]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * raid10.c : Multiple Devices driver for Linux | |
3 | * | |
4 | * Copyright (C) 2000-2004 Neil Brown | |
5 | * | |
6 | * RAID-10 support for md. | |
7 | * | |
25985edc | 8 | * Base on code in raid1.c. See raid1.c for further copyright information. |
1da177e4 LT |
9 | * |
10 | * | |
11 | * This program is free software; you can redistribute it and/or modify | |
12 | * it under the terms of the GNU General Public License as published by | |
13 | * the Free Software Foundation; either version 2, or (at your option) | |
14 | * any later version. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * (for example /usr/src/linux/COPYING); if not, write to the Free | |
18 | * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
20 | ||
5a0e3ad6 | 21 | #include <linux/slab.h> |
25570727 | 22 | #include <linux/delay.h> |
bff61975 | 23 | #include <linux/blkdev.h> |
056075c7 | 24 | #include <linux/module.h> |
bff61975 | 25 | #include <linux/seq_file.h> |
8bda470e | 26 | #include <linux/ratelimit.h> |
3ea7daa5 | 27 | #include <linux/kthread.h> |
109e3765 | 28 | #include <trace/events/block.h> |
43b2e5d8 | 29 | #include "md.h" |
ef740c37 | 30 | #include "raid10.h" |
dab8b292 | 31 | #include "raid0.h" |
ef740c37 | 32 | #include "bitmap.h" |
1da177e4 LT |
33 | |
34 | /* | |
35 | * RAID10 provides a combination of RAID0 and RAID1 functionality. | |
36 | * The layout of data is defined by | |
37 | * chunk_size | |
38 | * raid_disks | |
39 | * near_copies (stored in low byte of layout) | |
40 | * far_copies (stored in second byte of layout) | |
c93983bf | 41 | * far_offset (stored in bit 16 of layout ) |
475901af | 42 | * use_far_sets (stored in bit 17 of layout ) |
8bce6d35 | 43 | * use_far_sets_bugfixed (stored in bit 18 of layout ) |
1da177e4 | 44 | * |
475901af JB |
45 | * The data to be stored is divided into chunks using chunksize. Each device |
46 | * is divided into far_copies sections. In each section, chunks are laid out | |
47 | * in a style similar to raid0, but near_copies copies of each chunk is stored | |
48 | * (each on a different drive). The starting device for each section is offset | |
49 | * near_copies from the starting device of the previous section. Thus there | |
50 | * are (near_copies * far_copies) of each chunk, and each is on a different | |
51 | * drive. near_copies and far_copies must be at least one, and their product | |
52 | * is at most raid_disks. | |
c93983bf N |
53 | * |
54 | * If far_offset is true, then the far_copies are handled a bit differently. | |
475901af JB |
55 | * The copies are still in different stripes, but instead of being very far |
56 | * apart on disk, there are adjacent stripes. | |
57 | * | |
58 | * The far and offset algorithms are handled slightly differently if | |
59 | * 'use_far_sets' is true. In this case, the array's devices are grouped into | |
60 | * sets that are (near_copies * far_copies) in size. The far copied stripes | |
61 | * are still shifted by 'near_copies' devices, but this shifting stays confined | |
62 | * to the set rather than the entire array. This is done to improve the number | |
63 | * of device combinations that can fail without causing the array to fail. | |
64 | * Example 'far' algorithm w/o 'use_far_sets' (each letter represents a chunk | |
65 | * on a device): | |
66 | * A B C D A B C D E | |
67 | * ... ... | |
68 | * D A B C E A B C D | |
69 | * Example 'far' algorithm w/ 'use_far_sets' enabled (sets illustrated w/ []'s): | |
70 | * [A B] [C D] [A B] [C D E] | |
71 | * |...| |...| |...| | ... | | |
72 | * [B A] [D C] [B A] [E C D] | |
1da177e4 LT |
73 | */ |
74 | ||
75 | /* | |
76 | * Number of guaranteed r10bios in case of extreme VM load: | |
77 | */ | |
78 | #define NR_RAID10_BIOS 256 | |
79 | ||
473e87ce JB |
80 | /* when we get a read error on a read-only array, we redirect to another |
81 | * device without failing the first device, or trying to over-write to | |
82 | * correct the read error. To keep track of bad blocks on a per-bio | |
83 | * level, we store IO_BLOCKED in the appropriate 'bios' pointer | |
84 | */ | |
85 | #define IO_BLOCKED ((struct bio *)1) | |
86 | /* When we successfully write to a known bad-block, we need to remove the | |
87 | * bad-block marking which must be done from process context. So we record | |
88 | * the success by setting devs[n].bio to IO_MADE_GOOD | |
89 | */ | |
90 | #define IO_MADE_GOOD ((struct bio *)2) | |
91 | ||
92 | #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2) | |
93 | ||
94 | /* When there are this many requests queued to be written by | |
34db0cd6 N |
95 | * the raid10 thread, we become 'congested' to provide back-pressure |
96 | * for writeback. | |
97 | */ | |
98 | static int max_queued_requests = 1024; | |
99 | ||
e879a879 N |
100 | static void allow_barrier(struct r10conf *conf); |
101 | static void lower_barrier(struct r10conf *conf); | |
635f6416 | 102 | static int _enough(struct r10conf *conf, int previous, int ignore); |
1919cbb2 | 103 | static int enough(struct r10conf *conf, int ignore); |
3ea7daa5 N |
104 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, |
105 | int *skipped); | |
106 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio); | |
4246a0b6 | 107 | static void end_reshape_write(struct bio *bio); |
3ea7daa5 | 108 | static void end_reshape(struct r10conf *conf); |
0a27ec96 | 109 | |
578b54ad N |
110 | #define raid10_log(md, fmt, args...) \ |
111 | do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0) | |
112 | ||
dd0fc66f | 113 | static void * r10bio_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 114 | { |
e879a879 | 115 | struct r10conf *conf = data; |
9f2c9d12 | 116 | int size = offsetof(struct r10bio, devs[conf->copies]); |
1da177e4 | 117 | |
69335ef3 N |
118 | /* allocate a r10bio with room for raid_disks entries in the |
119 | * bios array */ | |
7eaceacc | 120 | return kzalloc(size, gfp_flags); |
1da177e4 LT |
121 | } |
122 | ||
123 | static void r10bio_pool_free(void *r10_bio, void *data) | |
124 | { | |
125 | kfree(r10_bio); | |
126 | } | |
127 | ||
0310fa21 | 128 | /* Maximum size of each resync request */ |
1da177e4 | 129 | #define RESYNC_BLOCK_SIZE (64*1024) |
1da177e4 | 130 | #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) |
0310fa21 N |
131 | /* amount of memory to reserve for resync requests */ |
132 | #define RESYNC_WINDOW (1024*1024) | |
133 | /* maximum number of concurrent requests, memory permitting */ | |
134 | #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE) | |
1da177e4 LT |
135 | |
136 | /* | |
137 | * When performing a resync, we need to read and compare, so | |
138 | * we need as many pages are there are copies. | |
139 | * When performing a recovery, we need 2 bios, one for read, | |
140 | * one for write (we recover only one drive per r10buf) | |
141 | * | |
142 | */ | |
dd0fc66f | 143 | static void * r10buf_pool_alloc(gfp_t gfp_flags, void *data) |
1da177e4 | 144 | { |
e879a879 | 145 | struct r10conf *conf = data; |
1da177e4 | 146 | struct page *page; |
9f2c9d12 | 147 | struct r10bio *r10_bio; |
1da177e4 LT |
148 | struct bio *bio; |
149 | int i, j; | |
150 | int nalloc; | |
151 | ||
152 | r10_bio = r10bio_pool_alloc(gfp_flags, conf); | |
7eaceacc | 153 | if (!r10_bio) |
1da177e4 | 154 | return NULL; |
1da177e4 | 155 | |
3ea7daa5 N |
156 | if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || |
157 | test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) | |
1da177e4 LT |
158 | nalloc = conf->copies; /* resync */ |
159 | else | |
160 | nalloc = 2; /* recovery */ | |
161 | ||
162 | /* | |
163 | * Allocate bios. | |
164 | */ | |
165 | for (j = nalloc ; j-- ; ) { | |
6746557f | 166 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); |
1da177e4 LT |
167 | if (!bio) |
168 | goto out_free_bio; | |
169 | r10_bio->devs[j].bio = bio; | |
69335ef3 N |
170 | if (!conf->have_replacement) |
171 | continue; | |
172 | bio = bio_kmalloc(gfp_flags, RESYNC_PAGES); | |
173 | if (!bio) | |
174 | goto out_free_bio; | |
175 | r10_bio->devs[j].repl_bio = bio; | |
1da177e4 LT |
176 | } |
177 | /* | |
178 | * Allocate RESYNC_PAGES data pages and attach them | |
179 | * where needed. | |
180 | */ | |
181 | for (j = 0 ; j < nalloc; j++) { | |
69335ef3 | 182 | struct bio *rbio = r10_bio->devs[j].repl_bio; |
1da177e4 LT |
183 | bio = r10_bio->devs[j].bio; |
184 | for (i = 0; i < RESYNC_PAGES; i++) { | |
3ea7daa5 N |
185 | if (j > 0 && !test_bit(MD_RECOVERY_SYNC, |
186 | &conf->mddev->recovery)) { | |
187 | /* we can share bv_page's during recovery | |
188 | * and reshape */ | |
c65060ad NK |
189 | struct bio *rbio = r10_bio->devs[0].bio; |
190 | page = rbio->bi_io_vec[i].bv_page; | |
191 | get_page(page); | |
192 | } else | |
193 | page = alloc_page(gfp_flags); | |
1da177e4 LT |
194 | if (unlikely(!page)) |
195 | goto out_free_pages; | |
196 | ||
197 | bio->bi_io_vec[i].bv_page = page; | |
69335ef3 N |
198 | if (rbio) |
199 | rbio->bi_io_vec[i].bv_page = page; | |
1da177e4 LT |
200 | } |
201 | } | |
202 | ||
203 | return r10_bio; | |
204 | ||
205 | out_free_pages: | |
206 | for ( ; i > 0 ; i--) | |
1345b1d8 | 207 | safe_put_page(bio->bi_io_vec[i-1].bv_page); |
1da177e4 LT |
208 | while (j--) |
209 | for (i = 0; i < RESYNC_PAGES ; i++) | |
1345b1d8 | 210 | safe_put_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); |
5fdd2cf8 | 211 | j = 0; |
1da177e4 | 212 | out_free_bio: |
5fdd2cf8 | 213 | for ( ; j < nalloc; j++) { |
214 | if (r10_bio->devs[j].bio) | |
215 | bio_put(r10_bio->devs[j].bio); | |
69335ef3 N |
216 | if (r10_bio->devs[j].repl_bio) |
217 | bio_put(r10_bio->devs[j].repl_bio); | |
218 | } | |
1da177e4 LT |
219 | r10bio_pool_free(r10_bio, conf); |
220 | return NULL; | |
221 | } | |
222 | ||
223 | static void r10buf_pool_free(void *__r10_bio, void *data) | |
224 | { | |
225 | int i; | |
e879a879 | 226 | struct r10conf *conf = data; |
9f2c9d12 | 227 | struct r10bio *r10bio = __r10_bio; |
1da177e4 LT |
228 | int j; |
229 | ||
230 | for (j=0; j < conf->copies; j++) { | |
231 | struct bio *bio = r10bio->devs[j].bio; | |
232 | if (bio) { | |
233 | for (i = 0; i < RESYNC_PAGES; i++) { | |
1345b1d8 | 234 | safe_put_page(bio->bi_io_vec[i].bv_page); |
1da177e4 LT |
235 | bio->bi_io_vec[i].bv_page = NULL; |
236 | } | |
237 | bio_put(bio); | |
238 | } | |
69335ef3 N |
239 | bio = r10bio->devs[j].repl_bio; |
240 | if (bio) | |
241 | bio_put(bio); | |
1da177e4 LT |
242 | } |
243 | r10bio_pool_free(r10bio, conf); | |
244 | } | |
245 | ||
e879a879 | 246 | static void put_all_bios(struct r10conf *conf, struct r10bio *r10_bio) |
1da177e4 LT |
247 | { |
248 | int i; | |
249 | ||
250 | for (i = 0; i < conf->copies; i++) { | |
251 | struct bio **bio = & r10_bio->devs[i].bio; | |
749c55e9 | 252 | if (!BIO_SPECIAL(*bio)) |
1da177e4 LT |
253 | bio_put(*bio); |
254 | *bio = NULL; | |
69335ef3 N |
255 | bio = &r10_bio->devs[i].repl_bio; |
256 | if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) | |
257 | bio_put(*bio); | |
258 | *bio = NULL; | |
1da177e4 LT |
259 | } |
260 | } | |
261 | ||
9f2c9d12 | 262 | static void free_r10bio(struct r10bio *r10_bio) |
1da177e4 | 263 | { |
e879a879 | 264 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 265 | |
1da177e4 LT |
266 | put_all_bios(conf, r10_bio); |
267 | mempool_free(r10_bio, conf->r10bio_pool); | |
268 | } | |
269 | ||
9f2c9d12 | 270 | static void put_buf(struct r10bio *r10_bio) |
1da177e4 | 271 | { |
e879a879 | 272 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
273 | |
274 | mempool_free(r10_bio, conf->r10buf_pool); | |
275 | ||
0a27ec96 | 276 | lower_barrier(conf); |
1da177e4 LT |
277 | } |
278 | ||
9f2c9d12 | 279 | static void reschedule_retry(struct r10bio *r10_bio) |
1da177e4 LT |
280 | { |
281 | unsigned long flags; | |
fd01b88c | 282 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 283 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
284 | |
285 | spin_lock_irqsave(&conf->device_lock, flags); | |
286 | list_add(&r10_bio->retry_list, &conf->retry_list); | |
4443ae10 | 287 | conf->nr_queued ++; |
1da177e4 LT |
288 | spin_unlock_irqrestore(&conf->device_lock, flags); |
289 | ||
388667be AJ |
290 | /* wake up frozen array... */ |
291 | wake_up(&conf->wait_barrier); | |
292 | ||
1da177e4 LT |
293 | md_wakeup_thread(mddev->thread); |
294 | } | |
295 | ||
296 | /* | |
297 | * raid_end_bio_io() is called when we have finished servicing a mirrored | |
298 | * operation and are ready to return a success/failure code to the buffer | |
299 | * cache layer. | |
300 | */ | |
9f2c9d12 | 301 | static void raid_end_bio_io(struct r10bio *r10_bio) |
1da177e4 LT |
302 | { |
303 | struct bio *bio = r10_bio->master_bio; | |
856e08e2 | 304 | int done; |
e879a879 | 305 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 306 | |
856e08e2 N |
307 | if (bio->bi_phys_segments) { |
308 | unsigned long flags; | |
309 | spin_lock_irqsave(&conf->device_lock, flags); | |
310 | bio->bi_phys_segments--; | |
311 | done = (bio->bi_phys_segments == 0); | |
312 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
313 | } else | |
314 | done = 1; | |
315 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4246a0b6 | 316 | bio->bi_error = -EIO; |
856e08e2 | 317 | if (done) { |
4246a0b6 | 318 | bio_endio(bio); |
856e08e2 N |
319 | /* |
320 | * Wake up any possible resync thread that waits for the device | |
321 | * to go idle. | |
322 | */ | |
323 | allow_barrier(conf); | |
324 | } | |
1da177e4 LT |
325 | free_r10bio(r10_bio); |
326 | } | |
327 | ||
328 | /* | |
329 | * Update disk head position estimator based on IRQ completion info. | |
330 | */ | |
9f2c9d12 | 331 | static inline void update_head_pos(int slot, struct r10bio *r10_bio) |
1da177e4 | 332 | { |
e879a879 | 333 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 LT |
334 | |
335 | conf->mirrors[r10_bio->devs[slot].devnum].head_position = | |
336 | r10_bio->devs[slot].addr + (r10_bio->sectors); | |
337 | } | |
338 | ||
778ca018 NK |
339 | /* |
340 | * Find the disk number which triggered given bio | |
341 | */ | |
e879a879 | 342 | static int find_bio_disk(struct r10conf *conf, struct r10bio *r10_bio, |
69335ef3 | 343 | struct bio *bio, int *slotp, int *replp) |
778ca018 NK |
344 | { |
345 | int slot; | |
69335ef3 | 346 | int repl = 0; |
778ca018 | 347 | |
69335ef3 | 348 | for (slot = 0; slot < conf->copies; slot++) { |
778ca018 NK |
349 | if (r10_bio->devs[slot].bio == bio) |
350 | break; | |
69335ef3 N |
351 | if (r10_bio->devs[slot].repl_bio == bio) { |
352 | repl = 1; | |
353 | break; | |
354 | } | |
355 | } | |
778ca018 NK |
356 | |
357 | BUG_ON(slot == conf->copies); | |
358 | update_head_pos(slot, r10_bio); | |
359 | ||
749c55e9 N |
360 | if (slotp) |
361 | *slotp = slot; | |
69335ef3 N |
362 | if (replp) |
363 | *replp = repl; | |
778ca018 NK |
364 | return r10_bio->devs[slot].devnum; |
365 | } | |
366 | ||
4246a0b6 | 367 | static void raid10_end_read_request(struct bio *bio) |
1da177e4 | 368 | { |
4246a0b6 | 369 | int uptodate = !bio->bi_error; |
9f2c9d12 | 370 | struct r10bio *r10_bio = bio->bi_private; |
1da177e4 | 371 | int slot, dev; |
abbf098e | 372 | struct md_rdev *rdev; |
e879a879 | 373 | struct r10conf *conf = r10_bio->mddev->private; |
1da177e4 | 374 | |
1da177e4 LT |
375 | slot = r10_bio->read_slot; |
376 | dev = r10_bio->devs[slot].devnum; | |
abbf098e | 377 | rdev = r10_bio->devs[slot].rdev; |
1da177e4 LT |
378 | /* |
379 | * this branch is our 'one mirror IO has finished' event handler: | |
380 | */ | |
4443ae10 N |
381 | update_head_pos(slot, r10_bio); |
382 | ||
383 | if (uptodate) { | |
1da177e4 LT |
384 | /* |
385 | * Set R10BIO_Uptodate in our master bio, so that | |
386 | * we will return a good error code to the higher | |
387 | * levels even if IO on some other mirrored buffer fails. | |
388 | * | |
389 | * The 'master' represents the composite IO operation to | |
390 | * user-side. So if something waits for IO, then it will | |
391 | * wait for the 'master' bio. | |
392 | */ | |
393 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
fae8cc5e N |
394 | } else { |
395 | /* If all other devices that store this block have | |
396 | * failed, we want to return the error upwards rather | |
397 | * than fail the last device. Here we redefine | |
398 | * "uptodate" to mean "Don't want to retry" | |
399 | */ | |
635f6416 N |
400 | if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), |
401 | rdev->raid_disk)) | |
fae8cc5e | 402 | uptodate = 1; |
fae8cc5e N |
403 | } |
404 | if (uptodate) { | |
1da177e4 | 405 | raid_end_bio_io(r10_bio); |
abbf098e | 406 | rdev_dec_pending(rdev, conf->mddev); |
4443ae10 | 407 | } else { |
1da177e4 | 408 | /* |
7c4e06ff | 409 | * oops, read error - keep the refcount on the rdev |
1da177e4 LT |
410 | */ |
411 | char b[BDEVNAME_SIZE]; | |
08464e09 | 412 | pr_err_ratelimited("md/raid10:%s: %s: rescheduling sector %llu\n", |
8bda470e | 413 | mdname(conf->mddev), |
abbf098e | 414 | bdevname(rdev->bdev, b), |
8bda470e | 415 | (unsigned long long)r10_bio->sector); |
856e08e2 | 416 | set_bit(R10BIO_ReadError, &r10_bio->state); |
1da177e4 LT |
417 | reschedule_retry(r10_bio); |
418 | } | |
1da177e4 LT |
419 | } |
420 | ||
9f2c9d12 | 421 | static void close_write(struct r10bio *r10_bio) |
bd870a16 N |
422 | { |
423 | /* clear the bitmap if all writes complete successfully */ | |
424 | bitmap_endwrite(r10_bio->mddev->bitmap, r10_bio->sector, | |
425 | r10_bio->sectors, | |
426 | !test_bit(R10BIO_Degraded, &r10_bio->state), | |
427 | 0); | |
428 | md_write_end(r10_bio->mddev); | |
429 | } | |
430 | ||
9f2c9d12 | 431 | static void one_write_done(struct r10bio *r10_bio) |
19d5f834 N |
432 | { |
433 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
434 | if (test_bit(R10BIO_WriteError, &r10_bio->state)) | |
435 | reschedule_retry(r10_bio); | |
436 | else { | |
437 | close_write(r10_bio); | |
438 | if (test_bit(R10BIO_MadeGood, &r10_bio->state)) | |
439 | reschedule_retry(r10_bio); | |
440 | else | |
441 | raid_end_bio_io(r10_bio); | |
442 | } | |
443 | } | |
444 | } | |
445 | ||
4246a0b6 | 446 | static void raid10_end_write_request(struct bio *bio) |
1da177e4 | 447 | { |
9f2c9d12 | 448 | struct r10bio *r10_bio = bio->bi_private; |
778ca018 | 449 | int dev; |
749c55e9 | 450 | int dec_rdev = 1; |
e879a879 | 451 | struct r10conf *conf = r10_bio->mddev->private; |
475b0321 | 452 | int slot, repl; |
4ca40c2c | 453 | struct md_rdev *rdev = NULL; |
1919cbb2 | 454 | struct bio *to_put = NULL; |
579ed34f SL |
455 | bool discard_error; |
456 | ||
457 | discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD; | |
1da177e4 | 458 | |
475b0321 | 459 | dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
1da177e4 | 460 | |
475b0321 N |
461 | if (repl) |
462 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c N |
463 | if (!rdev) { |
464 | smp_rmb(); | |
465 | repl = 0; | |
475b0321 | 466 | rdev = conf->mirrors[dev].rdev; |
4ca40c2c | 467 | } |
1da177e4 LT |
468 | /* |
469 | * this branch is our 'one mirror IO has finished' event handler: | |
470 | */ | |
579ed34f | 471 | if (bio->bi_error && !discard_error) { |
475b0321 N |
472 | if (repl) |
473 | /* Never record new bad blocks to replacement, | |
474 | * just fail it. | |
475 | */ | |
476 | md_error(rdev->mddev, rdev); | |
477 | else { | |
478 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
479 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
480 | set_bit(MD_RECOVERY_NEEDED, | |
481 | &rdev->mddev->recovery); | |
1919cbb2 | 482 | |
475b0321 | 483 | dec_rdev = 0; |
1919cbb2 N |
484 | if (test_bit(FailFast, &rdev->flags) && |
485 | (bio->bi_opf & MD_FAILFAST)) { | |
486 | md_error(rdev->mddev, rdev); | |
487 | if (!test_bit(Faulty, &rdev->flags)) | |
488 | /* This is the only remaining device, | |
489 | * We need to retry the write without | |
490 | * FailFast | |
491 | */ | |
492 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
493 | else { | |
494 | r10_bio->devs[slot].bio = NULL; | |
495 | to_put = bio; | |
496 | dec_rdev = 1; | |
497 | } | |
498 | } else | |
499 | set_bit(R10BIO_WriteError, &r10_bio->state); | |
475b0321 | 500 | } |
749c55e9 | 501 | } else { |
1da177e4 LT |
502 | /* |
503 | * Set R10BIO_Uptodate in our master bio, so that | |
504 | * we will return a good error code for to the higher | |
505 | * levels even if IO on some other mirrored buffer fails. | |
506 | * | |
507 | * The 'master' represents the composite IO operation to | |
508 | * user-side. So if something waits for IO, then it will | |
509 | * wait for the 'master' bio. | |
510 | */ | |
749c55e9 N |
511 | sector_t first_bad; |
512 | int bad_sectors; | |
513 | ||
3056e3ae AL |
514 | /* |
515 | * Do not set R10BIO_Uptodate if the current device is | |
516 | * rebuilding or Faulty. This is because we cannot use | |
517 | * such device for properly reading the data back (we could | |
518 | * potentially use it, if the current write would have felt | |
519 | * before rdev->recovery_offset, but for simplicity we don't | |
520 | * check this here. | |
521 | */ | |
522 | if (test_bit(In_sync, &rdev->flags) && | |
523 | !test_bit(Faulty, &rdev->flags)) | |
524 | set_bit(R10BIO_Uptodate, &r10_bio->state); | |
1da177e4 | 525 | |
749c55e9 | 526 | /* Maybe we can clear some bad blocks. */ |
475b0321 | 527 | if (is_badblock(rdev, |
749c55e9 N |
528 | r10_bio->devs[slot].addr, |
529 | r10_bio->sectors, | |
579ed34f | 530 | &first_bad, &bad_sectors) && !discard_error) { |
749c55e9 | 531 | bio_put(bio); |
475b0321 N |
532 | if (repl) |
533 | r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; | |
534 | else | |
535 | r10_bio->devs[slot].bio = IO_MADE_GOOD; | |
749c55e9 N |
536 | dec_rdev = 0; |
537 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
538 | } | |
539 | } | |
540 | ||
1da177e4 LT |
541 | /* |
542 | * | |
543 | * Let's see if all mirrored write operations have finished | |
544 | * already. | |
545 | */ | |
19d5f834 | 546 | one_write_done(r10_bio); |
749c55e9 | 547 | if (dec_rdev) |
884162df | 548 | rdev_dec_pending(rdev, conf->mddev); |
1919cbb2 N |
549 | if (to_put) |
550 | bio_put(to_put); | |
1da177e4 LT |
551 | } |
552 | ||
1da177e4 LT |
553 | /* |
554 | * RAID10 layout manager | |
25985edc | 555 | * As well as the chunksize and raid_disks count, there are two |
1da177e4 LT |
556 | * parameters: near_copies and far_copies. |
557 | * near_copies * far_copies must be <= raid_disks. | |
558 | * Normally one of these will be 1. | |
559 | * If both are 1, we get raid0. | |
560 | * If near_copies == raid_disks, we get raid1. | |
561 | * | |
25985edc | 562 | * Chunks are laid out in raid0 style with near_copies copies of the |
1da177e4 LT |
563 | * first chunk, followed by near_copies copies of the next chunk and |
564 | * so on. | |
565 | * If far_copies > 1, then after 1/far_copies of the array has been assigned | |
566 | * as described above, we start again with a device offset of near_copies. | |
567 | * So we effectively have another copy of the whole array further down all | |
568 | * the drives, but with blocks on different drives. | |
569 | * With this layout, and block is never stored twice on the one device. | |
570 | * | |
571 | * raid10_find_phys finds the sector offset of a given virtual sector | |
c93983bf | 572 | * on each device that it is on. |
1da177e4 LT |
573 | * |
574 | * raid10_find_virt does the reverse mapping, from a device and a | |
575 | * sector offset to a virtual address | |
576 | */ | |
577 | ||
f8c9e74f | 578 | static void __raid10_find_phys(struct geom *geo, struct r10bio *r10bio) |
1da177e4 LT |
579 | { |
580 | int n,f; | |
581 | sector_t sector; | |
582 | sector_t chunk; | |
583 | sector_t stripe; | |
584 | int dev; | |
1da177e4 | 585 | int slot = 0; |
9a3152ab JB |
586 | int last_far_set_start, last_far_set_size; |
587 | ||
588 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
589 | last_far_set_start *= geo->far_set_size; | |
590 | ||
591 | last_far_set_size = geo->far_set_size; | |
592 | last_far_set_size += (geo->raid_disks % geo->far_set_size); | |
1da177e4 LT |
593 | |
594 | /* now calculate first sector/dev */ | |
5cf00fcd N |
595 | chunk = r10bio->sector >> geo->chunk_shift; |
596 | sector = r10bio->sector & geo->chunk_mask; | |
1da177e4 | 597 | |
5cf00fcd | 598 | chunk *= geo->near_copies; |
1da177e4 | 599 | stripe = chunk; |
5cf00fcd N |
600 | dev = sector_div(stripe, geo->raid_disks); |
601 | if (geo->far_offset) | |
602 | stripe *= geo->far_copies; | |
1da177e4 | 603 | |
5cf00fcd | 604 | sector += stripe << geo->chunk_shift; |
1da177e4 LT |
605 | |
606 | /* and calculate all the others */ | |
5cf00fcd | 607 | for (n = 0; n < geo->near_copies; n++) { |
1da177e4 | 608 | int d = dev; |
475901af | 609 | int set; |
1da177e4 | 610 | sector_t s = sector; |
1da177e4 | 611 | r10bio->devs[slot].devnum = d; |
4c0ca26b | 612 | r10bio->devs[slot].addr = s; |
1da177e4 LT |
613 | slot++; |
614 | ||
5cf00fcd | 615 | for (f = 1; f < geo->far_copies; f++) { |
475901af | 616 | set = d / geo->far_set_size; |
5cf00fcd | 617 | d += geo->near_copies; |
475901af | 618 | |
9a3152ab JB |
619 | if ((geo->raid_disks % geo->far_set_size) && |
620 | (d > last_far_set_start)) { | |
621 | d -= last_far_set_start; | |
622 | d %= last_far_set_size; | |
623 | d += last_far_set_start; | |
624 | } else { | |
625 | d %= geo->far_set_size; | |
626 | d += geo->far_set_size * set; | |
627 | } | |
5cf00fcd | 628 | s += geo->stride; |
1da177e4 LT |
629 | r10bio->devs[slot].devnum = d; |
630 | r10bio->devs[slot].addr = s; | |
631 | slot++; | |
632 | } | |
633 | dev++; | |
5cf00fcd | 634 | if (dev >= geo->raid_disks) { |
1da177e4 | 635 | dev = 0; |
5cf00fcd | 636 | sector += (geo->chunk_mask + 1); |
1da177e4 LT |
637 | } |
638 | } | |
f8c9e74f N |
639 | } |
640 | ||
641 | static void raid10_find_phys(struct r10conf *conf, struct r10bio *r10bio) | |
642 | { | |
643 | struct geom *geo = &conf->geo; | |
644 | ||
645 | if (conf->reshape_progress != MaxSector && | |
646 | ((r10bio->sector >= conf->reshape_progress) != | |
647 | conf->mddev->reshape_backwards)) { | |
648 | set_bit(R10BIO_Previous, &r10bio->state); | |
649 | geo = &conf->prev; | |
650 | } else | |
651 | clear_bit(R10BIO_Previous, &r10bio->state); | |
652 | ||
653 | __raid10_find_phys(geo, r10bio); | |
1da177e4 LT |
654 | } |
655 | ||
e879a879 | 656 | static sector_t raid10_find_virt(struct r10conf *conf, sector_t sector, int dev) |
1da177e4 LT |
657 | { |
658 | sector_t offset, chunk, vchunk; | |
f8c9e74f N |
659 | /* Never use conf->prev as this is only called during resync |
660 | * or recovery, so reshape isn't happening | |
661 | */ | |
5cf00fcd | 662 | struct geom *geo = &conf->geo; |
475901af JB |
663 | int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; |
664 | int far_set_size = geo->far_set_size; | |
9a3152ab JB |
665 | int last_far_set_start; |
666 | ||
667 | if (geo->raid_disks % geo->far_set_size) { | |
668 | last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; | |
669 | last_far_set_start *= geo->far_set_size; | |
670 | ||
671 | if (dev >= last_far_set_start) { | |
672 | far_set_size = geo->far_set_size; | |
673 | far_set_size += (geo->raid_disks % geo->far_set_size); | |
674 | far_set_start = last_far_set_start; | |
675 | } | |
676 | } | |
1da177e4 | 677 | |
5cf00fcd N |
678 | offset = sector & geo->chunk_mask; |
679 | if (geo->far_offset) { | |
c93983bf | 680 | int fc; |
5cf00fcd N |
681 | chunk = sector >> geo->chunk_shift; |
682 | fc = sector_div(chunk, geo->far_copies); | |
683 | dev -= fc * geo->near_copies; | |
475901af JB |
684 | if (dev < far_set_start) |
685 | dev += far_set_size; | |
c93983bf | 686 | } else { |
5cf00fcd N |
687 | while (sector >= geo->stride) { |
688 | sector -= geo->stride; | |
475901af JB |
689 | if (dev < (geo->near_copies + far_set_start)) |
690 | dev += far_set_size - geo->near_copies; | |
c93983bf | 691 | else |
5cf00fcd | 692 | dev -= geo->near_copies; |
c93983bf | 693 | } |
5cf00fcd | 694 | chunk = sector >> geo->chunk_shift; |
c93983bf | 695 | } |
5cf00fcd N |
696 | vchunk = chunk * geo->raid_disks + dev; |
697 | sector_div(vchunk, geo->near_copies); | |
698 | return (vchunk << geo->chunk_shift) + offset; | |
1da177e4 LT |
699 | } |
700 | ||
1da177e4 LT |
701 | /* |
702 | * This routine returns the disk from which the requested read should | |
703 | * be done. There is a per-array 'next expected sequential IO' sector | |
704 | * number - if this matches on the next IO then we use the last disk. | |
705 | * There is also a per-disk 'last know head position' sector that is | |
706 | * maintained from IRQ contexts, both the normal and the resync IO | |
707 | * completion handlers update this position correctly. If there is no | |
708 | * perfect sequential match then we pick the disk whose head is closest. | |
709 | * | |
710 | * If there are 2 mirrors in the same 2 devices, performance degrades | |
711 | * because position is mirror, not device based. | |
712 | * | |
713 | * The rdev for the device selected will have nr_pending incremented. | |
714 | */ | |
715 | ||
716 | /* | |
717 | * FIXME: possibly should rethink readbalancing and do it differently | |
718 | * depending on near_copies / far_copies geometry. | |
719 | */ | |
96c3fd1f N |
720 | static struct md_rdev *read_balance(struct r10conf *conf, |
721 | struct r10bio *r10_bio, | |
722 | int *max_sectors) | |
1da177e4 | 723 | { |
af3a2cd6 | 724 | const sector_t this_sector = r10_bio->sector; |
56d99121 | 725 | int disk, slot; |
856e08e2 N |
726 | int sectors = r10_bio->sectors; |
727 | int best_good_sectors; | |
56d99121 | 728 | sector_t new_distance, best_dist; |
3bbae04b | 729 | struct md_rdev *best_rdev, *rdev = NULL; |
56d99121 N |
730 | int do_balance; |
731 | int best_slot; | |
5cf00fcd | 732 | struct geom *geo = &conf->geo; |
1da177e4 LT |
733 | |
734 | raid10_find_phys(conf, r10_bio); | |
735 | rcu_read_lock(); | |
856e08e2 | 736 | sectors = r10_bio->sectors; |
56d99121 | 737 | best_slot = -1; |
abbf098e | 738 | best_rdev = NULL; |
56d99121 | 739 | best_dist = MaxSector; |
856e08e2 | 740 | best_good_sectors = 0; |
56d99121 | 741 | do_balance = 1; |
8d3ca83d | 742 | clear_bit(R10BIO_FailFast, &r10_bio->state); |
1da177e4 LT |
743 | /* |
744 | * Check if we can balance. We can balance on the whole | |
6cce3b23 N |
745 | * device if no resync is going on (recovery is ok), or below |
746 | * the resync window. We take the first readable disk when | |
747 | * above the resync window. | |
1da177e4 LT |
748 | */ |
749 | if (conf->mddev->recovery_cp < MaxSector | |
56d99121 N |
750 | && (this_sector + sectors >= conf->next_resync)) |
751 | do_balance = 0; | |
1da177e4 | 752 | |
56d99121 | 753 | for (slot = 0; slot < conf->copies ; slot++) { |
856e08e2 N |
754 | sector_t first_bad; |
755 | int bad_sectors; | |
756 | sector_t dev_sector; | |
757 | ||
56d99121 N |
758 | if (r10_bio->devs[slot].bio == IO_BLOCKED) |
759 | continue; | |
1da177e4 | 760 | disk = r10_bio->devs[slot].devnum; |
abbf098e N |
761 | rdev = rcu_dereference(conf->mirrors[disk].replacement); |
762 | if (rdev == NULL || test_bit(Faulty, &rdev->flags) || | |
763 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
764 | rdev = rcu_dereference(conf->mirrors[disk].rdev); | |
050b6615 | 765 | if (rdev == NULL || |
8ae12666 | 766 | test_bit(Faulty, &rdev->flags)) |
abbf098e N |
767 | continue; |
768 | if (!test_bit(In_sync, &rdev->flags) && | |
769 | r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) | |
56d99121 N |
770 | continue; |
771 | ||
856e08e2 N |
772 | dev_sector = r10_bio->devs[slot].addr; |
773 | if (is_badblock(rdev, dev_sector, sectors, | |
774 | &first_bad, &bad_sectors)) { | |
775 | if (best_dist < MaxSector) | |
776 | /* Already have a better slot */ | |
777 | continue; | |
778 | if (first_bad <= dev_sector) { | |
779 | /* Cannot read here. If this is the | |
780 | * 'primary' device, then we must not read | |
781 | * beyond 'bad_sectors' from another device. | |
782 | */ | |
783 | bad_sectors -= (dev_sector - first_bad); | |
784 | if (!do_balance && sectors > bad_sectors) | |
785 | sectors = bad_sectors; | |
786 | if (best_good_sectors > sectors) | |
787 | best_good_sectors = sectors; | |
788 | } else { | |
789 | sector_t good_sectors = | |
790 | first_bad - dev_sector; | |
791 | if (good_sectors > best_good_sectors) { | |
792 | best_good_sectors = good_sectors; | |
793 | best_slot = slot; | |
abbf098e | 794 | best_rdev = rdev; |
856e08e2 N |
795 | } |
796 | if (!do_balance) | |
797 | /* Must read from here */ | |
798 | break; | |
799 | } | |
800 | continue; | |
801 | } else | |
802 | best_good_sectors = sectors; | |
803 | ||
56d99121 N |
804 | if (!do_balance) |
805 | break; | |
1da177e4 | 806 | |
8d3ca83d N |
807 | if (best_slot >= 0) |
808 | /* At least 2 disks to choose from so failfast is OK */ | |
809 | set_bit(R10BIO_FailFast, &r10_bio->state); | |
22dfdf52 N |
810 | /* This optimisation is debatable, and completely destroys |
811 | * sequential read speed for 'far copies' arrays. So only | |
812 | * keep it for 'near' arrays, and review those later. | |
813 | */ | |
5cf00fcd | 814 | if (geo->near_copies > 1 && !atomic_read(&rdev->nr_pending)) |
8d3ca83d | 815 | new_distance = 0; |
8ed3a195 KS |
816 | |
817 | /* for far > 1 always use the lowest address */ | |
8d3ca83d | 818 | else if (geo->far_copies > 1) |
56d99121 | 819 | new_distance = r10_bio->devs[slot].addr; |
8ed3a195 | 820 | else |
56d99121 N |
821 | new_distance = abs(r10_bio->devs[slot].addr - |
822 | conf->mirrors[disk].head_position); | |
823 | if (new_distance < best_dist) { | |
824 | best_dist = new_distance; | |
825 | best_slot = slot; | |
abbf098e | 826 | best_rdev = rdev; |
1da177e4 LT |
827 | } |
828 | } | |
abbf098e | 829 | if (slot >= conf->copies) { |
56d99121 | 830 | slot = best_slot; |
abbf098e N |
831 | rdev = best_rdev; |
832 | } | |
1da177e4 | 833 | |
56d99121 | 834 | if (slot >= 0) { |
56d99121 | 835 | atomic_inc(&rdev->nr_pending); |
56d99121 N |
836 | r10_bio->read_slot = slot; |
837 | } else | |
96c3fd1f | 838 | rdev = NULL; |
1da177e4 | 839 | rcu_read_unlock(); |
856e08e2 | 840 | *max_sectors = best_good_sectors; |
1da177e4 | 841 | |
96c3fd1f | 842 | return rdev; |
1da177e4 LT |
843 | } |
844 | ||
5c675f83 | 845 | static int raid10_congested(struct mddev *mddev, int bits) |
0d129228 | 846 | { |
e879a879 | 847 | struct r10conf *conf = mddev->private; |
0d129228 N |
848 | int i, ret = 0; |
849 | ||
4452226e | 850 | if ((bits & (1 << WB_async_congested)) && |
34db0cd6 N |
851 | conf->pending_count >= max_queued_requests) |
852 | return 1; | |
853 | ||
0d129228 | 854 | rcu_read_lock(); |
f8c9e74f N |
855 | for (i = 0; |
856 | (i < conf->geo.raid_disks || i < conf->prev.raid_disks) | |
857 | && ret == 0; | |
858 | i++) { | |
3cb03002 | 859 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); |
0d129228 | 860 | if (rdev && !test_bit(Faulty, &rdev->flags)) { |
165125e1 | 861 | struct request_queue *q = bdev_get_queue(rdev->bdev); |
0d129228 | 862 | |
dc3b17cc | 863 | ret |= bdi_congested(q->backing_dev_info, bits); |
0d129228 N |
864 | } |
865 | } | |
866 | rcu_read_unlock(); | |
867 | return ret; | |
868 | } | |
869 | ||
e879a879 | 870 | static void flush_pending_writes(struct r10conf *conf) |
a35e63ef N |
871 | { |
872 | /* Any writes that have been queued but are awaiting | |
873 | * bitmap updates get flushed here. | |
a35e63ef | 874 | */ |
a35e63ef N |
875 | spin_lock_irq(&conf->device_lock); |
876 | ||
877 | if (conf->pending_bio_list.head) { | |
878 | struct bio *bio; | |
879 | bio = bio_list_get(&conf->pending_bio_list); | |
34db0cd6 | 880 | conf->pending_count = 0; |
a35e63ef N |
881 | spin_unlock_irq(&conf->device_lock); |
882 | /* flush any pending bitmap writes to disk | |
883 | * before proceeding w/ I/O */ | |
884 | bitmap_unplug(conf->mddev->bitmap); | |
34db0cd6 | 885 | wake_up(&conf->wait_barrier); |
a35e63ef N |
886 | |
887 | while (bio) { /* submit pending writes */ | |
888 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 889 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
a35e63ef | 890 | bio->bi_next = NULL; |
a9ae93c8 N |
891 | bio->bi_bdev = rdev->bdev; |
892 | if (test_bit(Faulty, &rdev->flags)) { | |
893 | bio->bi_error = -EIO; | |
894 | bio_endio(bio); | |
895 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
896 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
532a2a3f | 897 | /* Just ignore it */ |
4246a0b6 | 898 | bio_endio(bio); |
532a2a3f SL |
899 | else |
900 | generic_make_request(bio); | |
a35e63ef N |
901 | bio = next; |
902 | } | |
a35e63ef N |
903 | } else |
904 | spin_unlock_irq(&conf->device_lock); | |
a35e63ef | 905 | } |
7eaceacc | 906 | |
0a27ec96 N |
907 | /* Barriers.... |
908 | * Sometimes we need to suspend IO while we do something else, | |
909 | * either some resync/recovery, or reconfigure the array. | |
910 | * To do this we raise a 'barrier'. | |
911 | * The 'barrier' is a counter that can be raised multiple times | |
912 | * to count how many activities are happening which preclude | |
913 | * normal IO. | |
914 | * We can only raise the barrier if there is no pending IO. | |
915 | * i.e. if nr_pending == 0. | |
916 | * We choose only to raise the barrier if no-one is waiting for the | |
917 | * barrier to go down. This means that as soon as an IO request | |
918 | * is ready, no other operations which require a barrier will start | |
919 | * until the IO request has had a chance. | |
920 | * | |
921 | * So: regular IO calls 'wait_barrier'. When that returns there | |
922 | * is no backgroup IO happening, It must arrange to call | |
923 | * allow_barrier when it has finished its IO. | |
924 | * backgroup IO calls must call raise_barrier. Once that returns | |
925 | * there is no normal IO happeing. It must arrange to call | |
926 | * lower_barrier when the particular background IO completes. | |
1da177e4 | 927 | */ |
1da177e4 | 928 | |
e879a879 | 929 | static void raise_barrier(struct r10conf *conf, int force) |
1da177e4 | 930 | { |
6cce3b23 | 931 | BUG_ON(force && !conf->barrier); |
1da177e4 | 932 | spin_lock_irq(&conf->resync_lock); |
0a27ec96 | 933 | |
6cce3b23 N |
934 | /* Wait until no block IO is waiting (unless 'force') */ |
935 | wait_event_lock_irq(conf->wait_barrier, force || !conf->nr_waiting, | |
eed8c02e | 936 | conf->resync_lock); |
0a27ec96 N |
937 | |
938 | /* block any new IO from starting */ | |
939 | conf->barrier++; | |
940 | ||
c3b328ac | 941 | /* Now wait for all pending IO to complete */ |
0a27ec96 | 942 | wait_event_lock_irq(conf->wait_barrier, |
0e5313e2 | 943 | !atomic_read(&conf->nr_pending) && conf->barrier < RESYNC_DEPTH, |
eed8c02e | 944 | conf->resync_lock); |
0a27ec96 N |
945 | |
946 | spin_unlock_irq(&conf->resync_lock); | |
947 | } | |
948 | ||
e879a879 | 949 | static void lower_barrier(struct r10conf *conf) |
0a27ec96 N |
950 | { |
951 | unsigned long flags; | |
952 | spin_lock_irqsave(&conf->resync_lock, flags); | |
953 | conf->barrier--; | |
954 | spin_unlock_irqrestore(&conf->resync_lock, flags); | |
955 | wake_up(&conf->wait_barrier); | |
956 | } | |
957 | ||
e879a879 | 958 | static void wait_barrier(struct r10conf *conf) |
0a27ec96 N |
959 | { |
960 | spin_lock_irq(&conf->resync_lock); | |
961 | if (conf->barrier) { | |
962 | conf->nr_waiting++; | |
d6b42dcb N |
963 | /* Wait for the barrier to drop. |
964 | * However if there are already pending | |
965 | * requests (preventing the barrier from | |
966 | * rising completely), and the | |
967 | * pre-process bio queue isn't empty, | |
968 | * then don't wait, as we need to empty | |
969 | * that queue to get the nr_pending | |
970 | * count down. | |
971 | */ | |
578b54ad | 972 | raid10_log(conf->mddev, "wait barrier"); |
d6b42dcb N |
973 | wait_event_lock_irq(conf->wait_barrier, |
974 | !conf->barrier || | |
0e5313e2 | 975 | (atomic_read(&conf->nr_pending) && |
d6b42dcb | 976 | current->bio_list && |
f5fe1b51 N |
977 | (!bio_list_empty(¤t->bio_list[0]) || |
978 | !bio_list_empty(¤t->bio_list[1]))), | |
eed8c02e | 979 | conf->resync_lock); |
0a27ec96 | 980 | conf->nr_waiting--; |
0e5313e2 TM |
981 | if (!conf->nr_waiting) |
982 | wake_up(&conf->wait_barrier); | |
1da177e4 | 983 | } |
0e5313e2 | 984 | atomic_inc(&conf->nr_pending); |
1da177e4 LT |
985 | spin_unlock_irq(&conf->resync_lock); |
986 | } | |
987 | ||
e879a879 | 988 | static void allow_barrier(struct r10conf *conf) |
0a27ec96 | 989 | { |
0e5313e2 TM |
990 | if ((atomic_dec_and_test(&conf->nr_pending)) || |
991 | (conf->array_freeze_pending)) | |
992 | wake_up(&conf->wait_barrier); | |
0a27ec96 N |
993 | } |
994 | ||
e2d59925 | 995 | static void freeze_array(struct r10conf *conf, int extra) |
4443ae10 N |
996 | { |
997 | /* stop syncio and normal IO and wait for everything to | |
f188593e | 998 | * go quiet. |
4443ae10 | 999 | * We increment barrier and nr_waiting, and then |
e2d59925 | 1000 | * wait until nr_pending match nr_queued+extra |
1c830532 N |
1001 | * This is called in the context of one normal IO request |
1002 | * that has failed. Thus any sync request that might be pending | |
1003 | * will be blocked by nr_pending, and we need to wait for | |
1004 | * pending IO requests to complete or be queued for re-try. | |
e2d59925 | 1005 | * Thus the number queued (nr_queued) plus this request (extra) |
1c830532 N |
1006 | * must match the number of pending IOs (nr_pending) before |
1007 | * we continue. | |
4443ae10 N |
1008 | */ |
1009 | spin_lock_irq(&conf->resync_lock); | |
0e5313e2 | 1010 | conf->array_freeze_pending++; |
4443ae10 N |
1011 | conf->barrier++; |
1012 | conf->nr_waiting++; | |
eed8c02e | 1013 | wait_event_lock_irq_cmd(conf->wait_barrier, |
0e5313e2 | 1014 | atomic_read(&conf->nr_pending) == conf->nr_queued+extra, |
eed8c02e LC |
1015 | conf->resync_lock, |
1016 | flush_pending_writes(conf)); | |
c3b328ac | 1017 | |
0e5313e2 | 1018 | conf->array_freeze_pending--; |
4443ae10 N |
1019 | spin_unlock_irq(&conf->resync_lock); |
1020 | } | |
1021 | ||
e879a879 | 1022 | static void unfreeze_array(struct r10conf *conf) |
4443ae10 N |
1023 | { |
1024 | /* reverse the effect of the freeze */ | |
1025 | spin_lock_irq(&conf->resync_lock); | |
1026 | conf->barrier--; | |
1027 | conf->nr_waiting--; | |
1028 | wake_up(&conf->wait_barrier); | |
1029 | spin_unlock_irq(&conf->resync_lock); | |
1030 | } | |
1031 | ||
f8c9e74f N |
1032 | static sector_t choose_data_offset(struct r10bio *r10_bio, |
1033 | struct md_rdev *rdev) | |
1034 | { | |
1035 | if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || | |
1036 | test_bit(R10BIO_Previous, &r10_bio->state)) | |
1037 | return rdev->data_offset; | |
1038 | else | |
1039 | return rdev->new_data_offset; | |
1040 | } | |
1041 | ||
57c67df4 N |
1042 | struct raid10_plug_cb { |
1043 | struct blk_plug_cb cb; | |
1044 | struct bio_list pending; | |
1045 | int pending_cnt; | |
1046 | }; | |
1047 | ||
1048 | static void raid10_unplug(struct blk_plug_cb *cb, bool from_schedule) | |
1049 | { | |
1050 | struct raid10_plug_cb *plug = container_of(cb, struct raid10_plug_cb, | |
1051 | cb); | |
1052 | struct mddev *mddev = plug->cb.data; | |
1053 | struct r10conf *conf = mddev->private; | |
1054 | struct bio *bio; | |
1055 | ||
874807a8 | 1056 | if (from_schedule || current->bio_list) { |
57c67df4 N |
1057 | spin_lock_irq(&conf->device_lock); |
1058 | bio_list_merge(&conf->pending_bio_list, &plug->pending); | |
1059 | conf->pending_count += plug->pending_cnt; | |
1060 | spin_unlock_irq(&conf->device_lock); | |
ee0b0244 | 1061 | wake_up(&conf->wait_barrier); |
57c67df4 N |
1062 | md_wakeup_thread(mddev->thread); |
1063 | kfree(plug); | |
1064 | return; | |
1065 | } | |
1066 | ||
1067 | /* we aren't scheduling, so we can do the write-out directly. */ | |
1068 | bio = bio_list_get(&plug->pending); | |
1069 | bitmap_unplug(mddev->bitmap); | |
1070 | wake_up(&conf->wait_barrier); | |
1071 | ||
1072 | while (bio) { /* submit pending writes */ | |
1073 | struct bio *next = bio->bi_next; | |
a9ae93c8 | 1074 | struct md_rdev *rdev = (void*)bio->bi_bdev; |
57c67df4 | 1075 | bio->bi_next = NULL; |
a9ae93c8 N |
1076 | bio->bi_bdev = rdev->bdev; |
1077 | if (test_bit(Faulty, &rdev->flags)) { | |
1078 | bio->bi_error = -EIO; | |
1079 | bio_endio(bio); | |
1080 | } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
1081 | !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) | |
32f9f570 | 1082 | /* Just ignore it */ |
4246a0b6 | 1083 | bio_endio(bio); |
32f9f570 SL |
1084 | else |
1085 | generic_make_request(bio); | |
57c67df4 N |
1086 | bio = next; |
1087 | } | |
1088 | kfree(plug); | |
1089 | } | |
1090 | ||
bb5f1ed7 RL |
1091 | static void raid10_read_request(struct mddev *mddev, struct bio *bio, |
1092 | struct r10bio *r10_bio) | |
1da177e4 | 1093 | { |
e879a879 | 1094 | struct r10conf *conf = mddev->private; |
1da177e4 | 1095 | struct bio *read_bio; |
bb5f1ed7 RL |
1096 | const int op = bio_op(bio); |
1097 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); | |
1098 | int sectors_handled; | |
1099 | int max_sectors; | |
1100 | sector_t sectors; | |
1101 | struct md_rdev *rdev; | |
1102 | int slot; | |
1103 | ||
1104 | /* | |
1105 | * Register the new request and wait if the reconstruction | |
1106 | * thread has put up a bar for new requests. | |
1107 | * Continue immediately if no resync is active currently. | |
1108 | */ | |
1109 | wait_barrier(conf); | |
1110 | ||
1111 | sectors = bio_sectors(bio); | |
1112 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && | |
1113 | bio->bi_iter.bi_sector < conf->reshape_progress && | |
1114 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
1115 | /* | |
1116 | * IO spans the reshape position. Need to wait for reshape to | |
1117 | * pass | |
1118 | */ | |
1119 | raid10_log(conf->mddev, "wait reshape"); | |
1120 | allow_barrier(conf); | |
1121 | wait_event(conf->wait_barrier, | |
1122 | conf->reshape_progress <= bio->bi_iter.bi_sector || | |
1123 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1124 | sectors); | |
1125 | wait_barrier(conf); | |
1126 | } | |
1127 | ||
1128 | read_again: | |
1129 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
1130 | if (!rdev) { | |
1131 | raid_end_bio_io(r10_bio); | |
1132 | return; | |
1133 | } | |
1134 | slot = r10_bio->read_slot; | |
1135 | ||
d7a10308 | 1136 | read_bio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
bb5f1ed7 RL |
1137 | bio_trim(read_bio, r10_bio->sector - bio->bi_iter.bi_sector, |
1138 | max_sectors); | |
1139 | ||
1140 | r10_bio->devs[slot].bio = read_bio; | |
1141 | r10_bio->devs[slot].rdev = rdev; | |
1142 | ||
1143 | read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + | |
1144 | choose_data_offset(r10_bio, rdev); | |
1145 | read_bio->bi_bdev = rdev->bdev; | |
1146 | read_bio->bi_end_io = raid10_end_read_request; | |
1147 | bio_set_op_attrs(read_bio, op, do_sync); | |
1148 | if (test_bit(FailFast, &rdev->flags) && | |
1149 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
1150 | read_bio->bi_opf |= MD_FAILFAST; | |
1151 | read_bio->bi_private = r10_bio; | |
1152 | ||
1153 | if (mddev->gendisk) | |
1154 | trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev), | |
1155 | read_bio, disk_devt(mddev->gendisk), | |
1156 | r10_bio->sector); | |
1157 | if (max_sectors < r10_bio->sectors) { | |
1158 | /* | |
1159 | * Could not read all from this device, so we will need another | |
1160 | * r10_bio. | |
1161 | */ | |
1162 | sectors_handled = (r10_bio->sector + max_sectors | |
1163 | - bio->bi_iter.bi_sector); | |
1164 | r10_bio->sectors = max_sectors; | |
1165 | spin_lock_irq(&conf->device_lock); | |
1166 | if (bio->bi_phys_segments == 0) | |
1167 | bio->bi_phys_segments = 2; | |
1168 | else | |
1169 | bio->bi_phys_segments++; | |
1170 | spin_unlock_irq(&conf->device_lock); | |
1171 | /* | |
1172 | * Cannot call generic_make_request directly as that will be | |
1173 | * queued in __generic_make_request and subsequent | |
1174 | * mempool_alloc might block waiting for it. so hand bio over | |
1175 | * to raid10d. | |
1176 | */ | |
1177 | reschedule_retry(r10_bio); | |
1178 | ||
1179 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1180 | ||
1181 | r10_bio->master_bio = bio; | |
1182 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; | |
1183 | r10_bio->state = 0; | |
1184 | r10_bio->mddev = mddev; | |
1185 | r10_bio->sector = bio->bi_iter.bi_sector + sectors_handled; | |
1186 | goto read_again; | |
1187 | } else | |
1188 | generic_make_request(read_bio); | |
1189 | return; | |
1190 | } | |
1191 | ||
1192 | static void raid10_write_request(struct mddev *mddev, struct bio *bio, | |
1193 | struct r10bio *r10_bio) | |
1194 | { | |
1195 | struct r10conf *conf = mddev->private; | |
1da177e4 | 1196 | int i; |
796a5cf0 | 1197 | const int op = bio_op(bio); |
1eff9d32 JA |
1198 | const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); |
1199 | const unsigned long do_fua = (bio->bi_opf & REQ_FUA); | |
6cce3b23 | 1200 | unsigned long flags; |
3cb03002 | 1201 | struct md_rdev *blocked_rdev; |
57c67df4 N |
1202 | struct blk_plug_cb *cb; |
1203 | struct raid10_plug_cb *plug = NULL; | |
bb5f1ed7 | 1204 | sector_t sectors; |
d4432c23 N |
1205 | int sectors_handled; |
1206 | int max_sectors; | |
1da177e4 | 1207 | |
9b622e2b TM |
1208 | md_write_start(mddev, bio); |
1209 | ||
cc13b1d1 N |
1210 | /* |
1211 | * Register the new request and wait if the reconstruction | |
1212 | * thread has put up a bar for new requests. | |
1213 | * Continue immediately if no resync is active currently. | |
1214 | */ | |
1215 | wait_barrier(conf); | |
1216 | ||
aa8b57aa | 1217 | sectors = bio_sectors(bio); |
3ea7daa5 | 1218 | while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
4f024f37 KO |
1219 | bio->bi_iter.bi_sector < conf->reshape_progress && |
1220 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { | |
bb5f1ed7 RL |
1221 | /* |
1222 | * IO spans the reshape position. Need to wait for reshape to | |
1223 | * pass | |
3ea7daa5 | 1224 | */ |
578b54ad | 1225 | raid10_log(conf->mddev, "wait reshape"); |
3ea7daa5 N |
1226 | allow_barrier(conf); |
1227 | wait_event(conf->wait_barrier, | |
4f024f37 KO |
1228 | conf->reshape_progress <= bio->bi_iter.bi_sector || |
1229 | conf->reshape_progress >= bio->bi_iter.bi_sector + | |
1230 | sectors); | |
3ea7daa5 N |
1231 | wait_barrier(conf); |
1232 | } | |
bb5f1ed7 | 1233 | |
3ea7daa5 | 1234 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
3ea7daa5 | 1235 | (mddev->reshape_backwards |
4f024f37 KO |
1236 | ? (bio->bi_iter.bi_sector < conf->reshape_safe && |
1237 | bio->bi_iter.bi_sector + sectors > conf->reshape_progress) | |
1238 | : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && | |
1239 | bio->bi_iter.bi_sector < conf->reshape_progress))) { | |
3ea7daa5 N |
1240 | /* Need to update reshape_position in metadata */ |
1241 | mddev->reshape_position = conf->reshape_progress; | |
2953079c SL |
1242 | set_mask_bits(&mddev->sb_flags, 0, |
1243 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
3ea7daa5 | 1244 | md_wakeup_thread(mddev->thread); |
578b54ad | 1245 | raid10_log(conf->mddev, "wait reshape metadata"); |
3ea7daa5 | 1246 | wait_event(mddev->sb_wait, |
2953079c | 1247 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); |
3ea7daa5 N |
1248 | |
1249 | conf->reshape_safe = mddev->reshape_position; | |
1250 | } | |
1251 | ||
34db0cd6 N |
1252 | if (conf->pending_count >= max_queued_requests) { |
1253 | md_wakeup_thread(mddev->thread); | |
578b54ad | 1254 | raid10_log(mddev, "wait queued"); |
34db0cd6 N |
1255 | wait_event(conf->wait_barrier, |
1256 | conf->pending_count < max_queued_requests); | |
1257 | } | |
6bfe0b49 | 1258 | /* first select target devices under rcu_lock and |
1da177e4 LT |
1259 | * inc refcount on their rdev. Record them by setting |
1260 | * bios[x] to bio | |
d4432c23 N |
1261 | * If there are known/acknowledged bad blocks on any device |
1262 | * on which we have seen a write error, we want to avoid | |
1263 | * writing to those blocks. This potentially requires several | |
1264 | * writes to write around the bad blocks. Each set of writes | |
1265 | * gets its own r10_bio with a set of bios attached. The number | |
1266 | * of r10_bios is recored in bio->bi_phys_segments just as with | |
1267 | * the read case. | |
1da177e4 | 1268 | */ |
c3b328ac | 1269 | |
69335ef3 | 1270 | r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ |
1da177e4 | 1271 | raid10_find_phys(conf, r10_bio); |
d4432c23 | 1272 | retry_write: |
cb6969e8 | 1273 | blocked_rdev = NULL; |
1da177e4 | 1274 | rcu_read_lock(); |
d4432c23 N |
1275 | max_sectors = r10_bio->sectors; |
1276 | ||
1da177e4 LT |
1277 | for (i = 0; i < conf->copies; i++) { |
1278 | int d = r10_bio->devs[i].devnum; | |
3cb03002 | 1279 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
475b0321 N |
1280 | struct md_rdev *rrdev = rcu_dereference( |
1281 | conf->mirrors[d].replacement); | |
4ca40c2c N |
1282 | if (rdev == rrdev) |
1283 | rrdev = NULL; | |
6bfe0b49 DW |
1284 | if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { |
1285 | atomic_inc(&rdev->nr_pending); | |
1286 | blocked_rdev = rdev; | |
1287 | break; | |
1288 | } | |
475b0321 N |
1289 | if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { |
1290 | atomic_inc(&rrdev->nr_pending); | |
1291 | blocked_rdev = rrdev; | |
1292 | break; | |
1293 | } | |
8ae12666 | 1294 | if (rdev && (test_bit(Faulty, &rdev->flags))) |
e7c0c3fa | 1295 | rdev = NULL; |
8ae12666 | 1296 | if (rrdev && (test_bit(Faulty, &rrdev->flags))) |
475b0321 N |
1297 | rrdev = NULL; |
1298 | ||
d4432c23 | 1299 | r10_bio->devs[i].bio = NULL; |
475b0321 | 1300 | r10_bio->devs[i].repl_bio = NULL; |
e7c0c3fa N |
1301 | |
1302 | if (!rdev && !rrdev) { | |
6cce3b23 | 1303 | set_bit(R10BIO_Degraded, &r10_bio->state); |
d4432c23 N |
1304 | continue; |
1305 | } | |
e7c0c3fa | 1306 | if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { |
d4432c23 N |
1307 | sector_t first_bad; |
1308 | sector_t dev_sector = r10_bio->devs[i].addr; | |
1309 | int bad_sectors; | |
1310 | int is_bad; | |
1311 | ||
bb5f1ed7 | 1312 | is_bad = is_badblock(rdev, dev_sector, max_sectors, |
d4432c23 N |
1313 | &first_bad, &bad_sectors); |
1314 | if (is_bad < 0) { | |
1315 | /* Mustn't write here until the bad block | |
1316 | * is acknowledged | |
1317 | */ | |
1318 | atomic_inc(&rdev->nr_pending); | |
1319 | set_bit(BlockedBadBlocks, &rdev->flags); | |
1320 | blocked_rdev = rdev; | |
1321 | break; | |
1322 | } | |
1323 | if (is_bad && first_bad <= dev_sector) { | |
1324 | /* Cannot write here at all */ | |
1325 | bad_sectors -= (dev_sector - first_bad); | |
1326 | if (bad_sectors < max_sectors) | |
1327 | /* Mustn't write more than bad_sectors | |
1328 | * to other devices yet | |
1329 | */ | |
1330 | max_sectors = bad_sectors; | |
1331 | /* We don't set R10BIO_Degraded as that | |
1332 | * only applies if the disk is missing, | |
1333 | * so it might be re-added, and we want to | |
1334 | * know to recover this chunk. | |
1335 | * In this case the device is here, and the | |
1336 | * fact that this chunk is not in-sync is | |
1337 | * recorded in the bad block log. | |
1338 | */ | |
1339 | continue; | |
1340 | } | |
1341 | if (is_bad) { | |
1342 | int good_sectors = first_bad - dev_sector; | |
1343 | if (good_sectors < max_sectors) | |
1344 | max_sectors = good_sectors; | |
1345 | } | |
6cce3b23 | 1346 | } |
e7c0c3fa N |
1347 | if (rdev) { |
1348 | r10_bio->devs[i].bio = bio; | |
1349 | atomic_inc(&rdev->nr_pending); | |
1350 | } | |
475b0321 N |
1351 | if (rrdev) { |
1352 | r10_bio->devs[i].repl_bio = bio; | |
1353 | atomic_inc(&rrdev->nr_pending); | |
1354 | } | |
1da177e4 LT |
1355 | } |
1356 | rcu_read_unlock(); | |
1357 | ||
6bfe0b49 DW |
1358 | if (unlikely(blocked_rdev)) { |
1359 | /* Have to wait for this device to get unblocked, then retry */ | |
1360 | int j; | |
1361 | int d; | |
1362 | ||
475b0321 | 1363 | for (j = 0; j < i; j++) { |
6bfe0b49 DW |
1364 | if (r10_bio->devs[j].bio) { |
1365 | d = r10_bio->devs[j].devnum; | |
1366 | rdev_dec_pending(conf->mirrors[d].rdev, mddev); | |
1367 | } | |
475b0321 | 1368 | if (r10_bio->devs[j].repl_bio) { |
4ca40c2c | 1369 | struct md_rdev *rdev; |
475b0321 | 1370 | d = r10_bio->devs[j].devnum; |
4ca40c2c N |
1371 | rdev = conf->mirrors[d].replacement; |
1372 | if (!rdev) { | |
1373 | /* Race with remove_disk */ | |
1374 | smp_mb(); | |
1375 | rdev = conf->mirrors[d].rdev; | |
1376 | } | |
1377 | rdev_dec_pending(rdev, mddev); | |
475b0321 N |
1378 | } |
1379 | } | |
6bfe0b49 | 1380 | allow_barrier(conf); |
578b54ad | 1381 | raid10_log(conf->mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); |
6bfe0b49 DW |
1382 | md_wait_for_blocked_rdev(blocked_rdev, mddev); |
1383 | wait_barrier(conf); | |
1384 | goto retry_write; | |
1385 | } | |
1386 | ||
d4432c23 N |
1387 | if (max_sectors < r10_bio->sectors) { |
1388 | /* We are splitting this into multiple parts, so | |
1389 | * we need to prepare for allocating another r10_bio. | |
1390 | */ | |
1391 | r10_bio->sectors = max_sectors; | |
1392 | spin_lock_irq(&conf->device_lock); | |
1393 | if (bio->bi_phys_segments == 0) | |
1394 | bio->bi_phys_segments = 2; | |
1395 | else | |
1396 | bio->bi_phys_segments++; | |
1397 | spin_unlock_irq(&conf->device_lock); | |
1398 | } | |
4f024f37 KO |
1399 | sectors_handled = r10_bio->sector + max_sectors - |
1400 | bio->bi_iter.bi_sector; | |
d4432c23 | 1401 | |
4e78064f | 1402 | atomic_set(&r10_bio->remaining, 1); |
d4432c23 | 1403 | bitmap_startwrite(mddev->bitmap, r10_bio->sector, r10_bio->sectors, 0); |
06d91a5f | 1404 | |
1da177e4 LT |
1405 | for (i = 0; i < conf->copies; i++) { |
1406 | struct bio *mbio; | |
1407 | int d = r10_bio->devs[i].devnum; | |
e7c0c3fa N |
1408 | if (r10_bio->devs[i].bio) { |
1409 | struct md_rdev *rdev = conf->mirrors[d].rdev; | |
d7a10308 | 1410 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 1411 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1412 | max_sectors); |
e7c0c3fa N |
1413 | r10_bio->devs[i].bio = mbio; |
1414 | ||
4f024f37 | 1415 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr+ |
bb5f1ed7 | 1416 | choose_data_offset(r10_bio, rdev)); |
109e3765 | 1417 | mbio->bi_bdev = rdev->bdev; |
e7c0c3fa | 1418 | mbio->bi_end_io = raid10_end_write_request; |
288dab8a | 1419 | bio_set_op_attrs(mbio, op, do_sync | do_fua); |
1919cbb2 N |
1420 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags) && |
1421 | enough(conf, d)) | |
1422 | mbio->bi_opf |= MD_FAILFAST; | |
e7c0c3fa N |
1423 | mbio->bi_private = r10_bio; |
1424 | ||
109e3765 N |
1425 | if (conf->mddev->gendisk) |
1426 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1427 | mbio, disk_devt(conf->mddev->gendisk), | |
1428 | r10_bio->sector); | |
1429 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1430 | mbio->bi_bdev = (void*)rdev; | |
1431 | ||
e7c0c3fa N |
1432 | atomic_inc(&r10_bio->remaining); |
1433 | ||
1434 | cb = blk_check_plugged(raid10_unplug, mddev, | |
1435 | sizeof(*plug)); | |
1436 | if (cb) | |
1437 | plug = container_of(cb, struct raid10_plug_cb, | |
1438 | cb); | |
1439 | else | |
1440 | plug = NULL; | |
1441 | spin_lock_irqsave(&conf->device_lock, flags); | |
1442 | if (plug) { | |
1443 | bio_list_add(&plug->pending, mbio); | |
1444 | plug->pending_cnt++; | |
1445 | } else { | |
1446 | bio_list_add(&conf->pending_bio_list, mbio); | |
1447 | conf->pending_count++; | |
1448 | } | |
1449 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1450 | if (!plug) | |
1451 | md_wakeup_thread(mddev->thread); | |
1452 | } | |
57c67df4 | 1453 | |
e7c0c3fa N |
1454 | if (r10_bio->devs[i].repl_bio) { |
1455 | struct md_rdev *rdev = conf->mirrors[d].replacement; | |
1456 | if (rdev == NULL) { | |
1457 | /* Replacement just got moved to main 'rdev' */ | |
1458 | smp_mb(); | |
1459 | rdev = conf->mirrors[d].rdev; | |
1460 | } | |
d7a10308 | 1461 | mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 1462 | bio_trim(mbio, r10_bio->sector - bio->bi_iter.bi_sector, |
6678d83f | 1463 | max_sectors); |
e7c0c3fa N |
1464 | r10_bio->devs[i].repl_bio = mbio; |
1465 | ||
4f024f37 | 1466 | mbio->bi_iter.bi_sector = (r10_bio->devs[i].addr + |
bb5f1ed7 | 1467 | choose_data_offset(r10_bio, rdev)); |
109e3765 | 1468 | mbio->bi_bdev = rdev->bdev; |
e7c0c3fa | 1469 | mbio->bi_end_io = raid10_end_write_request; |
288dab8a | 1470 | bio_set_op_attrs(mbio, op, do_sync | do_fua); |
e7c0c3fa N |
1471 | mbio->bi_private = r10_bio; |
1472 | ||
109e3765 N |
1473 | if (conf->mddev->gendisk) |
1474 | trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev), | |
1475 | mbio, disk_devt(conf->mddev->gendisk), | |
1476 | r10_bio->sector); | |
1477 | /* flush_pending_writes() needs access to the rdev so...*/ | |
1478 | mbio->bi_bdev = (void*)rdev; | |
1479 | ||
e7c0c3fa | 1480 | atomic_inc(&r10_bio->remaining); |
6d399783 SL |
1481 | |
1482 | cb = blk_check_plugged(raid10_unplug, mddev, | |
1483 | sizeof(*plug)); | |
1484 | if (cb) | |
1485 | plug = container_of(cb, struct raid10_plug_cb, | |
1486 | cb); | |
1487 | else | |
1488 | plug = NULL; | |
e7c0c3fa | 1489 | spin_lock_irqsave(&conf->device_lock, flags); |
6d399783 SL |
1490 | if (plug) { |
1491 | bio_list_add(&plug->pending, mbio); | |
1492 | plug->pending_cnt++; | |
1493 | } else { | |
1494 | bio_list_add(&conf->pending_bio_list, mbio); | |
1495 | conf->pending_count++; | |
1496 | } | |
e7c0c3fa | 1497 | spin_unlock_irqrestore(&conf->device_lock, flags); |
6d399783 | 1498 | if (!plug) |
e7c0c3fa | 1499 | md_wakeup_thread(mddev->thread); |
57c67df4 | 1500 | } |
1da177e4 LT |
1501 | } |
1502 | ||
079fa166 N |
1503 | /* Don't remove the bias on 'remaining' (one_write_done) until |
1504 | * after checking if we need to go around again. | |
1505 | */ | |
a35e63ef | 1506 | |
aa8b57aa | 1507 | if (sectors_handled < bio_sectors(bio)) { |
079fa166 | 1508 | one_write_done(r10_bio); |
5e570289 | 1509 | /* We need another r10_bio. It has already been counted |
d4432c23 N |
1510 | * in bio->bi_phys_segments. |
1511 | */ | |
1512 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1513 | ||
1514 | r10_bio->master_bio = bio; | |
aa8b57aa | 1515 | r10_bio->sectors = bio_sectors(bio) - sectors_handled; |
d4432c23 N |
1516 | |
1517 | r10_bio->mddev = mddev; | |
4f024f37 | 1518 | r10_bio->sector = bio->bi_iter.bi_sector + sectors_handled; |
d4432c23 N |
1519 | r10_bio->state = 0; |
1520 | goto retry_write; | |
1521 | } | |
079fa166 | 1522 | one_write_done(r10_bio); |
20d0189b KO |
1523 | } |
1524 | ||
bb5f1ed7 RL |
1525 | static void __make_request(struct mddev *mddev, struct bio *bio) |
1526 | { | |
1527 | struct r10conf *conf = mddev->private; | |
1528 | struct r10bio *r10_bio; | |
1529 | ||
1530 | r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); | |
1531 | ||
1532 | r10_bio->master_bio = bio; | |
1533 | r10_bio->sectors = bio_sectors(bio); | |
1534 | ||
1535 | r10_bio->mddev = mddev; | |
1536 | r10_bio->sector = bio->bi_iter.bi_sector; | |
1537 | r10_bio->state = 0; | |
1538 | ||
1539 | /* | |
1540 | * We might need to issue multiple reads to different devices if there | |
1541 | * are bad blocks around, so we keep track of the number of reads in | |
1542 | * bio->bi_phys_segments. If this is 0, there is only one r10_bio and | |
1543 | * no locking will be needed when the request completes. If it is | |
1544 | * non-zero, then it is the number of not-completed requests. | |
1545 | */ | |
1546 | bio->bi_phys_segments = 0; | |
1547 | bio_clear_flag(bio, BIO_SEG_VALID); | |
1548 | ||
1549 | if (bio_data_dir(bio) == READ) | |
1550 | raid10_read_request(mddev, bio, r10_bio); | |
1551 | else | |
1552 | raid10_write_request(mddev, bio, r10_bio); | |
1553 | } | |
1554 | ||
849674e4 | 1555 | static void raid10_make_request(struct mddev *mddev, struct bio *bio) |
20d0189b KO |
1556 | { |
1557 | struct r10conf *conf = mddev->private; | |
1558 | sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); | |
1559 | int chunk_sects = chunk_mask + 1; | |
1560 | ||
1561 | struct bio *split; | |
1562 | ||
1eff9d32 | 1563 | if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { |
20d0189b KO |
1564 | md_flush_request(mddev, bio); |
1565 | return; | |
1566 | } | |
1567 | ||
20d0189b KO |
1568 | do { |
1569 | ||
1570 | /* | |
1571 | * If this request crosses a chunk boundary, we need to split | |
1572 | * it. | |
1573 | */ | |
1574 | if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + | |
1575 | bio_sectors(bio) > chunk_sects | |
1576 | && (conf->geo.near_copies < conf->geo.raid_disks | |
1577 | || conf->prev.near_copies < | |
1578 | conf->prev.raid_disks))) { | |
1579 | split = bio_split(bio, chunk_sects - | |
1580 | (bio->bi_iter.bi_sector & | |
1581 | (chunk_sects - 1)), | |
1582 | GFP_NOIO, fs_bio_set); | |
1583 | bio_chain(split, bio); | |
1584 | } else { | |
1585 | split = bio; | |
1586 | } | |
1587 | ||
61eb2b43 SL |
1588 | /* |
1589 | * If a bio is splitted, the first part of bio will pass | |
1590 | * barrier but the bio is queued in current->bio_list (see | |
1591 | * generic_make_request). If there is a raise_barrier() called | |
1592 | * here, the second part of bio can't pass barrier. But since | |
1593 | * the first part bio isn't dispatched to underlaying disks | |
1594 | * yet, the barrier is never released, hence raise_barrier will | |
1595 | * alays wait. We have a deadlock. | |
1596 | * Note, this only happens in read path. For write path, the | |
1597 | * first part of bio is dispatched in a schedule() call | |
1598 | * (because of blk plug) or offloaded to raid10d. | |
1599 | * Quitting from the function immediately can change the bio | |
1600 | * order queued in bio_list and avoid the deadlock. | |
1601 | */ | |
20d0189b | 1602 | __make_request(mddev, split); |
61eb2b43 SL |
1603 | if (split != bio && bio_data_dir(bio) == READ) { |
1604 | generic_make_request(bio); | |
1605 | break; | |
1606 | } | |
20d0189b | 1607 | } while (split != bio); |
079fa166 N |
1608 | |
1609 | /* In case raid10d snuck in to freeze_array */ | |
1610 | wake_up(&conf->wait_barrier); | |
1da177e4 LT |
1611 | } |
1612 | ||
849674e4 | 1613 | static void raid10_status(struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 1614 | { |
e879a879 | 1615 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
1616 | int i; |
1617 | ||
5cf00fcd | 1618 | if (conf->geo.near_copies < conf->geo.raid_disks) |
9d8f0363 | 1619 | seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); |
5cf00fcd N |
1620 | if (conf->geo.near_copies > 1) |
1621 | seq_printf(seq, " %d near-copies", conf->geo.near_copies); | |
1622 | if (conf->geo.far_copies > 1) { | |
1623 | if (conf->geo.far_offset) | |
1624 | seq_printf(seq, " %d offset-copies", conf->geo.far_copies); | |
c93983bf | 1625 | else |
5cf00fcd | 1626 | seq_printf(seq, " %d far-copies", conf->geo.far_copies); |
8bce6d35 N |
1627 | if (conf->geo.far_set_size != conf->geo.raid_disks) |
1628 | seq_printf(seq, " %d devices per set", conf->geo.far_set_size); | |
c93983bf | 1629 | } |
5cf00fcd N |
1630 | seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, |
1631 | conf->geo.raid_disks - mddev->degraded); | |
d44b0a92 N |
1632 | rcu_read_lock(); |
1633 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
1634 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
1635 | seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); | |
1636 | } | |
1637 | rcu_read_unlock(); | |
1da177e4 LT |
1638 | seq_printf(seq, "]"); |
1639 | } | |
1640 | ||
700c7213 N |
1641 | /* check if there are enough drives for |
1642 | * every block to appear on atleast one. | |
1643 | * Don't consider the device numbered 'ignore' | |
1644 | * as we might be about to remove it. | |
1645 | */ | |
635f6416 | 1646 | static int _enough(struct r10conf *conf, int previous, int ignore) |
700c7213 N |
1647 | { |
1648 | int first = 0; | |
725d6e57 | 1649 | int has_enough = 0; |
635f6416 N |
1650 | int disks, ncopies; |
1651 | if (previous) { | |
1652 | disks = conf->prev.raid_disks; | |
1653 | ncopies = conf->prev.near_copies; | |
1654 | } else { | |
1655 | disks = conf->geo.raid_disks; | |
1656 | ncopies = conf->geo.near_copies; | |
1657 | } | |
700c7213 | 1658 | |
725d6e57 | 1659 | rcu_read_lock(); |
700c7213 N |
1660 | do { |
1661 | int n = conf->copies; | |
1662 | int cnt = 0; | |
80b48124 | 1663 | int this = first; |
700c7213 | 1664 | while (n--) { |
725d6e57 N |
1665 | struct md_rdev *rdev; |
1666 | if (this != ignore && | |
1667 | (rdev = rcu_dereference(conf->mirrors[this].rdev)) && | |
1668 | test_bit(In_sync, &rdev->flags)) | |
700c7213 | 1669 | cnt++; |
635f6416 | 1670 | this = (this+1) % disks; |
700c7213 N |
1671 | } |
1672 | if (cnt == 0) | |
725d6e57 | 1673 | goto out; |
635f6416 | 1674 | first = (first + ncopies) % disks; |
700c7213 | 1675 | } while (first != 0); |
725d6e57 N |
1676 | has_enough = 1; |
1677 | out: | |
1678 | rcu_read_unlock(); | |
1679 | return has_enough; | |
700c7213 N |
1680 | } |
1681 | ||
f8c9e74f N |
1682 | static int enough(struct r10conf *conf, int ignore) |
1683 | { | |
635f6416 N |
1684 | /* when calling 'enough', both 'prev' and 'geo' must |
1685 | * be stable. | |
1686 | * This is ensured if ->reconfig_mutex or ->device_lock | |
1687 | * is held. | |
1688 | */ | |
1689 | return _enough(conf, 0, ignore) && | |
1690 | _enough(conf, 1, ignore); | |
f8c9e74f N |
1691 | } |
1692 | ||
849674e4 | 1693 | static void raid10_error(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 LT |
1694 | { |
1695 | char b[BDEVNAME_SIZE]; | |
e879a879 | 1696 | struct r10conf *conf = mddev->private; |
635f6416 | 1697 | unsigned long flags; |
1da177e4 LT |
1698 | |
1699 | /* | |
1700 | * If it is not operational, then we have already marked it as dead | |
1701 | * else if it is the last working disks, ignore the error, let the | |
1702 | * next level up know. | |
1703 | * else mark the drive as failed | |
1704 | */ | |
635f6416 | 1705 | spin_lock_irqsave(&conf->device_lock, flags); |
b2d444d7 | 1706 | if (test_bit(In_sync, &rdev->flags) |
635f6416 | 1707 | && !enough(conf, rdev->raid_disk)) { |
1da177e4 LT |
1708 | /* |
1709 | * Don't fail the drive, just return an IO error. | |
1da177e4 | 1710 | */ |
635f6416 | 1711 | spin_unlock_irqrestore(&conf->device_lock, flags); |
1da177e4 | 1712 | return; |
635f6416 | 1713 | } |
2446dba0 | 1714 | if (test_and_clear_bit(In_sync, &rdev->flags)) |
1da177e4 | 1715 | mddev->degraded++; |
2446dba0 N |
1716 | /* |
1717 | * If recovery is running, make sure it aborts. | |
1718 | */ | |
1719 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); | |
de393cde | 1720 | set_bit(Blocked, &rdev->flags); |
b2d444d7 | 1721 | set_bit(Faulty, &rdev->flags); |
2953079c SL |
1722 | set_mask_bits(&mddev->sb_flags, 0, |
1723 | BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING)); | |
635f6416 | 1724 | spin_unlock_irqrestore(&conf->device_lock, flags); |
08464e09 N |
1725 | pr_crit("md/raid10:%s: Disk failure on %s, disabling device.\n" |
1726 | "md/raid10:%s: Operation continuing on %d devices.\n", | |
1727 | mdname(mddev), bdevname(rdev->bdev, b), | |
1728 | mdname(mddev), conf->geo.raid_disks - mddev->degraded); | |
1da177e4 LT |
1729 | } |
1730 | ||
e879a879 | 1731 | static void print_conf(struct r10conf *conf) |
1da177e4 LT |
1732 | { |
1733 | int i; | |
4056ca51 | 1734 | struct md_rdev *rdev; |
1da177e4 | 1735 | |
08464e09 | 1736 | pr_debug("RAID10 conf printout:\n"); |
1da177e4 | 1737 | if (!conf) { |
08464e09 | 1738 | pr_debug("(!conf)\n"); |
1da177e4 LT |
1739 | return; |
1740 | } | |
08464e09 N |
1741 | pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, |
1742 | conf->geo.raid_disks); | |
1da177e4 | 1743 | |
4056ca51 N |
1744 | /* This is only called with ->reconfix_mutex held, so |
1745 | * rcu protection of rdev is not needed */ | |
5cf00fcd | 1746 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1747 | char b[BDEVNAME_SIZE]; |
4056ca51 N |
1748 | rdev = conf->mirrors[i].rdev; |
1749 | if (rdev) | |
08464e09 N |
1750 | pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n", |
1751 | i, !test_bit(In_sync, &rdev->flags), | |
1752 | !test_bit(Faulty, &rdev->flags), | |
1753 | bdevname(rdev->bdev,b)); | |
1da177e4 LT |
1754 | } |
1755 | } | |
1756 | ||
e879a879 | 1757 | static void close_sync(struct r10conf *conf) |
1da177e4 | 1758 | { |
0a27ec96 N |
1759 | wait_barrier(conf); |
1760 | allow_barrier(conf); | |
1da177e4 LT |
1761 | |
1762 | mempool_destroy(conf->r10buf_pool); | |
1763 | conf->r10buf_pool = NULL; | |
1764 | } | |
1765 | ||
fd01b88c | 1766 | static int raid10_spare_active(struct mddev *mddev) |
1da177e4 LT |
1767 | { |
1768 | int i; | |
e879a879 | 1769 | struct r10conf *conf = mddev->private; |
dc280d98 | 1770 | struct raid10_info *tmp; |
6b965620 N |
1771 | int count = 0; |
1772 | unsigned long flags; | |
1da177e4 LT |
1773 | |
1774 | /* | |
1775 | * Find all non-in_sync disks within the RAID10 configuration | |
1776 | * and mark them in_sync | |
1777 | */ | |
5cf00fcd | 1778 | for (i = 0; i < conf->geo.raid_disks; i++) { |
1da177e4 | 1779 | tmp = conf->mirrors + i; |
4ca40c2c N |
1780 | if (tmp->replacement |
1781 | && tmp->replacement->recovery_offset == MaxSector | |
1782 | && !test_bit(Faulty, &tmp->replacement->flags) | |
1783 | && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { | |
1784 | /* Replacement has just become active */ | |
1785 | if (!tmp->rdev | |
1786 | || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) | |
1787 | count++; | |
1788 | if (tmp->rdev) { | |
1789 | /* Replaced device not technically faulty, | |
1790 | * but we need to be sure it gets removed | |
1791 | * and never re-added. | |
1792 | */ | |
1793 | set_bit(Faulty, &tmp->rdev->flags); | |
1794 | sysfs_notify_dirent_safe( | |
1795 | tmp->rdev->sysfs_state); | |
1796 | } | |
1797 | sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); | |
1798 | } else if (tmp->rdev | |
61e4947c | 1799 | && tmp->rdev->recovery_offset == MaxSector |
4ca40c2c N |
1800 | && !test_bit(Faulty, &tmp->rdev->flags) |
1801 | && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { | |
6b965620 | 1802 | count++; |
2863b9eb | 1803 | sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); |
1da177e4 LT |
1804 | } |
1805 | } | |
6b965620 N |
1806 | spin_lock_irqsave(&conf->device_lock, flags); |
1807 | mddev->degraded -= count; | |
1808 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 LT |
1809 | |
1810 | print_conf(conf); | |
6b965620 | 1811 | return count; |
1da177e4 LT |
1812 | } |
1813 | ||
fd01b88c | 1814 | static int raid10_add_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1815 | { |
e879a879 | 1816 | struct r10conf *conf = mddev->private; |
199050ea | 1817 | int err = -EEXIST; |
1da177e4 | 1818 | int mirror; |
6c2fce2e | 1819 | int first = 0; |
5cf00fcd | 1820 | int last = conf->geo.raid_disks - 1; |
1da177e4 LT |
1821 | |
1822 | if (mddev->recovery_cp < MaxSector) | |
1823 | /* only hot-add to in-sync arrays, as recovery is | |
1824 | * very different from resync | |
1825 | */ | |
199050ea | 1826 | return -EBUSY; |
635f6416 | 1827 | if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) |
199050ea | 1828 | return -EINVAL; |
1da177e4 | 1829 | |
1501efad DW |
1830 | if (md_integrity_add_rdev(rdev, mddev)) |
1831 | return -ENXIO; | |
1832 | ||
a53a6c85 | 1833 | if (rdev->raid_disk >= 0) |
6c2fce2e | 1834 | first = last = rdev->raid_disk; |
1da177e4 | 1835 | |
2c4193df | 1836 | if (rdev->saved_raid_disk >= first && |
6cce3b23 N |
1837 | conf->mirrors[rdev->saved_raid_disk].rdev == NULL) |
1838 | mirror = rdev->saved_raid_disk; | |
1839 | else | |
6c2fce2e | 1840 | mirror = first; |
2bb77736 | 1841 | for ( ; mirror <= last ; mirror++) { |
dc280d98 | 1842 | struct raid10_info *p = &conf->mirrors[mirror]; |
2bb77736 N |
1843 | if (p->recovery_disabled == mddev->recovery_disabled) |
1844 | continue; | |
b7044d41 N |
1845 | if (p->rdev) { |
1846 | if (!test_bit(WantReplacement, &p->rdev->flags) || | |
1847 | p->replacement != NULL) | |
1848 | continue; | |
1849 | clear_bit(In_sync, &rdev->flags); | |
1850 | set_bit(Replacement, &rdev->flags); | |
1851 | rdev->raid_disk = mirror; | |
1852 | err = 0; | |
9092c02d JB |
1853 | if (mddev->gendisk) |
1854 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1855 | rdev->data_offset << 9); | |
b7044d41 N |
1856 | conf->fullsync = 1; |
1857 | rcu_assign_pointer(p->replacement, rdev); | |
1858 | break; | |
1859 | } | |
1da177e4 | 1860 | |
9092c02d JB |
1861 | if (mddev->gendisk) |
1862 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
1863 | rdev->data_offset << 9); | |
1da177e4 | 1864 | |
2bb77736 | 1865 | p->head_position = 0; |
d890fa2b | 1866 | p->recovery_disabled = mddev->recovery_disabled - 1; |
2bb77736 N |
1867 | rdev->raid_disk = mirror; |
1868 | err = 0; | |
1869 | if (rdev->saved_raid_disk != mirror) | |
1870 | conf->fullsync = 1; | |
1871 | rcu_assign_pointer(p->rdev, rdev); | |
1872 | break; | |
1873 | } | |
ed30be07 | 1874 | if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) |
532a2a3f SL |
1875 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
1876 | ||
1da177e4 | 1877 | print_conf(conf); |
199050ea | 1878 | return err; |
1da177e4 LT |
1879 | } |
1880 | ||
b8321b68 | 1881 | static int raid10_remove_disk(struct mddev *mddev, struct md_rdev *rdev) |
1da177e4 | 1882 | { |
e879a879 | 1883 | struct r10conf *conf = mddev->private; |
1da177e4 | 1884 | int err = 0; |
b8321b68 | 1885 | int number = rdev->raid_disk; |
c8ab903e | 1886 | struct md_rdev **rdevp; |
dc280d98 | 1887 | struct raid10_info *p = conf->mirrors + number; |
1da177e4 LT |
1888 | |
1889 | print_conf(conf); | |
c8ab903e N |
1890 | if (rdev == p->rdev) |
1891 | rdevp = &p->rdev; | |
1892 | else if (rdev == p->replacement) | |
1893 | rdevp = &p->replacement; | |
1894 | else | |
1895 | return 0; | |
1896 | ||
1897 | if (test_bit(In_sync, &rdev->flags) || | |
1898 | atomic_read(&rdev->nr_pending)) { | |
1899 | err = -EBUSY; | |
1900 | goto abort; | |
1901 | } | |
d787be40 | 1902 | /* Only remove non-faulty devices if recovery |
c8ab903e N |
1903 | * is not possible. |
1904 | */ | |
1905 | if (!test_bit(Faulty, &rdev->flags) && | |
1906 | mddev->recovery_disabled != p->recovery_disabled && | |
4ca40c2c | 1907 | (!p->replacement || p->replacement == rdev) && |
63aced61 | 1908 | number < conf->geo.raid_disks && |
c8ab903e N |
1909 | enough(conf, -1)) { |
1910 | err = -EBUSY; | |
1911 | goto abort; | |
1da177e4 | 1912 | } |
c8ab903e | 1913 | *rdevp = NULL; |
d787be40 N |
1914 | if (!test_bit(RemoveSynchronized, &rdev->flags)) { |
1915 | synchronize_rcu(); | |
1916 | if (atomic_read(&rdev->nr_pending)) { | |
1917 | /* lost the race, try later */ | |
1918 | err = -EBUSY; | |
1919 | *rdevp = rdev; | |
1920 | goto abort; | |
1921 | } | |
1922 | } | |
1923 | if (p->replacement) { | |
4ca40c2c N |
1924 | /* We must have just cleared 'rdev' */ |
1925 | p->rdev = p->replacement; | |
1926 | clear_bit(Replacement, &p->replacement->flags); | |
1927 | smp_mb(); /* Make sure other CPUs may see both as identical | |
1928 | * but will never see neither -- if they are careful. | |
1929 | */ | |
1930 | p->replacement = NULL; | |
1931 | clear_bit(WantReplacement, &rdev->flags); | |
1932 | } else | |
1933 | /* We might have just remove the Replacement as faulty | |
1934 | * Clear the flag just in case | |
1935 | */ | |
1936 | clear_bit(WantReplacement, &rdev->flags); | |
1937 | ||
c8ab903e N |
1938 | err = md_integrity_register(mddev); |
1939 | ||
1da177e4 LT |
1940 | abort: |
1941 | ||
1942 | print_conf(conf); | |
1943 | return err; | |
1944 | } | |
1945 | ||
4246a0b6 | 1946 | static void end_sync_read(struct bio *bio) |
1da177e4 | 1947 | { |
9f2c9d12 | 1948 | struct r10bio *r10_bio = bio->bi_private; |
e879a879 | 1949 | struct r10conf *conf = r10_bio->mddev->private; |
778ca018 | 1950 | int d; |
1da177e4 | 1951 | |
3ea7daa5 N |
1952 | if (bio == r10_bio->master_bio) { |
1953 | /* this is a reshape read */ | |
1954 | d = r10_bio->read_slot; /* really the read dev */ | |
1955 | } else | |
1956 | d = find_bio_disk(conf, r10_bio, bio, NULL, NULL); | |
0eb3ff12 | 1957 | |
4246a0b6 | 1958 | if (!bio->bi_error) |
0eb3ff12 | 1959 | set_bit(R10BIO_Uptodate, &r10_bio->state); |
e684e41d N |
1960 | else |
1961 | /* The write handler will notice the lack of | |
1962 | * R10BIO_Uptodate and record any errors etc | |
1963 | */ | |
4dbcdc75 N |
1964 | atomic_add(r10_bio->sectors, |
1965 | &conf->mirrors[d].rdev->corrected_errors); | |
1da177e4 LT |
1966 | |
1967 | /* for reconstruct, we always reschedule after a read. | |
1968 | * for resync, only after all reads | |
1969 | */ | |
73d5c38a | 1970 | rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); |
1da177e4 LT |
1971 | if (test_bit(R10BIO_IsRecover, &r10_bio->state) || |
1972 | atomic_dec_and_test(&r10_bio->remaining)) { | |
1973 | /* we have read all the blocks, | |
1974 | * do the comparison in process context in raid10d | |
1975 | */ | |
1976 | reschedule_retry(r10_bio); | |
1977 | } | |
1da177e4 LT |
1978 | } |
1979 | ||
9f2c9d12 | 1980 | static void end_sync_request(struct r10bio *r10_bio) |
1da177e4 | 1981 | { |
fd01b88c | 1982 | struct mddev *mddev = r10_bio->mddev; |
dfc70645 | 1983 | |
1da177e4 LT |
1984 | while (atomic_dec_and_test(&r10_bio->remaining)) { |
1985 | if (r10_bio->master_bio == NULL) { | |
1986 | /* the primary of several recovery bios */ | |
73d5c38a | 1987 | sector_t s = r10_bio->sectors; |
1a0b7cd8 N |
1988 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1989 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1990 | reschedule_retry(r10_bio); |
1991 | else | |
1992 | put_buf(r10_bio); | |
73d5c38a | 1993 | md_done_sync(mddev, s, 1); |
1da177e4 LT |
1994 | break; |
1995 | } else { | |
9f2c9d12 | 1996 | struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; |
1a0b7cd8 N |
1997 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
1998 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 N |
1999 | reschedule_retry(r10_bio); |
2000 | else | |
2001 | put_buf(r10_bio); | |
1da177e4 LT |
2002 | r10_bio = r10_bio2; |
2003 | } | |
2004 | } | |
1da177e4 LT |
2005 | } |
2006 | ||
4246a0b6 | 2007 | static void end_sync_write(struct bio *bio) |
5e570289 | 2008 | { |
9f2c9d12 | 2009 | struct r10bio *r10_bio = bio->bi_private; |
fd01b88c | 2010 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2011 | struct r10conf *conf = mddev->private; |
5e570289 N |
2012 | int d; |
2013 | sector_t first_bad; | |
2014 | int bad_sectors; | |
2015 | int slot; | |
9ad1aefc | 2016 | int repl; |
4ca40c2c | 2017 | struct md_rdev *rdev = NULL; |
5e570289 | 2018 | |
9ad1aefc N |
2019 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); |
2020 | if (repl) | |
2021 | rdev = conf->mirrors[d].replacement; | |
547414d1 | 2022 | else |
9ad1aefc | 2023 | rdev = conf->mirrors[d].rdev; |
5e570289 | 2024 | |
4246a0b6 | 2025 | if (bio->bi_error) { |
9ad1aefc N |
2026 | if (repl) |
2027 | md_error(mddev, rdev); | |
2028 | else { | |
2029 | set_bit(WriteErrorSeen, &rdev->flags); | |
b7044d41 N |
2030 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2031 | set_bit(MD_RECOVERY_NEEDED, | |
2032 | &rdev->mddev->recovery); | |
9ad1aefc N |
2033 | set_bit(R10BIO_WriteError, &r10_bio->state); |
2034 | } | |
2035 | } else if (is_badblock(rdev, | |
5e570289 N |
2036 | r10_bio->devs[slot].addr, |
2037 | r10_bio->sectors, | |
2038 | &first_bad, &bad_sectors)) | |
2039 | set_bit(R10BIO_MadeGood, &r10_bio->state); | |
2040 | ||
9ad1aefc | 2041 | rdev_dec_pending(rdev, mddev); |
5e570289 N |
2042 | |
2043 | end_sync_request(r10_bio); | |
2044 | } | |
2045 | ||
1da177e4 LT |
2046 | /* |
2047 | * Note: sync and recover and handled very differently for raid10 | |
2048 | * This code is for resync. | |
2049 | * For resync, we read through virtual addresses and read all blocks. | |
2050 | * If there is any error, we schedule a write. The lowest numbered | |
2051 | * drive is authoritative. | |
2052 | * However requests come for physical address, so we need to map. | |
2053 | * For every physical address there are raid_disks/copies virtual addresses, | |
2054 | * which is always are least one, but is not necessarly an integer. | |
2055 | * This means that a physical address can span multiple chunks, so we may | |
2056 | * have to submit multiple io requests for a single sync request. | |
2057 | */ | |
2058 | /* | |
2059 | * We check if all blocks are in-sync and only write to blocks that | |
2060 | * aren't in sync | |
2061 | */ | |
9f2c9d12 | 2062 | static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2063 | { |
e879a879 | 2064 | struct r10conf *conf = mddev->private; |
1da177e4 LT |
2065 | int i, first; |
2066 | struct bio *tbio, *fbio; | |
f4380a91 | 2067 | int vcnt; |
1da177e4 LT |
2068 | |
2069 | atomic_set(&r10_bio->remaining, 1); | |
2070 | ||
2071 | /* find the first device with a block */ | |
2072 | for (i=0; i<conf->copies; i++) | |
4246a0b6 | 2073 | if (!r10_bio->devs[i].bio->bi_error) |
1da177e4 LT |
2074 | break; |
2075 | ||
2076 | if (i == conf->copies) | |
2077 | goto done; | |
2078 | ||
2079 | first = i; | |
2080 | fbio = r10_bio->devs[i].bio; | |
cc578588 AP |
2081 | fbio->bi_iter.bi_size = r10_bio->sectors << 9; |
2082 | fbio->bi_iter.bi_idx = 0; | |
1da177e4 | 2083 | |
f4380a91 | 2084 | vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); |
1da177e4 | 2085 | /* now find blocks with errors */ |
0eb3ff12 N |
2086 | for (i=0 ; i < conf->copies ; i++) { |
2087 | int j, d; | |
8d3ca83d | 2088 | struct md_rdev *rdev; |
1da177e4 | 2089 | |
1da177e4 | 2090 | tbio = r10_bio->devs[i].bio; |
0eb3ff12 N |
2091 | |
2092 | if (tbio->bi_end_io != end_sync_read) | |
2093 | continue; | |
2094 | if (i == first) | |
1da177e4 | 2095 | continue; |
8d3ca83d N |
2096 | d = r10_bio->devs[i].devnum; |
2097 | rdev = conf->mirrors[d].rdev; | |
4246a0b6 | 2098 | if (!r10_bio->devs[i].bio->bi_error) { |
0eb3ff12 N |
2099 | /* We know that the bi_io_vec layout is the same for |
2100 | * both 'first' and 'i', so we just compare them. | |
2101 | * All vec entries are PAGE_SIZE; | |
2102 | */ | |
7bb23c49 N |
2103 | int sectors = r10_bio->sectors; |
2104 | for (j = 0; j < vcnt; j++) { | |
2105 | int len = PAGE_SIZE; | |
2106 | if (sectors < (len / 512)) | |
2107 | len = sectors * 512; | |
0eb3ff12 N |
2108 | if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), |
2109 | page_address(tbio->bi_io_vec[j].bv_page), | |
7bb23c49 | 2110 | len)) |
0eb3ff12 | 2111 | break; |
7bb23c49 N |
2112 | sectors -= len/512; |
2113 | } | |
0eb3ff12 N |
2114 | if (j == vcnt) |
2115 | continue; | |
7f7583d4 | 2116 | atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); |
f84ee364 N |
2117 | if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) |
2118 | /* Don't fix anything. */ | |
2119 | continue; | |
8d3ca83d N |
2120 | } else if (test_bit(FailFast, &rdev->flags)) { |
2121 | /* Just give up on this device */ | |
2122 | md_error(rdev->mddev, rdev); | |
2123 | continue; | |
0eb3ff12 | 2124 | } |
f84ee364 N |
2125 | /* Ok, we need to write this bio, either to correct an |
2126 | * inconsistency or to correct an unreadable block. | |
1da177e4 LT |
2127 | * First we need to fixup bv_offset, bv_len and |
2128 | * bi_vecs, as the read request might have corrupted these | |
2129 | */ | |
8be185f2 KO |
2130 | bio_reset(tbio); |
2131 | ||
1da177e4 | 2132 | tbio->bi_vcnt = vcnt; |
cc578588 | 2133 | tbio->bi_iter.bi_size = fbio->bi_iter.bi_size; |
1da177e4 | 2134 | tbio->bi_private = r10_bio; |
4f024f37 | 2135 | tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; |
1da177e4 | 2136 | tbio->bi_end_io = end_sync_write; |
796a5cf0 | 2137 | bio_set_op_attrs(tbio, REQ_OP_WRITE, 0); |
1da177e4 | 2138 | |
c31df25f KO |
2139 | bio_copy_data(tbio, fbio); |
2140 | ||
1da177e4 LT |
2141 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); |
2142 | atomic_inc(&r10_bio->remaining); | |
aa8b57aa | 2143 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); |
1da177e4 | 2144 | |
1919cbb2 N |
2145 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
2146 | tbio->bi_opf |= MD_FAILFAST; | |
4f024f37 | 2147 | tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; |
1da177e4 LT |
2148 | tbio->bi_bdev = conf->mirrors[d].rdev->bdev; |
2149 | generic_make_request(tbio); | |
2150 | } | |
2151 | ||
9ad1aefc N |
2152 | /* Now write out to any replacement devices |
2153 | * that are active | |
2154 | */ | |
2155 | for (i = 0; i < conf->copies; i++) { | |
c31df25f | 2156 | int d; |
9ad1aefc N |
2157 | |
2158 | tbio = r10_bio->devs[i].repl_bio; | |
2159 | if (!tbio || !tbio->bi_end_io) | |
2160 | continue; | |
2161 | if (r10_bio->devs[i].bio->bi_end_io != end_sync_write | |
2162 | && r10_bio->devs[i].bio != fbio) | |
c31df25f | 2163 | bio_copy_data(tbio, fbio); |
9ad1aefc N |
2164 | d = r10_bio->devs[i].devnum; |
2165 | atomic_inc(&r10_bio->remaining); | |
2166 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2167 | bio_sectors(tbio)); |
9ad1aefc N |
2168 | generic_make_request(tbio); |
2169 | } | |
2170 | ||
1da177e4 LT |
2171 | done: |
2172 | if (atomic_dec_and_test(&r10_bio->remaining)) { | |
2173 | md_done_sync(mddev, r10_bio->sectors, 1); | |
2174 | put_buf(r10_bio); | |
2175 | } | |
2176 | } | |
2177 | ||
2178 | /* | |
2179 | * Now for the recovery code. | |
2180 | * Recovery happens across physical sectors. | |
2181 | * We recover all non-is_sync drives by finding the virtual address of | |
2182 | * each, and then choose a working drive that also has that virt address. | |
2183 | * There is a separate r10_bio for each non-in_sync drive. | |
2184 | * Only the first two slots are in use. The first for reading, | |
2185 | * The second for writing. | |
2186 | * | |
2187 | */ | |
9f2c9d12 | 2188 | static void fix_recovery_read_error(struct r10bio *r10_bio) |
5e570289 N |
2189 | { |
2190 | /* We got a read error during recovery. | |
2191 | * We repeat the read in smaller page-sized sections. | |
2192 | * If a read succeeds, write it to the new device or record | |
2193 | * a bad block if we cannot. | |
2194 | * If a read fails, record a bad block on both old and | |
2195 | * new devices. | |
2196 | */ | |
fd01b88c | 2197 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2198 | struct r10conf *conf = mddev->private; |
5e570289 N |
2199 | struct bio *bio = r10_bio->devs[0].bio; |
2200 | sector_t sect = 0; | |
2201 | int sectors = r10_bio->sectors; | |
2202 | int idx = 0; | |
2203 | int dr = r10_bio->devs[0].devnum; | |
2204 | int dw = r10_bio->devs[1].devnum; | |
2205 | ||
2206 | while (sectors) { | |
2207 | int s = sectors; | |
3cb03002 | 2208 | struct md_rdev *rdev; |
5e570289 N |
2209 | sector_t addr; |
2210 | int ok; | |
2211 | ||
2212 | if (s > (PAGE_SIZE>>9)) | |
2213 | s = PAGE_SIZE >> 9; | |
2214 | ||
2215 | rdev = conf->mirrors[dr].rdev; | |
2216 | addr = r10_bio->devs[0].addr + sect, | |
2217 | ok = sync_page_io(rdev, | |
2218 | addr, | |
2219 | s << 9, | |
2220 | bio->bi_io_vec[idx].bv_page, | |
796a5cf0 | 2221 | REQ_OP_READ, 0, false); |
5e570289 N |
2222 | if (ok) { |
2223 | rdev = conf->mirrors[dw].rdev; | |
2224 | addr = r10_bio->devs[1].addr + sect; | |
2225 | ok = sync_page_io(rdev, | |
2226 | addr, | |
2227 | s << 9, | |
2228 | bio->bi_io_vec[idx].bv_page, | |
796a5cf0 | 2229 | REQ_OP_WRITE, 0, false); |
b7044d41 | 2230 | if (!ok) { |
5e570289 | 2231 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2232 | if (!test_and_set_bit(WantReplacement, |
2233 | &rdev->flags)) | |
2234 | set_bit(MD_RECOVERY_NEEDED, | |
2235 | &rdev->mddev->recovery); | |
2236 | } | |
5e570289 N |
2237 | } |
2238 | if (!ok) { | |
2239 | /* We don't worry if we cannot set a bad block - | |
2240 | * it really is bad so there is no loss in not | |
2241 | * recording it yet | |
2242 | */ | |
2243 | rdev_set_badblocks(rdev, addr, s, 0); | |
2244 | ||
2245 | if (rdev != conf->mirrors[dw].rdev) { | |
2246 | /* need bad block on destination too */ | |
3cb03002 | 2247 | struct md_rdev *rdev2 = conf->mirrors[dw].rdev; |
5e570289 N |
2248 | addr = r10_bio->devs[1].addr + sect; |
2249 | ok = rdev_set_badblocks(rdev2, addr, s, 0); | |
2250 | if (!ok) { | |
2251 | /* just abort the recovery */ | |
08464e09 N |
2252 | pr_notice("md/raid10:%s: recovery aborted due to read error\n", |
2253 | mdname(mddev)); | |
5e570289 N |
2254 | |
2255 | conf->mirrors[dw].recovery_disabled | |
2256 | = mddev->recovery_disabled; | |
2257 | set_bit(MD_RECOVERY_INTR, | |
2258 | &mddev->recovery); | |
2259 | break; | |
2260 | } | |
2261 | } | |
2262 | } | |
2263 | ||
2264 | sectors -= s; | |
2265 | sect += s; | |
2266 | idx++; | |
2267 | } | |
2268 | } | |
1da177e4 | 2269 | |
9f2c9d12 | 2270 | static void recovery_request_write(struct mddev *mddev, struct r10bio *r10_bio) |
1da177e4 | 2271 | { |
e879a879 | 2272 | struct r10conf *conf = mddev->private; |
c65060ad | 2273 | int d; |
24afd80d | 2274 | struct bio *wbio, *wbio2; |
1da177e4 | 2275 | |
5e570289 N |
2276 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { |
2277 | fix_recovery_read_error(r10_bio); | |
2278 | end_sync_request(r10_bio); | |
2279 | return; | |
2280 | } | |
2281 | ||
c65060ad NK |
2282 | /* |
2283 | * share the pages with the first bio | |
1da177e4 LT |
2284 | * and submit the write request |
2285 | */ | |
1da177e4 | 2286 | d = r10_bio->devs[1].devnum; |
24afd80d N |
2287 | wbio = r10_bio->devs[1].bio; |
2288 | wbio2 = r10_bio->devs[1].repl_bio; | |
0eb25bb0 N |
2289 | /* Need to test wbio2->bi_end_io before we call |
2290 | * generic_make_request as if the former is NULL, | |
2291 | * the latter is free to free wbio2. | |
2292 | */ | |
2293 | if (wbio2 && !wbio2->bi_end_io) | |
2294 | wbio2 = NULL; | |
24afd80d N |
2295 | if (wbio->bi_end_io) { |
2296 | atomic_inc(&conf->mirrors[d].rdev->nr_pending); | |
aa8b57aa | 2297 | md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); |
24afd80d N |
2298 | generic_make_request(wbio); |
2299 | } | |
0eb25bb0 | 2300 | if (wbio2) { |
24afd80d N |
2301 | atomic_inc(&conf->mirrors[d].replacement->nr_pending); |
2302 | md_sync_acct(conf->mirrors[d].replacement->bdev, | |
aa8b57aa | 2303 | bio_sectors(wbio2)); |
24afd80d N |
2304 | generic_make_request(wbio2); |
2305 | } | |
1da177e4 LT |
2306 | } |
2307 | ||
1e50915f RB |
2308 | /* |
2309 | * Used by fix_read_error() to decay the per rdev read_errors. | |
2310 | * We halve the read error count for every hour that has elapsed | |
2311 | * since the last recorded read error. | |
2312 | * | |
2313 | */ | |
fd01b88c | 2314 | static void check_decay_read_errors(struct mddev *mddev, struct md_rdev *rdev) |
1e50915f | 2315 | { |
0e3ef49e | 2316 | long cur_time_mon; |
1e50915f RB |
2317 | unsigned long hours_since_last; |
2318 | unsigned int read_errors = atomic_read(&rdev->read_errors); | |
2319 | ||
0e3ef49e | 2320 | cur_time_mon = ktime_get_seconds(); |
1e50915f | 2321 | |
0e3ef49e | 2322 | if (rdev->last_read_error == 0) { |
1e50915f RB |
2323 | /* first time we've seen a read error */ |
2324 | rdev->last_read_error = cur_time_mon; | |
2325 | return; | |
2326 | } | |
2327 | ||
0e3ef49e AB |
2328 | hours_since_last = (long)(cur_time_mon - |
2329 | rdev->last_read_error) / 3600; | |
1e50915f RB |
2330 | |
2331 | rdev->last_read_error = cur_time_mon; | |
2332 | ||
2333 | /* | |
2334 | * if hours_since_last is > the number of bits in read_errors | |
2335 | * just set read errors to 0. We do this to avoid | |
2336 | * overflowing the shift of read_errors by hours_since_last. | |
2337 | */ | |
2338 | if (hours_since_last >= 8 * sizeof(read_errors)) | |
2339 | atomic_set(&rdev->read_errors, 0); | |
2340 | else | |
2341 | atomic_set(&rdev->read_errors, read_errors >> hours_since_last); | |
2342 | } | |
2343 | ||
3cb03002 | 2344 | static int r10_sync_page_io(struct md_rdev *rdev, sector_t sector, |
58c54fcc N |
2345 | int sectors, struct page *page, int rw) |
2346 | { | |
2347 | sector_t first_bad; | |
2348 | int bad_sectors; | |
2349 | ||
2350 | if (is_badblock(rdev, sector, sectors, &first_bad, &bad_sectors) | |
2351 | && (rw == READ || test_bit(WriteErrorSeen, &rdev->flags))) | |
2352 | return -1; | |
796a5cf0 | 2353 | if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false)) |
58c54fcc N |
2354 | /* success */ |
2355 | return 1; | |
b7044d41 | 2356 | if (rw == WRITE) { |
58c54fcc | 2357 | set_bit(WriteErrorSeen, &rdev->flags); |
b7044d41 N |
2358 | if (!test_and_set_bit(WantReplacement, &rdev->flags)) |
2359 | set_bit(MD_RECOVERY_NEEDED, | |
2360 | &rdev->mddev->recovery); | |
2361 | } | |
58c54fcc N |
2362 | /* need to record an error - either for the block or the device */ |
2363 | if (!rdev_set_badblocks(rdev, sector, sectors, 0)) | |
2364 | md_error(rdev->mddev, rdev); | |
2365 | return 0; | |
2366 | } | |
2367 | ||
1da177e4 LT |
2368 | /* |
2369 | * This is a kernel thread which: | |
2370 | * | |
2371 | * 1. Retries failed read operations on working mirrors. | |
2372 | * 2. Updates the raid superblock when problems encounter. | |
6814d536 | 2373 | * 3. Performs writes following reads for array synchronising. |
1da177e4 LT |
2374 | */ |
2375 | ||
e879a879 | 2376 | static void fix_read_error(struct r10conf *conf, struct mddev *mddev, struct r10bio *r10_bio) |
6814d536 N |
2377 | { |
2378 | int sect = 0; /* Offset from r10_bio->sector */ | |
2379 | int sectors = r10_bio->sectors; | |
3cb03002 | 2380 | struct md_rdev*rdev; |
1e50915f | 2381 | int max_read_errors = atomic_read(&mddev->max_corr_read_errors); |
0544a21d | 2382 | int d = r10_bio->devs[r10_bio->read_slot].devnum; |
1e50915f | 2383 | |
7c4e06ff N |
2384 | /* still own a reference to this rdev, so it cannot |
2385 | * have been cleared recently. | |
2386 | */ | |
2387 | rdev = conf->mirrors[d].rdev; | |
1e50915f | 2388 | |
7c4e06ff N |
2389 | if (test_bit(Faulty, &rdev->flags)) |
2390 | /* drive has already been failed, just ignore any | |
2391 | more fix_read_error() attempts */ | |
2392 | return; | |
1e50915f | 2393 | |
7c4e06ff N |
2394 | check_decay_read_errors(mddev, rdev); |
2395 | atomic_inc(&rdev->read_errors); | |
2396 | if (atomic_read(&rdev->read_errors) > max_read_errors) { | |
2397 | char b[BDEVNAME_SIZE]; | |
2398 | bdevname(rdev->bdev, b); | |
1e50915f | 2399 | |
08464e09 N |
2400 | pr_notice("md/raid10:%s: %s: Raid device exceeded read_error threshold [cur %d:max %d]\n", |
2401 | mdname(mddev), b, | |
2402 | atomic_read(&rdev->read_errors), max_read_errors); | |
2403 | pr_notice("md/raid10:%s: %s: Failing raid device\n", | |
2404 | mdname(mddev), b); | |
d683c8e0 | 2405 | md_error(mddev, rdev); |
fae8cc5e | 2406 | r10_bio->devs[r10_bio->read_slot].bio = IO_BLOCKED; |
7c4e06ff | 2407 | return; |
1e50915f | 2408 | } |
1e50915f | 2409 | |
6814d536 N |
2410 | while(sectors) { |
2411 | int s = sectors; | |
2412 | int sl = r10_bio->read_slot; | |
2413 | int success = 0; | |
2414 | int start; | |
2415 | ||
2416 | if (s > (PAGE_SIZE>>9)) | |
2417 | s = PAGE_SIZE >> 9; | |
2418 | ||
2419 | rcu_read_lock(); | |
2420 | do { | |
8dbed5ce N |
2421 | sector_t first_bad; |
2422 | int bad_sectors; | |
2423 | ||
0544a21d | 2424 | d = r10_bio->devs[sl].devnum; |
6814d536 N |
2425 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
2426 | if (rdev && | |
8dbed5ce | 2427 | test_bit(In_sync, &rdev->flags) && |
f5b67ae8 | 2428 | !test_bit(Faulty, &rdev->flags) && |
8dbed5ce N |
2429 | is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, |
2430 | &first_bad, &bad_sectors) == 0) { | |
6814d536 N |
2431 | atomic_inc(&rdev->nr_pending); |
2432 | rcu_read_unlock(); | |
2b193363 | 2433 | success = sync_page_io(rdev, |
6814d536 | 2434 | r10_bio->devs[sl].addr + |
ccebd4c4 | 2435 | sect, |
6814d536 | 2436 | s<<9, |
796a5cf0 MC |
2437 | conf->tmppage, |
2438 | REQ_OP_READ, 0, false); | |
6814d536 N |
2439 | rdev_dec_pending(rdev, mddev); |
2440 | rcu_read_lock(); | |
2441 | if (success) | |
2442 | break; | |
2443 | } | |
2444 | sl++; | |
2445 | if (sl == conf->copies) | |
2446 | sl = 0; | |
2447 | } while (!success && sl != r10_bio->read_slot); | |
2448 | rcu_read_unlock(); | |
2449 | ||
2450 | if (!success) { | |
58c54fcc N |
2451 | /* Cannot read from anywhere, just mark the block |
2452 | * as bad on the first device to discourage future | |
2453 | * reads. | |
2454 | */ | |
6814d536 | 2455 | int dn = r10_bio->devs[r10_bio->read_slot].devnum; |
58c54fcc N |
2456 | rdev = conf->mirrors[dn].rdev; |
2457 | ||
2458 | if (!rdev_set_badblocks( | |
2459 | rdev, | |
2460 | r10_bio->devs[r10_bio->read_slot].addr | |
2461 | + sect, | |
fae8cc5e | 2462 | s, 0)) { |
58c54fcc | 2463 | md_error(mddev, rdev); |
fae8cc5e N |
2464 | r10_bio->devs[r10_bio->read_slot].bio |
2465 | = IO_BLOCKED; | |
2466 | } | |
6814d536 N |
2467 | break; |
2468 | } | |
2469 | ||
2470 | start = sl; | |
2471 | /* write it back and re-read */ | |
2472 | rcu_read_lock(); | |
2473 | while (sl != r10_bio->read_slot) { | |
67b8dc4b | 2474 | char b[BDEVNAME_SIZE]; |
0544a21d | 2475 | |
6814d536 N |
2476 | if (sl==0) |
2477 | sl = conf->copies; | |
2478 | sl--; | |
2479 | d = r10_bio->devs[sl].devnum; | |
2480 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2481 | if (!rdev || |
f5b67ae8 | 2482 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2483 | !test_bit(In_sync, &rdev->flags)) |
2484 | continue; | |
2485 | ||
2486 | atomic_inc(&rdev->nr_pending); | |
2487 | rcu_read_unlock(); | |
58c54fcc N |
2488 | if (r10_sync_page_io(rdev, |
2489 | r10_bio->devs[sl].addr + | |
2490 | sect, | |
055d3747 | 2491 | s, conf->tmppage, WRITE) |
1294b9c9 N |
2492 | == 0) { |
2493 | /* Well, this device is dead */ | |
08464e09 N |
2494 | pr_notice("md/raid10:%s: read correction write failed (%d sectors at %llu on %s)\n", |
2495 | mdname(mddev), s, | |
2496 | (unsigned long long)( | |
2497 | sect + | |
2498 | choose_data_offset(r10_bio, | |
2499 | rdev)), | |
2500 | bdevname(rdev->bdev, b)); | |
2501 | pr_notice("md/raid10:%s: %s: failing drive\n", | |
2502 | mdname(mddev), | |
2503 | bdevname(rdev->bdev, b)); | |
6814d536 | 2504 | } |
1294b9c9 N |
2505 | rdev_dec_pending(rdev, mddev); |
2506 | rcu_read_lock(); | |
6814d536 N |
2507 | } |
2508 | sl = start; | |
2509 | while (sl != r10_bio->read_slot) { | |
1294b9c9 | 2510 | char b[BDEVNAME_SIZE]; |
0544a21d | 2511 | |
6814d536 N |
2512 | if (sl==0) |
2513 | sl = conf->copies; | |
2514 | sl--; | |
2515 | d = r10_bio->devs[sl].devnum; | |
2516 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
1294b9c9 | 2517 | if (!rdev || |
f5b67ae8 | 2518 | test_bit(Faulty, &rdev->flags) || |
1294b9c9 N |
2519 | !test_bit(In_sync, &rdev->flags)) |
2520 | continue; | |
6814d536 | 2521 | |
1294b9c9 N |
2522 | atomic_inc(&rdev->nr_pending); |
2523 | rcu_read_unlock(); | |
58c54fcc N |
2524 | switch (r10_sync_page_io(rdev, |
2525 | r10_bio->devs[sl].addr + | |
2526 | sect, | |
055d3747 | 2527 | s, conf->tmppage, |
58c54fcc N |
2528 | READ)) { |
2529 | case 0: | |
1294b9c9 | 2530 | /* Well, this device is dead */ |
08464e09 | 2531 | pr_notice("md/raid10:%s: unable to read back corrected sectors (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2532 | mdname(mddev), s, |
2533 | (unsigned long long)( | |
f8c9e74f N |
2534 | sect + |
2535 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 | 2536 | bdevname(rdev->bdev, b)); |
08464e09 | 2537 | pr_notice("md/raid10:%s: %s: failing drive\n", |
1294b9c9 N |
2538 | mdname(mddev), |
2539 | bdevname(rdev->bdev, b)); | |
58c54fcc N |
2540 | break; |
2541 | case 1: | |
08464e09 | 2542 | pr_info("md/raid10:%s: read error corrected (%d sectors at %llu on %s)\n", |
1294b9c9 N |
2543 | mdname(mddev), s, |
2544 | (unsigned long long)( | |
f8c9e74f N |
2545 | sect + |
2546 | choose_data_offset(r10_bio, rdev)), | |
1294b9c9 N |
2547 | bdevname(rdev->bdev, b)); |
2548 | atomic_add(s, &rdev->corrected_errors); | |
6814d536 | 2549 | } |
1294b9c9 N |
2550 | |
2551 | rdev_dec_pending(rdev, mddev); | |
2552 | rcu_read_lock(); | |
6814d536 N |
2553 | } |
2554 | rcu_read_unlock(); | |
2555 | ||
2556 | sectors -= s; | |
2557 | sect += s; | |
2558 | } | |
2559 | } | |
2560 | ||
9f2c9d12 | 2561 | static int narrow_write_error(struct r10bio *r10_bio, int i) |
bd870a16 N |
2562 | { |
2563 | struct bio *bio = r10_bio->master_bio; | |
fd01b88c | 2564 | struct mddev *mddev = r10_bio->mddev; |
e879a879 | 2565 | struct r10conf *conf = mddev->private; |
3cb03002 | 2566 | struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; |
bd870a16 N |
2567 | /* bio has the data to be written to slot 'i' where |
2568 | * we just recently had a write error. | |
2569 | * We repeatedly clone the bio and trim down to one block, | |
2570 | * then try the write. Where the write fails we record | |
2571 | * a bad block. | |
2572 | * It is conceivable that the bio doesn't exactly align with | |
2573 | * blocks. We must handle this. | |
2574 | * | |
2575 | * We currently own a reference to the rdev. | |
2576 | */ | |
2577 | ||
2578 | int block_sectors; | |
2579 | sector_t sector; | |
2580 | int sectors; | |
2581 | int sect_to_write = r10_bio->sectors; | |
2582 | int ok = 1; | |
2583 | ||
2584 | if (rdev->badblocks.shift < 0) | |
2585 | return 0; | |
2586 | ||
f04ebb0b N |
2587 | block_sectors = roundup(1 << rdev->badblocks.shift, |
2588 | bdev_logical_block_size(rdev->bdev) >> 9); | |
bd870a16 N |
2589 | sector = r10_bio->sector; |
2590 | sectors = ((r10_bio->sector + block_sectors) | |
2591 | & ~(sector_t)(block_sectors - 1)) | |
2592 | - sector; | |
2593 | ||
2594 | while (sect_to_write) { | |
2595 | struct bio *wbio; | |
27028626 | 2596 | sector_t wsector; |
bd870a16 N |
2597 | if (sectors > sect_to_write) |
2598 | sectors = sect_to_write; | |
2599 | /* Write at 'sector' for 'sectors' */ | |
d7a10308 | 2600 | wbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 2601 | bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); |
27028626 TM |
2602 | wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); |
2603 | wbio->bi_iter.bi_sector = wsector + | |
2604 | choose_data_offset(r10_bio, rdev); | |
bd870a16 | 2605 | wbio->bi_bdev = rdev->bdev; |
796a5cf0 | 2606 | bio_set_op_attrs(wbio, REQ_OP_WRITE, 0); |
4e49ea4a MC |
2607 | |
2608 | if (submit_bio_wait(wbio) < 0) | |
bd870a16 | 2609 | /* Failure! */ |
27028626 | 2610 | ok = rdev_set_badblocks(rdev, wsector, |
bd870a16 N |
2611 | sectors, 0) |
2612 | && ok; | |
2613 | ||
2614 | bio_put(wbio); | |
2615 | sect_to_write -= sectors; | |
2616 | sector += sectors; | |
2617 | sectors = block_sectors; | |
2618 | } | |
2619 | return ok; | |
2620 | } | |
2621 | ||
9f2c9d12 | 2622 | static void handle_read_error(struct mddev *mddev, struct r10bio *r10_bio) |
560f8e55 N |
2623 | { |
2624 | int slot = r10_bio->read_slot; | |
560f8e55 | 2625 | struct bio *bio; |
e879a879 | 2626 | struct r10conf *conf = mddev->private; |
abbf098e | 2627 | struct md_rdev *rdev = r10_bio->devs[slot].rdev; |
560f8e55 N |
2628 | char b[BDEVNAME_SIZE]; |
2629 | unsigned long do_sync; | |
856e08e2 | 2630 | int max_sectors; |
109e3765 N |
2631 | dev_t bio_dev; |
2632 | sector_t bio_last_sector; | |
560f8e55 N |
2633 | |
2634 | /* we got a read error. Maybe the drive is bad. Maybe just | |
2635 | * the block and we can fix it. | |
2636 | * We freeze all other IO, and try reading the block from | |
2637 | * other devices. When we find one, we re-write | |
2638 | * and check it that fixes the read error. | |
2639 | * This is all done synchronously while the array is | |
2640 | * frozen. | |
2641 | */ | |
fae8cc5e N |
2642 | bio = r10_bio->devs[slot].bio; |
2643 | bdevname(bio->bi_bdev, b); | |
109e3765 N |
2644 | bio_dev = bio->bi_bdev->bd_dev; |
2645 | bio_last_sector = r10_bio->devs[slot].addr + rdev->data_offset + r10_bio->sectors; | |
fae8cc5e N |
2646 | bio_put(bio); |
2647 | r10_bio->devs[slot].bio = NULL; | |
2648 | ||
8d3ca83d N |
2649 | if (mddev->ro) |
2650 | r10_bio->devs[slot].bio = IO_BLOCKED; | |
2651 | else if (!test_bit(FailFast, &rdev->flags)) { | |
e2d59925 | 2652 | freeze_array(conf, 1); |
560f8e55 N |
2653 | fix_read_error(conf, mddev, r10_bio); |
2654 | unfreeze_array(conf); | |
fae8cc5e | 2655 | } else |
8d3ca83d | 2656 | md_error(mddev, rdev); |
fae8cc5e | 2657 | |
abbf098e | 2658 | rdev_dec_pending(rdev, mddev); |
560f8e55 | 2659 | |
7399c31b | 2660 | read_more: |
96c3fd1f N |
2661 | rdev = read_balance(conf, r10_bio, &max_sectors); |
2662 | if (rdev == NULL) { | |
08464e09 N |
2663 | pr_crit_ratelimited("md/raid10:%s: %s: unrecoverable I/O read error for block %llu\n", |
2664 | mdname(mddev), b, | |
2665 | (unsigned long long)r10_bio->sector); | |
560f8e55 | 2666 | raid_end_bio_io(r10_bio); |
560f8e55 N |
2667 | return; |
2668 | } | |
2669 | ||
1eff9d32 | 2670 | do_sync = (r10_bio->master_bio->bi_opf & REQ_SYNC); |
560f8e55 | 2671 | slot = r10_bio->read_slot; |
08464e09 N |
2672 | pr_err_ratelimited("md/raid10:%s: %s: redirecting sector %llu to another mirror\n", |
2673 | mdname(mddev), | |
2674 | bdevname(rdev->bdev, b), | |
2675 | (unsigned long long)r10_bio->sector); | |
d7a10308 | 2676 | bio = bio_clone_fast(r10_bio->master_bio, GFP_NOIO, mddev->bio_set); |
4f024f37 | 2677 | bio_trim(bio, r10_bio->sector - bio->bi_iter.bi_sector, max_sectors); |
560f8e55 | 2678 | r10_bio->devs[slot].bio = bio; |
abbf098e | 2679 | r10_bio->devs[slot].rdev = rdev; |
4f024f37 | 2680 | bio->bi_iter.bi_sector = r10_bio->devs[slot].addr |
f8c9e74f | 2681 | + choose_data_offset(r10_bio, rdev); |
560f8e55 | 2682 | bio->bi_bdev = rdev->bdev; |
796a5cf0 | 2683 | bio_set_op_attrs(bio, REQ_OP_READ, do_sync); |
8d3ca83d N |
2684 | if (test_bit(FailFast, &rdev->flags) && |
2685 | test_bit(R10BIO_FailFast, &r10_bio->state)) | |
2686 | bio->bi_opf |= MD_FAILFAST; | |
560f8e55 N |
2687 | bio->bi_private = r10_bio; |
2688 | bio->bi_end_io = raid10_end_read_request; | |
109e3765 N |
2689 | trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), |
2690 | bio, bio_dev, | |
2691 | bio_last_sector - r10_bio->sectors); | |
2692 | ||
7399c31b N |
2693 | if (max_sectors < r10_bio->sectors) { |
2694 | /* Drat - have to split this up more */ | |
2695 | struct bio *mbio = r10_bio->master_bio; | |
2696 | int sectors_handled = | |
2697 | r10_bio->sector + max_sectors | |
4f024f37 | 2698 | - mbio->bi_iter.bi_sector; |
7399c31b N |
2699 | r10_bio->sectors = max_sectors; |
2700 | spin_lock_irq(&conf->device_lock); | |
2701 | if (mbio->bi_phys_segments == 0) | |
2702 | mbio->bi_phys_segments = 2; | |
2703 | else | |
2704 | mbio->bi_phys_segments++; | |
2705 | spin_unlock_irq(&conf->device_lock); | |
2706 | generic_make_request(bio); | |
7399c31b N |
2707 | |
2708 | r10_bio = mempool_alloc(conf->r10bio_pool, | |
2709 | GFP_NOIO); | |
2710 | r10_bio->master_bio = mbio; | |
aa8b57aa | 2711 | r10_bio->sectors = bio_sectors(mbio) - sectors_handled; |
7399c31b N |
2712 | r10_bio->state = 0; |
2713 | set_bit(R10BIO_ReadError, | |
2714 | &r10_bio->state); | |
2715 | r10_bio->mddev = mddev; | |
4f024f37 | 2716 | r10_bio->sector = mbio->bi_iter.bi_sector |
7399c31b N |
2717 | + sectors_handled; |
2718 | ||
2719 | goto read_more; | |
2720 | } else | |
2721 | generic_make_request(bio); | |
560f8e55 N |
2722 | } |
2723 | ||
e879a879 | 2724 | static void handle_write_completed(struct r10conf *conf, struct r10bio *r10_bio) |
749c55e9 N |
2725 | { |
2726 | /* Some sort of write request has finished and it | |
2727 | * succeeded in writing where we thought there was a | |
2728 | * bad block. So forget the bad block. | |
1a0b7cd8 N |
2729 | * Or possibly if failed and we need to record |
2730 | * a bad block. | |
749c55e9 N |
2731 | */ |
2732 | int m; | |
3cb03002 | 2733 | struct md_rdev *rdev; |
749c55e9 N |
2734 | |
2735 | if (test_bit(R10BIO_IsSync, &r10_bio->state) || | |
2736 | test_bit(R10BIO_IsRecover, &r10_bio->state)) { | |
1a0b7cd8 N |
2737 | for (m = 0; m < conf->copies; m++) { |
2738 | int dev = r10_bio->devs[m].devnum; | |
2739 | rdev = conf->mirrors[dev].rdev; | |
2740 | if (r10_bio->devs[m].bio == NULL) | |
2741 | continue; | |
4246a0b6 | 2742 | if (!r10_bio->devs[m].bio->bi_error) { |
749c55e9 N |
2743 | rdev_clear_badblocks( |
2744 | rdev, | |
2745 | r10_bio->devs[m].addr, | |
c6563a8c | 2746 | r10_bio->sectors, 0); |
1a0b7cd8 N |
2747 | } else { |
2748 | if (!rdev_set_badblocks( | |
2749 | rdev, | |
2750 | r10_bio->devs[m].addr, | |
2751 | r10_bio->sectors, 0)) | |
2752 | md_error(conf->mddev, rdev); | |
749c55e9 | 2753 | } |
9ad1aefc N |
2754 | rdev = conf->mirrors[dev].replacement; |
2755 | if (r10_bio->devs[m].repl_bio == NULL) | |
2756 | continue; | |
4246a0b6 CH |
2757 | |
2758 | if (!r10_bio->devs[m].repl_bio->bi_error) { | |
9ad1aefc N |
2759 | rdev_clear_badblocks( |
2760 | rdev, | |
2761 | r10_bio->devs[m].addr, | |
c6563a8c | 2762 | r10_bio->sectors, 0); |
9ad1aefc N |
2763 | } else { |
2764 | if (!rdev_set_badblocks( | |
2765 | rdev, | |
2766 | r10_bio->devs[m].addr, | |
2767 | r10_bio->sectors, 0)) | |
2768 | md_error(conf->mddev, rdev); | |
2769 | } | |
1a0b7cd8 | 2770 | } |
749c55e9 N |
2771 | put_buf(r10_bio); |
2772 | } else { | |
95af587e | 2773 | bool fail = false; |
bd870a16 N |
2774 | for (m = 0; m < conf->copies; m++) { |
2775 | int dev = r10_bio->devs[m].devnum; | |
2776 | struct bio *bio = r10_bio->devs[m].bio; | |
2777 | rdev = conf->mirrors[dev].rdev; | |
2778 | if (bio == IO_MADE_GOOD) { | |
749c55e9 N |
2779 | rdev_clear_badblocks( |
2780 | rdev, | |
2781 | r10_bio->devs[m].addr, | |
c6563a8c | 2782 | r10_bio->sectors, 0); |
749c55e9 | 2783 | rdev_dec_pending(rdev, conf->mddev); |
4246a0b6 | 2784 | } else if (bio != NULL && bio->bi_error) { |
95af587e | 2785 | fail = true; |
bd870a16 N |
2786 | if (!narrow_write_error(r10_bio, m)) { |
2787 | md_error(conf->mddev, rdev); | |
2788 | set_bit(R10BIO_Degraded, | |
2789 | &r10_bio->state); | |
2790 | } | |
2791 | rdev_dec_pending(rdev, conf->mddev); | |
749c55e9 | 2792 | } |
475b0321 N |
2793 | bio = r10_bio->devs[m].repl_bio; |
2794 | rdev = conf->mirrors[dev].replacement; | |
4ca40c2c | 2795 | if (rdev && bio == IO_MADE_GOOD) { |
475b0321 N |
2796 | rdev_clear_badblocks( |
2797 | rdev, | |
2798 | r10_bio->devs[m].addr, | |
c6563a8c | 2799 | r10_bio->sectors, 0); |
475b0321 N |
2800 | rdev_dec_pending(rdev, conf->mddev); |
2801 | } | |
bd870a16 | 2802 | } |
95af587e N |
2803 | if (fail) { |
2804 | spin_lock_irq(&conf->device_lock); | |
2805 | list_add(&r10_bio->retry_list, &conf->bio_end_io_list); | |
23ddba80 | 2806 | conf->nr_queued++; |
95af587e N |
2807 | spin_unlock_irq(&conf->device_lock); |
2808 | md_wakeup_thread(conf->mddev->thread); | |
c340702c N |
2809 | } else { |
2810 | if (test_bit(R10BIO_WriteError, | |
2811 | &r10_bio->state)) | |
2812 | close_write(r10_bio); | |
95af587e | 2813 | raid_end_bio_io(r10_bio); |
c340702c | 2814 | } |
749c55e9 N |
2815 | } |
2816 | } | |
2817 | ||
4ed8731d | 2818 | static void raid10d(struct md_thread *thread) |
1da177e4 | 2819 | { |
4ed8731d | 2820 | struct mddev *mddev = thread->mddev; |
9f2c9d12 | 2821 | struct r10bio *r10_bio; |
1da177e4 | 2822 | unsigned long flags; |
e879a879 | 2823 | struct r10conf *conf = mddev->private; |
1da177e4 | 2824 | struct list_head *head = &conf->retry_list; |
e1dfa0a2 | 2825 | struct blk_plug plug; |
1da177e4 LT |
2826 | |
2827 | md_check_recovery(mddev); | |
1da177e4 | 2828 | |
95af587e | 2829 | if (!list_empty_careful(&conf->bio_end_io_list) && |
2953079c | 2830 | !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
95af587e N |
2831 | LIST_HEAD(tmp); |
2832 | spin_lock_irqsave(&conf->device_lock, flags); | |
2953079c | 2833 | if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { |
23ddba80 SL |
2834 | while (!list_empty(&conf->bio_end_io_list)) { |
2835 | list_move(conf->bio_end_io_list.prev, &tmp); | |
2836 | conf->nr_queued--; | |
2837 | } | |
95af587e N |
2838 | } |
2839 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
2840 | while (!list_empty(&tmp)) { | |
a452744b MP |
2841 | r10_bio = list_first_entry(&tmp, struct r10bio, |
2842 | retry_list); | |
95af587e | 2843 | list_del(&r10_bio->retry_list); |
c340702c N |
2844 | if (mddev->degraded) |
2845 | set_bit(R10BIO_Degraded, &r10_bio->state); | |
2846 | ||
2847 | if (test_bit(R10BIO_WriteError, | |
2848 | &r10_bio->state)) | |
2849 | close_write(r10_bio); | |
95af587e N |
2850 | raid_end_bio_io(r10_bio); |
2851 | } | |
2852 | } | |
2853 | ||
e1dfa0a2 | 2854 | blk_start_plug(&plug); |
1da177e4 | 2855 | for (;;) { |
6cce3b23 | 2856 | |
0021b7bc | 2857 | flush_pending_writes(conf); |
6cce3b23 | 2858 | |
a35e63ef N |
2859 | spin_lock_irqsave(&conf->device_lock, flags); |
2860 | if (list_empty(head)) { | |
2861 | spin_unlock_irqrestore(&conf->device_lock, flags); | |
1da177e4 | 2862 | break; |
a35e63ef | 2863 | } |
9f2c9d12 | 2864 | r10_bio = list_entry(head->prev, struct r10bio, retry_list); |
1da177e4 | 2865 | list_del(head->prev); |
4443ae10 | 2866 | conf->nr_queued--; |
1da177e4 LT |
2867 | spin_unlock_irqrestore(&conf->device_lock, flags); |
2868 | ||
2869 | mddev = r10_bio->mddev; | |
070ec55d | 2870 | conf = mddev->private; |
bd870a16 N |
2871 | if (test_bit(R10BIO_MadeGood, &r10_bio->state) || |
2872 | test_bit(R10BIO_WriteError, &r10_bio->state)) | |
749c55e9 | 2873 | handle_write_completed(conf, r10_bio); |
3ea7daa5 N |
2874 | else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) |
2875 | reshape_request_write(mddev, r10_bio); | |
749c55e9 | 2876 | else if (test_bit(R10BIO_IsSync, &r10_bio->state)) |
1da177e4 | 2877 | sync_request_write(mddev, r10_bio); |
7eaceacc | 2878 | else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) |
1da177e4 | 2879 | recovery_request_write(mddev, r10_bio); |
856e08e2 | 2880 | else if (test_bit(R10BIO_ReadError, &r10_bio->state)) |
560f8e55 | 2881 | handle_read_error(mddev, r10_bio); |
856e08e2 N |
2882 | else { |
2883 | /* just a partial read to be scheduled from a | |
2884 | * separate context | |
2885 | */ | |
2886 | int slot = r10_bio->read_slot; | |
2887 | generic_make_request(r10_bio->devs[slot].bio); | |
2888 | } | |
560f8e55 | 2889 | |
1d9d5241 | 2890 | cond_resched(); |
2953079c | 2891 | if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) |
de393cde | 2892 | md_check_recovery(mddev); |
1da177e4 | 2893 | } |
e1dfa0a2 | 2894 | blk_finish_plug(&plug); |
1da177e4 LT |
2895 | } |
2896 | ||
e879a879 | 2897 | static int init_resync(struct r10conf *conf) |
1da177e4 LT |
2898 | { |
2899 | int buffs; | |
69335ef3 | 2900 | int i; |
1da177e4 LT |
2901 | |
2902 | buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; | |
b6385483 | 2903 | BUG_ON(conf->r10buf_pool); |
69335ef3 | 2904 | conf->have_replacement = 0; |
5cf00fcd | 2905 | for (i = 0; i < conf->geo.raid_disks; i++) |
69335ef3 N |
2906 | if (conf->mirrors[i].replacement) |
2907 | conf->have_replacement = 1; | |
1da177e4 LT |
2908 | conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); |
2909 | if (!conf->r10buf_pool) | |
2910 | return -ENOMEM; | |
2911 | conf->next_resync = 0; | |
2912 | return 0; | |
2913 | } | |
2914 | ||
2915 | /* | |
2916 | * perform a "sync" on one "block" | |
2917 | * | |
2918 | * We need to make sure that no normal I/O request - particularly write | |
2919 | * requests - conflict with active sync requests. | |
2920 | * | |
2921 | * This is achieved by tracking pending requests and a 'barrier' concept | |
2922 | * that can be installed to exclude normal IO requests. | |
2923 | * | |
2924 | * Resync and recovery are handled very differently. | |
2925 | * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. | |
2926 | * | |
2927 | * For resync, we iterate over virtual addresses, read all copies, | |
2928 | * and update if there are differences. If only one copy is live, | |
2929 | * skip it. | |
2930 | * For recovery, we iterate over physical addresses, read a good | |
2931 | * value for each non-in_sync drive, and over-write. | |
2932 | * | |
2933 | * So, for recovery we may have several outstanding complex requests for a | |
2934 | * given address, one for each out-of-sync device. We model this by allocating | |
2935 | * a number of r10_bio structures, one for each out-of-sync device. | |
2936 | * As we setup these structures, we collect all bio's together into a list | |
2937 | * which we then process collectively to add pages, and then process again | |
2938 | * to pass to generic_make_request. | |
2939 | * | |
2940 | * The r10_bio structures are linked using a borrowed master_bio pointer. | |
2941 | * This link is counted in ->remaining. When the r10_bio that points to NULL | |
2942 | * has its remaining count decremented to 0, the whole complex operation | |
2943 | * is complete. | |
2944 | * | |
2945 | */ | |
2946 | ||
849674e4 | 2947 | static sector_t raid10_sync_request(struct mddev *mddev, sector_t sector_nr, |
09314799 | 2948 | int *skipped) |
1da177e4 | 2949 | { |
e879a879 | 2950 | struct r10conf *conf = mddev->private; |
9f2c9d12 | 2951 | struct r10bio *r10_bio; |
1da177e4 LT |
2952 | struct bio *biolist = NULL, *bio; |
2953 | sector_t max_sector, nr_sectors; | |
1da177e4 | 2954 | int i; |
6cce3b23 | 2955 | int max_sync; |
57dab0bd | 2956 | sector_t sync_blocks; |
1da177e4 LT |
2957 | sector_t sectors_skipped = 0; |
2958 | int chunks_skipped = 0; | |
5cf00fcd | 2959 | sector_t chunk_mask = conf->geo.chunk_mask; |
1da177e4 LT |
2960 | |
2961 | if (!conf->r10buf_pool) | |
2962 | if (init_resync(conf)) | |
57afd89f | 2963 | return 0; |
1da177e4 | 2964 | |
7e83ccbe MW |
2965 | /* |
2966 | * Allow skipping a full rebuild for incremental assembly | |
2967 | * of a clean array, like RAID1 does. | |
2968 | */ | |
2969 | if (mddev->bitmap == NULL && | |
2970 | mddev->recovery_cp == MaxSector && | |
13765120 N |
2971 | mddev->reshape_position == MaxSector && |
2972 | !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && | |
7e83ccbe | 2973 | !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && |
13765120 | 2974 | !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && |
7e83ccbe MW |
2975 | conf->fullsync == 0) { |
2976 | *skipped = 1; | |
13765120 | 2977 | return mddev->dev_sectors - sector_nr; |
7e83ccbe MW |
2978 | } |
2979 | ||
1da177e4 | 2980 | skipped: |
58c0fed4 | 2981 | max_sector = mddev->dev_sectors; |
3ea7daa5 N |
2982 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || |
2983 | test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
1da177e4 LT |
2984 | max_sector = mddev->resync_max_sectors; |
2985 | if (sector_nr >= max_sector) { | |
6cce3b23 N |
2986 | /* If we aborted, we need to abort the |
2987 | * sync on the 'current' bitmap chucks (there can | |
2988 | * be several when recovering multiple devices). | |
2989 | * as we may have started syncing it but not finished. | |
2990 | * We can find the current address in | |
2991 | * mddev->curr_resync, but for recovery, | |
2992 | * we need to convert that to several | |
2993 | * virtual addresses. | |
2994 | */ | |
3ea7daa5 N |
2995 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { |
2996 | end_reshape(conf); | |
b3968552 | 2997 | close_sync(conf); |
3ea7daa5 N |
2998 | return 0; |
2999 | } | |
3000 | ||
6cce3b23 N |
3001 | if (mddev->curr_resync < max_sector) { /* aborted */ |
3002 | if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) | |
3003 | bitmap_end_sync(mddev->bitmap, mddev->curr_resync, | |
3004 | &sync_blocks, 1); | |
5cf00fcd | 3005 | else for (i = 0; i < conf->geo.raid_disks; i++) { |
6cce3b23 N |
3006 | sector_t sect = |
3007 | raid10_find_virt(conf, mddev->curr_resync, i); | |
3008 | bitmap_end_sync(mddev->bitmap, sect, | |
3009 | &sync_blocks, 1); | |
3010 | } | |
9ad1aefc N |
3011 | } else { |
3012 | /* completed sync */ | |
3013 | if ((!mddev->bitmap || conf->fullsync) | |
3014 | && conf->have_replacement | |
3015 | && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { | |
3016 | /* Completed a full sync so the replacements | |
3017 | * are now fully recovered. | |
3018 | */ | |
f90145f3 N |
3019 | rcu_read_lock(); |
3020 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
3021 | struct md_rdev *rdev = | |
3022 | rcu_dereference(conf->mirrors[i].replacement); | |
3023 | if (rdev) | |
3024 | rdev->recovery_offset = MaxSector; | |
3025 | } | |
3026 | rcu_read_unlock(); | |
9ad1aefc | 3027 | } |
6cce3b23 | 3028 | conf->fullsync = 0; |
9ad1aefc | 3029 | } |
6cce3b23 | 3030 | bitmap_close_sync(mddev->bitmap); |
1da177e4 | 3031 | close_sync(conf); |
57afd89f | 3032 | *skipped = 1; |
1da177e4 LT |
3033 | return sectors_skipped; |
3034 | } | |
3ea7daa5 N |
3035 | |
3036 | if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) | |
3037 | return reshape_request(mddev, sector_nr, skipped); | |
3038 | ||
5cf00fcd | 3039 | if (chunks_skipped >= conf->geo.raid_disks) { |
1da177e4 LT |
3040 | /* if there has been nothing to do on any drive, |
3041 | * then there is nothing to do at all.. | |
3042 | */ | |
57afd89f N |
3043 | *skipped = 1; |
3044 | return (max_sector - sector_nr) + sectors_skipped; | |
1da177e4 LT |
3045 | } |
3046 | ||
c6207277 N |
3047 | if (max_sector > mddev->resync_max) |
3048 | max_sector = mddev->resync_max; /* Don't do IO beyond here */ | |
3049 | ||
1da177e4 LT |
3050 | /* make sure whole request will fit in a chunk - if chunks |
3051 | * are meaningful | |
3052 | */ | |
5cf00fcd N |
3053 | if (conf->geo.near_copies < conf->geo.raid_disks && |
3054 | max_sector > (sector_nr | chunk_mask)) | |
3055 | max_sector = (sector_nr | chunk_mask) + 1; | |
1da177e4 | 3056 | |
7ac50447 TM |
3057 | /* |
3058 | * If there is non-resync activity waiting for a turn, then let it | |
3059 | * though before starting on this new sync request. | |
3060 | */ | |
3061 | if (conf->nr_waiting) | |
3062 | schedule_timeout_uninterruptible(1); | |
3063 | ||
1da177e4 LT |
3064 | /* Again, very different code for resync and recovery. |
3065 | * Both must result in an r10bio with a list of bios that | |
3066 | * have bi_end_io, bi_sector, bi_bdev set, | |
3067 | * and bi_private set to the r10bio. | |
3068 | * For recovery, we may actually create several r10bios | |
3069 | * with 2 bios in each, that correspond to the bios in the main one. | |
3070 | * In this case, the subordinate r10bios link back through a | |
3071 | * borrowed master_bio pointer, and the counter in the master | |
3072 | * includes a ref from each subordinate. | |
3073 | */ | |
3074 | /* First, we decide what to do and set ->bi_end_io | |
3075 | * To end_sync_read if we want to read, and | |
3076 | * end_sync_write if we will want to write. | |
3077 | */ | |
3078 | ||
6cce3b23 | 3079 | max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); |
1da177e4 LT |
3080 | if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { |
3081 | /* recovery... the complicated one */ | |
e875ecea | 3082 | int j; |
1da177e4 LT |
3083 | r10_bio = NULL; |
3084 | ||
5cf00fcd | 3085 | for (i = 0 ; i < conf->geo.raid_disks; i++) { |
ab9d47e9 | 3086 | int still_degraded; |
9f2c9d12 | 3087 | struct r10bio *rb2; |
ab9d47e9 N |
3088 | sector_t sect; |
3089 | int must_sync; | |
e875ecea | 3090 | int any_working; |
dc280d98 | 3091 | struct raid10_info *mirror = &conf->mirrors[i]; |
f90145f3 | 3092 | struct md_rdev *mrdev, *mreplace; |
24afd80d | 3093 | |
f90145f3 N |
3094 | rcu_read_lock(); |
3095 | mrdev = rcu_dereference(mirror->rdev); | |
3096 | mreplace = rcu_dereference(mirror->replacement); | |
3097 | ||
3098 | if ((mrdev == NULL || | |
f5b67ae8 | 3099 | test_bit(Faulty, &mrdev->flags) || |
f90145f3 N |
3100 | test_bit(In_sync, &mrdev->flags)) && |
3101 | (mreplace == NULL || | |
3102 | test_bit(Faulty, &mreplace->flags))) { | |
3103 | rcu_read_unlock(); | |
ab9d47e9 | 3104 | continue; |
f90145f3 | 3105 | } |
1da177e4 | 3106 | |
ab9d47e9 N |
3107 | still_degraded = 0; |
3108 | /* want to reconstruct this device */ | |
3109 | rb2 = r10_bio; | |
3110 | sect = raid10_find_virt(conf, sector_nr, i); | |
fc448a18 N |
3111 | if (sect >= mddev->resync_max_sectors) { |
3112 | /* last stripe is not complete - don't | |
3113 | * try to recover this sector. | |
3114 | */ | |
f90145f3 | 3115 | rcu_read_unlock(); |
fc448a18 N |
3116 | continue; |
3117 | } | |
f5b67ae8 N |
3118 | if (mreplace && test_bit(Faulty, &mreplace->flags)) |
3119 | mreplace = NULL; | |
24afd80d N |
3120 | /* Unless we are doing a full sync, or a replacement |
3121 | * we only need to recover the block if it is set in | |
3122 | * the bitmap | |
ab9d47e9 N |
3123 | */ |
3124 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3125 | &sync_blocks, 1); | |
3126 | if (sync_blocks < max_sync) | |
3127 | max_sync = sync_blocks; | |
3128 | if (!must_sync && | |
f90145f3 | 3129 | mreplace == NULL && |
ab9d47e9 N |
3130 | !conf->fullsync) { |
3131 | /* yep, skip the sync_blocks here, but don't assume | |
3132 | * that there will never be anything to do here | |
3133 | */ | |
3134 | chunks_skipped = -1; | |
f90145f3 | 3135 | rcu_read_unlock(); |
ab9d47e9 N |
3136 | continue; |
3137 | } | |
f90145f3 N |
3138 | atomic_inc(&mrdev->nr_pending); |
3139 | if (mreplace) | |
3140 | atomic_inc(&mreplace->nr_pending); | |
3141 | rcu_read_unlock(); | |
6cce3b23 | 3142 | |
ab9d47e9 | 3143 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3144 | r10_bio->state = 0; |
ab9d47e9 N |
3145 | raise_barrier(conf, rb2 != NULL); |
3146 | atomic_set(&r10_bio->remaining, 0); | |
18055569 | 3147 | |
ab9d47e9 N |
3148 | r10_bio->master_bio = (struct bio*)rb2; |
3149 | if (rb2) | |
3150 | atomic_inc(&rb2->remaining); | |
3151 | r10_bio->mddev = mddev; | |
3152 | set_bit(R10BIO_IsRecover, &r10_bio->state); | |
3153 | r10_bio->sector = sect; | |
1da177e4 | 3154 | |
ab9d47e9 N |
3155 | raid10_find_phys(conf, r10_bio); |
3156 | ||
3157 | /* Need to check if the array will still be | |
3158 | * degraded | |
3159 | */ | |
f90145f3 N |
3160 | rcu_read_lock(); |
3161 | for (j = 0; j < conf->geo.raid_disks; j++) { | |
3162 | struct md_rdev *rdev = rcu_dereference( | |
3163 | conf->mirrors[j].rdev); | |
3164 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
ab9d47e9 | 3165 | still_degraded = 1; |
87fc767b | 3166 | break; |
1da177e4 | 3167 | } |
f90145f3 | 3168 | } |
ab9d47e9 N |
3169 | |
3170 | must_sync = bitmap_start_sync(mddev->bitmap, sect, | |
3171 | &sync_blocks, still_degraded); | |
3172 | ||
e875ecea | 3173 | any_working = 0; |
ab9d47e9 | 3174 | for (j=0; j<conf->copies;j++) { |
e875ecea | 3175 | int k; |
ab9d47e9 | 3176 | int d = r10_bio->devs[j].devnum; |
5e570289 | 3177 | sector_t from_addr, to_addr; |
f90145f3 N |
3178 | struct md_rdev *rdev = |
3179 | rcu_dereference(conf->mirrors[d].rdev); | |
40c356ce N |
3180 | sector_t sector, first_bad; |
3181 | int bad_sectors; | |
f90145f3 N |
3182 | if (!rdev || |
3183 | !test_bit(In_sync, &rdev->flags)) | |
ab9d47e9 N |
3184 | continue; |
3185 | /* This is where we read from */ | |
e875ecea | 3186 | any_working = 1; |
40c356ce N |
3187 | sector = r10_bio->devs[j].addr; |
3188 | ||
3189 | if (is_badblock(rdev, sector, max_sync, | |
3190 | &first_bad, &bad_sectors)) { | |
3191 | if (first_bad > sector) | |
3192 | max_sync = first_bad - sector; | |
3193 | else { | |
3194 | bad_sectors -= (sector | |
3195 | - first_bad); | |
3196 | if (max_sync > bad_sectors) | |
3197 | max_sync = bad_sectors; | |
3198 | continue; | |
3199 | } | |
3200 | } | |
ab9d47e9 | 3201 | bio = r10_bio->devs[0].bio; |
8be185f2 | 3202 | bio_reset(bio); |
ab9d47e9 N |
3203 | bio->bi_next = biolist; |
3204 | biolist = bio; | |
3205 | bio->bi_private = r10_bio; | |
3206 | bio->bi_end_io = end_sync_read; | |
796a5cf0 | 3207 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3208 | if (test_bit(FailFast, &rdev->flags)) |
3209 | bio->bi_opf |= MD_FAILFAST; | |
5e570289 | 3210 | from_addr = r10_bio->devs[j].addr; |
4f024f37 KO |
3211 | bio->bi_iter.bi_sector = from_addr + |
3212 | rdev->data_offset; | |
24afd80d N |
3213 | bio->bi_bdev = rdev->bdev; |
3214 | atomic_inc(&rdev->nr_pending); | |
3215 | /* and we write to 'i' (if not in_sync) */ | |
ab9d47e9 N |
3216 | |
3217 | for (k=0; k<conf->copies; k++) | |
3218 | if (r10_bio->devs[k].devnum == i) | |
3219 | break; | |
3220 | BUG_ON(k == conf->copies); | |
5e570289 | 3221 | to_addr = r10_bio->devs[k].addr; |
ab9d47e9 | 3222 | r10_bio->devs[0].devnum = d; |
5e570289 | 3223 | r10_bio->devs[0].addr = from_addr; |
ab9d47e9 | 3224 | r10_bio->devs[1].devnum = i; |
5e570289 | 3225 | r10_bio->devs[1].addr = to_addr; |
ab9d47e9 | 3226 | |
f90145f3 | 3227 | if (!test_bit(In_sync, &mrdev->flags)) { |
24afd80d | 3228 | bio = r10_bio->devs[1].bio; |
8be185f2 | 3229 | bio_reset(bio); |
24afd80d N |
3230 | bio->bi_next = biolist; |
3231 | biolist = bio; | |
3232 | bio->bi_private = r10_bio; | |
3233 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3234 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3235 | bio->bi_iter.bi_sector = to_addr |
f90145f3 N |
3236 | + mrdev->data_offset; |
3237 | bio->bi_bdev = mrdev->bdev; | |
24afd80d N |
3238 | atomic_inc(&r10_bio->remaining); |
3239 | } else | |
3240 | r10_bio->devs[1].bio->bi_end_io = NULL; | |
3241 | ||
3242 | /* and maybe write to replacement */ | |
3243 | bio = r10_bio->devs[1].repl_bio; | |
3244 | if (bio) | |
3245 | bio->bi_end_io = NULL; | |
f90145f3 | 3246 | /* Note: if mreplace != NULL, then bio |
24afd80d N |
3247 | * cannot be NULL as r10buf_pool_alloc will |
3248 | * have allocated it. | |
3249 | * So the second test here is pointless. | |
3250 | * But it keeps semantic-checkers happy, and | |
3251 | * this comment keeps human reviewers | |
3252 | * happy. | |
3253 | */ | |
f90145f3 N |
3254 | if (mreplace == NULL || bio == NULL || |
3255 | test_bit(Faulty, &mreplace->flags)) | |
24afd80d | 3256 | break; |
8be185f2 | 3257 | bio_reset(bio); |
24afd80d N |
3258 | bio->bi_next = biolist; |
3259 | biolist = bio; | |
3260 | bio->bi_private = r10_bio; | |
3261 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3262 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
4f024f37 | 3263 | bio->bi_iter.bi_sector = to_addr + |
f90145f3 N |
3264 | mreplace->data_offset; |
3265 | bio->bi_bdev = mreplace->bdev; | |
24afd80d | 3266 | atomic_inc(&r10_bio->remaining); |
ab9d47e9 N |
3267 | break; |
3268 | } | |
f90145f3 | 3269 | rcu_read_unlock(); |
ab9d47e9 | 3270 | if (j == conf->copies) { |
e875ecea N |
3271 | /* Cannot recover, so abort the recovery or |
3272 | * record a bad block */ | |
e875ecea N |
3273 | if (any_working) { |
3274 | /* problem is that there are bad blocks | |
3275 | * on other device(s) | |
3276 | */ | |
3277 | int k; | |
3278 | for (k = 0; k < conf->copies; k++) | |
3279 | if (r10_bio->devs[k].devnum == i) | |
3280 | break; | |
24afd80d | 3281 | if (!test_bit(In_sync, |
f90145f3 | 3282 | &mrdev->flags) |
24afd80d | 3283 | && !rdev_set_badblocks( |
f90145f3 | 3284 | mrdev, |
24afd80d N |
3285 | r10_bio->devs[k].addr, |
3286 | max_sync, 0)) | |
3287 | any_working = 0; | |
f90145f3 | 3288 | if (mreplace && |
24afd80d | 3289 | !rdev_set_badblocks( |
f90145f3 | 3290 | mreplace, |
e875ecea N |
3291 | r10_bio->devs[k].addr, |
3292 | max_sync, 0)) | |
3293 | any_working = 0; | |
3294 | } | |
3295 | if (!any_working) { | |
3296 | if (!test_and_set_bit(MD_RECOVERY_INTR, | |
3297 | &mddev->recovery)) | |
08464e09 | 3298 | pr_warn("md/raid10:%s: insufficient working devices for recovery.\n", |
e875ecea | 3299 | mdname(mddev)); |
24afd80d | 3300 | mirror->recovery_disabled |
e875ecea N |
3301 | = mddev->recovery_disabled; |
3302 | } | |
e8b84915 N |
3303 | put_buf(r10_bio); |
3304 | if (rb2) | |
3305 | atomic_dec(&rb2->remaining); | |
3306 | r10_bio = rb2; | |
f90145f3 N |
3307 | rdev_dec_pending(mrdev, mddev); |
3308 | if (mreplace) | |
3309 | rdev_dec_pending(mreplace, mddev); | |
ab9d47e9 | 3310 | break; |
1da177e4 | 3311 | } |
f90145f3 N |
3312 | rdev_dec_pending(mrdev, mddev); |
3313 | if (mreplace) | |
3314 | rdev_dec_pending(mreplace, mddev); | |
8d3ca83d N |
3315 | if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { |
3316 | /* Only want this if there is elsewhere to | |
3317 | * read from. 'j' is currently the first | |
3318 | * readable copy. | |
3319 | */ | |
3320 | int targets = 1; | |
3321 | for (; j < conf->copies; j++) { | |
3322 | int d = r10_bio->devs[j].devnum; | |
3323 | if (conf->mirrors[d].rdev && | |
3324 | test_bit(In_sync, | |
3325 | &conf->mirrors[d].rdev->flags)) | |
3326 | targets++; | |
3327 | } | |
3328 | if (targets == 1) | |
3329 | r10_bio->devs[0].bio->bi_opf | |
3330 | &= ~MD_FAILFAST; | |
3331 | } | |
ab9d47e9 | 3332 | } |
1da177e4 LT |
3333 | if (biolist == NULL) { |
3334 | while (r10_bio) { | |
9f2c9d12 N |
3335 | struct r10bio *rb2 = r10_bio; |
3336 | r10_bio = (struct r10bio*) rb2->master_bio; | |
1da177e4 LT |
3337 | rb2->master_bio = NULL; |
3338 | put_buf(rb2); | |
3339 | } | |
3340 | goto giveup; | |
3341 | } | |
3342 | } else { | |
3343 | /* resync. Schedule a read for every block at this virt offset */ | |
3344 | int count = 0; | |
6cce3b23 | 3345 | |
c40f341f | 3346 | bitmap_cond_end_sync(mddev->bitmap, sector_nr, 0); |
78200d45 | 3347 | |
6cce3b23 N |
3348 | if (!bitmap_start_sync(mddev->bitmap, sector_nr, |
3349 | &sync_blocks, mddev->degraded) && | |
ab9d47e9 N |
3350 | !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, |
3351 | &mddev->recovery)) { | |
6cce3b23 N |
3352 | /* We can skip this block */ |
3353 | *skipped = 1; | |
3354 | return sync_blocks + sectors_skipped; | |
3355 | } | |
3356 | if (sync_blocks < max_sync) | |
3357 | max_sync = sync_blocks; | |
1da177e4 | 3358 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); |
cb8b12b5 | 3359 | r10_bio->state = 0; |
1da177e4 | 3360 | |
1da177e4 LT |
3361 | r10_bio->mddev = mddev; |
3362 | atomic_set(&r10_bio->remaining, 0); | |
6cce3b23 N |
3363 | raise_barrier(conf, 0); |
3364 | conf->next_resync = sector_nr; | |
1da177e4 LT |
3365 | |
3366 | r10_bio->master_bio = NULL; | |
3367 | r10_bio->sector = sector_nr; | |
3368 | set_bit(R10BIO_IsSync, &r10_bio->state); | |
3369 | raid10_find_phys(conf, r10_bio); | |
5cf00fcd | 3370 | r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; |
1da177e4 | 3371 | |
5cf00fcd | 3372 | for (i = 0; i < conf->copies; i++) { |
1da177e4 | 3373 | int d = r10_bio->devs[i].devnum; |
40c356ce N |
3374 | sector_t first_bad, sector; |
3375 | int bad_sectors; | |
f90145f3 | 3376 | struct md_rdev *rdev; |
40c356ce | 3377 | |
9ad1aefc N |
3378 | if (r10_bio->devs[i].repl_bio) |
3379 | r10_bio->devs[i].repl_bio->bi_end_io = NULL; | |
3380 | ||
1da177e4 | 3381 | bio = r10_bio->devs[i].bio; |
8be185f2 | 3382 | bio_reset(bio); |
4246a0b6 | 3383 | bio->bi_error = -EIO; |
f90145f3 N |
3384 | rcu_read_lock(); |
3385 | rdev = rcu_dereference(conf->mirrors[d].rdev); | |
3386 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3387 | rcu_read_unlock(); | |
1da177e4 | 3388 | continue; |
f90145f3 | 3389 | } |
40c356ce | 3390 | sector = r10_bio->devs[i].addr; |
f90145f3 | 3391 | if (is_badblock(rdev, sector, max_sync, |
40c356ce N |
3392 | &first_bad, &bad_sectors)) { |
3393 | if (first_bad > sector) | |
3394 | max_sync = first_bad - sector; | |
3395 | else { | |
3396 | bad_sectors -= (sector - first_bad); | |
3397 | if (max_sync > bad_sectors) | |
91502f09 | 3398 | max_sync = bad_sectors; |
f90145f3 | 3399 | rcu_read_unlock(); |
40c356ce N |
3400 | continue; |
3401 | } | |
3402 | } | |
f90145f3 | 3403 | atomic_inc(&rdev->nr_pending); |
1da177e4 LT |
3404 | atomic_inc(&r10_bio->remaining); |
3405 | bio->bi_next = biolist; | |
3406 | biolist = bio; | |
3407 | bio->bi_private = r10_bio; | |
3408 | bio->bi_end_io = end_sync_read; | |
796a5cf0 | 3409 | bio_set_op_attrs(bio, REQ_OP_READ, 0); |
8d3ca83d N |
3410 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
3411 | bio->bi_opf |= MD_FAILFAST; | |
f90145f3 N |
3412 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3413 | bio->bi_bdev = rdev->bdev; | |
1da177e4 | 3414 | count++; |
9ad1aefc | 3415 | |
f90145f3 N |
3416 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3417 | if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { | |
3418 | rcu_read_unlock(); | |
9ad1aefc | 3419 | continue; |
f90145f3 N |
3420 | } |
3421 | atomic_inc(&rdev->nr_pending); | |
3422 | rcu_read_unlock(); | |
9ad1aefc N |
3423 | |
3424 | /* Need to set up for writing to the replacement */ | |
3425 | bio = r10_bio->devs[i].repl_bio; | |
8be185f2 | 3426 | bio_reset(bio); |
4246a0b6 | 3427 | bio->bi_error = -EIO; |
9ad1aefc N |
3428 | |
3429 | sector = r10_bio->devs[i].addr; | |
9ad1aefc N |
3430 | bio->bi_next = biolist; |
3431 | biolist = bio; | |
3432 | bio->bi_private = r10_bio; | |
3433 | bio->bi_end_io = end_sync_write; | |
796a5cf0 | 3434 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
1919cbb2 N |
3435 | if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) |
3436 | bio->bi_opf |= MD_FAILFAST; | |
f90145f3 N |
3437 | bio->bi_iter.bi_sector = sector + rdev->data_offset; |
3438 | bio->bi_bdev = rdev->bdev; | |
9ad1aefc | 3439 | count++; |
1da177e4 LT |
3440 | } |
3441 | ||
3442 | if (count < 2) { | |
3443 | for (i=0; i<conf->copies; i++) { | |
3444 | int d = r10_bio->devs[i].devnum; | |
3445 | if (r10_bio->devs[i].bio->bi_end_io) | |
ab9d47e9 N |
3446 | rdev_dec_pending(conf->mirrors[d].rdev, |
3447 | mddev); | |
9ad1aefc N |
3448 | if (r10_bio->devs[i].repl_bio && |
3449 | r10_bio->devs[i].repl_bio->bi_end_io) | |
3450 | rdev_dec_pending( | |
3451 | conf->mirrors[d].replacement, | |
3452 | mddev); | |
1da177e4 LT |
3453 | } |
3454 | put_buf(r10_bio); | |
3455 | biolist = NULL; | |
3456 | goto giveup; | |
3457 | } | |
3458 | } | |
3459 | ||
1da177e4 | 3460 | nr_sectors = 0; |
6cce3b23 N |
3461 | if (sector_nr + max_sync < max_sector) |
3462 | max_sector = sector_nr + max_sync; | |
1da177e4 LT |
3463 | do { |
3464 | struct page *page; | |
3465 | int len = PAGE_SIZE; | |
1da177e4 LT |
3466 | if (sector_nr + (len>>9) > max_sector) |
3467 | len = (max_sector - sector_nr) << 9; | |
3468 | if (len == 0) | |
3469 | break; | |
3470 | for (bio= biolist ; bio ; bio=bio->bi_next) { | |
ab9d47e9 | 3471 | struct bio *bio2; |
1da177e4 | 3472 | page = bio->bi_io_vec[bio->bi_vcnt].bv_page; |
ab9d47e9 N |
3473 | if (bio_add_page(bio, page, len, 0)) |
3474 | continue; | |
3475 | ||
3476 | /* stop here */ | |
3477 | bio->bi_io_vec[bio->bi_vcnt].bv_page = page; | |
3478 | for (bio2 = biolist; | |
3479 | bio2 && bio2 != bio; | |
3480 | bio2 = bio2->bi_next) { | |
3481 | /* remove last page from this bio */ | |
3482 | bio2->bi_vcnt--; | |
4f024f37 | 3483 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 3484 | bio_clear_flag(bio2, BIO_SEG_VALID); |
1da177e4 | 3485 | } |
ab9d47e9 | 3486 | goto bio_full; |
1da177e4 LT |
3487 | } |
3488 | nr_sectors += len>>9; | |
3489 | sector_nr += len>>9; | |
3490 | } while (biolist->bi_vcnt < RESYNC_PAGES); | |
3491 | bio_full: | |
3492 | r10_bio->sectors = nr_sectors; | |
3493 | ||
3494 | while (biolist) { | |
3495 | bio = biolist; | |
3496 | biolist = biolist->bi_next; | |
3497 | ||
3498 | bio->bi_next = NULL; | |
3499 | r10_bio = bio->bi_private; | |
3500 | r10_bio->sectors = nr_sectors; | |
3501 | ||
3502 | if (bio->bi_end_io == end_sync_read) { | |
3503 | md_sync_acct(bio->bi_bdev, nr_sectors); | |
4246a0b6 | 3504 | bio->bi_error = 0; |
1da177e4 LT |
3505 | generic_make_request(bio); |
3506 | } | |
3507 | } | |
3508 | ||
57afd89f N |
3509 | if (sectors_skipped) |
3510 | /* pretend they weren't skipped, it makes | |
3511 | * no important difference in this case | |
3512 | */ | |
3513 | md_done_sync(mddev, sectors_skipped, 1); | |
3514 | ||
1da177e4 LT |
3515 | return sectors_skipped + nr_sectors; |
3516 | giveup: | |
3517 | /* There is nowhere to write, so all non-sync | |
e875ecea N |
3518 | * drives must be failed or in resync, all drives |
3519 | * have a bad block, so try the next chunk... | |
1da177e4 | 3520 | */ |
09b4068a N |
3521 | if (sector_nr + max_sync < max_sector) |
3522 | max_sector = sector_nr + max_sync; | |
3523 | ||
3524 | sectors_skipped += (max_sector - sector_nr); | |
1da177e4 LT |
3525 | chunks_skipped ++; |
3526 | sector_nr = max_sector; | |
1da177e4 | 3527 | goto skipped; |
1da177e4 LT |
3528 | } |
3529 | ||
80c3a6ce | 3530 | static sector_t |
fd01b88c | 3531 | raid10_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce DW |
3532 | { |
3533 | sector_t size; | |
e879a879 | 3534 | struct r10conf *conf = mddev->private; |
80c3a6ce DW |
3535 | |
3536 | if (!raid_disks) | |
3ea7daa5 N |
3537 | raid_disks = min(conf->geo.raid_disks, |
3538 | conf->prev.raid_disks); | |
80c3a6ce | 3539 | if (!sectors) |
dab8b292 | 3540 | sectors = conf->dev_sectors; |
80c3a6ce | 3541 | |
5cf00fcd N |
3542 | size = sectors >> conf->geo.chunk_shift; |
3543 | sector_div(size, conf->geo.far_copies); | |
80c3a6ce | 3544 | size = size * raid_disks; |
5cf00fcd | 3545 | sector_div(size, conf->geo.near_copies); |
80c3a6ce | 3546 | |
5cf00fcd | 3547 | return size << conf->geo.chunk_shift; |
80c3a6ce DW |
3548 | } |
3549 | ||
6508fdbf N |
3550 | static void calc_sectors(struct r10conf *conf, sector_t size) |
3551 | { | |
3552 | /* Calculate the number of sectors-per-device that will | |
3553 | * actually be used, and set conf->dev_sectors and | |
3554 | * conf->stride | |
3555 | */ | |
3556 | ||
5cf00fcd N |
3557 | size = size >> conf->geo.chunk_shift; |
3558 | sector_div(size, conf->geo.far_copies); | |
3559 | size = size * conf->geo.raid_disks; | |
3560 | sector_div(size, conf->geo.near_copies); | |
6508fdbf N |
3561 | /* 'size' is now the number of chunks in the array */ |
3562 | /* calculate "used chunks per device" */ | |
3563 | size = size * conf->copies; | |
3564 | ||
3565 | /* We need to round up when dividing by raid_disks to | |
3566 | * get the stride size. | |
3567 | */ | |
5cf00fcd | 3568 | size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); |
6508fdbf | 3569 | |
5cf00fcd | 3570 | conf->dev_sectors = size << conf->geo.chunk_shift; |
6508fdbf | 3571 | |
5cf00fcd N |
3572 | if (conf->geo.far_offset) |
3573 | conf->geo.stride = 1 << conf->geo.chunk_shift; | |
6508fdbf | 3574 | else { |
5cf00fcd N |
3575 | sector_div(size, conf->geo.far_copies); |
3576 | conf->geo.stride = size << conf->geo.chunk_shift; | |
6508fdbf N |
3577 | } |
3578 | } | |
dab8b292 | 3579 | |
deb200d0 N |
3580 | enum geo_type {geo_new, geo_old, geo_start}; |
3581 | static int setup_geo(struct geom *geo, struct mddev *mddev, enum geo_type new) | |
3582 | { | |
3583 | int nc, fc, fo; | |
3584 | int layout, chunk, disks; | |
3585 | switch (new) { | |
3586 | case geo_old: | |
3587 | layout = mddev->layout; | |
3588 | chunk = mddev->chunk_sectors; | |
3589 | disks = mddev->raid_disks - mddev->delta_disks; | |
3590 | break; | |
3591 | case geo_new: | |
3592 | layout = mddev->new_layout; | |
3593 | chunk = mddev->new_chunk_sectors; | |
3594 | disks = mddev->raid_disks; | |
3595 | break; | |
3596 | default: /* avoid 'may be unused' warnings */ | |
3597 | case geo_start: /* new when starting reshape - raid_disks not | |
3598 | * updated yet. */ | |
3599 | layout = mddev->new_layout; | |
3600 | chunk = mddev->new_chunk_sectors; | |
3601 | disks = mddev->raid_disks + mddev->delta_disks; | |
3602 | break; | |
3603 | } | |
8bce6d35 | 3604 | if (layout >> 19) |
deb200d0 N |
3605 | return -1; |
3606 | if (chunk < (PAGE_SIZE >> 9) || | |
3607 | !is_power_of_2(chunk)) | |
3608 | return -2; | |
3609 | nc = layout & 255; | |
3610 | fc = (layout >> 8) & 255; | |
3611 | fo = layout & (1<<16); | |
3612 | geo->raid_disks = disks; | |
3613 | geo->near_copies = nc; | |
3614 | geo->far_copies = fc; | |
3615 | geo->far_offset = fo; | |
8bce6d35 N |
3616 | switch (layout >> 17) { |
3617 | case 0: /* original layout. simple but not always optimal */ | |
3618 | geo->far_set_size = disks; | |
3619 | break; | |
3620 | case 1: /* "improved" layout which was buggy. Hopefully no-one is | |
3621 | * actually using this, but leave code here just in case.*/ | |
3622 | geo->far_set_size = disks/fc; | |
3623 | WARN(geo->far_set_size < fc, | |
3624 | "This RAID10 layout does not provide data safety - please backup and create new array\n"); | |
3625 | break; | |
3626 | case 2: /* "improved" layout fixed to match documentation */ | |
3627 | geo->far_set_size = fc * nc; | |
3628 | break; | |
3629 | default: /* Not a valid layout */ | |
3630 | return -1; | |
3631 | } | |
deb200d0 N |
3632 | geo->chunk_mask = chunk - 1; |
3633 | geo->chunk_shift = ffz(~chunk); | |
3634 | return nc*fc; | |
3635 | } | |
3636 | ||
e879a879 | 3637 | static struct r10conf *setup_conf(struct mddev *mddev) |
1da177e4 | 3638 | { |
e879a879 | 3639 | struct r10conf *conf = NULL; |
dab8b292 | 3640 | int err = -EINVAL; |
deb200d0 N |
3641 | struct geom geo; |
3642 | int copies; | |
3643 | ||
3644 | copies = setup_geo(&geo, mddev, geo_new); | |
1da177e4 | 3645 | |
deb200d0 | 3646 | if (copies == -2) { |
08464e09 N |
3647 | pr_warn("md/raid10:%s: chunk size must be at least PAGE_SIZE(%ld) and be a power of 2.\n", |
3648 | mdname(mddev), PAGE_SIZE); | |
dab8b292 | 3649 | goto out; |
1da177e4 | 3650 | } |
2604b703 | 3651 | |
deb200d0 | 3652 | if (copies < 2 || copies > mddev->raid_disks) { |
08464e09 N |
3653 | pr_warn("md/raid10:%s: unsupported raid10 layout: 0x%8x\n", |
3654 | mdname(mddev), mddev->new_layout); | |
1da177e4 LT |
3655 | goto out; |
3656 | } | |
dab8b292 TM |
3657 | |
3658 | err = -ENOMEM; | |
e879a879 | 3659 | conf = kzalloc(sizeof(struct r10conf), GFP_KERNEL); |
dab8b292 | 3660 | if (!conf) |
1da177e4 | 3661 | goto out; |
dab8b292 | 3662 | |
3ea7daa5 | 3663 | /* FIXME calc properly */ |
dc280d98 | 3664 | conf->mirrors = kzalloc(sizeof(struct raid10_info)*(mddev->raid_disks + |
78eaa0d4 | 3665 | max(0,-mddev->delta_disks)), |
dab8b292 TM |
3666 | GFP_KERNEL); |
3667 | if (!conf->mirrors) | |
3668 | goto out; | |
4443ae10 N |
3669 | |
3670 | conf->tmppage = alloc_page(GFP_KERNEL); | |
3671 | if (!conf->tmppage) | |
dab8b292 TM |
3672 | goto out; |
3673 | ||
deb200d0 N |
3674 | conf->geo = geo; |
3675 | conf->copies = copies; | |
dab8b292 TM |
3676 | conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, |
3677 | r10bio_pool_free, conf); | |
3678 | if (!conf->r10bio_pool) | |
3679 | goto out; | |
3680 | ||
6508fdbf | 3681 | calc_sectors(conf, mddev->dev_sectors); |
3ea7daa5 N |
3682 | if (mddev->reshape_position == MaxSector) { |
3683 | conf->prev = conf->geo; | |
3684 | conf->reshape_progress = MaxSector; | |
3685 | } else { | |
3686 | if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { | |
3687 | err = -EINVAL; | |
3688 | goto out; | |
3689 | } | |
3690 | conf->reshape_progress = mddev->reshape_position; | |
3691 | if (conf->prev.far_offset) | |
3692 | conf->prev.stride = 1 << conf->prev.chunk_shift; | |
3693 | else | |
3694 | /* far_copies must be 1 */ | |
3695 | conf->prev.stride = conf->dev_sectors; | |
3696 | } | |
299b0685 | 3697 | conf->reshape_safe = conf->reshape_progress; |
e7e72bf6 | 3698 | spin_lock_init(&conf->device_lock); |
dab8b292 | 3699 | INIT_LIST_HEAD(&conf->retry_list); |
95af587e | 3700 | INIT_LIST_HEAD(&conf->bio_end_io_list); |
dab8b292 TM |
3701 | |
3702 | spin_lock_init(&conf->resync_lock); | |
3703 | init_waitqueue_head(&conf->wait_barrier); | |
0e5313e2 | 3704 | atomic_set(&conf->nr_pending, 0); |
dab8b292 | 3705 | |
0232605d | 3706 | conf->thread = md_register_thread(raid10d, mddev, "raid10"); |
dab8b292 TM |
3707 | if (!conf->thread) |
3708 | goto out; | |
3709 | ||
dab8b292 TM |
3710 | conf->mddev = mddev; |
3711 | return conf; | |
3712 | ||
3713 | out: | |
dab8b292 | 3714 | if (conf) { |
644df1a8 | 3715 | mempool_destroy(conf->r10bio_pool); |
dab8b292 TM |
3716 | kfree(conf->mirrors); |
3717 | safe_put_page(conf->tmppage); | |
3718 | kfree(conf); | |
3719 | } | |
3720 | return ERR_PTR(err); | |
3721 | } | |
3722 | ||
849674e4 | 3723 | static int raid10_run(struct mddev *mddev) |
dab8b292 | 3724 | { |
e879a879 | 3725 | struct r10conf *conf; |
dab8b292 | 3726 | int i, disk_idx, chunk_size; |
dc280d98 | 3727 | struct raid10_info *disk; |
3cb03002 | 3728 | struct md_rdev *rdev; |
dab8b292 | 3729 | sector_t size; |
3ea7daa5 N |
3730 | sector_t min_offset_diff = 0; |
3731 | int first = 1; | |
532a2a3f | 3732 | bool discard_supported = false; |
dab8b292 TM |
3733 | |
3734 | if (mddev->private == NULL) { | |
3735 | conf = setup_conf(mddev); | |
3736 | if (IS_ERR(conf)) | |
3737 | return PTR_ERR(conf); | |
3738 | mddev->private = conf; | |
3739 | } | |
3740 | conf = mddev->private; | |
3741 | if (!conf) | |
3742 | goto out; | |
3743 | ||
dab8b292 TM |
3744 | mddev->thread = conf->thread; |
3745 | conf->thread = NULL; | |
3746 | ||
8f6c2e4b | 3747 | chunk_size = mddev->chunk_sectors << 9; |
cc4d1efd | 3748 | if (mddev->queue) { |
532a2a3f SL |
3749 | blk_queue_max_discard_sectors(mddev->queue, |
3750 | mddev->chunk_sectors); | |
5026d7a9 | 3751 | blk_queue_max_write_same_sectors(mddev->queue, 0); |
cc4d1efd JB |
3752 | blk_queue_io_min(mddev->queue, chunk_size); |
3753 | if (conf->geo.raid_disks % conf->geo.near_copies) | |
3754 | blk_queue_io_opt(mddev->queue, chunk_size * conf->geo.raid_disks); | |
3755 | else | |
3756 | blk_queue_io_opt(mddev->queue, chunk_size * | |
3757 | (conf->geo.raid_disks / conf->geo.near_copies)); | |
3758 | } | |
8f6c2e4b | 3759 | |
dafb20fa | 3760 | rdev_for_each(rdev, mddev) { |
3ea7daa5 | 3761 | long long diff; |
aba336bd | 3762 | struct request_queue *q; |
34b343cf | 3763 | |
1da177e4 | 3764 | disk_idx = rdev->raid_disk; |
f8c9e74f N |
3765 | if (disk_idx < 0) |
3766 | continue; | |
3767 | if (disk_idx >= conf->geo.raid_disks && | |
3768 | disk_idx >= conf->prev.raid_disks) | |
1da177e4 LT |
3769 | continue; |
3770 | disk = conf->mirrors + disk_idx; | |
3771 | ||
56a2559b N |
3772 | if (test_bit(Replacement, &rdev->flags)) { |
3773 | if (disk->replacement) | |
3774 | goto out_free_conf; | |
3775 | disk->replacement = rdev; | |
3776 | } else { | |
3777 | if (disk->rdev) | |
3778 | goto out_free_conf; | |
3779 | disk->rdev = rdev; | |
3780 | } | |
aba336bd | 3781 | q = bdev_get_queue(rdev->bdev); |
3ea7daa5 N |
3782 | diff = (rdev->new_data_offset - rdev->data_offset); |
3783 | if (!mddev->reshape_backwards) | |
3784 | diff = -diff; | |
3785 | if (diff < 0) | |
3786 | diff = 0; | |
3787 | if (first || diff < min_offset_diff) | |
3788 | min_offset_diff = diff; | |
56a2559b | 3789 | |
cc4d1efd JB |
3790 | if (mddev->gendisk) |
3791 | disk_stack_limits(mddev->gendisk, rdev->bdev, | |
3792 | rdev->data_offset << 9); | |
1da177e4 LT |
3793 | |
3794 | disk->head_position = 0; | |
532a2a3f SL |
3795 | |
3796 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) | |
3797 | discard_supported = true; | |
1da177e4 | 3798 | } |
3ea7daa5 | 3799 | |
ed30be07 JB |
3800 | if (mddev->queue) { |
3801 | if (discard_supported) | |
3802 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, | |
3803 | mddev->queue); | |
3804 | else | |
3805 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, | |
3806 | mddev->queue); | |
3807 | } | |
6d508242 | 3808 | /* need to check that every block has at least one working mirror */ |
700c7213 | 3809 | if (!enough(conf, -1)) { |
08464e09 | 3810 | pr_err("md/raid10:%s: not enough operational mirrors.\n", |
6d508242 | 3811 | mdname(mddev)); |
1da177e4 LT |
3812 | goto out_free_conf; |
3813 | } | |
3814 | ||
3ea7daa5 N |
3815 | if (conf->reshape_progress != MaxSector) { |
3816 | /* must ensure that shape change is supported */ | |
3817 | if (conf->geo.far_copies != 1 && | |
3818 | conf->geo.far_offset == 0) | |
3819 | goto out_free_conf; | |
3820 | if (conf->prev.far_copies != 1 && | |
78eaa0d4 | 3821 | conf->prev.far_offset == 0) |
3ea7daa5 N |
3822 | goto out_free_conf; |
3823 | } | |
3824 | ||
1da177e4 | 3825 | mddev->degraded = 0; |
f8c9e74f N |
3826 | for (i = 0; |
3827 | i < conf->geo.raid_disks | |
3828 | || i < conf->prev.raid_disks; | |
3829 | i++) { | |
1da177e4 LT |
3830 | |
3831 | disk = conf->mirrors + i; | |
3832 | ||
56a2559b N |
3833 | if (!disk->rdev && disk->replacement) { |
3834 | /* The replacement is all we have - use it */ | |
3835 | disk->rdev = disk->replacement; | |
3836 | disk->replacement = NULL; | |
3837 | clear_bit(Replacement, &disk->rdev->flags); | |
3838 | } | |
3839 | ||
5fd6c1dc | 3840 | if (!disk->rdev || |
2e333e89 | 3841 | !test_bit(In_sync, &disk->rdev->flags)) { |
1da177e4 LT |
3842 | disk->head_position = 0; |
3843 | mddev->degraded++; | |
0b59bb64 N |
3844 | if (disk->rdev && |
3845 | disk->rdev->saved_raid_disk < 0) | |
8c2e870a | 3846 | conf->fullsync = 1; |
1da177e4 | 3847 | } |
d890fa2b | 3848 | disk->recovery_disabled = mddev->recovery_disabled - 1; |
1da177e4 LT |
3849 | } |
3850 | ||
8c6ac868 | 3851 | if (mddev->recovery_cp != MaxSector) |
08464e09 N |
3852 | pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", |
3853 | mdname(mddev)); | |
3854 | pr_info("md/raid10:%s: active with %d out of %d devices\n", | |
5cf00fcd N |
3855 | mdname(mddev), conf->geo.raid_disks - mddev->degraded, |
3856 | conf->geo.raid_disks); | |
1da177e4 LT |
3857 | /* |
3858 | * Ok, everything is just fine now | |
3859 | */ | |
dab8b292 TM |
3860 | mddev->dev_sectors = conf->dev_sectors; |
3861 | size = raid10_size(mddev, 0, 0); | |
3862 | md_set_array_sectors(mddev, size); | |
3863 | mddev->resync_max_sectors = size; | |
46533ff7 | 3864 | set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); |
1da177e4 | 3865 | |
cc4d1efd | 3866 | if (mddev->queue) { |
5cf00fcd | 3867 | int stripe = conf->geo.raid_disks * |
9d8f0363 | 3868 | ((mddev->chunk_sectors << 9) / PAGE_SIZE); |
cc4d1efd JB |
3869 | |
3870 | /* Calculate max read-ahead size. | |
3871 | * We need to readahead at least twice a whole stripe.... | |
3872 | * maybe... | |
3873 | */ | |
5cf00fcd | 3874 | stripe /= conf->geo.near_copies; |
dc3b17cc JK |
3875 | if (mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
3876 | mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
1da177e4 LT |
3877 | } |
3878 | ||
a91a2785 MP |
3879 | if (md_integrity_register(mddev)) |
3880 | goto out_free_conf; | |
3881 | ||
3ea7daa5 N |
3882 | if (conf->reshape_progress != MaxSector) { |
3883 | unsigned long before_length, after_length; | |
3884 | ||
3885 | before_length = ((1 << conf->prev.chunk_shift) * | |
3886 | conf->prev.far_copies); | |
3887 | after_length = ((1 << conf->geo.chunk_shift) * | |
3888 | conf->geo.far_copies); | |
3889 | ||
3890 | if (max(before_length, after_length) > min_offset_diff) { | |
3891 | /* This cannot work */ | |
08464e09 | 3892 | pr_warn("md/raid10: offset difference not enough to continue reshape\n"); |
3ea7daa5 N |
3893 | goto out_free_conf; |
3894 | } | |
3895 | conf->offset_diff = min_offset_diff; | |
3896 | ||
3ea7daa5 N |
3897 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); |
3898 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
3899 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); | |
3900 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
3901 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
3902 | "reshape"); | |
3903 | } | |
3904 | ||
1da177e4 LT |
3905 | return 0; |
3906 | ||
3907 | out_free_conf: | |
01f96c0a | 3908 | md_unregister_thread(&mddev->thread); |
644df1a8 | 3909 | mempool_destroy(conf->r10bio_pool); |
1345b1d8 | 3910 | safe_put_page(conf->tmppage); |
990a8baf | 3911 | kfree(conf->mirrors); |
1da177e4 LT |
3912 | kfree(conf); |
3913 | mddev->private = NULL; | |
3914 | out: | |
3915 | return -EIO; | |
3916 | } | |
3917 | ||
afa0f557 | 3918 | static void raid10_free(struct mddev *mddev, void *priv) |
1da177e4 | 3919 | { |
afa0f557 | 3920 | struct r10conf *conf = priv; |
1da177e4 | 3921 | |
644df1a8 | 3922 | mempool_destroy(conf->r10bio_pool); |
0fea7ed8 | 3923 | safe_put_page(conf->tmppage); |
990a8baf | 3924 | kfree(conf->mirrors); |
c4796e21 N |
3925 | kfree(conf->mirrors_old); |
3926 | kfree(conf->mirrors_new); | |
1da177e4 | 3927 | kfree(conf); |
1da177e4 LT |
3928 | } |
3929 | ||
fd01b88c | 3930 | static void raid10_quiesce(struct mddev *mddev, int state) |
6cce3b23 | 3931 | { |
e879a879 | 3932 | struct r10conf *conf = mddev->private; |
6cce3b23 N |
3933 | |
3934 | switch(state) { | |
3935 | case 1: | |
3936 | raise_barrier(conf, 0); | |
3937 | break; | |
3938 | case 0: | |
3939 | lower_barrier(conf); | |
3940 | break; | |
3941 | } | |
6cce3b23 | 3942 | } |
1da177e4 | 3943 | |
006a09a0 N |
3944 | static int raid10_resize(struct mddev *mddev, sector_t sectors) |
3945 | { | |
3946 | /* Resize of 'far' arrays is not supported. | |
3947 | * For 'near' and 'offset' arrays we can set the | |
3948 | * number of sectors used to be an appropriate multiple | |
3949 | * of the chunk size. | |
3950 | * For 'offset', this is far_copies*chunksize. | |
3951 | * For 'near' the multiplier is the LCM of | |
3952 | * near_copies and raid_disks. | |
3953 | * So if far_copies > 1 && !far_offset, fail. | |
3954 | * Else find LCM(raid_disks, near_copy)*far_copies and | |
3955 | * multiply by chunk_size. Then round to this number. | |
3956 | * This is mostly done by raid10_size() | |
3957 | */ | |
3958 | struct r10conf *conf = mddev->private; | |
3959 | sector_t oldsize, size; | |
3960 | ||
f8c9e74f N |
3961 | if (mddev->reshape_position != MaxSector) |
3962 | return -EBUSY; | |
3963 | ||
5cf00fcd | 3964 | if (conf->geo.far_copies > 1 && !conf->geo.far_offset) |
006a09a0 N |
3965 | return -EINVAL; |
3966 | ||
3967 | oldsize = raid10_size(mddev, 0, 0); | |
3968 | size = raid10_size(mddev, sectors, 0); | |
a4a6125a N |
3969 | if (mddev->external_size && |
3970 | mddev->array_sectors > size) | |
006a09a0 | 3971 | return -EINVAL; |
a4a6125a N |
3972 | if (mddev->bitmap) { |
3973 | int ret = bitmap_resize(mddev->bitmap, size, 0, 0); | |
3974 | if (ret) | |
3975 | return ret; | |
3976 | } | |
3977 | md_set_array_sectors(mddev, size); | |
006a09a0 N |
3978 | if (sectors > mddev->dev_sectors && |
3979 | mddev->recovery_cp > oldsize) { | |
3980 | mddev->recovery_cp = oldsize; | |
3981 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
3982 | } | |
6508fdbf N |
3983 | calc_sectors(conf, sectors); |
3984 | mddev->dev_sectors = conf->dev_sectors; | |
006a09a0 N |
3985 | mddev->resync_max_sectors = size; |
3986 | return 0; | |
3987 | } | |
3988 | ||
53a6ab4d | 3989 | static void *raid10_takeover_raid0(struct mddev *mddev, sector_t size, int devs) |
dab8b292 | 3990 | { |
3cb03002 | 3991 | struct md_rdev *rdev; |
e879a879 | 3992 | struct r10conf *conf; |
dab8b292 TM |
3993 | |
3994 | if (mddev->degraded > 0) { | |
08464e09 N |
3995 | pr_warn("md/raid10:%s: Error: degraded raid0!\n", |
3996 | mdname(mddev)); | |
dab8b292 TM |
3997 | return ERR_PTR(-EINVAL); |
3998 | } | |
53a6ab4d | 3999 | sector_div(size, devs); |
dab8b292 | 4000 | |
dab8b292 TM |
4001 | /* Set new parameters */ |
4002 | mddev->new_level = 10; | |
4003 | /* new layout: far_copies = 1, near_copies = 2 */ | |
4004 | mddev->new_layout = (1<<8) + 2; | |
4005 | mddev->new_chunk_sectors = mddev->chunk_sectors; | |
4006 | mddev->delta_disks = mddev->raid_disks; | |
dab8b292 TM |
4007 | mddev->raid_disks *= 2; |
4008 | /* make sure it will be not marked as dirty */ | |
4009 | mddev->recovery_cp = MaxSector; | |
53a6ab4d | 4010 | mddev->dev_sectors = size; |
dab8b292 TM |
4011 | |
4012 | conf = setup_conf(mddev); | |
02214dc5 | 4013 | if (!IS_ERR(conf)) { |
dafb20fa | 4014 | rdev_for_each(rdev, mddev) |
53a6ab4d | 4015 | if (rdev->raid_disk >= 0) { |
e93f68a1 | 4016 | rdev->new_raid_disk = rdev->raid_disk * 2; |
53a6ab4d N |
4017 | rdev->sectors = size; |
4018 | } | |
02214dc5 KW |
4019 | conf->barrier = 1; |
4020 | } | |
4021 | ||
dab8b292 TM |
4022 | return conf; |
4023 | } | |
4024 | ||
fd01b88c | 4025 | static void *raid10_takeover(struct mddev *mddev) |
dab8b292 | 4026 | { |
e373ab10 | 4027 | struct r0conf *raid0_conf; |
dab8b292 TM |
4028 | |
4029 | /* raid10 can take over: | |
4030 | * raid0 - providing it has only two drives | |
4031 | */ | |
4032 | if (mddev->level == 0) { | |
4033 | /* for raid0 takeover only one zone is supported */ | |
e373ab10 N |
4034 | raid0_conf = mddev->private; |
4035 | if (raid0_conf->nr_strip_zones > 1) { | |
08464e09 N |
4036 | pr_warn("md/raid10:%s: cannot takeover raid 0 with more than one zone.\n", |
4037 | mdname(mddev)); | |
dab8b292 TM |
4038 | return ERR_PTR(-EINVAL); |
4039 | } | |
53a6ab4d N |
4040 | return raid10_takeover_raid0(mddev, |
4041 | raid0_conf->strip_zone->zone_end, | |
4042 | raid0_conf->strip_zone->nb_dev); | |
dab8b292 TM |
4043 | } |
4044 | return ERR_PTR(-EINVAL); | |
4045 | } | |
4046 | ||
3ea7daa5 N |
4047 | static int raid10_check_reshape(struct mddev *mddev) |
4048 | { | |
4049 | /* Called when there is a request to change | |
4050 | * - layout (to ->new_layout) | |
4051 | * - chunk size (to ->new_chunk_sectors) | |
4052 | * - raid_disks (by delta_disks) | |
4053 | * or when trying to restart a reshape that was ongoing. | |
4054 | * | |
4055 | * We need to validate the request and possibly allocate | |
4056 | * space if that might be an issue later. | |
4057 | * | |
4058 | * Currently we reject any reshape of a 'far' mode array, | |
4059 | * allow chunk size to change if new is generally acceptable, | |
4060 | * allow raid_disks to increase, and allow | |
4061 | * a switch between 'near' mode and 'offset' mode. | |
4062 | */ | |
4063 | struct r10conf *conf = mddev->private; | |
4064 | struct geom geo; | |
4065 | ||
4066 | if (conf->geo.far_copies != 1 && !conf->geo.far_offset) | |
4067 | return -EINVAL; | |
4068 | ||
4069 | if (setup_geo(&geo, mddev, geo_start) != conf->copies) | |
4070 | /* mustn't change number of copies */ | |
4071 | return -EINVAL; | |
4072 | if (geo.far_copies > 1 && !geo.far_offset) | |
4073 | /* Cannot switch to 'far' mode */ | |
4074 | return -EINVAL; | |
4075 | ||
4076 | if (mddev->array_sectors & geo.chunk_mask) | |
4077 | /* not factor of array size */ | |
4078 | return -EINVAL; | |
4079 | ||
3ea7daa5 N |
4080 | if (!enough(conf, -1)) |
4081 | return -EINVAL; | |
4082 | ||
4083 | kfree(conf->mirrors_new); | |
4084 | conf->mirrors_new = NULL; | |
4085 | if (mddev->delta_disks > 0) { | |
4086 | /* allocate new 'mirrors' list */ | |
4087 | conf->mirrors_new = kzalloc( | |
dc280d98 | 4088 | sizeof(struct raid10_info) |
3ea7daa5 N |
4089 | *(mddev->raid_disks + |
4090 | mddev->delta_disks), | |
4091 | GFP_KERNEL); | |
4092 | if (!conf->mirrors_new) | |
4093 | return -ENOMEM; | |
4094 | } | |
4095 | return 0; | |
4096 | } | |
4097 | ||
4098 | /* | |
4099 | * Need to check if array has failed when deciding whether to: | |
4100 | * - start an array | |
4101 | * - remove non-faulty devices | |
4102 | * - add a spare | |
4103 | * - allow a reshape | |
4104 | * This determination is simple when no reshape is happening. | |
4105 | * However if there is a reshape, we need to carefully check | |
4106 | * both the before and after sections. | |
4107 | * This is because some failed devices may only affect one | |
4108 | * of the two sections, and some non-in_sync devices may | |
4109 | * be insync in the section most affected by failed devices. | |
4110 | */ | |
4111 | static int calc_degraded(struct r10conf *conf) | |
4112 | { | |
4113 | int degraded, degraded2; | |
4114 | int i; | |
4115 | ||
4116 | rcu_read_lock(); | |
4117 | degraded = 0; | |
4118 | /* 'prev' section first */ | |
4119 | for (i = 0; i < conf->prev.raid_disks; i++) { | |
4120 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4121 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4122 | degraded++; | |
4123 | else if (!test_bit(In_sync, &rdev->flags)) | |
4124 | /* When we can reduce the number of devices in | |
4125 | * an array, this might not contribute to | |
4126 | * 'degraded'. It does now. | |
4127 | */ | |
4128 | degraded++; | |
4129 | } | |
4130 | rcu_read_unlock(); | |
4131 | if (conf->geo.raid_disks == conf->prev.raid_disks) | |
4132 | return degraded; | |
4133 | rcu_read_lock(); | |
4134 | degraded2 = 0; | |
4135 | for (i = 0; i < conf->geo.raid_disks; i++) { | |
4136 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); | |
4137 | if (!rdev || test_bit(Faulty, &rdev->flags)) | |
4138 | degraded2++; | |
4139 | else if (!test_bit(In_sync, &rdev->flags)) { | |
4140 | /* If reshape is increasing the number of devices, | |
4141 | * this section has already been recovered, so | |
4142 | * it doesn't contribute to degraded. | |
4143 | * else it does. | |
4144 | */ | |
4145 | if (conf->geo.raid_disks <= conf->prev.raid_disks) | |
4146 | degraded2++; | |
4147 | } | |
4148 | } | |
4149 | rcu_read_unlock(); | |
4150 | if (degraded2 > degraded) | |
4151 | return degraded2; | |
4152 | return degraded; | |
4153 | } | |
4154 | ||
4155 | static int raid10_start_reshape(struct mddev *mddev) | |
4156 | { | |
4157 | /* A 'reshape' has been requested. This commits | |
4158 | * the various 'new' fields and sets MD_RECOVER_RESHAPE | |
4159 | * This also checks if there are enough spares and adds them | |
4160 | * to the array. | |
4161 | * We currently require enough spares to make the final | |
4162 | * array non-degraded. We also require that the difference | |
4163 | * between old and new data_offset - on each device - is | |
4164 | * enough that we never risk over-writing. | |
4165 | */ | |
4166 | ||
4167 | unsigned long before_length, after_length; | |
4168 | sector_t min_offset_diff = 0; | |
4169 | int first = 1; | |
4170 | struct geom new; | |
4171 | struct r10conf *conf = mddev->private; | |
4172 | struct md_rdev *rdev; | |
4173 | int spares = 0; | |
bb63a701 | 4174 | int ret; |
3ea7daa5 N |
4175 | |
4176 | if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) | |
4177 | return -EBUSY; | |
4178 | ||
4179 | if (setup_geo(&new, mddev, geo_start) != conf->copies) | |
4180 | return -EINVAL; | |
4181 | ||
4182 | before_length = ((1 << conf->prev.chunk_shift) * | |
4183 | conf->prev.far_copies); | |
4184 | after_length = ((1 << conf->geo.chunk_shift) * | |
4185 | conf->geo.far_copies); | |
4186 | ||
4187 | rdev_for_each(rdev, mddev) { | |
4188 | if (!test_bit(In_sync, &rdev->flags) | |
4189 | && !test_bit(Faulty, &rdev->flags)) | |
4190 | spares++; | |
4191 | if (rdev->raid_disk >= 0) { | |
4192 | long long diff = (rdev->new_data_offset | |
4193 | - rdev->data_offset); | |
4194 | if (!mddev->reshape_backwards) | |
4195 | diff = -diff; | |
4196 | if (diff < 0) | |
4197 | diff = 0; | |
4198 | if (first || diff < min_offset_diff) | |
4199 | min_offset_diff = diff; | |
4200 | } | |
4201 | } | |
4202 | ||
4203 | if (max(before_length, after_length) > min_offset_diff) | |
4204 | return -EINVAL; | |
4205 | ||
4206 | if (spares < mddev->delta_disks) | |
4207 | return -EINVAL; | |
4208 | ||
4209 | conf->offset_diff = min_offset_diff; | |
4210 | spin_lock_irq(&conf->device_lock); | |
4211 | if (conf->mirrors_new) { | |
4212 | memcpy(conf->mirrors_new, conf->mirrors, | |
dc280d98 | 4213 | sizeof(struct raid10_info)*conf->prev.raid_disks); |
3ea7daa5 | 4214 | smp_mb(); |
c4796e21 | 4215 | kfree(conf->mirrors_old); |
3ea7daa5 N |
4216 | conf->mirrors_old = conf->mirrors; |
4217 | conf->mirrors = conf->mirrors_new; | |
4218 | conf->mirrors_new = NULL; | |
4219 | } | |
4220 | setup_geo(&conf->geo, mddev, geo_start); | |
4221 | smp_mb(); | |
4222 | if (mddev->reshape_backwards) { | |
4223 | sector_t size = raid10_size(mddev, 0, 0); | |
4224 | if (size < mddev->array_sectors) { | |
4225 | spin_unlock_irq(&conf->device_lock); | |
08464e09 N |
4226 | pr_warn("md/raid10:%s: array size must be reduce before number of disks\n", |
4227 | mdname(mddev)); | |
3ea7daa5 N |
4228 | return -EINVAL; |
4229 | } | |
4230 | mddev->resync_max_sectors = size; | |
4231 | conf->reshape_progress = size; | |
4232 | } else | |
4233 | conf->reshape_progress = 0; | |
299b0685 | 4234 | conf->reshape_safe = conf->reshape_progress; |
3ea7daa5 N |
4235 | spin_unlock_irq(&conf->device_lock); |
4236 | ||
bb63a701 N |
4237 | if (mddev->delta_disks && mddev->bitmap) { |
4238 | ret = bitmap_resize(mddev->bitmap, | |
4239 | raid10_size(mddev, 0, | |
4240 | conf->geo.raid_disks), | |
4241 | 0, 0); | |
4242 | if (ret) | |
4243 | goto abort; | |
4244 | } | |
3ea7daa5 N |
4245 | if (mddev->delta_disks > 0) { |
4246 | rdev_for_each(rdev, mddev) | |
4247 | if (rdev->raid_disk < 0 && | |
4248 | !test_bit(Faulty, &rdev->flags)) { | |
4249 | if (raid10_add_disk(mddev, rdev) == 0) { | |
4250 | if (rdev->raid_disk >= | |
4251 | conf->prev.raid_disks) | |
4252 | set_bit(In_sync, &rdev->flags); | |
4253 | else | |
4254 | rdev->recovery_offset = 0; | |
4255 | ||
4256 | if (sysfs_link_rdev(mddev, rdev)) | |
4257 | /* Failure here is OK */; | |
4258 | } | |
4259 | } else if (rdev->raid_disk >= conf->prev.raid_disks | |
4260 | && !test_bit(Faulty, &rdev->flags)) { | |
4261 | /* This is a spare that was manually added */ | |
4262 | set_bit(In_sync, &rdev->flags); | |
4263 | } | |
4264 | } | |
4265 | /* When a reshape changes the number of devices, | |
4266 | * ->degraded is measured against the larger of the | |
4267 | * pre and post numbers. | |
4268 | */ | |
4269 | spin_lock_irq(&conf->device_lock); | |
4270 | mddev->degraded = calc_degraded(conf); | |
4271 | spin_unlock_irq(&conf->device_lock); | |
4272 | mddev->raid_disks = conf->geo.raid_disks; | |
4273 | mddev->reshape_position = conf->reshape_progress; | |
2953079c | 4274 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 N |
4275 | |
4276 | clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); | |
4277 | clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); | |
ea358cd0 | 4278 | clear_bit(MD_RECOVERY_DONE, &mddev->recovery); |
3ea7daa5 N |
4279 | set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); |
4280 | set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); | |
4281 | ||
4282 | mddev->sync_thread = md_register_thread(md_do_sync, mddev, | |
4283 | "reshape"); | |
4284 | if (!mddev->sync_thread) { | |
bb63a701 N |
4285 | ret = -EAGAIN; |
4286 | goto abort; | |
3ea7daa5 N |
4287 | } |
4288 | conf->reshape_checkpoint = jiffies; | |
4289 | md_wakeup_thread(mddev->sync_thread); | |
4290 | md_new_event(mddev); | |
4291 | return 0; | |
bb63a701 N |
4292 | |
4293 | abort: | |
4294 | mddev->recovery = 0; | |
4295 | spin_lock_irq(&conf->device_lock); | |
4296 | conf->geo = conf->prev; | |
4297 | mddev->raid_disks = conf->geo.raid_disks; | |
4298 | rdev_for_each(rdev, mddev) | |
4299 | rdev->new_data_offset = rdev->data_offset; | |
4300 | smp_wmb(); | |
4301 | conf->reshape_progress = MaxSector; | |
299b0685 | 4302 | conf->reshape_safe = MaxSector; |
bb63a701 N |
4303 | mddev->reshape_position = MaxSector; |
4304 | spin_unlock_irq(&conf->device_lock); | |
4305 | return ret; | |
3ea7daa5 N |
4306 | } |
4307 | ||
4308 | /* Calculate the last device-address that could contain | |
4309 | * any block from the chunk that includes the array-address 's' | |
4310 | * and report the next address. | |
4311 | * i.e. the address returned will be chunk-aligned and after | |
4312 | * any data that is in the chunk containing 's'. | |
4313 | */ | |
4314 | static sector_t last_dev_address(sector_t s, struct geom *geo) | |
4315 | { | |
4316 | s = (s | geo->chunk_mask) + 1; | |
4317 | s >>= geo->chunk_shift; | |
4318 | s *= geo->near_copies; | |
4319 | s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); | |
4320 | s *= geo->far_copies; | |
4321 | s <<= geo->chunk_shift; | |
4322 | return s; | |
4323 | } | |
4324 | ||
4325 | /* Calculate the first device-address that could contain | |
4326 | * any block from the chunk that includes the array-address 's'. | |
4327 | * This too will be the start of a chunk | |
4328 | */ | |
4329 | static sector_t first_dev_address(sector_t s, struct geom *geo) | |
4330 | { | |
4331 | s >>= geo->chunk_shift; | |
4332 | s *= geo->near_copies; | |
4333 | sector_div(s, geo->raid_disks); | |
4334 | s *= geo->far_copies; | |
4335 | s <<= geo->chunk_shift; | |
4336 | return s; | |
4337 | } | |
4338 | ||
4339 | static sector_t reshape_request(struct mddev *mddev, sector_t sector_nr, | |
4340 | int *skipped) | |
4341 | { | |
4342 | /* We simply copy at most one chunk (smallest of old and new) | |
4343 | * at a time, possibly less if that exceeds RESYNC_PAGES, | |
4344 | * or we hit a bad block or something. | |
4345 | * This might mean we pause for normal IO in the middle of | |
02ec5026 | 4346 | * a chunk, but that is not a problem as mddev->reshape_position |
3ea7daa5 N |
4347 | * can record any location. |
4348 | * | |
4349 | * If we will want to write to a location that isn't | |
4350 | * yet recorded as 'safe' (i.e. in metadata on disk) then | |
4351 | * we need to flush all reshape requests and update the metadata. | |
4352 | * | |
4353 | * When reshaping forwards (e.g. to more devices), we interpret | |
4354 | * 'safe' as the earliest block which might not have been copied | |
4355 | * down yet. We divide this by previous stripe size and multiply | |
4356 | * by previous stripe length to get lowest device offset that we | |
4357 | * cannot write to yet. | |
4358 | * We interpret 'sector_nr' as an address that we want to write to. | |
4359 | * From this we use last_device_address() to find where we might | |
4360 | * write to, and first_device_address on the 'safe' position. | |
4361 | * If this 'next' write position is after the 'safe' position, | |
4362 | * we must update the metadata to increase the 'safe' position. | |
4363 | * | |
4364 | * When reshaping backwards, we round in the opposite direction | |
4365 | * and perform the reverse test: next write position must not be | |
4366 | * less than current safe position. | |
4367 | * | |
4368 | * In all this the minimum difference in data offsets | |
4369 | * (conf->offset_diff - always positive) allows a bit of slack, | |
02ec5026 | 4370 | * so next can be after 'safe', but not by more than offset_diff |
3ea7daa5 N |
4371 | * |
4372 | * We need to prepare all the bios here before we start any IO | |
4373 | * to ensure the size we choose is acceptable to all devices. | |
4374 | * The means one for each copy for write-out and an extra one for | |
4375 | * read-in. | |
4376 | * We store the read-in bio in ->master_bio and the others in | |
4377 | * ->devs[x].bio and ->devs[x].repl_bio. | |
4378 | */ | |
4379 | struct r10conf *conf = mddev->private; | |
4380 | struct r10bio *r10_bio; | |
4381 | sector_t next, safe, last; | |
4382 | int max_sectors; | |
4383 | int nr_sectors; | |
4384 | int s; | |
4385 | struct md_rdev *rdev; | |
4386 | int need_flush = 0; | |
4387 | struct bio *blist; | |
4388 | struct bio *bio, *read_bio; | |
4389 | int sectors_done = 0; | |
4390 | ||
4391 | if (sector_nr == 0) { | |
4392 | /* If restarting in the middle, skip the initial sectors */ | |
4393 | if (mddev->reshape_backwards && | |
4394 | conf->reshape_progress < raid10_size(mddev, 0, 0)) { | |
4395 | sector_nr = (raid10_size(mddev, 0, 0) | |
4396 | - conf->reshape_progress); | |
4397 | } else if (!mddev->reshape_backwards && | |
4398 | conf->reshape_progress > 0) | |
4399 | sector_nr = conf->reshape_progress; | |
4400 | if (sector_nr) { | |
4401 | mddev->curr_resync_completed = sector_nr; | |
4402 | sysfs_notify(&mddev->kobj, NULL, "sync_completed"); | |
4403 | *skipped = 1; | |
4404 | return sector_nr; | |
4405 | } | |
4406 | } | |
4407 | ||
4408 | /* We don't use sector_nr to track where we are up to | |
4409 | * as that doesn't work well for ->reshape_backwards. | |
4410 | * So just use ->reshape_progress. | |
4411 | */ | |
4412 | if (mddev->reshape_backwards) { | |
4413 | /* 'next' is the earliest device address that we might | |
4414 | * write to for this chunk in the new layout | |
4415 | */ | |
4416 | next = first_dev_address(conf->reshape_progress - 1, | |
4417 | &conf->geo); | |
4418 | ||
4419 | /* 'safe' is the last device address that we might read from | |
4420 | * in the old layout after a restart | |
4421 | */ | |
4422 | safe = last_dev_address(conf->reshape_safe - 1, | |
4423 | &conf->prev); | |
4424 | ||
4425 | if (next + conf->offset_diff < safe) | |
4426 | need_flush = 1; | |
4427 | ||
4428 | last = conf->reshape_progress - 1; | |
4429 | sector_nr = last & ~(sector_t)(conf->geo.chunk_mask | |
4430 | & conf->prev.chunk_mask); | |
4431 | if (sector_nr + RESYNC_BLOCK_SIZE/512 < last) | |
4432 | sector_nr = last + 1 - RESYNC_BLOCK_SIZE/512; | |
4433 | } else { | |
4434 | /* 'next' is after the last device address that we | |
4435 | * might write to for this chunk in the new layout | |
4436 | */ | |
4437 | next = last_dev_address(conf->reshape_progress, &conf->geo); | |
4438 | ||
4439 | /* 'safe' is the earliest device address that we might | |
4440 | * read from in the old layout after a restart | |
4441 | */ | |
4442 | safe = first_dev_address(conf->reshape_safe, &conf->prev); | |
4443 | ||
4444 | /* Need to update metadata if 'next' might be beyond 'safe' | |
4445 | * as that would possibly corrupt data | |
4446 | */ | |
4447 | if (next > safe + conf->offset_diff) | |
4448 | need_flush = 1; | |
4449 | ||
4450 | sector_nr = conf->reshape_progress; | |
4451 | last = sector_nr | (conf->geo.chunk_mask | |
4452 | & conf->prev.chunk_mask); | |
4453 | ||
4454 | if (sector_nr + RESYNC_BLOCK_SIZE/512 <= last) | |
4455 | last = sector_nr + RESYNC_BLOCK_SIZE/512 - 1; | |
4456 | } | |
4457 | ||
4458 | if (need_flush || | |
4459 | time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { | |
4460 | /* Need to update reshape_position in metadata */ | |
4461 | wait_barrier(conf); | |
4462 | mddev->reshape_position = conf->reshape_progress; | |
4463 | if (mddev->reshape_backwards) | |
4464 | mddev->curr_resync_completed = raid10_size(mddev, 0, 0) | |
4465 | - conf->reshape_progress; | |
4466 | else | |
4467 | mddev->curr_resync_completed = conf->reshape_progress; | |
4468 | conf->reshape_checkpoint = jiffies; | |
2953079c | 4469 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
3ea7daa5 | 4470 | md_wakeup_thread(mddev->thread); |
2953079c | 4471 | wait_event(mddev->sb_wait, mddev->sb_flags == 0 || |
c91abf5a N |
4472 | test_bit(MD_RECOVERY_INTR, &mddev->recovery)); |
4473 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { | |
4474 | allow_barrier(conf); | |
4475 | return sectors_done; | |
4476 | } | |
3ea7daa5 N |
4477 | conf->reshape_safe = mddev->reshape_position; |
4478 | allow_barrier(conf); | |
4479 | } | |
4480 | ||
4481 | read_more: | |
4482 | /* Now schedule reads for blocks from sector_nr to last */ | |
4483 | r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); | |
cb8b12b5 | 4484 | r10_bio->state = 0; |
3ea7daa5 N |
4485 | raise_barrier(conf, sectors_done != 0); |
4486 | atomic_set(&r10_bio->remaining, 0); | |
4487 | r10_bio->mddev = mddev; | |
4488 | r10_bio->sector = sector_nr; | |
4489 | set_bit(R10BIO_IsReshape, &r10_bio->state); | |
4490 | r10_bio->sectors = last - sector_nr + 1; | |
4491 | rdev = read_balance(conf, r10_bio, &max_sectors); | |
4492 | BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); | |
4493 | ||
4494 | if (!rdev) { | |
4495 | /* Cannot read from here, so need to record bad blocks | |
4496 | * on all the target devices. | |
4497 | */ | |
4498 | // FIXME | |
e337aead | 4499 | mempool_free(r10_bio, conf->r10buf_pool); |
3ea7daa5 N |
4500 | set_bit(MD_RECOVERY_INTR, &mddev->recovery); |
4501 | return sectors_done; | |
4502 | } | |
4503 | ||
4504 | read_bio = bio_alloc_mddev(GFP_KERNEL, RESYNC_PAGES, mddev); | |
4505 | ||
4506 | read_bio->bi_bdev = rdev->bdev; | |
4f024f37 | 4507 | read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr |
3ea7daa5 N |
4508 | + rdev->data_offset); |
4509 | read_bio->bi_private = r10_bio; | |
4510 | read_bio->bi_end_io = end_sync_read; | |
796a5cf0 | 4511 | bio_set_op_attrs(read_bio, REQ_OP_READ, 0); |
ce0b0a46 | 4512 | read_bio->bi_flags &= (~0UL << BIO_RESET_BITS); |
4246a0b6 | 4513 | read_bio->bi_error = 0; |
3ea7daa5 | 4514 | read_bio->bi_vcnt = 0; |
4f024f37 | 4515 | read_bio->bi_iter.bi_size = 0; |
3ea7daa5 N |
4516 | r10_bio->master_bio = read_bio; |
4517 | r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; | |
4518 | ||
4519 | /* Now find the locations in the new layout */ | |
4520 | __raid10_find_phys(&conf->geo, r10_bio); | |
4521 | ||
4522 | blist = read_bio; | |
4523 | read_bio->bi_next = NULL; | |
4524 | ||
d094d686 | 4525 | rcu_read_lock(); |
3ea7daa5 N |
4526 | for (s = 0; s < conf->copies*2; s++) { |
4527 | struct bio *b; | |
4528 | int d = r10_bio->devs[s/2].devnum; | |
4529 | struct md_rdev *rdev2; | |
4530 | if (s&1) { | |
d094d686 | 4531 | rdev2 = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4532 | b = r10_bio->devs[s/2].repl_bio; |
4533 | } else { | |
d094d686 | 4534 | rdev2 = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4535 | b = r10_bio->devs[s/2].bio; |
4536 | } | |
4537 | if (!rdev2 || test_bit(Faulty, &rdev2->flags)) | |
4538 | continue; | |
8be185f2 KO |
4539 | |
4540 | bio_reset(b); | |
3ea7daa5 | 4541 | b->bi_bdev = rdev2->bdev; |
4f024f37 KO |
4542 | b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + |
4543 | rdev2->new_data_offset; | |
3ea7daa5 N |
4544 | b->bi_private = r10_bio; |
4545 | b->bi_end_io = end_reshape_write; | |
796a5cf0 | 4546 | bio_set_op_attrs(b, REQ_OP_WRITE, 0); |
3ea7daa5 | 4547 | b->bi_next = blist; |
3ea7daa5 N |
4548 | blist = b; |
4549 | } | |
4550 | ||
4551 | /* Now add as many pages as possible to all of these bios. */ | |
4552 | ||
4553 | nr_sectors = 0; | |
4554 | for (s = 0 ; s < max_sectors; s += PAGE_SIZE >> 9) { | |
4555 | struct page *page = r10_bio->devs[0].bio->bi_io_vec[s/(PAGE_SIZE>>9)].bv_page; | |
4556 | int len = (max_sectors - s) << 9; | |
4557 | if (len > PAGE_SIZE) | |
4558 | len = PAGE_SIZE; | |
4559 | for (bio = blist; bio ; bio = bio->bi_next) { | |
4560 | struct bio *bio2; | |
4561 | if (bio_add_page(bio, page, len, 0)) | |
4562 | continue; | |
4563 | ||
4564 | /* Didn't fit, must stop */ | |
4565 | for (bio2 = blist; | |
4566 | bio2 && bio2 != bio; | |
4567 | bio2 = bio2->bi_next) { | |
4568 | /* Remove last page from this bio */ | |
4569 | bio2->bi_vcnt--; | |
4f024f37 | 4570 | bio2->bi_iter.bi_size -= len; |
b7c44ed9 | 4571 | bio_clear_flag(bio2, BIO_SEG_VALID); |
3ea7daa5 N |
4572 | } |
4573 | goto bio_full; | |
4574 | } | |
4575 | sector_nr += len >> 9; | |
4576 | nr_sectors += len >> 9; | |
4577 | } | |
4578 | bio_full: | |
d094d686 | 4579 | rcu_read_unlock(); |
3ea7daa5 N |
4580 | r10_bio->sectors = nr_sectors; |
4581 | ||
4582 | /* Now submit the read */ | |
4583 | md_sync_acct(read_bio->bi_bdev, r10_bio->sectors); | |
4584 | atomic_inc(&r10_bio->remaining); | |
4585 | read_bio->bi_next = NULL; | |
4586 | generic_make_request(read_bio); | |
4587 | sector_nr += nr_sectors; | |
4588 | sectors_done += nr_sectors; | |
4589 | if (sector_nr <= last) | |
4590 | goto read_more; | |
4591 | ||
4592 | /* Now that we have done the whole section we can | |
4593 | * update reshape_progress | |
4594 | */ | |
4595 | if (mddev->reshape_backwards) | |
4596 | conf->reshape_progress -= sectors_done; | |
4597 | else | |
4598 | conf->reshape_progress += sectors_done; | |
4599 | ||
4600 | return sectors_done; | |
4601 | } | |
4602 | ||
4603 | static void end_reshape_request(struct r10bio *r10_bio); | |
4604 | static int handle_reshape_read_error(struct mddev *mddev, | |
4605 | struct r10bio *r10_bio); | |
4606 | static void reshape_request_write(struct mddev *mddev, struct r10bio *r10_bio) | |
4607 | { | |
4608 | /* Reshape read completed. Hopefully we have a block | |
4609 | * to write out. | |
4610 | * If we got a read error then we do sync 1-page reads from | |
4611 | * elsewhere until we find the data - or give up. | |
4612 | */ | |
4613 | struct r10conf *conf = mddev->private; | |
4614 | int s; | |
4615 | ||
4616 | if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) | |
4617 | if (handle_reshape_read_error(mddev, r10_bio) < 0) { | |
4618 | /* Reshape has been aborted */ | |
4619 | md_done_sync(mddev, r10_bio->sectors, 0); | |
4620 | return; | |
4621 | } | |
4622 | ||
4623 | /* We definitely have the data in the pages, schedule the | |
4624 | * writes. | |
4625 | */ | |
4626 | atomic_set(&r10_bio->remaining, 1); | |
4627 | for (s = 0; s < conf->copies*2; s++) { | |
4628 | struct bio *b; | |
4629 | int d = r10_bio->devs[s/2].devnum; | |
4630 | struct md_rdev *rdev; | |
d094d686 | 4631 | rcu_read_lock(); |
3ea7daa5 | 4632 | if (s&1) { |
d094d686 | 4633 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
3ea7daa5 N |
4634 | b = r10_bio->devs[s/2].repl_bio; |
4635 | } else { | |
d094d686 | 4636 | rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4637 | b = r10_bio->devs[s/2].bio; |
4638 | } | |
d094d686 N |
4639 | if (!rdev || test_bit(Faulty, &rdev->flags)) { |
4640 | rcu_read_unlock(); | |
3ea7daa5 | 4641 | continue; |
d094d686 | 4642 | } |
3ea7daa5 | 4643 | atomic_inc(&rdev->nr_pending); |
d094d686 | 4644 | rcu_read_unlock(); |
3ea7daa5 N |
4645 | md_sync_acct(b->bi_bdev, r10_bio->sectors); |
4646 | atomic_inc(&r10_bio->remaining); | |
4647 | b->bi_next = NULL; | |
4648 | generic_make_request(b); | |
4649 | } | |
4650 | end_reshape_request(r10_bio); | |
4651 | } | |
4652 | ||
4653 | static void end_reshape(struct r10conf *conf) | |
4654 | { | |
4655 | if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) | |
4656 | return; | |
4657 | ||
4658 | spin_lock_irq(&conf->device_lock); | |
4659 | conf->prev = conf->geo; | |
4660 | md_finish_reshape(conf->mddev); | |
4661 | smp_wmb(); | |
4662 | conf->reshape_progress = MaxSector; | |
299b0685 | 4663 | conf->reshape_safe = MaxSector; |
3ea7daa5 N |
4664 | spin_unlock_irq(&conf->device_lock); |
4665 | ||
4666 | /* read-ahead size must cover two whole stripes, which is | |
4667 | * 2 * (datadisks) * chunksize where 'n' is the number of raid devices | |
4668 | */ | |
4669 | if (conf->mddev->queue) { | |
4670 | int stripe = conf->geo.raid_disks * | |
4671 | ((conf->mddev->chunk_sectors << 9) / PAGE_SIZE); | |
4672 | stripe /= conf->geo.near_copies; | |
dc3b17cc JK |
4673 | if (conf->mddev->queue->backing_dev_info->ra_pages < 2 * stripe) |
4674 | conf->mddev->queue->backing_dev_info->ra_pages = 2 * stripe; | |
3ea7daa5 N |
4675 | } |
4676 | conf->fullsync = 0; | |
4677 | } | |
4678 | ||
3ea7daa5 N |
4679 | static int handle_reshape_read_error(struct mddev *mddev, |
4680 | struct r10bio *r10_bio) | |
4681 | { | |
4682 | /* Use sync reads to get the blocks from somewhere else */ | |
4683 | int sectors = r10_bio->sectors; | |
3ea7daa5 | 4684 | struct r10conf *conf = mddev->private; |
e0ee7785 N |
4685 | struct { |
4686 | struct r10bio r10_bio; | |
4687 | struct r10dev devs[conf->copies]; | |
4688 | } on_stack; | |
4689 | struct r10bio *r10b = &on_stack.r10_bio; | |
3ea7daa5 N |
4690 | int slot = 0; |
4691 | int idx = 0; | |
4692 | struct bio_vec *bvec = r10_bio->master_bio->bi_io_vec; | |
4693 | ||
e0ee7785 N |
4694 | r10b->sector = r10_bio->sector; |
4695 | __raid10_find_phys(&conf->prev, r10b); | |
3ea7daa5 N |
4696 | |
4697 | while (sectors) { | |
4698 | int s = sectors; | |
4699 | int success = 0; | |
4700 | int first_slot = slot; | |
4701 | ||
4702 | if (s > (PAGE_SIZE >> 9)) | |
4703 | s = PAGE_SIZE >> 9; | |
4704 | ||
d094d686 | 4705 | rcu_read_lock(); |
3ea7daa5 | 4706 | while (!success) { |
e0ee7785 | 4707 | int d = r10b->devs[slot].devnum; |
d094d686 | 4708 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
3ea7daa5 N |
4709 | sector_t addr; |
4710 | if (rdev == NULL || | |
4711 | test_bit(Faulty, &rdev->flags) || | |
4712 | !test_bit(In_sync, &rdev->flags)) | |
4713 | goto failed; | |
4714 | ||
e0ee7785 | 4715 | addr = r10b->devs[slot].addr + idx * PAGE_SIZE; |
d094d686 N |
4716 | atomic_inc(&rdev->nr_pending); |
4717 | rcu_read_unlock(); | |
3ea7daa5 N |
4718 | success = sync_page_io(rdev, |
4719 | addr, | |
4720 | s << 9, | |
4721 | bvec[idx].bv_page, | |
796a5cf0 | 4722 | REQ_OP_READ, 0, false); |
d094d686 N |
4723 | rdev_dec_pending(rdev, mddev); |
4724 | rcu_read_lock(); | |
3ea7daa5 N |
4725 | if (success) |
4726 | break; | |
4727 | failed: | |
4728 | slot++; | |
4729 | if (slot >= conf->copies) | |
4730 | slot = 0; | |
4731 | if (slot == first_slot) | |
4732 | break; | |
4733 | } | |
d094d686 | 4734 | rcu_read_unlock(); |
3ea7daa5 N |
4735 | if (!success) { |
4736 | /* couldn't read this block, must give up */ | |
4737 | set_bit(MD_RECOVERY_INTR, | |
4738 | &mddev->recovery); | |
4739 | return -EIO; | |
4740 | } | |
4741 | sectors -= s; | |
4742 | idx++; | |
4743 | } | |
4744 | return 0; | |
4745 | } | |
4746 | ||
4246a0b6 | 4747 | static void end_reshape_write(struct bio *bio) |
3ea7daa5 | 4748 | { |
3ea7daa5 N |
4749 | struct r10bio *r10_bio = bio->bi_private; |
4750 | struct mddev *mddev = r10_bio->mddev; | |
4751 | struct r10conf *conf = mddev->private; | |
4752 | int d; | |
4753 | int slot; | |
4754 | int repl; | |
4755 | struct md_rdev *rdev = NULL; | |
4756 | ||
4757 | d = find_bio_disk(conf, r10_bio, bio, &slot, &repl); | |
4758 | if (repl) | |
4759 | rdev = conf->mirrors[d].replacement; | |
4760 | if (!rdev) { | |
4761 | smp_mb(); | |
4762 | rdev = conf->mirrors[d].rdev; | |
4763 | } | |
4764 | ||
4246a0b6 | 4765 | if (bio->bi_error) { |
3ea7daa5 N |
4766 | /* FIXME should record badblock */ |
4767 | md_error(mddev, rdev); | |
4768 | } | |
4769 | ||
4770 | rdev_dec_pending(rdev, mddev); | |
4771 | end_reshape_request(r10_bio); | |
4772 | } | |
4773 | ||
4774 | static void end_reshape_request(struct r10bio *r10_bio) | |
4775 | { | |
4776 | if (!atomic_dec_and_test(&r10_bio->remaining)) | |
4777 | return; | |
4778 | md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); | |
4779 | bio_put(r10_bio->master_bio); | |
4780 | put_buf(r10_bio); | |
4781 | } | |
4782 | ||
4783 | static void raid10_finish_reshape(struct mddev *mddev) | |
4784 | { | |
4785 | struct r10conf *conf = mddev->private; | |
4786 | ||
4787 | if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) | |
4788 | return; | |
4789 | ||
4790 | if (mddev->delta_disks > 0) { | |
4791 | sector_t size = raid10_size(mddev, 0, 0); | |
4792 | md_set_array_sectors(mddev, size); | |
4793 | if (mddev->recovery_cp > mddev->resync_max_sectors) { | |
4794 | mddev->recovery_cp = mddev->resync_max_sectors; | |
4795 | set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); | |
4796 | } | |
4797 | mddev->resync_max_sectors = size; | |
859644f0 HM |
4798 | if (mddev->queue) { |
4799 | set_capacity(mddev->gendisk, mddev->array_sectors); | |
4800 | revalidate_disk(mddev->gendisk); | |
4801 | } | |
63aced61 N |
4802 | } else { |
4803 | int d; | |
d094d686 | 4804 | rcu_read_lock(); |
63aced61 N |
4805 | for (d = conf->geo.raid_disks ; |
4806 | d < conf->geo.raid_disks - mddev->delta_disks; | |
4807 | d++) { | |
d094d686 | 4808 | struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); |
63aced61 N |
4809 | if (rdev) |
4810 | clear_bit(In_sync, &rdev->flags); | |
d094d686 | 4811 | rdev = rcu_dereference(conf->mirrors[d].replacement); |
63aced61 N |
4812 | if (rdev) |
4813 | clear_bit(In_sync, &rdev->flags); | |
4814 | } | |
d094d686 | 4815 | rcu_read_unlock(); |
3ea7daa5 N |
4816 | } |
4817 | mddev->layout = mddev->new_layout; | |
4818 | mddev->chunk_sectors = 1 << conf->geo.chunk_shift; | |
4819 | mddev->reshape_position = MaxSector; | |
4820 | mddev->delta_disks = 0; | |
4821 | mddev->reshape_backwards = 0; | |
4822 | } | |
4823 | ||
84fc4b56 | 4824 | static struct md_personality raid10_personality = |
1da177e4 LT |
4825 | { |
4826 | .name = "raid10", | |
2604b703 | 4827 | .level = 10, |
1da177e4 | 4828 | .owner = THIS_MODULE, |
849674e4 SL |
4829 | .make_request = raid10_make_request, |
4830 | .run = raid10_run, | |
afa0f557 | 4831 | .free = raid10_free, |
849674e4 SL |
4832 | .status = raid10_status, |
4833 | .error_handler = raid10_error, | |
1da177e4 LT |
4834 | .hot_add_disk = raid10_add_disk, |
4835 | .hot_remove_disk= raid10_remove_disk, | |
4836 | .spare_active = raid10_spare_active, | |
849674e4 | 4837 | .sync_request = raid10_sync_request, |
6cce3b23 | 4838 | .quiesce = raid10_quiesce, |
80c3a6ce | 4839 | .size = raid10_size, |
006a09a0 | 4840 | .resize = raid10_resize, |
dab8b292 | 4841 | .takeover = raid10_takeover, |
3ea7daa5 N |
4842 | .check_reshape = raid10_check_reshape, |
4843 | .start_reshape = raid10_start_reshape, | |
4844 | .finish_reshape = raid10_finish_reshape, | |
5c675f83 | 4845 | .congested = raid10_congested, |
1da177e4 LT |
4846 | }; |
4847 | ||
4848 | static int __init raid_init(void) | |
4849 | { | |
2604b703 | 4850 | return register_md_personality(&raid10_personality); |
1da177e4 LT |
4851 | } |
4852 | ||
4853 | static void raid_exit(void) | |
4854 | { | |
2604b703 | 4855 | unregister_md_personality(&raid10_personality); |
1da177e4 LT |
4856 | } |
4857 | ||
4858 | module_init(raid_init); | |
4859 | module_exit(raid_exit); | |
4860 | MODULE_LICENSE("GPL"); | |
0efb9e61 | 4861 | MODULE_DESCRIPTION("RAID10 (striped mirror) personality for MD"); |
1da177e4 | 4862 | MODULE_ALIAS("md-personality-9"); /* RAID10 */ |
d9d166c2 | 4863 | MODULE_ALIAS("md-raid10"); |
2604b703 | 4864 | MODULE_ALIAS("md-level-10"); |
34db0cd6 N |
4865 | |
4866 | module_param(max_queued_requests, int, S_IRUGO|S_IWUSR); |