | /* This is a spare that was manually added */ | |
4528 | set_bit(In_sync, &rdev->flags); | |
4529 | } | |
4530 | } | |
4531 | /* When a reshape changes the number of devices, | |
4532 | * ->degraded is measured against the larger of the | |
4533 | * pre and post numbers. | |
4534 | */ | |
4535 | spin_lock_irq(&conf->device_lock); | |
4536 | mddev->degraded = calc_degraded(conf); | |
4537 | spin_unlock_irq(&conf->device_lock); | |
4538 | mddev->raid_disks = conf->geo.raid_disks; | |
4539 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4540 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4541 | |
4542 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4543 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4544 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 | 4545 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
ad39c081 | 4546 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); |
3ea7daa5 | 4547 | conf->reshape_checkpoint = jiffies; |
54679486 | 4548 | md_new_event(); |
3ea7daa5 | 4549 | return 0; |
bb63a701 N |
4550 | |
4551 | abort: | |
4552 | mddev->recovery = 0; | |
4553 | spin_lock_irq(&conf->device_lock); | |
4554 | conf->geo = conf->prev; | |
4555 | mddev->raid_disks = conf->geo.raid_disks; | |
4556 | rdev_for_each(rdev, mddev) | |
4557 | rdev->new_data_offset = rdev->data_offset; | |
4558 | smp_wmb(); | |
4559 | conf->reshape_progress = MaxSector; | |
299b0685 | 4560 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4561 | mddev->reshape_position = MaxSector; |
4562 | spin_unlock_irq(&conf->device_lock); | |
4563 | return ret; | |
3ea7daa5 N |
4564 | } |
4565 | ||
4566 | /* Calculate the last device-address that could contain | |
4567 | * any block from the chunk that includes the array-address 's' | |
4568 | * and report the next address. | |
4569 | * i.e. the address returned will be chunk-aligned and after | |
4570 | * any data that is in the chunk containing 's'. | |
4571 | */ | |
4572 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4573 | { | |
4574 | s = (s | geo->chunk_mask) + 1; | |
4575 | s >>= geo->chunk_shift; | |
4576 | s *= geo->near_copies; | |
4577 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4578 | s *= geo->far_copies; | |
4579 | s <<= geo->chunk_shift; | |
4580 | return s; | |
4581 | } | |
4582 | ||
4583 | /* Calculate the first device-address that could contain | |
4584 | * any block from the chunk that includes the array-address 's'. | |
4585 | * This too will be the start of a chunk | |
4586 | */ | |
4587 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4588 | { | |
4589 | s >>= geo->chunk_shift; | |
4590 | s *= geo->near_copies; | |
4591 | sector_div(s, geo->raid_disks); | |
4592 | s *= geo->far_copies; | |
4593 | s <<= geo->chunk_shift; | |
4594 | return s; | |
4595 | } | |
4596 | ||
4597 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4598 | int *skipped) | |
4599 | { | |
4600 | /* We simply copy at most one chunk (smallest of old and new) | |
4601 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4602 | * or we hit a bad block or something. | |
4603 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4604 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4605 | * can record any location. |
4606 | * | |
4607 | * If we will want to write to a location that isn't | |
4608 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4609 | * we need to flush all reshape requests and update the metadata. | |
4610 | * | |
4611 | * When reshaping forwards (e.g. to more devices), we interpret | |
4612 | * 'safe' as the earliest block which might not have been copied | |
4613 | * down yet. We divide this by previous stripe size and multiply | |
4614 | * by previous stripe length to get lowest device offset that we | |
4615 | * cannot write to yet. | |
4616 | * We interpret 'sector_nr' as an address that we want to write to. | |
4617 | * From this we use last_device_address() to find where we might | |
4618 | * write to, and first_device_address on the 'safe' position. | |
4619 | * If this 'next' write position is after the 'safe' position, | |
4620 | * we must update the metadata to increase the 'safe' position. | |
4621 | * | |
4622 | * When reshaping backwards, we round in the opposite direction | |
4623 | * and perform the reverse test: next write position must not be | |
4624 | * less than current safe position. | |
4625 | * | |
4626 | * In all this the minimum difference in data offsets | |
4627 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4628 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4629 | * |
4630 | * We need to prepare all the bios here before we start any IO | |
4631 | * to ensure the size we choose is acceptable to all devices. | |
4632 | * The means one for each copy for write-out and an extra one for | |
4633 | * read-in. | |
4634 | * We store the read-in bio in ->master_bio and the others in | |
4635 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4636 | */ | |
4637 | struct r10conf *conf = mddev->private; | |
4638 | struct r10bio *r10_bio; | |
4639 | sector_t next, safe, last; | |
4640 | int max_sectors; | |
4641 | int nr_sectors; | |
4642 | int s; | |
4643 | struct md_rdev *rdev; | |
4644 | int need_flush = 0; | |
4645 | struct bio *blist; | |
4646 | struct bio *bio, *read_bio; | |
4647 | int sectors_done = 0; | |
f0250618 | 4648 | struct page **pages; |
3ea7daa5 N |
4649 | |
4650 | if (sector_nr == 0) { | |
4651 | /* If restarting in the middle, skip the initial sectors */ | |
4652 | if (mddev->reshape_backwards && | |
4653 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4654 | sector_nr = (raid10_size(mddev, 0, 0) | |
4655 | - conf->reshape_progress); | |
4656 | } else if (!mddev->reshape_backwards && | |
4657 | conf->reshape_progress > 0) | |
4658 | sector_nr = conf->reshape_progress; | |
4659 | if (sector_nr) { | |
4660 | mddev->curr_resync_completed = sector_nr; | |
e1a86dbb | 4661 | sysfs_notify_dirent_safe(mddev->sysfs_completed); |
3ea7daa5 N |
4662 | *skipped = 1; |
4663 | return sector_nr; | |
4664 | } | |
4665 | } | |
4666 | ||
4667 | /* We don't use sector_nr to track where we are up to | |
4668 | * as that doesn't work well for ->reshape_backwards. | |
4669 | * So just use ->reshape_progress. | |
4670 | */ | |
4671 | if (mddev->reshape_backwards) { | |
4672 | /* 'next' is the earliest device address that we might | |
4673 | * write to for this chunk in the new layout | |
4674 | */ | |
4675 | next = first_dev_address(conf->reshape_progress - 1, | |
4676 | &conf->geo); | |
4677 | ||
4678 | /* 'safe' is the last device address that we might read from | |
4679 | * in the old layout after a restart | |
4680 | */ | |
4681 | safe = last_dev_address(conf->reshape_safe - 1, | |
4682 | &conf->prev); | |
4683 | ||
4684 | if (next + conf->offset_diff < safe) | |
4685 | need_flush = 1; | |
4686 | ||
4687 | last = conf->reshape_progress - 1; | |
4688 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4689 | & conf->prev.chunk_mask); | |
e287308b ZL |
4690 | if (sector_nr + RESYNC_SECTORS < last) |
4691 | sector_nr = last + 1 - RESYNC_SECTORS; | |
3ea7daa5 N |
4692 | } else { |
4693 | /* 'next' is after the last device address that we | |
4694 | * might write to for this chunk in the new layout | |
4695 | */ | |
4696 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4697 | ||
4698 | /* 'safe' is the earliest device address that we might | |
4699 | * read from in the old layout after a restart | |
4700 | */ | |
4701 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4702 | ||
4703 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4704 | * as that would possibly corrupt data | |
4705 | */ | |
4706 | if (next > safe + conf->offset_diff) | |
4707 | need_flush = 1; | |
4708 | ||
4709 | sector_nr = conf->reshape_progress; | |
4710 | last = sector_nr | (conf->geo.chunk_mask | |
4711 | & conf->prev.chunk_mask); | |
4712 | ||
e287308b ZL |
4713 | if (sector_nr + RESYNC_SECTORS <= last) |
4714 | last = sector_nr + RESYNC_SECTORS - 1; | |
3ea7daa5 N |
4715 | } |
4716 | ||
4717 | if (need_flush || | |
4718 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4719 | /* Need to update reshape_position in metadata */ | |
c9aa889b | 4720 | wait_barrier(conf, false); |
3ea7daa5 N |
4721 | mddev->reshape_position = conf->reshape_progress; |
4722 | if (mddev->reshape_backwards) | |
4723 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4724 | - conf->reshape_progress; | |
4725 | else | |
4726 | mddev->curr_resync_completed = conf->reshape_progress; | |
4727 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4728 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4729 | md_wakeup_thread(mddev->thread); |
2953079c | 4730 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4731 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4732 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4733 | allow_barrier(conf); | |
4734 | return sectors_done; | |
4735 | } | |
3ea7daa5 N |
4736 | conf->reshape_safe = mddev->reshape_position; |
4737 | allow_barrier(conf); | |
4738 | } | |
4739 | ||
1d0ffd26 | 4740 | raise_barrier(conf, 0); |
3ea7daa5 N |
4741 | read_more: |
4742 | /* Now schedule reads for blocks from sector_nr to last */ | |
208410b5 | 4743 | r10_bio = raid10_alloc_init_r10buf(conf); |
cb8b12b5 | 4744 | r10_bio->state = 0; |
1d0ffd26 | 4745 | raise_barrier(conf, 1); |
3ea7daa5 N |
4746 | atomic_set(&r10_bio->remaining, 0); |
4747 | r10_bio->mddev = mddev; | |
4748 | r10_bio->sector = sector_nr; | |
4749 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4750 | r10_bio->sectors = last - sector_nr + 1; | |
4751 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4752 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4753 | ||
4754 | if (!rdev) { | |
4755 | /* Cannot read from here, so need to record bad blocks | |
4756 | * on all the target devices. | |
4757 | */ | |
4758 | // FIXME | |
afeee514 | 4759 | mempool_free(r10_bio, &conf->r10buf_pool); |
3ea7daa5 N |
4760 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4761 | return sectors_done; | |
4762 | } | |
4763 | ||
609be106 CH |
4764 | read_bio = bio_alloc_bioset(rdev->bdev, RESYNC_PAGES, REQ_OP_READ, |
4765 | GFP_KERNEL, &mddev->bio_set); | |
4f024f37 | 4766 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4767 | + rdev->data_offset); |
4768 | read_bio->bi_private = r10_bio; | |
81fa1520 | 4769 | read_bio->bi_end_io = end_reshape_read; |
3ea7daa5 N |
4770 | r10_bio->master_bio = read_bio; |
4771 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4772 | ||
7564beda GJ |
4773 | /* |
4774 | * Broadcast RESYNC message to other nodes, so all nodes would not | |
4775 | * write to the region to avoid conflict. | |
4776 | */ | |
4777 | if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { | |
4778 | struct mdp_superblock_1 *sb = NULL; | |
4779 | int sb_reshape_pos = 0; | |
4780 | ||
4781 | conf->cluster_sync_low = sector_nr; | |
4782 | conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; | |
4783 | sb = page_address(rdev->sb_page); | |
4784 | if (sb) { | |
4785 | sb_reshape_pos = le64_to_cpu(sb->reshape_position); | |
4786 | /* | |
4787 | * Set cluster_sync_low again if next address for array | |
4788 | * reshape is less than cluster_sync_low. Since we can't | |
4789 | * update cluster_sync_low until it has finished reshape. | |
4790 | */ | |
4791 | if (sb_reshape_pos < conf->cluster_sync_low) | |
4792 | conf->cluster_sync_low = sb_reshape_pos; | |
4793 | } | |
4794 | ||
4795 | md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, | |
4796 | conf->cluster_sync_high); | |
4797 | } | |
4798 | ||
3ea7daa5 N |
4799 | /* Now find the locations in the new layout */ |
4800 | __raid10_find_phys(&conf->geo, r10_bio); | |
4801 | ||
4802 | blist = read_bio; | |
4803 | read_bio->bi_next = NULL; | |
4804 | ||
4805 | for (s = 0; s < conf->copies*2; s++) { | |
4806 | struct bio *b; | |
4807 | int d = r10_bio->devs[s/2].devnum; | |
4808 | struct md_rdev *rdev2; | |
4809 | if (s&1) { | |
a448af25 | 4810 | rdev2 = conf->mirrors[d].replacement; |
3ea7daa5 N |
4811 | b = r10_bio->devs[s/2].repl_bio; |
4812 | } else { | |
a448af25 | 4813 | rdev2 = conf->mirrors[d].rdev; |
3ea7daa5 N |
4814 | b = r10_bio->devs[s/2].bio; |
4815 | } | |
4816 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4817 | continue; | |
8be185f2 | 4818 | |
74d46992 | 4819 | bio_set_dev(b, rdev2->bdev); |
4f024f37 KO |
4820 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4821 | rdev2->new_data_offset; | |
3ea7daa5 | 4822 | b->bi_end_io = end_reshape_write; |
c34b7ac6 | 4823 | b->bi_opf = REQ_OP_WRITE; |
3ea7daa5 | 4824 | b->bi_next = blist; |
3ea7daa5 N |
4825 | blist = b; |
4826 | } | |
4827 | ||
4828 | /* Now add as many pages as possible to all of these bios. */ | |
4829 | ||
4830 | nr_sectors = 0; | |
f0250618 | 4831 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; |
3ea7daa5 | 4832 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { |
f0250618 | 4833 | struct page *page = pages[s / (PAGE_SIZE >> 9)]; |
3ea7daa5 N |
4834 | int len = (max_sectors - s) << 9; |
4835 | if (len > PAGE_SIZE) | |
4836 | len = PAGE_SIZE; | |
4837 | for (bio = blist; bio ; bio = bio->bi_next) { | |
0c67dd64 JT |
4838 | if (WARN_ON(!bio_add_page(bio, page, len, 0))) { |
4839 | bio->bi_status = BLK_STS_RESOURCE; | |
4840 | bio_endio(bio); | |
4841 | return sectors_done; | |
4842 | } | |
3ea7daa5 N |
4843 | } |
4844 | sector_nr += len >> 9; | |
4845 | nr_sectors += len >> 9; | |
4846 | } | |
3ea7daa5 N |
4847 | r10_bio->sectors = nr_sectors; |
4848 | ||
4849 | /* Now submit the read */ | |
74d46992 | 4850 | md_sync_acct_bio(read_bio, r10_bio->sectors); |
3ea7daa5 N |
4851 | atomic_inc(&r10_bio->remaining); |
4852 | read_bio->bi_next = NULL; | |
ed00aabd | 4853 | submit_bio_noacct(read_bio); |
3ea7daa5 N |
4854 | sectors_done += nr_sectors; |
4855 | if (sector_nr <= last) | |
4856 | goto read_more; | |
4857 | ||
1d0ffd26 XN |
4858 | lower_barrier(conf); |
4859 | ||
3ea7daa5 N |
4860 | /* Now that we have done the whole section we can |
4861 | * update reshape_progress | |
4862 | */ | |
4863 | if (mddev->reshape_backwards) | |
4864 | conf->reshape_progress -= sectors_done; | |
4865 | else | |
4866 | conf->reshape_progress += sectors_done; | |
4867 | ||
4868 | return sectors_done; | |
4869 | } | |
4870 | ||
4871 | static void end_reshape_request(struct r10bio *r10_bio); | |
4872 | static int handle_reshape_read_error(struct mddev *mddev, | |
4873 | struct r10bio *r10_bio); | |
4874 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4875 | { | |
4876 | /* Reshape read completed. Hopefully we have a block | |
4877 | * to write out. | |
4878 | * If we got a read error then we do sync 1-page reads from | |
4879 | * elsewhere until we find the data - or give up. | |
4880 | */ | |
4881 | struct r10conf *conf = mddev->private; | |
4882 | int s; | |
4883 | ||
4884 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4885 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4886 | /* Reshape has been aborted */ | |
4887 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4888 | return; | |
4889 | } | |
4890 | ||
4891 | /* We definitely have the data in the pages, schedule the | |
4892 | * writes. | |
4893 | */ | |
4894 | atomic_set(&r10_bio->remaining, 1); | |
4895 | for (s = 0; s < conf->copies*2; s++) { | |
4896 | struct bio *b; | |
4897 | int d = r10_bio->devs[s/2].devnum; | |
4898 | struct md_rdev *rdev; | |
4899 | if (s&1) { | |
a448af25 | 4900 | rdev = conf->mirrors[d].replacement; |
3ea7daa5 N |
4901 | b = r10_bio->devs[s/2].repl_bio; |
4902 | } else { | |
a448af25 | 4903 | rdev = conf->mirrors[d].rdev; |
3ea7daa5 N |
4904 | b = r10_bio->devs[s/2].bio; |
4905 | } | |
a448af25 | 4906 | if (!rdev || test_bit(Faulty, &rdev->flags)) |
3ea7daa5 | 4907 | continue; |
a448af25 | 4908 | |
3ea7daa5 | 4909 | atomic_inc(&rdev->nr_pending); |
74d46992 | 4910 | md_sync_acct_bio(b, r10_bio->sectors); |
3ea7daa5 N |
4911 | atomic_inc(&r10_bio->remaining); |
4912 | b->bi_next = NULL; | |
ed00aabd | 4913 | submit_bio_noacct(b); |
3ea7daa5 N |
4914 | } |
4915 | end_reshape_request(r10_bio); | |
4916 | } | |
4917 | ||
4918 | static void end_reshape(struct r10conf *conf) | |
4919 | { | |
4920 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4921 | return; | |
4922 | ||
4923 | spin_lock_irq(&conf->device_lock); | |
4924 | conf->prev = conf->geo; | |
4925 | md_finish_reshape(conf->mddev); | |
4926 | smp_wmb(); | |
4927 | conf->reshape_progress = MaxSector; | |
299b0685 | 4928 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4929 | spin_unlock_irq(&conf->device_lock); |
4930 | ||
3d8466ba | 4931 | mddev_update_io_opt(conf->mddev, raid10_nr_stripes(conf)); |
3ea7daa5 N |
4932 | conf->fullsync = 0; |
4933 | } | |
4934 | ||
7564beda GJ |
4935 | static void raid10_update_reshape_pos(struct mddev *mddev) |
4936 | { | |
4937 | struct r10conf *conf = mddev->private; | |
5ebaf80b | 4938 | sector_t lo, hi; |
7564beda | 4939 | |
5ebaf80b GJ |
4940 | md_cluster_ops->resync_info_get(mddev, &lo, &hi); |
4941 | if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) | |
4942 | || mddev->reshape_position == MaxSector) | |
4943 | conf->reshape_progress = mddev->reshape_position; | |
4944 | else | |
4945 | WARN_ON_ONCE(1); | |
7564beda GJ |
4946 | } |
4947 | ||
3ea7daa5 N |
4948 | static int handle_reshape_read_error(struct mddev *mddev, |
4949 | struct r10bio *r10_bio) | |
4950 | { | |
4951 | /* Use sync reads to get the blocks from somewhere else */ | |
4952 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4953 | struct r10conf *conf = mddev->private; |
584ed9fa | 4954 | struct r10bio *r10b; |
3ea7daa5 N |
4955 | int slot = 0; |
4956 | int idx = 0; | |
2d06e3b7 ML |
4957 | struct page **pages; |
4958 | ||
8cf05a78 | 4959 | r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); |
584ed9fa MK |
4960 | if (!r10b) { |
4961 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
4962 | return -ENOMEM; | |
4963 | } | |
4964 | ||
2d06e3b7 ML |
4965 | /* reshape IOs share pages from .devs[0].bio */ |
4966 | pages = get_resync_pages(r10_bio->devs[0].bio)->pages; | |
3ea7daa5 | 4967 | |
e0ee7785 N |
4968 | r10b->sector = r10_bio->sector; |
4969 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4970 | |
4971 | while (sectors) { | |
4972 | int s = sectors; | |
4973 | int success = 0; | |
4974 | int first_slot = slot; | |
4975 | ||
4976 | if (s > (PAGE_SIZE >> 9)) | |
4977 | s = PAGE_SIZE >> 9; | |
4978 | ||
4979 | while (!success) { | |
e0ee7785 | 4980 | int d = r10b->devs[slot].devnum; |
a448af25 | 4981 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
3ea7daa5 N |
4982 | sector_t addr; |
4983 | if (rdev == NULL || | |
4984 | test_bit(Faulty, &rdev->flags) || | |
4985 | !test_bit(In_sync, &rdev->flags)) | |
4986 | goto failed; | |
4987 | ||
e0ee7785 | 4988 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 | 4989 | atomic_inc(&rdev->nr_pending); |
3ea7daa5 N |
4990 | success = sync_page_io(rdev, |
4991 | addr, | |
4992 | s << 9, | |
2d06e3b7 | 4993 | pages[idx], |
4ce4c73f | 4994 | REQ_OP_READ, false); |
d094d686 | 4995 | rdev_dec_pending(rdev, mddev); |
3ea7daa5 N |
4996 | if (success) |
4997 | break; | |
4998 | failed: | |
4999 | slot++; | |
5000 | if (slot >= conf->copies) | |
5001 | slot = 0; | |
5002 | if (slot == first_slot) | |
5003 | break; | |
5004 | } | |
5005 | if (!success) { | |
5006 | /* couldn't read this block, must give up */ | |
5007 | set_bit(MD_RECOVERY_INTR, | |
5008 | &mddev->recovery); | |
584ed9fa | 5009 | kfree(r10b); |
3ea7daa5 N |
5010 | return -EIO; |
5011 | } | |
5012 | sectors -= s; | |
5013 | idx++; | |
5014 | } | |
584ed9fa | 5015 | kfree(r10b); |
3ea7daa5 N |
5016 | return 0; |
5017 | } | |
5018 | ||
4246a0b6 | 5019 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 5020 | { |
f0250618 | 5021 | struct r10bio *r10_bio = get_resync_r10bio(bio); |
3ea7daa5 N |
5022 | struct mddev *mddev = r10_bio->mddev; |
5023 | struct r10conf *conf = mddev->private; | |
5024 | int d; | |
5025 | int slot; | |
5026 | int repl; | |
5027 | struct md_rdev *rdev = NULL; | |
5028 | ||
5029 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
a448af25 YK |
5030 | rdev = repl ? conf->mirrors[d].replacement : |
5031 | conf->mirrors[d].rdev; | |
3ea7daa5 | 5032 | |
4e4cbee9 | 5033 | if (bio->bi_status) { |
3ea7daa5 N |
5034 | /* FIXME should record badblock */ |
5035 | md_error(mddev, rdev); | |
5036 | } | |
5037 | ||
5038 | rdev_dec_pending(rdev, mddev); | |
5039 | end_reshape_request(r10_bio); | |
5040 | } | |
5041 | ||
5042 | static void end_reshape_request(struct r10bio *r10_bio) | |
5043 | { | |
5044 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
5045 | return; | |
5046 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
5047 | bio_put(r10_bio->master_bio); | |
5048 | put_buf(r10_bio); | |
5049 | } | |
5050 | ||
5051 | static void raid10_finish_reshape(struct mddev *mddev) | |
5052 | { | |
5053 | struct r10conf *conf = mddev->private; | |
5054 | ||
5055 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
5056 | return; | |
5057 | ||
5058 | if (mddev->delta_disks > 0) { | |
3ea7daa5 N |
5059 | if (mddev->recovery_cp > mddev->resync_max_sectors) { |
5060 | mddev->recovery_cp = mddev->resync_max_sectors; | |
5061 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
5062 | } | |
8876391e | 5063 | mddev->resync_max_sectors = mddev->array_sectors; |
63aced61 N |
5064 | } else { |
5065 | int d; | |
5066 | for (d = conf->geo.raid_disks ; | |
5067 | d < conf->geo.raid_disks - mddev->delta_disks; | |
5068 | d++) { | |
a448af25 | 5069 | struct md_rdev *rdev = conf->mirrors[d].rdev; |
63aced61 N |
5070 | if (rdev) |
5071 | clear_bit(In_sync, &rdev->flags); | |
a448af25 | 5072 | rdev = conf->mirrors[d].replacement; |
63aced61 N |
5073 | if (rdev) |
5074 | clear_bit(In_sync, &rdev->flags); | |
5075 | } | |
3ea7daa5 N |
5076 | } |
5077 | mddev->layout = mddev->new_layout; | |
5078 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
5079 | mddev->reshape_position = MaxSector; | |
5080 | mddev->delta_disks = 0; | |
5081 | mddev->reshape_backwards = 0; | |
5082 | } | |
5083 | ||
84fc4b56 | 5084 | static struct md_personality raid10_personality = |
1da177e4 LT |
5085 | { |
5086 | .name = "raid10", | |
2604b703 | 5087 | .level = 10, |
1da177e4 | 5088 | .owner = THIS_MODULE, |
849674e4 SL |
5089 | .make_request = raid10_make_request, |
5090 | .run = raid10_run, | |
afa0f557 | 5091 | .free = raid10_free, |
849674e4 SL |
5092 | .status = raid10_status, |
5093 | .error_handler = raid10_error, | |
1da177e4 LT |
5094 | .hot_add_disk = raid10_add_disk, |
5095 | .hot_remove_disk= raid10_remove_disk, | |
5096 | .spare_active = raid10_spare_active, | |
849674e4 | 5097 | .sync_request = raid10_sync_request, |
6cce3b23 | 5098 | .quiesce = raid10_quiesce, |
80c3a6ce | 5099 | .size = raid10_size, |
006a09a0 | 5100 | .resize = raid10_resize, |
dab8b292 | 5101 | .takeover = raid10_takeover, |
3ea7daa5 N |
5102 | .check_reshape = raid10_check_reshape, |
5103 | .start_reshape = raid10_start_reshape, | |
5104 | .finish_reshape = raid10_finish_reshape, | |
7564beda | 5105 | .update_reshape_pos = raid10_update_reshape_pos, |
1da177e4 LT |
5106 | }; |
5107 | ||
5108 | static int __init raid_init(void) | |
5109 | { | |
2604b703 | 5110 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
5111 | } |
5112 | ||
5113 | static void raid_exit(void) | |
5114 | { | |
2604b703 | 5115 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
5116 | } |
5117 | ||
5118 | module_init(raid_init); | |
5119 | module_exit(raid_exit); | |
5120 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 5121 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 5122 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 5123 | MODULE_ALIAS("md-raid10"); |
2604b703 | 5124 | MODULE_ALIAS("md-level-10"); |